Back to home page

OSCL-LXR

 
 

    


0001 /*
0002  * This file is part of the Chelsio T4 Ethernet driver for Linux.
0003  *
0004  * Copyright (c) 2003-2016 Chelsio Communications, Inc. All rights reserved.
0005  *
0006  * This software is available to you under a choice of one of two
0007  * licenses.  You may choose to be licensed under the terms of the GNU
0008  * General Public License (GPL) Version 2, available from the file
0009  * COPYING in the main directory of this source tree, or the
0010  * OpenIB.org BSD license below:
0011  *
0012  *     Redistribution and use in source and binary forms, with or
0013  *     without modification, are permitted provided that the following
0014  *     conditions are met:
0015  *
0016  *      - Redistributions of source code must retain the above
0017  *        copyright notice, this list of conditions and the following
0018  *        disclaimer.
0019  *
0020  *      - Redistributions in binary form must reproduce the above
0021  *        copyright notice, this list of conditions and the following
0022  *        disclaimer in the documentation and/or other materials
0023  *        provided with the distribution.
0024  *
0025  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
0026  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
0027  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
0028  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
0029  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
0030  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
0031  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
0032  * SOFTWARE.
0033  */
0034 
0035 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
0036 
0037 #include <linux/bitmap.h>
0038 #include <linux/crc32.h>
0039 #include <linux/ctype.h>
0040 #include <linux/debugfs.h>
0041 #include <linux/err.h>
0042 #include <linux/etherdevice.h>
0043 #include <linux/firmware.h>
0044 #include <linux/if.h>
0045 #include <linux/if_vlan.h>
0046 #include <linux/init.h>
0047 #include <linux/log2.h>
0048 #include <linux/mdio.h>
0049 #include <linux/module.h>
0050 #include <linux/moduleparam.h>
0051 #include <linux/mutex.h>
0052 #include <linux/netdevice.h>
0053 #include <linux/pci.h>
0054 #include <linux/aer.h>
0055 #include <linux/rtnetlink.h>
0056 #include <linux/sched.h>
0057 #include <linux/seq_file.h>
0058 #include <linux/sockios.h>
0059 #include <linux/vmalloc.h>
0060 #include <linux/workqueue.h>
0061 #include <net/neighbour.h>
0062 #include <net/netevent.h>
0063 #include <net/addrconf.h>
0064 #include <net/bonding.h>
0065 #include <linux/uaccess.h>
0066 #include <linux/crash_dump.h>
0067 #include <net/udp_tunnel.h>
0068 #include <net/xfrm.h>
0069 #if IS_ENABLED(CONFIG_CHELSIO_TLS_DEVICE)
0070 #include <net/tls.h>
0071 #endif
0072 
0073 #include "cxgb4.h"
0074 #include "cxgb4_filter.h"
0075 #include "t4_regs.h"
0076 #include "t4_values.h"
0077 #include "t4_msg.h"
0078 #include "t4fw_api.h"
0079 #include "t4fw_version.h"
0080 #include "cxgb4_dcb.h"
0081 #include "srq.h"
0082 #include "cxgb4_debugfs.h"
0083 #include "clip_tbl.h"
0084 #include "l2t.h"
0085 #include "smt.h"
0086 #include "sched.h"
0087 #include "cxgb4_tc_u32.h"
0088 #include "cxgb4_tc_flower.h"
0089 #include "cxgb4_tc_mqprio.h"
0090 #include "cxgb4_tc_matchall.h"
0091 #include "cxgb4_ptp.h"
0092 #include "cxgb4_cudbg.h"
0093 
0094 char cxgb4_driver_name[] = KBUILD_MODNAME;
0095 
0096 #define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
0097 
0098 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
0099              NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
0100              NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
0101 
0102 /* Macros needed to support the PCI Device ID Table ...
0103  */
0104 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
0105     static const struct pci_device_id cxgb4_pci_tbl[] = {
0106 #define CXGB4_UNIFIED_PF 0x4
0107 
0108 #define CH_PCI_DEVICE_ID_FUNCTION CXGB4_UNIFIED_PF
0109 
0110 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
0111  * called for both.
0112  */
0113 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
0114 
0115 #define CH_PCI_ID_TABLE_ENTRY(devid) \
0116         {PCI_VDEVICE(CHELSIO, (devid)), CXGB4_UNIFIED_PF}
0117 
0118 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
0119         { 0, } \
0120     }
0121 
0122 #include "t4_pci_id_tbl.h"
0123 
0124 #define FW4_FNAME "cxgb4/t4fw.bin"
0125 #define FW5_FNAME "cxgb4/t5fw.bin"
0126 #define FW6_FNAME "cxgb4/t6fw.bin"
0127 #define FW4_CFNAME "cxgb4/t4-config.txt"
0128 #define FW5_CFNAME "cxgb4/t5-config.txt"
0129 #define FW6_CFNAME "cxgb4/t6-config.txt"
0130 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
0131 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
0132 #define PHY_AQ1202_DEVICEID 0x4409
0133 #define PHY_BCM84834_DEVICEID 0x4486
0134 
0135 MODULE_DESCRIPTION(DRV_DESC);
0136 MODULE_AUTHOR("Chelsio Communications");
0137 MODULE_LICENSE("Dual BSD/GPL");
0138 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
0139 MODULE_FIRMWARE(FW4_FNAME);
0140 MODULE_FIRMWARE(FW5_FNAME);
0141 MODULE_FIRMWARE(FW6_FNAME);
0142 
0143 /*
0144  * The driver uses the best interrupt scheme available on a platform in the
0145  * order MSI-X, MSI, legacy INTx interrupts.  This parameter determines which
0146  * of these schemes the driver may consider as follows:
0147  *
0148  * msi = 2: choose from among all three options
0149  * msi = 1: only consider MSI and INTx interrupts
0150  * msi = 0: force INTx interrupts
0151  */
0152 static int msi = 2;
0153 
0154 module_param(msi, int, 0644);
0155 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
0156 
0157 /*
0158  * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
0159  * offset by 2 bytes in order to have the IP headers line up on 4-byte
0160  * boundaries.  This is a requirement for many architectures which will throw
0161  * a machine check fault if an attempt is made to access one of the 4-byte IP
0162  * header fields on a non-4-byte boundary.  And it's a major performance issue
0163  * even on some architectures which allow it like some implementations of the
0164  * x86 ISA.  However, some architectures don't mind this and for some very
0165  * edge-case performance sensitive applications (like forwarding large volumes
0166  * of small packets), setting this DMA offset to 0 will decrease the number of
0167  * PCI-E Bus transfers enough to measurably affect performance.
0168  */
0169 static int rx_dma_offset = 2;
0170 
0171 /* TX Queue select used to determine what algorithm to use for selecting TX
0172  * queue. Select between the kernel provided function (select_queue=0) or user
0173  * cxgb_select_queue function (select_queue=1)
0174  *
0175  * Default: select_queue=0
0176  */
0177 static int select_queue;
0178 module_param(select_queue, int, 0644);
0179 MODULE_PARM_DESC(select_queue,
0180          "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
0181 
0182 static struct dentry *cxgb4_debugfs_root;
0183 
0184 LIST_HEAD(adapter_list);
0185 DEFINE_MUTEX(uld_mutex);
0186 LIST_HEAD(uld_list);
0187 
0188 static int cfg_queues(struct adapter *adap);
0189 
0190 static void link_report(struct net_device *dev)
0191 {
0192     if (!netif_carrier_ok(dev))
0193         netdev_info(dev, "link down\n");
0194     else {
0195         static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
0196 
0197         const char *s;
0198         const struct port_info *p = netdev_priv(dev);
0199 
0200         switch (p->link_cfg.speed) {
0201         case 100:
0202             s = "100Mbps";
0203             break;
0204         case 1000:
0205             s = "1Gbps";
0206             break;
0207         case 10000:
0208             s = "10Gbps";
0209             break;
0210         case 25000:
0211             s = "25Gbps";
0212             break;
0213         case 40000:
0214             s = "40Gbps";
0215             break;
0216         case 50000:
0217             s = "50Gbps";
0218             break;
0219         case 100000:
0220             s = "100Gbps";
0221             break;
0222         default:
0223             pr_info("%s: unsupported speed: %d\n",
0224                 dev->name, p->link_cfg.speed);
0225             return;
0226         }
0227 
0228         netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
0229                 fc[p->link_cfg.fc]);
0230     }
0231 }
0232 
0233 #ifdef CONFIG_CHELSIO_T4_DCB
0234 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
0235 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
0236 {
0237     struct port_info *pi = netdev_priv(dev);
0238     struct adapter *adap = pi->adapter;
0239     struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
0240     int i;
0241 
0242     /* We use a simple mapping of Port TX Queue Index to DCB
0243      * Priority when we're enabling DCB.
0244      */
0245     for (i = 0; i < pi->nqsets; i++, txq++) {
0246         u32 name, value;
0247         int err;
0248 
0249         name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
0250             FW_PARAMS_PARAM_X_V(
0251                 FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
0252             FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
0253         value = enable ? i : 0xffffffff;
0254 
0255         /* Since we can be called while atomic (from "interrupt
0256          * level") we need to issue the Set Parameters Commannd
0257          * without sleeping (timeout < 0).
0258          */
0259         err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
0260                         &name, &value,
0261                         -FW_CMD_MAX_TIMEOUT);
0262 
0263         if (err)
0264             dev_err(adap->pdev_dev,
0265                 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
0266                 enable ? "set" : "unset", pi->port_id, i, -err);
0267         else
0268             txq->dcb_prio = enable ? value : 0;
0269     }
0270 }
0271 
0272 int cxgb4_dcb_enabled(const struct net_device *dev)
0273 {
0274     struct port_info *pi = netdev_priv(dev);
0275 
0276     if (!pi->dcb.enabled)
0277         return 0;
0278 
0279     return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
0280         (pi->dcb.state == CXGB4_DCB_STATE_HOST));
0281 }
0282 #endif /* CONFIG_CHELSIO_T4_DCB */
0283 
0284 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
0285 {
0286     struct net_device *dev = adapter->port[port_id];
0287 
0288     /* Skip changes from disabled ports. */
0289     if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
0290         if (link_stat)
0291             netif_carrier_on(dev);
0292         else {
0293 #ifdef CONFIG_CHELSIO_T4_DCB
0294             if (cxgb4_dcb_enabled(dev)) {
0295                 cxgb4_dcb_reset(dev);
0296                 dcb_tx_queue_prio_enable(dev, false);
0297             }
0298 #endif /* CONFIG_CHELSIO_T4_DCB */
0299             netif_carrier_off(dev);
0300         }
0301 
0302         link_report(dev);
0303     }
0304 }
0305 
0306 void t4_os_portmod_changed(struct adapter *adap, int port_id)
0307 {
0308     static const char *mod_str[] = {
0309         NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
0310     };
0311 
0312     struct net_device *dev = adap->port[port_id];
0313     struct port_info *pi = netdev_priv(dev);
0314 
0315     if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
0316         netdev_info(dev, "port module unplugged\n");
0317     else if (pi->mod_type < ARRAY_SIZE(mod_str))
0318         netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
0319     else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
0320         netdev_info(dev, "%s: unsupported port module inserted\n",
0321                 dev->name);
0322     else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
0323         netdev_info(dev, "%s: unknown port module inserted\n",
0324                 dev->name);
0325     else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
0326         netdev_info(dev, "%s: transceiver module error\n", dev->name);
0327     else
0328         netdev_info(dev, "%s: unknown module type %d inserted\n",
0329                 dev->name, pi->mod_type);
0330 
0331     /* If the interface is running, then we'll need any "sticky" Link
0332      * Parameters redone with a new Transceiver Module.
0333      */
0334     pi->link_cfg.redo_l1cfg = netif_running(dev);
0335 }
0336 
0337 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
0338 module_param(dbfifo_int_thresh, int, 0644);
0339 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
0340 
0341 /*
0342  * usecs to sleep while draining the dbfifo
0343  */
0344 static int dbfifo_drain_delay = 1000;
0345 module_param(dbfifo_drain_delay, int, 0644);
0346 MODULE_PARM_DESC(dbfifo_drain_delay,
0347          "usecs to sleep while draining the dbfifo");
0348 
0349 static inline int cxgb4_set_addr_hash(struct port_info *pi)
0350 {
0351     struct adapter *adap = pi->adapter;
0352     u64 vec = 0;
0353     bool ucast = false;
0354     struct hash_mac_addr *entry;
0355 
0356     /* Calculate the hash vector for the updated list and program it */
0357     list_for_each_entry(entry, &adap->mac_hlist, list) {
0358         ucast |= is_unicast_ether_addr(entry->addr);
0359         vec |= (1ULL << hash_mac_addr(entry->addr));
0360     }
0361     return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
0362                 vec, false);
0363 }
0364 
0365 static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
0366 {
0367     struct port_info *pi = netdev_priv(netdev);
0368     struct adapter *adap = pi->adapter;
0369     int ret;
0370     u64 mhash = 0;
0371     u64 uhash = 0;
0372     /* idx stores the index of allocated filters,
0373      * its size should be modified based on the number of
0374      * MAC addresses that we allocate filters for
0375      */
0376 
0377     u16 idx[1] = {};
0378     bool free = false;
0379     bool ucast = is_unicast_ether_addr(mac_addr);
0380     const u8 *maclist[1] = {mac_addr};
0381     struct hash_mac_addr *new_entry;
0382 
0383     ret = cxgb4_alloc_mac_filt(adap, pi->viid, free, 1, maclist,
0384                    idx, ucast ? &uhash : &mhash, false);
0385     if (ret < 0)
0386         goto out;
0387     /* if hash != 0, then add the addr to hash addr list
0388      * so on the end we will calculate the hash for the
0389      * list and program it
0390      */
0391     if (uhash || mhash) {
0392         new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
0393         if (!new_entry)
0394             return -ENOMEM;
0395         ether_addr_copy(new_entry->addr, mac_addr);
0396         list_add_tail(&new_entry->list, &adap->mac_hlist);
0397         ret = cxgb4_set_addr_hash(pi);
0398     }
0399 out:
0400     return ret < 0 ? ret : 0;
0401 }
0402 
0403 static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
0404 {
0405     struct port_info *pi = netdev_priv(netdev);
0406     struct adapter *adap = pi->adapter;
0407     int ret;
0408     const u8 *maclist[1] = {mac_addr};
0409     struct hash_mac_addr *entry, *tmp;
0410 
0411     /* If the MAC address to be removed is in the hash addr
0412      * list, delete it from the list and update hash vector
0413      */
0414     list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
0415         if (ether_addr_equal(entry->addr, mac_addr)) {
0416             list_del(&entry->list);
0417             kfree(entry);
0418             return cxgb4_set_addr_hash(pi);
0419         }
0420     }
0421 
0422     ret = cxgb4_free_mac_filt(adap, pi->viid, 1, maclist, false);
0423     return ret < 0 ? -EINVAL : 0;
0424 }
0425 
0426 /*
0427  * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
0428  * If @mtu is -1 it is left unchanged.
0429  */
0430 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
0431 {
0432     struct port_info *pi = netdev_priv(dev);
0433     struct adapter *adapter = pi->adapter;
0434 
0435     __dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
0436     __dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
0437 
0438     return t4_set_rxmode(adapter, adapter->mbox, pi->viid, pi->viid_mirror,
0439                  mtu, (dev->flags & IFF_PROMISC) ? 1 : 0,
0440                  (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
0441                  sleep_ok);
0442 }
0443 
0444 /**
0445  *  cxgb4_change_mac - Update match filter for a MAC address.
0446  *  @pi: the port_info
0447  *  @viid: the VI id
0448  *  @tcam_idx: TCAM index of existing filter for old value of MAC address,
0449  *         or -1
0450  *  @addr: the new MAC address value
0451  *  @persist: whether a new MAC allocation should be persistent
0452  *  @smt_idx: the destination to store the new SMT index.
0453  *
0454  *  Modifies an MPS filter and sets it to the new MAC address if
0455  *  @tcam_idx >= 0, or adds the MAC address to a new filter if
0456  *  @tcam_idx < 0. In the latter case the address is added persistently
0457  *  if @persist is %true.
0458  *  Addresses are programmed to hash region, if tcam runs out of entries.
0459  *
0460  */
0461 int cxgb4_change_mac(struct port_info *pi, unsigned int viid,
0462              int *tcam_idx, const u8 *addr, bool persist,
0463              u8 *smt_idx)
0464 {
0465     struct adapter *adapter = pi->adapter;
0466     struct hash_mac_addr *entry, *new_entry;
0467     int ret;
0468 
0469     ret = t4_change_mac(adapter, adapter->mbox, viid,
0470                 *tcam_idx, addr, persist, smt_idx);
0471     /* We ran out of TCAM entries. try programming hash region. */
0472     if (ret == -ENOMEM) {
0473         /* If the MAC address to be updated is in the hash addr
0474          * list, update it from the list
0475          */
0476         list_for_each_entry(entry, &adapter->mac_hlist, list) {
0477             if (entry->iface_mac) {
0478                 ether_addr_copy(entry->addr, addr);
0479                 goto set_hash;
0480             }
0481         }
0482         new_entry = kzalloc(sizeof(*new_entry), GFP_KERNEL);
0483         if (!new_entry)
0484             return -ENOMEM;
0485         ether_addr_copy(new_entry->addr, addr);
0486         new_entry->iface_mac = true;
0487         list_add_tail(&new_entry->list, &adapter->mac_hlist);
0488 set_hash:
0489         ret = cxgb4_set_addr_hash(pi);
0490     } else if (ret >= 0) {
0491         *tcam_idx = ret;
0492         ret = 0;
0493     }
0494 
0495     return ret;
0496 }
0497 
0498 /*
0499  *  link_start - enable a port
0500  *  @dev: the port to enable
0501  *
0502  *  Performs the MAC and PHY actions needed to enable a port.
0503  */
0504 static int link_start(struct net_device *dev)
0505 {
0506     struct port_info *pi = netdev_priv(dev);
0507     unsigned int mb = pi->adapter->mbox;
0508     int ret;
0509 
0510     /*
0511      * We do not set address filters and promiscuity here, the stack does
0512      * that step explicitly.
0513      */
0514     ret = t4_set_rxmode(pi->adapter, mb, pi->viid, pi->viid_mirror,
0515                 dev->mtu, -1, -1, -1,
0516                 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
0517     if (ret == 0)
0518         ret = cxgb4_update_mac_filt(pi, pi->viid, &pi->xact_addr_filt,
0519                         dev->dev_addr, true, &pi->smt_idx);
0520     if (ret == 0)
0521         ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
0522                     &pi->link_cfg);
0523     if (ret == 0) {
0524         local_bh_disable();
0525         ret = t4_enable_pi_params(pi->adapter, mb, pi, true,
0526                       true, CXGB4_DCB_ENABLED);
0527         local_bh_enable();
0528     }
0529 
0530     return ret;
0531 }
0532 
0533 #ifdef CONFIG_CHELSIO_T4_DCB
0534 /* Handle a Data Center Bridging update message from the firmware. */
0535 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
0536 {
0537     int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
0538     struct net_device *dev = adap->port[adap->chan_map[port]];
0539     int old_dcb_enabled = cxgb4_dcb_enabled(dev);
0540     int new_dcb_enabled;
0541 
0542     cxgb4_dcb_handle_fw_update(adap, pcmd);
0543     new_dcb_enabled = cxgb4_dcb_enabled(dev);
0544 
0545     /* If the DCB has become enabled or disabled on the port then we're
0546      * going to need to set up/tear down DCB Priority parameters for the
0547      * TX Queues associated with the port.
0548      */
0549     if (new_dcb_enabled != old_dcb_enabled)
0550         dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
0551 }
0552 #endif /* CONFIG_CHELSIO_T4_DCB */
0553 
0554 /* Response queue handler for the FW event queue.
0555  */
0556 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
0557               const struct pkt_gl *gl)
0558 {
0559     u8 opcode = ((const struct rss_header *)rsp)->opcode;
0560 
0561     rsp++;                                          /* skip RSS header */
0562 
0563     /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
0564      */
0565     if (unlikely(opcode == CPL_FW4_MSG &&
0566        ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
0567         rsp++;
0568         opcode = ((const struct rss_header *)rsp)->opcode;
0569         rsp++;
0570         if (opcode != CPL_SGE_EGR_UPDATE) {
0571             dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
0572                 , opcode);
0573             goto out;
0574         }
0575     }
0576 
0577     if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
0578         const struct cpl_sge_egr_update *p = (void *)rsp;
0579         unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
0580         struct sge_txq *txq;
0581 
0582         txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
0583         txq->restarts++;
0584         if (txq->q_type == CXGB4_TXQ_ETH) {
0585             struct sge_eth_txq *eq;
0586 
0587             eq = container_of(txq, struct sge_eth_txq, q);
0588             t4_sge_eth_txq_egress_update(q->adap, eq, -1);
0589         } else {
0590             struct sge_uld_txq *oq;
0591 
0592             oq = container_of(txq, struct sge_uld_txq, q);
0593             tasklet_schedule(&oq->qresume_tsk);
0594         }
0595     } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
0596         const struct cpl_fw6_msg *p = (void *)rsp;
0597 
0598 #ifdef CONFIG_CHELSIO_T4_DCB
0599         const struct fw_port_cmd *pcmd = (const void *)p->data;
0600         unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
0601         unsigned int action =
0602             FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
0603 
0604         if (cmd == FW_PORT_CMD &&
0605             (action == FW_PORT_ACTION_GET_PORT_INFO ||
0606              action == FW_PORT_ACTION_GET_PORT_INFO32)) {
0607             int port = FW_PORT_CMD_PORTID_G(
0608                     be32_to_cpu(pcmd->op_to_portid));
0609             struct net_device *dev;
0610             int dcbxdis, state_input;
0611 
0612             dev = q->adap->port[q->adap->chan_map[port]];
0613             dcbxdis = (action == FW_PORT_ACTION_GET_PORT_INFO
0614               ? !!(pcmd->u.info.dcbxdis_pkd & FW_PORT_CMD_DCBXDIS_F)
0615               : !!(be32_to_cpu(pcmd->u.info32.lstatus32_to_cbllen32)
0616                    & FW_PORT_CMD_DCBXDIS32_F));
0617             state_input = (dcbxdis
0618                        ? CXGB4_DCB_INPUT_FW_DISABLED
0619                        : CXGB4_DCB_INPUT_FW_ENABLED);
0620 
0621             cxgb4_dcb_state_fsm(dev, state_input);
0622         }
0623 
0624         if (cmd == FW_PORT_CMD &&
0625             action == FW_PORT_ACTION_L2_DCB_CFG)
0626             dcb_rpl(q->adap, pcmd);
0627         else
0628 #endif
0629             if (p->type == 0)
0630                 t4_handle_fw_rpl(q->adap, p->data);
0631     } else if (opcode == CPL_L2T_WRITE_RPL) {
0632         const struct cpl_l2t_write_rpl *p = (void *)rsp;
0633 
0634         do_l2t_write_rpl(q->adap, p);
0635     } else if (opcode == CPL_SMT_WRITE_RPL) {
0636         const struct cpl_smt_write_rpl *p = (void *)rsp;
0637 
0638         do_smt_write_rpl(q->adap, p);
0639     } else if (opcode == CPL_SET_TCB_RPL) {
0640         const struct cpl_set_tcb_rpl *p = (void *)rsp;
0641 
0642         filter_rpl(q->adap, p);
0643     } else if (opcode == CPL_ACT_OPEN_RPL) {
0644         const struct cpl_act_open_rpl *p = (void *)rsp;
0645 
0646         hash_filter_rpl(q->adap, p);
0647     } else if (opcode == CPL_ABORT_RPL_RSS) {
0648         const struct cpl_abort_rpl_rss *p = (void *)rsp;
0649 
0650         hash_del_filter_rpl(q->adap, p);
0651     } else if (opcode == CPL_SRQ_TABLE_RPL) {
0652         const struct cpl_srq_table_rpl *p = (void *)rsp;
0653 
0654         do_srq_table_rpl(q->adap, p);
0655     } else
0656         dev_err(q->adap->pdev_dev,
0657             "unexpected CPL %#x on FW event queue\n", opcode);
0658 out:
0659     return 0;
0660 }
0661 
0662 static void disable_msi(struct adapter *adapter)
0663 {
0664     if (adapter->flags & CXGB4_USING_MSIX) {
0665         pci_disable_msix(adapter->pdev);
0666         adapter->flags &= ~CXGB4_USING_MSIX;
0667     } else if (adapter->flags & CXGB4_USING_MSI) {
0668         pci_disable_msi(adapter->pdev);
0669         adapter->flags &= ~CXGB4_USING_MSI;
0670     }
0671 }
0672 
0673 /*
0674  * Interrupt handler for non-data events used with MSI-X.
0675  */
0676 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
0677 {
0678     struct adapter *adap = cookie;
0679     u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
0680 
0681     if (v & PFSW_F) {
0682         adap->swintr = 1;
0683         t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
0684     }
0685     if (adap->flags & CXGB4_MASTER_PF)
0686         t4_slow_intr_handler(adap);
0687     return IRQ_HANDLED;
0688 }
0689 
0690 int cxgb4_set_msix_aff(struct adapter *adap, unsigned short vec,
0691                cpumask_var_t *aff_mask, int idx)
0692 {
0693     int rv;
0694 
0695     if (!zalloc_cpumask_var(aff_mask, GFP_KERNEL)) {
0696         dev_err(adap->pdev_dev, "alloc_cpumask_var failed\n");
0697         return -ENOMEM;
0698     }
0699 
0700     cpumask_set_cpu(cpumask_local_spread(idx, dev_to_node(adap->pdev_dev)),
0701             *aff_mask);
0702 
0703     rv = irq_set_affinity_hint(vec, *aff_mask);
0704     if (rv)
0705         dev_warn(adap->pdev_dev,
0706              "irq_set_affinity_hint %u failed %d\n",
0707              vec, rv);
0708 
0709     return 0;
0710 }
0711 
0712 void cxgb4_clear_msix_aff(unsigned short vec, cpumask_var_t aff_mask)
0713 {
0714     irq_set_affinity_hint(vec, NULL);
0715     free_cpumask_var(aff_mask);
0716 }
0717 
0718 static int request_msix_queue_irqs(struct adapter *adap)
0719 {
0720     struct sge *s = &adap->sge;
0721     struct msix_info *minfo;
0722     int err, ethqidx;
0723 
0724     if (s->fwevtq_msix_idx < 0)
0725         return -ENOMEM;
0726 
0727     err = request_irq(adap->msix_info[s->fwevtq_msix_idx].vec,
0728               t4_sge_intr_msix, 0,
0729               adap->msix_info[s->fwevtq_msix_idx].desc,
0730               &s->fw_evtq);
0731     if (err)
0732         return err;
0733 
0734     for_each_ethrxq(s, ethqidx) {
0735         minfo = s->ethrxq[ethqidx].msix;
0736         err = request_irq(minfo->vec,
0737                   t4_sge_intr_msix, 0,
0738                   minfo->desc,
0739                   &s->ethrxq[ethqidx].rspq);
0740         if (err)
0741             goto unwind;
0742 
0743         cxgb4_set_msix_aff(adap, minfo->vec,
0744                    &minfo->aff_mask, ethqidx);
0745     }
0746     return 0;
0747 
0748 unwind:
0749     while (--ethqidx >= 0) {
0750         minfo = s->ethrxq[ethqidx].msix;
0751         cxgb4_clear_msix_aff(minfo->vec, minfo->aff_mask);
0752         free_irq(minfo->vec, &s->ethrxq[ethqidx].rspq);
0753     }
0754     free_irq(adap->msix_info[s->fwevtq_msix_idx].vec, &s->fw_evtq);
0755     return err;
0756 }
0757 
0758 static void free_msix_queue_irqs(struct adapter *adap)
0759 {
0760     struct sge *s = &adap->sge;
0761     struct msix_info *minfo;
0762     int i;
0763 
0764     free_irq(adap->msix_info[s->fwevtq_msix_idx].vec, &s->fw_evtq);
0765     for_each_ethrxq(s, i) {
0766         minfo = s->ethrxq[i].msix;
0767         cxgb4_clear_msix_aff(minfo->vec, minfo->aff_mask);
0768         free_irq(minfo->vec, &s->ethrxq[i].rspq);
0769     }
0770 }
0771 
0772 static int setup_ppod_edram(struct adapter *adap)
0773 {
0774     unsigned int param, val;
0775     int ret;
0776 
0777     /* Driver sends FW_PARAMS_PARAM_DEV_PPOD_EDRAM read command to check
0778      * if firmware supports ppod edram feature or not. If firmware
0779      * returns 1, then driver can enable this feature by sending
0780      * FW_PARAMS_PARAM_DEV_PPOD_EDRAM write command with value 1 to
0781      * enable ppod edram feature.
0782      */
0783     param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
0784         FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PPOD_EDRAM));
0785 
0786     ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &param, &val);
0787     if (ret < 0) {
0788         dev_warn(adap->pdev_dev,
0789              "querying PPOD_EDRAM support failed: %d\n",
0790              ret);
0791         return -1;
0792     }
0793 
0794     if (val != 1)
0795         return -1;
0796 
0797     ret = t4_set_params(adap, adap->mbox, adap->pf, 0, 1, &param, &val);
0798     if (ret < 0) {
0799         dev_err(adap->pdev_dev,
0800             "setting PPOD_EDRAM failed: %d\n", ret);
0801         return -1;
0802     }
0803     return 0;
0804 }
0805 
0806 static void adap_config_hpfilter(struct adapter *adapter)
0807 {
0808     u32 param, val = 0;
0809     int ret;
0810 
0811     /* Enable HP filter region. Older fw will fail this request and
0812      * it is fine.
0813      */
0814     param = FW_PARAM_DEV(HPFILTER_REGION_SUPPORT);
0815     ret = t4_set_params(adapter, adapter->mbox, adapter->pf, 0,
0816                 1, &param, &val);
0817 
0818     /* An error means FW doesn't know about HP filter support,
0819      * it's not a problem, don't return an error.
0820      */
0821     if (ret < 0)
0822         dev_err(adapter->pdev_dev,
0823             "HP filter region isn't supported by FW\n");
0824 }
0825 
0826 static int cxgb4_config_rss(const struct port_info *pi, u16 *rss,
0827                 u16 rss_size, u16 viid)
0828 {
0829     struct adapter *adap = pi->adapter;
0830     int ret;
0831 
0832     ret = t4_config_rss_range(adap, adap->mbox, viid, 0, rss_size, rss,
0833                   rss_size);
0834     if (ret)
0835         return ret;
0836 
0837     /* If Tunnel All Lookup isn't specified in the global RSS
0838      * Configuration, then we need to specify a default Ingress
0839      * Queue for any ingress packets which aren't hashed.  We'll
0840      * use our first ingress queue ...
0841      */
0842     return t4_config_vi_rss(adap, adap->mbox, viid,
0843                 FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
0844                 FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
0845                 FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
0846                 FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
0847                 FW_RSS_VI_CONFIG_CMD_UDPEN_F,
0848                 rss[0]);
0849 }
0850 
0851 /**
0852  *  cxgb4_write_rss - write the RSS table for a given port
0853  *  @pi: the port
0854  *  @queues: array of queue indices for RSS
0855  *
0856  *  Sets up the portion of the HW RSS table for the port's VI to distribute
0857  *  packets to the Rx queues in @queues.
0858  *  Should never be called before setting up sge eth rx queues
0859  */
0860 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
0861 {
0862     struct adapter *adapter = pi->adapter;
0863     const struct sge_eth_rxq *rxq;
0864     int i, err;
0865     u16 *rss;
0866 
0867     rxq = &adapter->sge.ethrxq[pi->first_qset];
0868     rss = kmalloc_array(pi->rss_size, sizeof(u16), GFP_KERNEL);
0869     if (!rss)
0870         return -ENOMEM;
0871 
0872     /* map the queue indices to queue ids */
0873     for (i = 0; i < pi->rss_size; i++, queues++)
0874         rss[i] = rxq[*queues].rspq.abs_id;
0875 
0876     err = cxgb4_config_rss(pi, rss, pi->rss_size, pi->viid);
0877     kfree(rss);
0878     return err;
0879 }
0880 
0881 /**
0882  *  setup_rss - configure RSS
0883  *  @adap: the adapter
0884  *
0885  *  Sets up RSS for each port.
0886  */
0887 static int setup_rss(struct adapter *adap)
0888 {
0889     int i, j, err;
0890 
0891     for_each_port(adap, i) {
0892         const struct port_info *pi = adap2pinfo(adap, i);
0893 
0894         /* Fill default values with equal distribution */
0895         for (j = 0; j < pi->rss_size; j++)
0896             pi->rss[j] = j % pi->nqsets;
0897 
0898         err = cxgb4_write_rss(pi, pi->rss);
0899         if (err)
0900             return err;
0901     }
0902     return 0;
0903 }
0904 
0905 /*
0906  * Return the channel of the ingress queue with the given qid.
0907  */
0908 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
0909 {
0910     qid -= p->ingr_start;
0911     return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
0912 }
0913 
0914 void cxgb4_quiesce_rx(struct sge_rspq *q)
0915 {
0916     if (q->handler)
0917         napi_disable(&q->napi);
0918 }
0919 
0920 /*
0921  * Wait until all NAPI handlers are descheduled.
0922  */
0923 static void quiesce_rx(struct adapter *adap)
0924 {
0925     int i;
0926 
0927     for (i = 0; i < adap->sge.ingr_sz; i++) {
0928         struct sge_rspq *q = adap->sge.ingr_map[i];
0929 
0930         if (!q)
0931             continue;
0932 
0933         cxgb4_quiesce_rx(q);
0934     }
0935 }
0936 
0937 /* Disable interrupt and napi handler */
0938 static void disable_interrupts(struct adapter *adap)
0939 {
0940     struct sge *s = &adap->sge;
0941 
0942     if (adap->flags & CXGB4_FULL_INIT_DONE) {
0943         t4_intr_disable(adap);
0944         if (adap->flags & CXGB4_USING_MSIX) {
0945             free_msix_queue_irqs(adap);
0946             free_irq(adap->msix_info[s->nd_msix_idx].vec,
0947                  adap);
0948         } else {
0949             free_irq(adap->pdev->irq, adap);
0950         }
0951         quiesce_rx(adap);
0952     }
0953 }
0954 
0955 void cxgb4_enable_rx(struct adapter *adap, struct sge_rspq *q)
0956 {
0957     if (q->handler)
0958         napi_enable(&q->napi);
0959 
0960     /* 0-increment GTS to start the timer and enable interrupts */
0961     t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
0962              SEINTARM_V(q->intr_params) |
0963              INGRESSQID_V(q->cntxt_id));
0964 }
0965 
0966 /*
0967  * Enable NAPI scheduling and interrupt generation for all Rx queues.
0968  */
0969 static void enable_rx(struct adapter *adap)
0970 {
0971     int i;
0972 
0973     for (i = 0; i < adap->sge.ingr_sz; i++) {
0974         struct sge_rspq *q = adap->sge.ingr_map[i];
0975 
0976         if (!q)
0977             continue;
0978 
0979         cxgb4_enable_rx(adap, q);
0980     }
0981 }
0982 
0983 static int setup_non_data_intr(struct adapter *adap)
0984 {
0985     int msix;
0986 
0987     adap->sge.nd_msix_idx = -1;
0988     if (!(adap->flags & CXGB4_USING_MSIX))
0989         return 0;
0990 
0991     /* Request MSI-X vector for non-data interrupt */
0992     msix = cxgb4_get_msix_idx_from_bmap(adap);
0993     if (msix < 0)
0994         return -ENOMEM;
0995 
0996     snprintf(adap->msix_info[msix].desc,
0997          sizeof(adap->msix_info[msix].desc),
0998          "%s", adap->port[0]->name);
0999 
1000     adap->sge.nd_msix_idx = msix;
1001     return 0;
1002 }
1003 
1004 static int setup_fw_sge_queues(struct adapter *adap)
1005 {
1006     struct sge *s = &adap->sge;
1007     int msix, err = 0;
1008 
1009     bitmap_zero(s->starving_fl, s->egr_sz);
1010     bitmap_zero(s->txq_maperr, s->egr_sz);
1011 
1012     if (adap->flags & CXGB4_USING_MSIX) {
1013         s->fwevtq_msix_idx = -1;
1014         msix = cxgb4_get_msix_idx_from_bmap(adap);
1015         if (msix < 0)
1016             return -ENOMEM;
1017 
1018         snprintf(adap->msix_info[msix].desc,
1019              sizeof(adap->msix_info[msix].desc),
1020              "%s-FWeventq", adap->port[0]->name);
1021     } else {
1022         err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
1023                        NULL, NULL, NULL, -1);
1024         if (err)
1025             return err;
1026         msix = -((int)s->intrq.abs_id + 1);
1027     }
1028 
1029     err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
1030                    msix, NULL, fwevtq_handler, NULL, -1);
1031     if (err && msix >= 0)
1032         cxgb4_free_msix_idx_in_bmap(adap, msix);
1033 
1034     s->fwevtq_msix_idx = msix;
1035     return err;
1036 }
1037 
1038 /**
1039  *  setup_sge_queues - configure SGE Tx/Rx/response queues
1040  *  @adap: the adapter
1041  *
1042  *  Determines how many sets of SGE queues to use and initializes them.
1043  *  We support multiple queue sets per port if we have MSI-X, otherwise
1044  *  just one queue set per port.
1045  */
1046 static int setup_sge_queues(struct adapter *adap)
1047 {
1048     struct sge_uld_rxq_info *rxq_info = NULL;
1049     struct sge *s = &adap->sge;
1050     unsigned int cmplqid = 0;
1051     int err, i, j, msix = 0;
1052 
1053     if (is_uld(adap))
1054         rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
1055 
1056     if (!(adap->flags & CXGB4_USING_MSIX))
1057         msix = -((int)s->intrq.abs_id + 1);
1058 
1059     for_each_port(adap, i) {
1060         struct net_device *dev = adap->port[i];
1061         struct port_info *pi = netdev_priv(dev);
1062         struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
1063         struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
1064 
1065         for (j = 0; j < pi->nqsets; j++, q++) {
1066             if (msix >= 0) {
1067                 msix = cxgb4_get_msix_idx_from_bmap(adap);
1068                 if (msix < 0) {
1069                     err = msix;
1070                     goto freeout;
1071                 }
1072 
1073                 snprintf(adap->msix_info[msix].desc,
1074                      sizeof(adap->msix_info[msix].desc),
1075                      "%s-Rx%d", dev->name, j);
1076                 q->msix = &adap->msix_info[msix];
1077             }
1078 
1079             err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
1080                            msix, &q->fl,
1081                            t4_ethrx_handler,
1082                            NULL,
1083                            t4_get_tp_ch_map(adap,
1084                                 pi->tx_chan));
1085             if (err)
1086                 goto freeout;
1087             q->rspq.idx = j;
1088             memset(&q->stats, 0, sizeof(q->stats));
1089         }
1090 
1091         q = &s->ethrxq[pi->first_qset];
1092         for (j = 0; j < pi->nqsets; j++, t++, q++) {
1093             err = t4_sge_alloc_eth_txq(adap, t, dev,
1094                     netdev_get_tx_queue(dev, j),
1095                     q->rspq.cntxt_id,
1096                     !!(adap->flags & CXGB4_SGE_DBQ_TIMER));
1097             if (err)
1098                 goto freeout;
1099         }
1100     }
1101 
1102     for_each_port(adap, i) {
1103         /* Note that cmplqid below is 0 if we don't
1104          * have RDMA queues, and that's the right value.
1105          */
1106         if (rxq_info)
1107             cmplqid = rxq_info->uldrxq[i].rspq.cntxt_id;
1108 
1109         err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
1110                         s->fw_evtq.cntxt_id, cmplqid);
1111         if (err)
1112             goto freeout;
1113     }
1114 
1115     if (!is_t4(adap->params.chip)) {
1116         err = t4_sge_alloc_eth_txq(adap, &s->ptptxq, adap->port[0],
1117                        netdev_get_tx_queue(adap->port[0], 0)
1118                        , s->fw_evtq.cntxt_id, false);
1119         if (err)
1120             goto freeout;
1121     }
1122 
1123     t4_write_reg(adap, is_t4(adap->params.chip) ?
1124                 MPS_TRC_RSS_CONTROL_A :
1125                 MPS_T5_TRC_RSS_CONTROL_A,
1126              RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
1127              QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
1128     return 0;
1129 freeout:
1130     dev_err(adap->pdev_dev, "Can't allocate queues, err=%d\n", -err);
1131     t4_free_sge_resources(adap);
1132     return err;
1133 }
1134 
1135 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
1136                  struct net_device *sb_dev)
1137 {
1138     int txq;
1139 
1140 #ifdef CONFIG_CHELSIO_T4_DCB
1141     /* If a Data Center Bridging has been successfully negotiated on this
1142      * link then we'll use the skb's priority to map it to a TX Queue.
1143      * The skb's priority is determined via the VLAN Tag Priority Code
1144      * Point field.
1145      */
1146     if (cxgb4_dcb_enabled(dev) && !is_kdump_kernel()) {
1147         u16 vlan_tci;
1148         int err;
1149 
1150         err = vlan_get_tag(skb, &vlan_tci);
1151         if (unlikely(err)) {
1152             if (net_ratelimit())
1153                 netdev_warn(dev,
1154                         "TX Packet without VLAN Tag on DCB Link\n");
1155             txq = 0;
1156         } else {
1157             txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1158 #ifdef CONFIG_CHELSIO_T4_FCOE
1159             if (skb->protocol == htons(ETH_P_FCOE))
1160                 txq = skb->priority & 0x7;
1161 #endif /* CONFIG_CHELSIO_T4_FCOE */
1162         }
1163         return txq;
1164     }
1165 #endif /* CONFIG_CHELSIO_T4_DCB */
1166 
1167     if (dev->num_tc) {
1168         struct port_info *pi = netdev2pinfo(dev);
1169         u8 ver, proto;
1170 
1171         ver = ip_hdr(skb)->version;
1172         proto = (ver == 6) ? ipv6_hdr(skb)->nexthdr :
1173                      ip_hdr(skb)->protocol;
1174 
1175         /* Send unsupported traffic pattern to normal NIC queues. */
1176         txq = netdev_pick_tx(dev, skb, sb_dev);
1177         if (xfrm_offload(skb) || is_ptp_enabled(skb, dev) ||
1178             skb->encapsulation ||
1179             cxgb4_is_ktls_skb(skb) ||
1180             (proto != IPPROTO_TCP && proto != IPPROTO_UDP))
1181             txq = txq % pi->nqsets;
1182 
1183         return txq;
1184     }
1185 
1186     if (select_queue) {
1187         txq = (skb_rx_queue_recorded(skb)
1188             ? skb_get_rx_queue(skb)
1189             : smp_processor_id());
1190 
1191         while (unlikely(txq >= dev->real_num_tx_queues))
1192             txq -= dev->real_num_tx_queues;
1193 
1194         return txq;
1195     }
1196 
1197     return netdev_pick_tx(dev, skb, NULL) % dev->real_num_tx_queues;
1198 }
1199 
1200 static int closest_timer(const struct sge *s, int time)
1201 {
1202     int i, delta, match = 0, min_delta = INT_MAX;
1203 
1204     for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1205         delta = time - s->timer_val[i];
1206         if (delta < 0)
1207             delta = -delta;
1208         if (delta < min_delta) {
1209             min_delta = delta;
1210             match = i;
1211         }
1212     }
1213     return match;
1214 }
1215 
1216 static int closest_thres(const struct sge *s, int thres)
1217 {
1218     int i, delta, match = 0, min_delta = INT_MAX;
1219 
1220     for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1221         delta = thres - s->counter_val[i];
1222         if (delta < 0)
1223             delta = -delta;
1224         if (delta < min_delta) {
1225             min_delta = delta;
1226             match = i;
1227         }
1228     }
1229     return match;
1230 }
1231 
1232 /**
1233  *  cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
1234  *  @q: the Rx queue
1235  *  @us: the hold-off time in us, or 0 to disable timer
1236  *  @cnt: the hold-off packet count, or 0 to disable counter
1237  *
1238  *  Sets an Rx queue's interrupt hold-off time and packet count.  At least
1239  *  one of the two needs to be enabled for the queue to generate interrupts.
1240  */
1241 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
1242                    unsigned int us, unsigned int cnt)
1243 {
1244     struct adapter *adap = q->adap;
1245 
1246     if ((us | cnt) == 0)
1247         cnt = 1;
1248 
1249     if (cnt) {
1250         int err;
1251         u32 v, new_idx;
1252 
1253         new_idx = closest_thres(&adap->sge, cnt);
1254         if (q->desc && q->pktcnt_idx != new_idx) {
1255             /* the queue has already been created, update it */
1256             v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1257                 FW_PARAMS_PARAM_X_V(
1258                     FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1259                 FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
1260             err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
1261                         &v, &new_idx);
1262             if (err)
1263                 return err;
1264         }
1265         q->pktcnt_idx = new_idx;
1266     }
1267 
1268     us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1269     q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
1270     return 0;
1271 }
1272 
1273 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1274 {
1275     netdev_features_t changed = dev->features ^ features;
1276     const struct port_info *pi = netdev_priv(dev);
1277     int err;
1278 
1279     if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1280         return 0;
1281 
1282     err = t4_set_rxmode(pi->adapter, pi->adapter->mbox, pi->viid,
1283                 pi->viid_mirror, -1, -1, -1, -1,
1284                 !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1285     if (unlikely(err))
1286         dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1287     return err;
1288 }
1289 
1290 static int setup_debugfs(struct adapter *adap)
1291 {
1292     if (IS_ERR_OR_NULL(adap->debugfs_root))
1293         return -1;
1294 
1295 #ifdef CONFIG_DEBUG_FS
1296     t4_setup_debugfs(adap);
1297 #endif
1298     return 0;
1299 }
1300 
1301 static void cxgb4_port_mirror_free_rxq(struct adapter *adap,
1302                        struct sge_eth_rxq *mirror_rxq)
1303 {
1304     if ((adap->flags & CXGB4_FULL_INIT_DONE) &&
1305         !(adap->flags & CXGB4_SHUTTING_DOWN))
1306         cxgb4_quiesce_rx(&mirror_rxq->rspq);
1307 
1308     if (adap->flags & CXGB4_USING_MSIX) {
1309         cxgb4_clear_msix_aff(mirror_rxq->msix->vec,
1310                      mirror_rxq->msix->aff_mask);
1311         free_irq(mirror_rxq->msix->vec, &mirror_rxq->rspq);
1312         cxgb4_free_msix_idx_in_bmap(adap, mirror_rxq->msix->idx);
1313     }
1314 
1315     free_rspq_fl(adap, &mirror_rxq->rspq, &mirror_rxq->fl);
1316 }
1317 
1318 static int cxgb4_port_mirror_alloc_queues(struct net_device *dev)
1319 {
1320     struct port_info *pi = netdev2pinfo(dev);
1321     struct adapter *adap = netdev2adap(dev);
1322     struct sge_eth_rxq *mirror_rxq;
1323     struct sge *s = &adap->sge;
1324     int ret = 0, msix = 0;
1325     u16 i, rxqid;
1326     u16 *rss;
1327 
1328     if (!pi->vi_mirror_count)
1329         return 0;
1330 
1331     if (s->mirror_rxq[pi->port_id])
1332         return 0;
1333 
1334     mirror_rxq = kcalloc(pi->nmirrorqsets, sizeof(*mirror_rxq), GFP_KERNEL);
1335     if (!mirror_rxq)
1336         return -ENOMEM;
1337 
1338     s->mirror_rxq[pi->port_id] = mirror_rxq;
1339 
1340     if (!(adap->flags & CXGB4_USING_MSIX))
1341         msix = -((int)adap->sge.intrq.abs_id + 1);
1342 
1343     for (i = 0, rxqid = 0; i < pi->nmirrorqsets; i++, rxqid++) {
1344         mirror_rxq = &s->mirror_rxq[pi->port_id][i];
1345 
1346         /* Allocate Mirror Rxqs */
1347         if (msix >= 0) {
1348             msix = cxgb4_get_msix_idx_from_bmap(adap);
1349             if (msix < 0) {
1350                 ret = msix;
1351                 goto out_free_queues;
1352             }
1353 
1354             mirror_rxq->msix = &adap->msix_info[msix];
1355             snprintf(mirror_rxq->msix->desc,
1356                  sizeof(mirror_rxq->msix->desc),
1357                  "%s-mirrorrxq%d", dev->name, i);
1358         }
1359 
1360         init_rspq(adap, &mirror_rxq->rspq,
1361               CXGB4_MIRROR_RXQ_DEFAULT_INTR_USEC,
1362               CXGB4_MIRROR_RXQ_DEFAULT_PKT_CNT,
1363               CXGB4_MIRROR_RXQ_DEFAULT_DESC_NUM,
1364               CXGB4_MIRROR_RXQ_DEFAULT_DESC_SIZE);
1365 
1366         mirror_rxq->fl.size = CXGB4_MIRROR_FLQ_DEFAULT_DESC_NUM;
1367 
1368         ret = t4_sge_alloc_rxq(adap, &mirror_rxq->rspq, false,
1369                        dev, msix, &mirror_rxq->fl,
1370                        t4_ethrx_handler, NULL, 0);
1371         if (ret)
1372             goto out_free_msix_idx;
1373 
1374         /* Setup MSI-X vectors for Mirror Rxqs */
1375         if (adap->flags & CXGB4_USING_MSIX) {
1376             ret = request_irq(mirror_rxq->msix->vec,
1377                       t4_sge_intr_msix, 0,
1378                       mirror_rxq->msix->desc,
1379                       &mirror_rxq->rspq);
1380             if (ret)
1381                 goto out_free_rxq;
1382 
1383             cxgb4_set_msix_aff(adap, mirror_rxq->msix->vec,
1384                        &mirror_rxq->msix->aff_mask, i);
1385         }
1386 
1387         /* Start NAPI for Mirror Rxqs */
1388         cxgb4_enable_rx(adap, &mirror_rxq->rspq);
1389     }
1390 
1391     /* Setup RSS for Mirror Rxqs */
1392     rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
1393     if (!rss) {
1394         ret = -ENOMEM;
1395         goto out_free_queues;
1396     }
1397 
1398     mirror_rxq = &s->mirror_rxq[pi->port_id][0];
1399     for (i = 0; i < pi->rss_size; i++)
1400         rss[i] = mirror_rxq[i % pi->nmirrorqsets].rspq.abs_id;
1401 
1402     ret = cxgb4_config_rss(pi, rss, pi->rss_size, pi->viid_mirror);
1403     kfree(rss);
1404     if (ret)
1405         goto out_free_queues;
1406 
1407     return 0;
1408 
1409 out_free_rxq:
1410     free_rspq_fl(adap, &mirror_rxq->rspq, &mirror_rxq->fl);
1411 
1412 out_free_msix_idx:
1413     cxgb4_free_msix_idx_in_bmap(adap, mirror_rxq->msix->idx);
1414 
1415 out_free_queues:
1416     while (rxqid-- > 0)
1417         cxgb4_port_mirror_free_rxq(adap,
1418                        &s->mirror_rxq[pi->port_id][rxqid]);
1419 
1420     kfree(s->mirror_rxq[pi->port_id]);
1421     s->mirror_rxq[pi->port_id] = NULL;
1422     return ret;
1423 }
1424 
1425 static void cxgb4_port_mirror_free_queues(struct net_device *dev)
1426 {
1427     struct port_info *pi = netdev2pinfo(dev);
1428     struct adapter *adap = netdev2adap(dev);
1429     struct sge *s = &adap->sge;
1430     u16 i;
1431 
1432     if (!pi->vi_mirror_count)
1433         return;
1434 
1435     if (!s->mirror_rxq[pi->port_id])
1436         return;
1437 
1438     for (i = 0; i < pi->nmirrorqsets; i++)
1439         cxgb4_port_mirror_free_rxq(adap,
1440                        &s->mirror_rxq[pi->port_id][i]);
1441 
1442     kfree(s->mirror_rxq[pi->port_id]);
1443     s->mirror_rxq[pi->port_id] = NULL;
1444 }
1445 
1446 static int cxgb4_port_mirror_start(struct net_device *dev)
1447 {
1448     struct port_info *pi = netdev2pinfo(dev);
1449     struct adapter *adap = netdev2adap(dev);
1450     int ret, idx = -1;
1451 
1452     if (!pi->vi_mirror_count)
1453         return 0;
1454 
1455     /* Mirror VIs can be created dynamically after stack had
1456      * already setup Rx modes like MTU, promisc, allmulti, etc.
1457      * on main VI. So, parse what the stack had setup on the
1458      * main VI and update the same on the mirror VI.
1459      */
1460     ret = t4_set_rxmode(adap, adap->mbox, pi->viid, pi->viid_mirror,
1461                 dev->mtu, (dev->flags & IFF_PROMISC) ? 1 : 0,
1462                 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1,
1463                 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
1464     if (ret) {
1465         dev_err(adap->pdev_dev,
1466             "Failed start up Rx mode for Mirror VI 0x%x, ret: %d\n",
1467             pi->viid_mirror, ret);
1468         return ret;
1469     }
1470 
1471     /* Enable replication bit for the device's MAC address
1472      * in MPS TCAM, so that the packets for the main VI are
1473      * replicated to mirror VI.
1474      */
1475     ret = cxgb4_update_mac_filt(pi, pi->viid_mirror, &idx,
1476                     dev->dev_addr, true, NULL);
1477     if (ret) {
1478         dev_err(adap->pdev_dev,
1479             "Failed updating MAC filter for Mirror VI 0x%x, ret: %d\n",
1480             pi->viid_mirror, ret);
1481         return ret;
1482     }
1483 
1484     /* Enabling a Virtual Interface can result in an interrupt
1485      * during the processing of the VI Enable command and, in some
1486      * paths, result in an attempt to issue another command in the
1487      * interrupt context. Thus, we disable interrupts during the
1488      * course of the VI Enable command ...
1489      */
1490     local_bh_disable();
1491     ret = t4_enable_vi_params(adap, adap->mbox, pi->viid_mirror, true, true,
1492                   false);
1493     local_bh_enable();
1494     if (ret)
1495         dev_err(adap->pdev_dev,
1496             "Failed starting Mirror VI 0x%x, ret: %d\n",
1497             pi->viid_mirror, ret);
1498 
1499     return ret;
1500 }
1501 
1502 static void cxgb4_port_mirror_stop(struct net_device *dev)
1503 {
1504     struct port_info *pi = netdev2pinfo(dev);
1505     struct adapter *adap = netdev2adap(dev);
1506 
1507     if (!pi->vi_mirror_count)
1508         return;
1509 
1510     t4_enable_vi_params(adap, adap->mbox, pi->viid_mirror, false, false,
1511                 false);
1512 }
1513 
1514 int cxgb4_port_mirror_alloc(struct net_device *dev)
1515 {
1516     struct port_info *pi = netdev2pinfo(dev);
1517     struct adapter *adap = netdev2adap(dev);
1518     int ret = 0;
1519 
1520     if (!pi->nmirrorqsets)
1521         return -EOPNOTSUPP;
1522 
1523     mutex_lock(&pi->vi_mirror_mutex);
1524     if (pi->viid_mirror) {
1525         pi->vi_mirror_count++;
1526         goto out_unlock;
1527     }
1528 
1529     ret = t4_init_port_mirror(pi, adap->mbox, pi->port_id, adap->pf, 0,
1530                   &pi->viid_mirror);
1531     if (ret)
1532         goto out_unlock;
1533 
1534     pi->vi_mirror_count = 1;
1535 
1536     if (adap->flags & CXGB4_FULL_INIT_DONE) {
1537         ret = cxgb4_port_mirror_alloc_queues(dev);
1538         if (ret)
1539             goto out_free_vi;
1540 
1541         ret = cxgb4_port_mirror_start(dev);
1542         if (ret)
1543             goto out_free_queues;
1544     }
1545 
1546     mutex_unlock(&pi->vi_mirror_mutex);
1547     return 0;
1548 
1549 out_free_queues:
1550     cxgb4_port_mirror_free_queues(dev);
1551 
1552 out_free_vi:
1553     pi->vi_mirror_count = 0;
1554     t4_free_vi(adap, adap->mbox, adap->pf, 0, pi->viid_mirror);
1555     pi->viid_mirror = 0;
1556 
1557 out_unlock:
1558     mutex_unlock(&pi->vi_mirror_mutex);
1559     return ret;
1560 }
1561 
1562 void cxgb4_port_mirror_free(struct net_device *dev)
1563 {
1564     struct port_info *pi = netdev2pinfo(dev);
1565     struct adapter *adap = netdev2adap(dev);
1566 
1567     mutex_lock(&pi->vi_mirror_mutex);
1568     if (!pi->viid_mirror)
1569         goto out_unlock;
1570 
1571     if (pi->vi_mirror_count > 1) {
1572         pi->vi_mirror_count--;
1573         goto out_unlock;
1574     }
1575 
1576     cxgb4_port_mirror_stop(dev);
1577     cxgb4_port_mirror_free_queues(dev);
1578 
1579     pi->vi_mirror_count = 0;
1580     t4_free_vi(adap, adap->mbox, adap->pf, 0, pi->viid_mirror);
1581     pi->viid_mirror = 0;
1582 
1583 out_unlock:
1584     mutex_unlock(&pi->vi_mirror_mutex);
1585 }
1586 
1587 /*
1588  * upper-layer driver support
1589  */
1590 
1591 /*
1592  * Allocate an active-open TID and set it to the supplied value.
1593  */
1594 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1595 {
1596     int atid = -1;
1597 
1598     spin_lock_bh(&t->atid_lock);
1599     if (t->afree) {
1600         union aopen_entry *p = t->afree;
1601 
1602         atid = (p - t->atid_tab) + t->atid_base;
1603         t->afree = p->next;
1604         p->data = data;
1605         t->atids_in_use++;
1606     }
1607     spin_unlock_bh(&t->atid_lock);
1608     return atid;
1609 }
1610 EXPORT_SYMBOL(cxgb4_alloc_atid);
1611 
1612 /*
1613  * Release an active-open TID.
1614  */
1615 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1616 {
1617     union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1618 
1619     spin_lock_bh(&t->atid_lock);
1620     p->next = t->afree;
1621     t->afree = p;
1622     t->atids_in_use--;
1623     spin_unlock_bh(&t->atid_lock);
1624 }
1625 EXPORT_SYMBOL(cxgb4_free_atid);
1626 
1627 /*
1628  * Allocate a server TID and set it to the supplied value.
1629  */
1630 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1631 {
1632     int stid;
1633 
1634     spin_lock_bh(&t->stid_lock);
1635     if (family == PF_INET) {
1636         stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1637         if (stid < t->nstids)
1638             __set_bit(stid, t->stid_bmap);
1639         else
1640             stid = -1;
1641     } else {
1642         stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
1643         if (stid < 0)
1644             stid = -1;
1645     }
1646     if (stid >= 0) {
1647         t->stid_tab[stid].data = data;
1648         stid += t->stid_base;
1649         /* IPv6 requires max of 520 bits or 16 cells in TCAM
1650          * This is equivalent to 4 TIDs. With CLIP enabled it
1651          * needs 2 TIDs.
1652          */
1653         if (family == PF_INET6) {
1654             t->stids_in_use += 2;
1655             t->v6_stids_in_use += 2;
1656         } else {
1657             t->stids_in_use++;
1658         }
1659     }
1660     spin_unlock_bh(&t->stid_lock);
1661     return stid;
1662 }
1663 EXPORT_SYMBOL(cxgb4_alloc_stid);
1664 
1665 /* Allocate a server filter TID and set it to the supplied value.
1666  */
1667 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1668 {
1669     int stid;
1670 
1671     spin_lock_bh(&t->stid_lock);
1672     if (family == PF_INET) {
1673         stid = find_next_zero_bit(t->stid_bmap,
1674                 t->nstids + t->nsftids, t->nstids);
1675         if (stid < (t->nstids + t->nsftids))
1676             __set_bit(stid, t->stid_bmap);
1677         else
1678             stid = -1;
1679     } else {
1680         stid = -1;
1681     }
1682     if (stid >= 0) {
1683         t->stid_tab[stid].data = data;
1684         stid -= t->nstids;
1685         stid += t->sftid_base;
1686         t->sftids_in_use++;
1687     }
1688     spin_unlock_bh(&t->stid_lock);
1689     return stid;
1690 }
1691 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1692 
1693 /* Release a server TID.
1694  */
1695 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1696 {
1697     /* Is it a server filter TID? */
1698     if (t->nsftids && (stid >= t->sftid_base)) {
1699         stid -= t->sftid_base;
1700         stid += t->nstids;
1701     } else {
1702         stid -= t->stid_base;
1703     }
1704 
1705     spin_lock_bh(&t->stid_lock);
1706     if (family == PF_INET)
1707         __clear_bit(stid, t->stid_bmap);
1708     else
1709         bitmap_release_region(t->stid_bmap, stid, 1);
1710     t->stid_tab[stid].data = NULL;
1711     if (stid < t->nstids) {
1712         if (family == PF_INET6) {
1713             t->stids_in_use -= 2;
1714             t->v6_stids_in_use -= 2;
1715         } else {
1716             t->stids_in_use--;
1717         }
1718     } else {
1719         t->sftids_in_use--;
1720     }
1721 
1722     spin_unlock_bh(&t->stid_lock);
1723 }
1724 EXPORT_SYMBOL(cxgb4_free_stid);
1725 
1726 /*
1727  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
1728  */
1729 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1730                unsigned int tid)
1731 {
1732     struct cpl_tid_release *req;
1733 
1734     set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1735     req = __skb_put(skb, sizeof(*req));
1736     INIT_TP_WR(req, tid);
1737     OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1738 }
1739 
1740 /*
1741  * Queue a TID release request and if necessary schedule a work queue to
1742  * process it.
1743  */
1744 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1745                     unsigned int tid)
1746 {
1747     struct adapter *adap = container_of(t, struct adapter, tids);
1748     void **p = &t->tid_tab[tid - t->tid_base];
1749 
1750     spin_lock_bh(&adap->tid_release_lock);
1751     *p = adap->tid_release_head;
1752     /* Low 2 bits encode the Tx channel number */
1753     adap->tid_release_head = (void **)((uintptr_t)p | chan);
1754     if (!adap->tid_release_task_busy) {
1755         adap->tid_release_task_busy = true;
1756         queue_work(adap->workq, &adap->tid_release_task);
1757     }
1758     spin_unlock_bh(&adap->tid_release_lock);
1759 }
1760 
1761 /*
1762  * Process the list of pending TID release requests.
1763  */
1764 static void process_tid_release_list(struct work_struct *work)
1765 {
1766     struct sk_buff *skb;
1767     struct adapter *adap;
1768 
1769     adap = container_of(work, struct adapter, tid_release_task);
1770 
1771     spin_lock_bh(&adap->tid_release_lock);
1772     while (adap->tid_release_head) {
1773         void **p = adap->tid_release_head;
1774         unsigned int chan = (uintptr_t)p & 3;
1775         p = (void *)p - chan;
1776 
1777         adap->tid_release_head = *p;
1778         *p = NULL;
1779         spin_unlock_bh(&adap->tid_release_lock);
1780 
1781         while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1782                      GFP_KERNEL)))
1783             schedule_timeout_uninterruptible(1);
1784 
1785         mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1786         t4_ofld_send(adap, skb);
1787         spin_lock_bh(&adap->tid_release_lock);
1788     }
1789     adap->tid_release_task_busy = false;
1790     spin_unlock_bh(&adap->tid_release_lock);
1791 }
1792 
1793 /*
1794  * Release a TID and inform HW.  If we are unable to allocate the release
1795  * message we defer to a work queue.
1796  */
1797 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
1798               unsigned short family)
1799 {
1800     struct adapter *adap = container_of(t, struct adapter, tids);
1801     struct sk_buff *skb;
1802 
1803     WARN_ON(tid_out_of_range(&adap->tids, tid));
1804 
1805     if (t->tid_tab[tid - adap->tids.tid_base]) {
1806         t->tid_tab[tid - adap->tids.tid_base] = NULL;
1807         atomic_dec(&t->conns_in_use);
1808         if (t->hash_base && (tid >= t->hash_base)) {
1809             if (family == AF_INET6)
1810                 atomic_sub(2, &t->hash_tids_in_use);
1811             else
1812                 atomic_dec(&t->hash_tids_in_use);
1813         } else {
1814             if (family == AF_INET6)
1815                 atomic_sub(2, &t->tids_in_use);
1816             else
1817                 atomic_dec(&t->tids_in_use);
1818         }
1819     }
1820 
1821     skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1822     if (likely(skb)) {
1823         mk_tid_release(skb, chan, tid);
1824         t4_ofld_send(adap, skb);
1825     } else
1826         cxgb4_queue_tid_release(t, chan, tid);
1827 }
1828 EXPORT_SYMBOL(cxgb4_remove_tid);
1829 
1830 /*
1831  * Allocate and initialize the TID tables.  Returns 0 on success.
1832  */
1833 static int tid_init(struct tid_info *t)
1834 {
1835     struct adapter *adap = container_of(t, struct adapter, tids);
1836     unsigned int max_ftids = t->nftids + t->nsftids;
1837     unsigned int natids = t->natids;
1838     unsigned int hpftid_bmap_size;
1839     unsigned int eotid_bmap_size;
1840     unsigned int stid_bmap_size;
1841     unsigned int ftid_bmap_size;
1842     size_t size;
1843 
1844     stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1845     ftid_bmap_size = BITS_TO_LONGS(t->nftids);
1846     hpftid_bmap_size = BITS_TO_LONGS(t->nhpftids);
1847     eotid_bmap_size = BITS_TO_LONGS(t->neotids);
1848     size = t->ntids * sizeof(*t->tid_tab) +
1849            natids * sizeof(*t->atid_tab) +
1850            t->nstids * sizeof(*t->stid_tab) +
1851            t->nsftids * sizeof(*t->stid_tab) +
1852            stid_bmap_size * sizeof(long) +
1853            t->nhpftids * sizeof(*t->hpftid_tab) +
1854            hpftid_bmap_size * sizeof(long) +
1855            max_ftids * sizeof(*t->ftid_tab) +
1856            ftid_bmap_size * sizeof(long) +
1857            t->neotids * sizeof(*t->eotid_tab) +
1858            eotid_bmap_size * sizeof(long);
1859 
1860     t->tid_tab = kvzalloc(size, GFP_KERNEL);
1861     if (!t->tid_tab)
1862         return -ENOMEM;
1863 
1864     t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1865     t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1866     t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1867     t->hpftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1868     t->hpftid_bmap = (unsigned long *)&t->hpftid_tab[t->nhpftids];
1869     t->ftid_tab = (struct filter_entry *)&t->hpftid_bmap[hpftid_bmap_size];
1870     t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
1871     t->eotid_tab = (struct eotid_entry *)&t->ftid_bmap[ftid_bmap_size];
1872     t->eotid_bmap = (unsigned long *)&t->eotid_tab[t->neotids];
1873     spin_lock_init(&t->stid_lock);
1874     spin_lock_init(&t->atid_lock);
1875     spin_lock_init(&t->ftid_lock);
1876 
1877     t->stids_in_use = 0;
1878     t->v6_stids_in_use = 0;
1879     t->sftids_in_use = 0;
1880     t->afree = NULL;
1881     t->atids_in_use = 0;
1882     atomic_set(&t->tids_in_use, 0);
1883     atomic_set(&t->conns_in_use, 0);
1884     atomic_set(&t->hash_tids_in_use, 0);
1885     atomic_set(&t->eotids_in_use, 0);
1886 
1887     /* Setup the free list for atid_tab and clear the stid bitmap. */
1888     if (natids) {
1889         while (--natids)
1890             t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1891         t->afree = t->atid_tab;
1892     }
1893 
1894     if (is_offload(adap)) {
1895         bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1896         /* Reserve stid 0 for T4/T5 adapters */
1897         if (!t->stid_base &&
1898             CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1899             __set_bit(0, t->stid_bmap);
1900 
1901         if (t->neotids)
1902             bitmap_zero(t->eotid_bmap, t->neotids);
1903     }
1904 
1905     if (t->nhpftids)
1906         bitmap_zero(t->hpftid_bmap, t->nhpftids);
1907     bitmap_zero(t->ftid_bmap, t->nftids);
1908     return 0;
1909 }
1910 
1911 /**
1912  *  cxgb4_create_server - create an IP server
1913  *  @dev: the device
1914  *  @stid: the server TID
1915  *  @sip: local IP address to bind server to
1916  *  @sport: the server's TCP port
1917  *  @vlan: the VLAN header information
1918  *  @queue: queue to direct messages from this server to
1919  *
1920  *  Create an IP server for the given port and address.
1921  *  Returns <0 on error and one of the %NET_XMIT_* values on success.
1922  */
1923 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1924             __be32 sip, __be16 sport, __be16 vlan,
1925             unsigned int queue)
1926 {
1927     unsigned int chan;
1928     struct sk_buff *skb;
1929     struct adapter *adap;
1930     struct cpl_pass_open_req *req;
1931     int ret;
1932 
1933     skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1934     if (!skb)
1935         return -ENOMEM;
1936 
1937     adap = netdev2adap(dev);
1938     req = __skb_put(skb, sizeof(*req));
1939     INIT_TP_WR(req, 0);
1940     OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1941     req->local_port = sport;
1942     req->peer_port = htons(0);
1943     req->local_ip = sip;
1944     req->peer_ip = htonl(0);
1945     chan = rxq_to_chan(&adap->sge, queue);
1946     req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1947     req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1948                 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1949     ret = t4_mgmt_tx(adap, skb);
1950     return net_xmit_eval(ret);
1951 }
1952 EXPORT_SYMBOL(cxgb4_create_server);
1953 
1954 /*  cxgb4_create_server6 - create an IPv6 server
1955  *  @dev: the device
1956  *  @stid: the server TID
1957  *  @sip: local IPv6 address to bind server to
1958  *  @sport: the server's TCP port
1959  *  @queue: queue to direct messages from this server to
1960  *
1961  *  Create an IPv6 server for the given port and address.
1962  *  Returns <0 on error and one of the %NET_XMIT_* values on success.
1963  */
1964 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1965              const struct in6_addr *sip, __be16 sport,
1966              unsigned int queue)
1967 {
1968     unsigned int chan;
1969     struct sk_buff *skb;
1970     struct adapter *adap;
1971     struct cpl_pass_open_req6 *req;
1972     int ret;
1973 
1974     skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1975     if (!skb)
1976         return -ENOMEM;
1977 
1978     adap = netdev2adap(dev);
1979     req = __skb_put(skb, sizeof(*req));
1980     INIT_TP_WR(req, 0);
1981     OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1982     req->local_port = sport;
1983     req->peer_port = htons(0);
1984     req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1985     req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1986     req->peer_ip_hi = cpu_to_be64(0);
1987     req->peer_ip_lo = cpu_to_be64(0);
1988     chan = rxq_to_chan(&adap->sge, queue);
1989     req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1990     req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1991                 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1992     ret = t4_mgmt_tx(adap, skb);
1993     return net_xmit_eval(ret);
1994 }
1995 EXPORT_SYMBOL(cxgb4_create_server6);
1996 
1997 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1998             unsigned int queue, bool ipv6)
1999 {
2000     struct sk_buff *skb;
2001     struct adapter *adap;
2002     struct cpl_close_listsvr_req *req;
2003     int ret;
2004 
2005     adap = netdev2adap(dev);
2006 
2007     skb = alloc_skb(sizeof(*req), GFP_KERNEL);
2008     if (!skb)
2009         return -ENOMEM;
2010 
2011     req = __skb_put(skb, sizeof(*req));
2012     INIT_TP_WR(req, 0);
2013     OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
2014     req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
2015                 LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
2016     ret = t4_mgmt_tx(adap, skb);
2017     return net_xmit_eval(ret);
2018 }
2019 EXPORT_SYMBOL(cxgb4_remove_server);
2020 
2021 /**
2022  *  cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
2023  *  @mtus: the HW MTU table
2024  *  @mtu: the target MTU
2025  *  @idx: index of selected entry in the MTU table
2026  *
2027  *  Returns the index and the value in the HW MTU table that is closest to
2028  *  but does not exceed @mtu, unless @mtu is smaller than any value in the
2029  *  table, in which case that smallest available value is selected.
2030  */
2031 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
2032                 unsigned int *idx)
2033 {
2034     unsigned int i = 0;
2035 
2036     while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
2037         ++i;
2038     if (idx)
2039         *idx = i;
2040     return mtus[i];
2041 }
2042 EXPORT_SYMBOL(cxgb4_best_mtu);
2043 
2044 /**
2045  *     cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
2046  *     @mtus: the HW MTU table
2047  *     @header_size: Header Size
2048  *     @data_size_max: maximum Data Segment Size
2049  *     @data_size_align: desired Data Segment Size Alignment (2^N)
2050  *     @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
2051  *
2052  *     Similar to cxgb4_best_mtu() but instead of searching the Hardware
2053  *     MTU Table based solely on a Maximum MTU parameter, we break that
2054  *     parameter up into a Header Size and Maximum Data Segment Size, and
2055  *     provide a desired Data Segment Size Alignment.  If we find an MTU in
2056  *     the Hardware MTU Table which will result in a Data Segment Size with
2057  *     the requested alignment _and_ that MTU isn't "too far" from the
2058  *     closest MTU, then we'll return that rather than the closest MTU.
2059  */
2060 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
2061                     unsigned short header_size,
2062                     unsigned short data_size_max,
2063                     unsigned short data_size_align,
2064                     unsigned int *mtu_idxp)
2065 {
2066     unsigned short max_mtu = header_size + data_size_max;
2067     unsigned short data_size_align_mask = data_size_align - 1;
2068     int mtu_idx, aligned_mtu_idx;
2069 
2070     /* Scan the MTU Table till we find an MTU which is larger than our
2071      * Maximum MTU or we reach the end of the table.  Along the way,
2072      * record the last MTU found, if any, which will result in a Data
2073      * Segment Length matching the requested alignment.
2074      */
2075     for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
2076         unsigned short data_size = mtus[mtu_idx] - header_size;
2077 
2078         /* If this MTU minus the Header Size would result in a
2079          * Data Segment Size of the desired alignment, remember it.
2080          */
2081         if ((data_size & data_size_align_mask) == 0)
2082             aligned_mtu_idx = mtu_idx;
2083 
2084         /* If we're not at the end of the Hardware MTU Table and the
2085          * next element is larger than our Maximum MTU, drop out of
2086          * the loop.
2087          */
2088         if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
2089             break;
2090     }
2091 
2092     /* If we fell out of the loop because we ran to the end of the table,
2093      * then we just have to use the last [largest] entry.
2094      */
2095     if (mtu_idx == NMTUS)
2096         mtu_idx--;
2097 
2098     /* If we found an MTU which resulted in the requested Data Segment
2099      * Length alignment and that's "not far" from the largest MTU which is
2100      * less than or equal to the maximum MTU, then use that.
2101      */
2102     if (aligned_mtu_idx >= 0 &&
2103         mtu_idx - aligned_mtu_idx <= 1)
2104         mtu_idx = aligned_mtu_idx;
2105 
2106     /* If the caller has passed in an MTU Index pointer, pass the
2107      * MTU Index back.  Return the MTU value.
2108      */
2109     if (mtu_idxp)
2110         *mtu_idxp = mtu_idx;
2111     return mtus[mtu_idx];
2112 }
2113 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
2114 
2115 /**
2116  *  cxgb4_port_chan - get the HW channel of a port
2117  *  @dev: the net device for the port
2118  *
2119  *  Return the HW Tx channel of the given port.
2120  */
2121 unsigned int cxgb4_port_chan(const struct net_device *dev)
2122 {
2123     return netdev2pinfo(dev)->tx_chan;
2124 }
2125 EXPORT_SYMBOL(cxgb4_port_chan);
2126 
2127 /**
2128  *      cxgb4_port_e2cchan - get the HW c-channel of a port
2129  *      @dev: the net device for the port
2130  *
2131  *      Return the HW RX c-channel of the given port.
2132  */
2133 unsigned int cxgb4_port_e2cchan(const struct net_device *dev)
2134 {
2135     return netdev2pinfo(dev)->rx_cchan;
2136 }
2137 EXPORT_SYMBOL(cxgb4_port_e2cchan);
2138 
2139 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
2140 {
2141     struct adapter *adap = netdev2adap(dev);
2142     u32 v1, v2, lp_count, hp_count;
2143 
2144     v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
2145     v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
2146     if (is_t4(adap->params.chip)) {
2147         lp_count = LP_COUNT_G(v1);
2148         hp_count = HP_COUNT_G(v1);
2149     } else {
2150         lp_count = LP_COUNT_T5_G(v1);
2151         hp_count = HP_COUNT_T5_G(v2);
2152     }
2153     return lpfifo ? lp_count : hp_count;
2154 }
2155 EXPORT_SYMBOL(cxgb4_dbfifo_count);
2156 
2157 /**
2158  *  cxgb4_port_viid - get the VI id of a port
2159  *  @dev: the net device for the port
2160  *
2161  *  Return the VI id of the given port.
2162  */
2163 unsigned int cxgb4_port_viid(const struct net_device *dev)
2164 {
2165     return netdev2pinfo(dev)->viid;
2166 }
2167 EXPORT_SYMBOL(cxgb4_port_viid);
2168 
2169 /**
2170  *  cxgb4_port_idx - get the index of a port
2171  *  @dev: the net device for the port
2172  *
2173  *  Return the index of the given port.
2174  */
2175 unsigned int cxgb4_port_idx(const struct net_device *dev)
2176 {
2177     return netdev2pinfo(dev)->port_id;
2178 }
2179 EXPORT_SYMBOL(cxgb4_port_idx);
2180 
2181 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
2182              struct tp_tcp_stats *v6)
2183 {
2184     struct adapter *adap = pci_get_drvdata(pdev);
2185 
2186     spin_lock(&adap->stats_lock);
2187     t4_tp_get_tcp_stats(adap, v4, v6, false);
2188     spin_unlock(&adap->stats_lock);
2189 }
2190 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
2191 
2192 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
2193               const unsigned int *pgsz_order)
2194 {
2195     struct adapter *adap = netdev2adap(dev);
2196 
2197     t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
2198     t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
2199              HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
2200              HPZ3_V(pgsz_order[3]));
2201 }
2202 EXPORT_SYMBOL(cxgb4_iscsi_init);
2203 
2204 int cxgb4_flush_eq_cache(struct net_device *dev)
2205 {
2206     struct adapter *adap = netdev2adap(dev);
2207 
2208     return t4_sge_ctxt_flush(adap, adap->mbox, CTXT_EGRESS);
2209 }
2210 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
2211 
2212 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
2213 {
2214     u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
2215     __be64 indices;
2216     int ret;
2217 
2218     spin_lock(&adap->win0_lock);
2219     ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
2220                sizeof(indices), (__be32 *)&indices,
2221                T4_MEMORY_READ);
2222     spin_unlock(&adap->win0_lock);
2223     if (!ret) {
2224         *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
2225         *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
2226     }
2227     return ret;
2228 }
2229 
2230 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
2231             u16 size)
2232 {
2233     struct adapter *adap = netdev2adap(dev);
2234     u16 hw_pidx, hw_cidx;
2235     int ret;
2236 
2237     ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
2238     if (ret)
2239         goto out;
2240 
2241     if (pidx != hw_pidx) {
2242         u16 delta;
2243         u32 val;
2244 
2245         if (pidx >= hw_pidx)
2246             delta = pidx - hw_pidx;
2247         else
2248             delta = size - hw_pidx + pidx;
2249 
2250         if (is_t4(adap->params.chip))
2251             val = PIDX_V(delta);
2252         else
2253             val = PIDX_T5_V(delta);
2254         wmb();
2255         t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2256                  QID_V(qid) | val);
2257     }
2258 out:
2259     return ret;
2260 }
2261 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
2262 
2263 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
2264 {
2265     u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
2266     u32 edc0_end, edc1_end, mc0_end, mc1_end;
2267     u32 offset, memtype, memaddr;
2268     struct adapter *adap;
2269     u32 hma_size = 0;
2270     int ret;
2271 
2272     adap = netdev2adap(dev);
2273 
2274     offset = ((stag >> 8) * 32) + adap->vres.stag.start;
2275 
2276     /* Figure out where the offset lands in the Memory Type/Address scheme.
2277      * This code assumes that the memory is laid out starting at offset 0
2278      * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
2279      * and EDC1.  Some cards will have neither MC0 nor MC1, most cards have
2280      * MC0, and some have both MC0 and MC1.
2281      */
2282     size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
2283     edc0_size = EDRAM0_SIZE_G(size) << 20;
2284     size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
2285     edc1_size = EDRAM1_SIZE_G(size) << 20;
2286     size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
2287     mc0_size = EXT_MEM0_SIZE_G(size) << 20;
2288 
2289     if (t4_read_reg(adap, MA_TARGET_MEM_ENABLE_A) & HMA_MUX_F) {
2290         size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
2291         hma_size = EXT_MEM1_SIZE_G(size) << 20;
2292     }
2293     edc0_end = edc0_size;
2294     edc1_end = edc0_end + edc1_size;
2295     mc0_end = edc1_end + mc0_size;
2296 
2297     if (offset < edc0_end) {
2298         memtype = MEM_EDC0;
2299         memaddr = offset;
2300     } else if (offset < edc1_end) {
2301         memtype = MEM_EDC1;
2302         memaddr = offset - edc0_end;
2303     } else {
2304         if (hma_size && (offset < (edc1_end + hma_size))) {
2305             memtype = MEM_HMA;
2306             memaddr = offset - edc1_end;
2307         } else if (offset < mc0_end) {
2308             memtype = MEM_MC0;
2309             memaddr = offset - edc1_end;
2310         } else if (is_t5(adap->params.chip)) {
2311             size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
2312             mc1_size = EXT_MEM1_SIZE_G(size) << 20;
2313             mc1_end = mc0_end + mc1_size;
2314             if (offset < mc1_end) {
2315                 memtype = MEM_MC1;
2316                 memaddr = offset - mc0_end;
2317             } else {
2318                 /* offset beyond the end of any memory */
2319                 goto err;
2320             }
2321         } else {
2322             /* T4/T6 only has a single memory channel */
2323             goto err;
2324         }
2325     }
2326 
2327     spin_lock(&adap->win0_lock);
2328     ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
2329     spin_unlock(&adap->win0_lock);
2330     return ret;
2331 
2332 err:
2333     dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
2334         stag, offset);
2335     return -EINVAL;
2336 }
2337 EXPORT_SYMBOL(cxgb4_read_tpte);
2338 
2339 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
2340 {
2341     u32 hi, lo;
2342     struct adapter *adap;
2343 
2344     adap = netdev2adap(dev);
2345     lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
2346     hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
2347 
2348     return ((u64)hi << 32) | (u64)lo;
2349 }
2350 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
2351 
2352 int cxgb4_bar2_sge_qregs(struct net_device *dev,
2353              unsigned int qid,
2354              enum cxgb4_bar2_qtype qtype,
2355              int user,
2356              u64 *pbar2_qoffset,
2357              unsigned int *pbar2_qid)
2358 {
2359     return t4_bar2_sge_qregs(netdev2adap(dev),
2360                  qid,
2361                  (qtype == CXGB4_BAR2_QTYPE_EGRESS
2362                   ? T4_BAR2_QTYPE_EGRESS
2363                   : T4_BAR2_QTYPE_INGRESS),
2364                  user,
2365                  pbar2_qoffset,
2366                  pbar2_qid);
2367 }
2368 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
2369 
2370 static struct pci_driver cxgb4_driver;
2371 
2372 static void check_neigh_update(struct neighbour *neigh)
2373 {
2374     const struct device *parent;
2375     const struct net_device *netdev = neigh->dev;
2376 
2377     if (is_vlan_dev(netdev))
2378         netdev = vlan_dev_real_dev(netdev);
2379     parent = netdev->dev.parent;
2380     if (parent && parent->driver == &cxgb4_driver.driver)
2381         t4_l2t_update(dev_get_drvdata(parent), neigh);
2382 }
2383 
2384 static int netevent_cb(struct notifier_block *nb, unsigned long event,
2385                void *data)
2386 {
2387     switch (event) {
2388     case NETEVENT_NEIGH_UPDATE:
2389         check_neigh_update(data);
2390         break;
2391     case NETEVENT_REDIRECT:
2392     default:
2393         break;
2394     }
2395     return 0;
2396 }
2397 
2398 static bool netevent_registered;
2399 static struct notifier_block cxgb4_netevent_nb = {
2400     .notifier_call = netevent_cb
2401 };
2402 
2403 static void drain_db_fifo(struct adapter *adap, int usecs)
2404 {
2405     u32 v1, v2, lp_count, hp_count;
2406 
2407     do {
2408         v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
2409         v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
2410         if (is_t4(adap->params.chip)) {
2411             lp_count = LP_COUNT_G(v1);
2412             hp_count = HP_COUNT_G(v1);
2413         } else {
2414             lp_count = LP_COUNT_T5_G(v1);
2415             hp_count = HP_COUNT_T5_G(v2);
2416         }
2417 
2418         if (lp_count == 0 && hp_count == 0)
2419             break;
2420         set_current_state(TASK_UNINTERRUPTIBLE);
2421         schedule_timeout(usecs_to_jiffies(usecs));
2422     } while (1);
2423 }
2424 
2425 static void disable_txq_db(struct sge_txq *q)
2426 {
2427     unsigned long flags;
2428 
2429     spin_lock_irqsave(&q->db_lock, flags);
2430     q->db_disabled = 1;
2431     spin_unlock_irqrestore(&q->db_lock, flags);
2432 }
2433 
2434 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
2435 {
2436     spin_lock_irq(&q->db_lock);
2437     if (q->db_pidx_inc) {
2438         /* Make sure that all writes to the TX descriptors
2439          * are committed before we tell HW about them.
2440          */
2441         wmb();
2442         t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2443                  QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
2444         q->db_pidx_inc = 0;
2445     }
2446     q->db_disabled = 0;
2447     spin_unlock_irq(&q->db_lock);
2448 }
2449 
2450 static void disable_dbs(struct adapter *adap)
2451 {
2452     int i;
2453 
2454     for_each_ethrxq(&adap->sge, i)
2455         disable_txq_db(&adap->sge.ethtxq[i].q);
2456     if (is_offload(adap)) {
2457         struct sge_uld_txq_info *txq_info =
2458             adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2459 
2460         if (txq_info) {
2461             for_each_ofldtxq(&adap->sge, i) {
2462                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2463 
2464                 disable_txq_db(&txq->q);
2465             }
2466         }
2467     }
2468     for_each_port(adap, i)
2469         disable_txq_db(&adap->sge.ctrlq[i].q);
2470 }
2471 
2472 static void enable_dbs(struct adapter *adap)
2473 {
2474     int i;
2475 
2476     for_each_ethrxq(&adap->sge, i)
2477         enable_txq_db(adap, &adap->sge.ethtxq[i].q);
2478     if (is_offload(adap)) {
2479         struct sge_uld_txq_info *txq_info =
2480             adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2481 
2482         if (txq_info) {
2483             for_each_ofldtxq(&adap->sge, i) {
2484                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2485 
2486                 enable_txq_db(adap, &txq->q);
2487             }
2488         }
2489     }
2490     for_each_port(adap, i)
2491         enable_txq_db(adap, &adap->sge.ctrlq[i].q);
2492 }
2493 
2494 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
2495 {
2496     enum cxgb4_uld type = CXGB4_ULD_RDMA;
2497 
2498     if (adap->uld && adap->uld[type].handle)
2499         adap->uld[type].control(adap->uld[type].handle, cmd);
2500 }
2501 
2502 static void process_db_full(struct work_struct *work)
2503 {
2504     struct adapter *adap;
2505 
2506     adap = container_of(work, struct adapter, db_full_task);
2507 
2508     drain_db_fifo(adap, dbfifo_drain_delay);
2509     enable_dbs(adap);
2510     notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2511     if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2512         t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2513                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
2514                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
2515     else
2516         t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2517                  DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
2518 }
2519 
2520 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
2521 {
2522     u16 hw_pidx, hw_cidx;
2523     int ret;
2524 
2525     spin_lock_irq(&q->db_lock);
2526     ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
2527     if (ret)
2528         goto out;
2529     if (q->db_pidx != hw_pidx) {
2530         u16 delta;
2531         u32 val;
2532 
2533         if (q->db_pidx >= hw_pidx)
2534             delta = q->db_pidx - hw_pidx;
2535         else
2536             delta = q->size - hw_pidx + q->db_pidx;
2537 
2538         if (is_t4(adap->params.chip))
2539             val = PIDX_V(delta);
2540         else
2541             val = PIDX_T5_V(delta);
2542         wmb();
2543         t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2544                  QID_V(q->cntxt_id) | val);
2545     }
2546 out:
2547     q->db_disabled = 0;
2548     q->db_pidx_inc = 0;
2549     spin_unlock_irq(&q->db_lock);
2550     if (ret)
2551         CH_WARN(adap, "DB drop recovery failed.\n");
2552 }
2553 
2554 static void recover_all_queues(struct adapter *adap)
2555 {
2556     int i;
2557 
2558     for_each_ethrxq(&adap->sge, i)
2559         sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
2560     if (is_offload(adap)) {
2561         struct sge_uld_txq_info *txq_info =
2562             adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2563         if (txq_info) {
2564             for_each_ofldtxq(&adap->sge, i) {
2565                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2566 
2567                 sync_txq_pidx(adap, &txq->q);
2568             }
2569         }
2570     }
2571     for_each_port(adap, i)
2572         sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2573 }
2574 
2575 static void process_db_drop(struct work_struct *work)
2576 {
2577     struct adapter *adap;
2578 
2579     adap = container_of(work, struct adapter, db_drop_task);
2580 
2581     if (is_t4(adap->params.chip)) {
2582         drain_db_fifo(adap, dbfifo_drain_delay);
2583         notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
2584         drain_db_fifo(adap, dbfifo_drain_delay);
2585         recover_all_queues(adap);
2586         drain_db_fifo(adap, dbfifo_drain_delay);
2587         enable_dbs(adap);
2588         notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2589     } else if (is_t5(adap->params.chip)) {
2590         u32 dropped_db = t4_read_reg(adap, 0x010ac);
2591         u16 qid = (dropped_db >> 15) & 0x1ffff;
2592         u16 pidx_inc = dropped_db & 0x1fff;
2593         u64 bar2_qoffset;
2594         unsigned int bar2_qid;
2595         int ret;
2596 
2597         ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2598                     0, &bar2_qoffset, &bar2_qid);
2599         if (ret)
2600             dev_err(adap->pdev_dev, "doorbell drop recovery: "
2601                 "qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2602         else
2603             writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2604                    adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2605 
2606         /* Re-enable BAR2 WC */
2607         t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2608     }
2609 
2610     if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2611         t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2612 }
2613 
2614 void t4_db_full(struct adapter *adap)
2615 {
2616     if (is_t4(adap->params.chip)) {
2617         disable_dbs(adap);
2618         notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2619         t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2620                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2621         queue_work(adap->workq, &adap->db_full_task);
2622     }
2623 }
2624 
2625 void t4_db_dropped(struct adapter *adap)
2626 {
2627     if (is_t4(adap->params.chip)) {
2628         disable_dbs(adap);
2629         notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2630     }
2631     queue_work(adap->workq, &adap->db_drop_task);
2632 }
2633 
2634 void t4_register_netevent_notifier(void)
2635 {
2636     if (!netevent_registered) {
2637         register_netevent_notifier(&cxgb4_netevent_nb);
2638         netevent_registered = true;
2639     }
2640 }
2641 
2642 static void detach_ulds(struct adapter *adap)
2643 {
2644     unsigned int i;
2645 
2646     if (!is_uld(adap))
2647         return;
2648 
2649     mutex_lock(&uld_mutex);
2650     list_del(&adap->list_node);
2651 
2652     for (i = 0; i < CXGB4_ULD_MAX; i++)
2653         if (adap->uld && adap->uld[i].handle)
2654             adap->uld[i].state_change(adap->uld[i].handle,
2655                          CXGB4_STATE_DETACH);
2656 
2657     if (netevent_registered && list_empty(&adapter_list)) {
2658         unregister_netevent_notifier(&cxgb4_netevent_nb);
2659         netevent_registered = false;
2660     }
2661     mutex_unlock(&uld_mutex);
2662 }
2663 
2664 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2665 {
2666     unsigned int i;
2667 
2668     mutex_lock(&uld_mutex);
2669     for (i = 0; i < CXGB4_ULD_MAX; i++)
2670         if (adap->uld && adap->uld[i].handle)
2671             adap->uld[i].state_change(adap->uld[i].handle,
2672                           new_state);
2673     mutex_unlock(&uld_mutex);
2674 }
2675 
2676 #if IS_ENABLED(CONFIG_IPV6)
2677 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2678                    unsigned long event, void *data)
2679 {
2680     struct inet6_ifaddr *ifa = data;
2681     struct net_device *event_dev = ifa->idev->dev;
2682     const struct device *parent = NULL;
2683 #if IS_ENABLED(CONFIG_BONDING)
2684     struct adapter *adap;
2685 #endif
2686     if (is_vlan_dev(event_dev))
2687         event_dev = vlan_dev_real_dev(event_dev);
2688 #if IS_ENABLED(CONFIG_BONDING)
2689     if (event_dev->flags & IFF_MASTER) {
2690         list_for_each_entry(adap, &adapter_list, list_node) {
2691             switch (event) {
2692             case NETDEV_UP:
2693                 cxgb4_clip_get(adap->port[0],
2694                            (const u32 *)ifa, 1);
2695                 break;
2696             case NETDEV_DOWN:
2697                 cxgb4_clip_release(adap->port[0],
2698                            (const u32 *)ifa, 1);
2699                 break;
2700             default:
2701                 break;
2702             }
2703         }
2704         return NOTIFY_OK;
2705     }
2706 #endif
2707 
2708     if (event_dev)
2709         parent = event_dev->dev.parent;
2710 
2711     if (parent && parent->driver == &cxgb4_driver.driver) {
2712         switch (event) {
2713         case NETDEV_UP:
2714             cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2715             break;
2716         case NETDEV_DOWN:
2717             cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2718             break;
2719         default:
2720             break;
2721         }
2722     }
2723     return NOTIFY_OK;
2724 }
2725 
2726 static bool inet6addr_registered;
2727 static struct notifier_block cxgb4_inet6addr_notifier = {
2728     .notifier_call = cxgb4_inet6addr_handler
2729 };
2730 
2731 static void update_clip(const struct adapter *adap)
2732 {
2733     int i;
2734     struct net_device *dev;
2735     int ret;
2736 
2737     rcu_read_lock();
2738 
2739     for (i = 0; i < MAX_NPORTS; i++) {
2740         dev = adap->port[i];
2741         ret = 0;
2742 
2743         if (dev)
2744             ret = cxgb4_update_root_dev_clip(dev);
2745 
2746         if (ret < 0)
2747             break;
2748     }
2749     rcu_read_unlock();
2750 }
2751 #endif /* IS_ENABLED(CONFIG_IPV6) */
2752 
2753 /**
2754  *  cxgb_up - enable the adapter
2755  *  @adap: adapter being enabled
2756  *
2757  *  Called when the first port is enabled, this function performs the
2758  *  actions necessary to make an adapter operational, such as completing
2759  *  the initialization of HW modules, and enabling interrupts.
2760  *
2761  *  Must be called with the rtnl lock held.
2762  */
2763 static int cxgb_up(struct adapter *adap)
2764 {
2765     struct sge *s = &adap->sge;
2766     int err;
2767 
2768     mutex_lock(&uld_mutex);
2769     err = setup_sge_queues(adap);
2770     if (err)
2771         goto rel_lock;
2772     err = setup_rss(adap);
2773     if (err)
2774         goto freeq;
2775 
2776     if (adap->flags & CXGB4_USING_MSIX) {
2777         if (s->nd_msix_idx < 0) {
2778             err = -ENOMEM;
2779             goto irq_err;
2780         }
2781 
2782         err = request_irq(adap->msix_info[s->nd_msix_idx].vec,
2783                   t4_nondata_intr, 0,
2784                   adap->msix_info[s->nd_msix_idx].desc, adap);
2785         if (err)
2786             goto irq_err;
2787 
2788         err = request_msix_queue_irqs(adap);
2789         if (err)
2790             goto irq_err_free_nd_msix;
2791     } else {
2792         err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2793                   (adap->flags & CXGB4_USING_MSI) ? 0
2794                                   : IRQF_SHARED,
2795                   adap->port[0]->name, adap);
2796         if (err)
2797             goto irq_err;
2798     }
2799 
2800     enable_rx(adap);
2801     t4_sge_start(adap);
2802     t4_intr_enable(adap);
2803     adap->flags |= CXGB4_FULL_INIT_DONE;
2804     mutex_unlock(&uld_mutex);
2805 
2806     notify_ulds(adap, CXGB4_STATE_UP);
2807 #if IS_ENABLED(CONFIG_IPV6)
2808     update_clip(adap);
2809 #endif
2810     return err;
2811 
2812 irq_err_free_nd_msix:
2813     free_irq(adap->msix_info[s->nd_msix_idx].vec, adap);
2814 irq_err:
2815     dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2816 freeq:
2817     t4_free_sge_resources(adap);
2818 rel_lock:
2819     mutex_unlock(&uld_mutex);
2820     return err;
2821 }
2822 
2823 static void cxgb_down(struct adapter *adapter)
2824 {
2825     cancel_work_sync(&adapter->tid_release_task);
2826     cancel_work_sync(&adapter->db_full_task);
2827     cancel_work_sync(&adapter->db_drop_task);
2828     adapter->tid_release_task_busy = false;
2829     adapter->tid_release_head = NULL;
2830 
2831     t4_sge_stop(adapter);
2832     t4_free_sge_resources(adapter);
2833 
2834     adapter->flags &= ~CXGB4_FULL_INIT_DONE;
2835 }
2836 
2837 /*
2838  * net_device operations
2839  */
2840 static int cxgb_open(struct net_device *dev)
2841 {
2842     struct port_info *pi = netdev_priv(dev);
2843     struct adapter *adapter = pi->adapter;
2844     int err;
2845 
2846     netif_carrier_off(dev);
2847 
2848     if (!(adapter->flags & CXGB4_FULL_INIT_DONE)) {
2849         err = cxgb_up(adapter);
2850         if (err < 0)
2851             return err;
2852     }
2853 
2854     /* It's possible that the basic port information could have
2855      * changed since we first read it.
2856      */
2857     err = t4_update_port_info(pi);
2858     if (err < 0)
2859         return err;
2860 
2861     err = link_start(dev);
2862     if (err)
2863         return err;
2864 
2865     if (pi->nmirrorqsets) {
2866         mutex_lock(&pi->vi_mirror_mutex);
2867         err = cxgb4_port_mirror_alloc_queues(dev);
2868         if (err)
2869             goto out_unlock;
2870 
2871         err = cxgb4_port_mirror_start(dev);
2872         if (err)
2873             goto out_free_queues;
2874         mutex_unlock(&pi->vi_mirror_mutex);
2875     }
2876 
2877     netif_tx_start_all_queues(dev);
2878     return 0;
2879 
2880 out_free_queues:
2881     cxgb4_port_mirror_free_queues(dev);
2882 
2883 out_unlock:
2884     mutex_unlock(&pi->vi_mirror_mutex);
2885     return err;
2886 }
2887 
2888 static int cxgb_close(struct net_device *dev)
2889 {
2890     struct port_info *pi = netdev_priv(dev);
2891     struct adapter *adapter = pi->adapter;
2892     int ret;
2893 
2894     netif_tx_stop_all_queues(dev);
2895     netif_carrier_off(dev);
2896     ret = t4_enable_pi_params(adapter, adapter->pf, pi,
2897                   false, false, false);
2898 #ifdef CONFIG_CHELSIO_T4_DCB
2899     cxgb4_dcb_reset(dev);
2900     dcb_tx_queue_prio_enable(dev, false);
2901 #endif
2902     if (ret)
2903         return ret;
2904 
2905     if (pi->nmirrorqsets) {
2906         mutex_lock(&pi->vi_mirror_mutex);
2907         cxgb4_port_mirror_stop(dev);
2908         cxgb4_port_mirror_free_queues(dev);
2909         mutex_unlock(&pi->vi_mirror_mutex);
2910     }
2911 
2912     return 0;
2913 }
2914 
2915 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2916         __be32 sip, __be16 sport, __be16 vlan,
2917         unsigned int queue, unsigned char port, unsigned char mask)
2918 {
2919     int ret;
2920     struct filter_entry *f;
2921     struct adapter *adap;
2922     int i;
2923     u8 *val;
2924 
2925     adap = netdev2adap(dev);
2926 
2927     /* Adjust stid to correct filter index */
2928     stid -= adap->tids.sftid_base;
2929     stid += adap->tids.nftids;
2930 
2931     /* Check to make sure the filter requested is writable ...
2932      */
2933     f = &adap->tids.ftid_tab[stid];
2934     ret = writable_filter(f);
2935     if (ret)
2936         return ret;
2937 
2938     /* Clear out any old resources being used by the filter before
2939      * we start constructing the new filter.
2940      */
2941     if (f->valid)
2942         clear_filter(adap, f);
2943 
2944     /* Clear out filter specifications */
2945     memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2946     f->fs.val.lport = be16_to_cpu(sport);
2947     f->fs.mask.lport  = ~0;
2948     val = (u8 *)&sip;
2949     if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2950         for (i = 0; i < 4; i++) {
2951             f->fs.val.lip[i] = val[i];
2952             f->fs.mask.lip[i] = ~0;
2953         }
2954         if (adap->params.tp.vlan_pri_map & PORT_F) {
2955             f->fs.val.iport = port;
2956             f->fs.mask.iport = mask;
2957         }
2958     }
2959 
2960     if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2961         f->fs.val.proto = IPPROTO_TCP;
2962         f->fs.mask.proto = ~0;
2963     }
2964 
2965     f->fs.dirsteer = 1;
2966     f->fs.iq = queue;
2967     /* Mark filter as locked */
2968     f->locked = 1;
2969     f->fs.rpttid = 1;
2970 
2971     /* Save the actual tid. We need this to get the corresponding
2972      * filter entry structure in filter_rpl.
2973      */
2974     f->tid = stid + adap->tids.ftid_base;
2975     ret = set_filter_wr(adap, stid);
2976     if (ret) {
2977         clear_filter(adap, f);
2978         return ret;
2979     }
2980 
2981     return 0;
2982 }
2983 EXPORT_SYMBOL(cxgb4_create_server_filter);
2984 
2985 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2986         unsigned int queue, bool ipv6)
2987 {
2988     struct filter_entry *f;
2989     struct adapter *adap;
2990 
2991     adap = netdev2adap(dev);
2992 
2993     /* Adjust stid to correct filter index */
2994     stid -= adap->tids.sftid_base;
2995     stid += adap->tids.nftids;
2996 
2997     f = &adap->tids.ftid_tab[stid];
2998     /* Unlock the filter */
2999     f->locked = 0;
3000 
3001     return delete_filter(adap, stid);
3002 }
3003 EXPORT_SYMBOL(cxgb4_remove_server_filter);
3004 
3005 static void cxgb_get_stats(struct net_device *dev,
3006                struct rtnl_link_stats64 *ns)
3007 {
3008     struct port_stats stats;
3009     struct port_info *p = netdev_priv(dev);
3010     struct adapter *adapter = p->adapter;
3011 
3012     /* Block retrieving statistics during EEH error
3013      * recovery. Otherwise, the recovery might fail
3014      * and the PCI device will be removed permanently
3015      */
3016     spin_lock(&adapter->stats_lock);
3017     if (!netif_device_present(dev)) {
3018         spin_unlock(&adapter->stats_lock);
3019         return;
3020     }
3021     t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
3022                  &p->stats_base);
3023     spin_unlock(&adapter->stats_lock);
3024 
3025     ns->tx_bytes   = stats.tx_octets;
3026     ns->tx_packets = stats.tx_frames;
3027     ns->rx_bytes   = stats.rx_octets;
3028     ns->rx_packets = stats.rx_frames;
3029     ns->multicast  = stats.rx_mcast_frames;
3030 
3031     /* detailed rx_errors */
3032     ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
3033                    stats.rx_runt;
3034     ns->rx_over_errors   = 0;
3035     ns->rx_crc_errors    = stats.rx_fcs_err;
3036     ns->rx_frame_errors  = stats.rx_symbol_err;
3037     ns->rx_dropped       = stats.rx_ovflow0 + stats.rx_ovflow1 +
3038                    stats.rx_ovflow2 + stats.rx_ovflow3 +
3039                    stats.rx_trunc0 + stats.rx_trunc1 +
3040                    stats.rx_trunc2 + stats.rx_trunc3;
3041     ns->rx_missed_errors = 0;
3042 
3043     /* detailed tx_errors */
3044     ns->tx_aborted_errors   = 0;
3045     ns->tx_carrier_errors   = 0;
3046     ns->tx_fifo_errors      = 0;
3047     ns->tx_heartbeat_errors = 0;
3048     ns->tx_window_errors    = 0;
3049 
3050     ns->tx_errors = stats.tx_error_frames;
3051     ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
3052         ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
3053 }
3054 
3055 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
3056 {
3057     unsigned int mbox;
3058     int ret = 0, prtad, devad;
3059     struct port_info *pi = netdev_priv(dev);
3060     struct adapter *adapter = pi->adapter;
3061     struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
3062 
3063     switch (cmd) {
3064     case SIOCGMIIPHY:
3065         if (pi->mdio_addr < 0)
3066             return -EOPNOTSUPP;
3067         data->phy_id = pi->mdio_addr;
3068         break;
3069     case SIOCGMIIREG:
3070     case SIOCSMIIREG:
3071         if (mdio_phy_id_is_c45(data->phy_id)) {
3072             prtad = mdio_phy_id_prtad(data->phy_id);
3073             devad = mdio_phy_id_devad(data->phy_id);
3074         } else if (data->phy_id < 32) {
3075             prtad = data->phy_id;
3076             devad = 0;
3077             data->reg_num &= 0x1f;
3078         } else
3079             return -EINVAL;
3080 
3081         mbox = pi->adapter->pf;
3082         if (cmd == SIOCGMIIREG)
3083             ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
3084                      data->reg_num, &data->val_out);
3085         else
3086             ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
3087                      data->reg_num, data->val_in);
3088         break;
3089     case SIOCGHWTSTAMP:
3090         return copy_to_user(req->ifr_data, &pi->tstamp_config,
3091                     sizeof(pi->tstamp_config)) ?
3092             -EFAULT : 0;
3093     case SIOCSHWTSTAMP:
3094         if (copy_from_user(&pi->tstamp_config, req->ifr_data,
3095                    sizeof(pi->tstamp_config)))
3096             return -EFAULT;
3097 
3098         if (!is_t4(adapter->params.chip)) {
3099             switch (pi->tstamp_config.tx_type) {
3100             case HWTSTAMP_TX_OFF:
3101             case HWTSTAMP_TX_ON:
3102                 break;
3103             default:
3104                 return -ERANGE;
3105             }
3106 
3107             switch (pi->tstamp_config.rx_filter) {
3108             case HWTSTAMP_FILTER_NONE:
3109                 pi->rxtstamp = false;
3110                 break;
3111             case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3112             case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3113                 cxgb4_ptprx_timestamping(pi, pi->port_id,
3114                              PTP_TS_L4);
3115                 break;
3116             case HWTSTAMP_FILTER_PTP_V2_EVENT:
3117                 cxgb4_ptprx_timestamping(pi, pi->port_id,
3118                              PTP_TS_L2_L4);
3119                 break;
3120             case HWTSTAMP_FILTER_ALL:
3121             case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3122             case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3123             case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3124             case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3125                 pi->rxtstamp = true;
3126                 break;
3127             default:
3128                 pi->tstamp_config.rx_filter =
3129                     HWTSTAMP_FILTER_NONE;
3130                 return -ERANGE;
3131             }
3132 
3133             if ((pi->tstamp_config.tx_type == HWTSTAMP_TX_OFF) &&
3134                 (pi->tstamp_config.rx_filter ==
3135                 HWTSTAMP_FILTER_NONE)) {
3136                 if (cxgb4_ptp_txtype(adapter, pi->port_id) >= 0)
3137                     pi->ptp_enable = false;
3138             }
3139 
3140             if (pi->tstamp_config.rx_filter !=
3141                 HWTSTAMP_FILTER_NONE) {
3142                 if (cxgb4_ptp_redirect_rx_packet(adapter,
3143                                  pi) >= 0)
3144                     pi->ptp_enable = true;
3145             }
3146         } else {
3147             /* For T4 Adapters */
3148             switch (pi->tstamp_config.rx_filter) {
3149             case HWTSTAMP_FILTER_NONE:
3150             pi->rxtstamp = false;
3151             break;
3152             case HWTSTAMP_FILTER_ALL:
3153             pi->rxtstamp = true;
3154             break;
3155             default:
3156             pi->tstamp_config.rx_filter =
3157             HWTSTAMP_FILTER_NONE;
3158             return -ERANGE;
3159             }
3160         }
3161         return copy_to_user(req->ifr_data, &pi->tstamp_config,
3162                     sizeof(pi->tstamp_config)) ?
3163             -EFAULT : 0;
3164     default:
3165         return -EOPNOTSUPP;
3166     }
3167     return ret;
3168 }
3169 
3170 static void cxgb_set_rxmode(struct net_device *dev)
3171 {
3172     /* unfortunately we can't return errors to the stack */
3173     set_rxmode(dev, -1, false);
3174 }
3175 
3176 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
3177 {
3178     struct port_info *pi = netdev_priv(dev);
3179     int ret;
3180 
3181     ret = t4_set_rxmode(pi->adapter, pi->adapter->mbox, pi->viid,
3182                 pi->viid_mirror, new_mtu, -1, -1, -1, -1, true);
3183     if (!ret)
3184         dev->mtu = new_mtu;
3185     return ret;
3186 }
3187 
3188 #ifdef CONFIG_PCI_IOV
3189 static int cxgb4_mgmt_open(struct net_device *dev)
3190 {
3191     /* Turn carrier off since we don't have to transmit anything on this
3192      * interface.
3193      */
3194     netif_carrier_off(dev);
3195     return 0;
3196 }
3197 
3198 /* Fill MAC address that will be assigned by the FW */
3199 static void cxgb4_mgmt_fill_vf_station_mac_addr(struct adapter *adap)
3200 {
3201     u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
3202     unsigned int i, vf, nvfs;
3203     u16 a, b;
3204     int err;
3205     u8 *na;
3206 
3207     err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
3208     if (err)
3209         return;
3210 
3211     na = adap->params.vpd.na;
3212     for (i = 0; i < ETH_ALEN; i++)
3213         hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
3214                   hex2val(na[2 * i + 1]));
3215 
3216     a = (hw_addr[0] << 8) | hw_addr[1];
3217     b = (hw_addr[1] << 8) | hw_addr[2];
3218     a ^= b;
3219     a |= 0x0200;    /* locally assigned Ethernet MAC address */
3220     a &= ~0x0100;   /* not a multicast Ethernet MAC address */
3221     macaddr[0] = a >> 8;
3222     macaddr[1] = a & 0xff;
3223 
3224     for (i = 2; i < 5; i++)
3225         macaddr[i] = hw_addr[i + 1];
3226 
3227     for (vf = 0, nvfs = pci_sriov_get_totalvfs(adap->pdev);
3228         vf < nvfs; vf++) {
3229         macaddr[5] = adap->pf * nvfs + vf;
3230         ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, macaddr);
3231     }
3232 }
3233 
3234 static int cxgb4_mgmt_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
3235 {
3236     struct port_info *pi = netdev_priv(dev);
3237     struct adapter *adap = pi->adapter;
3238     int ret;
3239 
3240     /* verify MAC addr is valid */
3241     if (!is_valid_ether_addr(mac)) {
3242         dev_err(pi->adapter->pdev_dev,
3243             "Invalid Ethernet address %pM for VF %d\n",
3244             mac, vf);
3245         return -EINVAL;
3246     }
3247 
3248     dev_info(pi->adapter->pdev_dev,
3249          "Setting MAC %pM on VF %d\n", mac, vf);
3250     ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
3251     if (!ret)
3252         ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
3253     return ret;
3254 }
3255 
3256 static int cxgb4_mgmt_get_vf_config(struct net_device *dev,
3257                     int vf, struct ifla_vf_info *ivi)
3258 {
3259     struct port_info *pi = netdev_priv(dev);
3260     struct adapter *adap = pi->adapter;
3261     struct vf_info *vfinfo;
3262 
3263     if (vf >= adap->num_vfs)
3264         return -EINVAL;
3265     vfinfo = &adap->vfinfo[vf];
3266 
3267     ivi->vf = vf;
3268     ivi->max_tx_rate = vfinfo->tx_rate;
3269     ivi->min_tx_rate = 0;
3270     ether_addr_copy(ivi->mac, vfinfo->vf_mac_addr);
3271     ivi->vlan = vfinfo->vlan;
3272     ivi->linkstate = vfinfo->link_state;
3273     return 0;
3274 }
3275 
3276 static int cxgb4_mgmt_get_phys_port_id(struct net_device *dev,
3277                        struct netdev_phys_item_id *ppid)
3278 {
3279     struct port_info *pi = netdev_priv(dev);
3280     unsigned int phy_port_id;
3281 
3282     phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
3283     ppid->id_len = sizeof(phy_port_id);
3284     memcpy(ppid->id, &phy_port_id, ppid->id_len);
3285     return 0;
3286 }
3287 
3288 static int cxgb4_mgmt_set_vf_rate(struct net_device *dev, int vf,
3289                   int min_tx_rate, int max_tx_rate)
3290 {
3291     struct port_info *pi = netdev_priv(dev);
3292     struct adapter *adap = pi->adapter;
3293     unsigned int link_ok, speed, mtu;
3294     u32 fw_pfvf, fw_class;
3295     int class_id = vf;
3296     int ret;
3297     u16 pktsize;
3298 
3299     if (vf >= adap->num_vfs)
3300         return -EINVAL;
3301 
3302     if (min_tx_rate) {
3303         dev_err(adap->pdev_dev,
3304             "Min tx rate (%d) (> 0) for VF %d is Invalid.\n",
3305             min_tx_rate, vf);
3306         return -EINVAL;
3307     }
3308 
3309     if (max_tx_rate == 0) {
3310         /* unbind VF to to any Traffic Class */
3311         fw_pfvf =
3312             (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
3313              FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
3314         fw_class = 0xffffffff;
3315         ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1,
3316                     &fw_pfvf, &fw_class);
3317         if (ret) {
3318             dev_err(adap->pdev_dev,
3319                 "Err %d in unbinding PF %d VF %d from TX Rate Limiting\n",
3320                 ret, adap->pf, vf);
3321             return -EINVAL;
3322         }
3323         dev_info(adap->pdev_dev,
3324              "PF %d VF %d is unbound from TX Rate Limiting\n",
3325              adap->pf, vf);
3326         adap->vfinfo[vf].tx_rate = 0;
3327         return 0;
3328     }
3329 
3330     ret = t4_get_link_params(pi, &link_ok, &speed, &mtu);
3331     if (ret != FW_SUCCESS) {
3332         dev_err(adap->pdev_dev,
3333             "Failed to get link information for VF %d\n", vf);
3334         return -EINVAL;
3335     }
3336 
3337     if (!link_ok) {
3338         dev_err(adap->pdev_dev, "Link down for VF %d\n", vf);
3339         return -EINVAL;
3340     }
3341 
3342     if (max_tx_rate > speed) {
3343         dev_err(adap->pdev_dev,
3344             "Max tx rate %d for VF %d can't be > link-speed %u",
3345             max_tx_rate, vf, speed);
3346         return -EINVAL;
3347     }
3348 
3349     pktsize = mtu;
3350     /* subtract ethhdr size and 4 bytes crc since, f/w appends it */
3351     pktsize = pktsize - sizeof(struct ethhdr) - 4;
3352     /* subtract ipv4 hdr size, tcp hdr size to get typical IPv4 MSS size */
3353     pktsize = pktsize - sizeof(struct iphdr) - sizeof(struct tcphdr);
3354     /* configure Traffic Class for rate-limiting */
3355     ret = t4_sched_params(adap, SCHED_CLASS_TYPE_PACKET,
3356                   SCHED_CLASS_LEVEL_CL_RL,
3357                   SCHED_CLASS_MODE_CLASS,
3358                   SCHED_CLASS_RATEUNIT_BITS,
3359                   SCHED_CLASS_RATEMODE_ABS,
3360                   pi->tx_chan, class_id, 0,
3361                   max_tx_rate * 1000, 0, pktsize, 0);
3362     if (ret) {
3363         dev_err(adap->pdev_dev, "Err %d for Traffic Class config\n",
3364             ret);
3365         return -EINVAL;
3366     }
3367     dev_info(adap->pdev_dev,
3368          "Class %d with MSS %u configured with rate %u\n",
3369          class_id, pktsize, max_tx_rate);
3370 
3371     /* bind VF to configured Traffic Class */
3372     fw_pfvf = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
3373            FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
3374     fw_class = class_id;
3375     ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1, &fw_pfvf,
3376                 &fw_class);
3377     if (ret) {
3378         dev_err(adap->pdev_dev,
3379             "Err %d in binding PF %d VF %d to Traffic Class %d\n",
3380             ret, adap->pf, vf, class_id);
3381         return -EINVAL;
3382     }
3383     dev_info(adap->pdev_dev, "PF %d VF %d is bound to Class %d\n",
3384          adap->pf, vf, class_id);
3385     adap->vfinfo[vf].tx_rate = max_tx_rate;
3386     return 0;
3387 }
3388 
3389 static int cxgb4_mgmt_set_vf_vlan(struct net_device *dev, int vf,
3390                   u16 vlan, u8 qos, __be16 vlan_proto)
3391 {
3392     struct port_info *pi = netdev_priv(dev);
3393     struct adapter *adap = pi->adapter;
3394     int ret;
3395 
3396     if (vf >= adap->num_vfs || vlan > 4095 || qos > 7)
3397         return -EINVAL;
3398 
3399     if (vlan_proto != htons(ETH_P_8021Q) || qos != 0)
3400         return -EPROTONOSUPPORT;
3401 
3402     ret = t4_set_vlan_acl(adap, adap->mbox, vf + 1, vlan);
3403     if (!ret) {
3404         adap->vfinfo[vf].vlan = vlan;
3405         return 0;
3406     }
3407 
3408     dev_err(adap->pdev_dev, "Err %d %s VLAN ACL for PF/VF %d/%d\n",
3409         ret, (vlan ? "setting" : "clearing"), adap->pf, vf);
3410     return ret;
3411 }
3412 
3413 static int cxgb4_mgmt_set_vf_link_state(struct net_device *dev, int vf,
3414                     int link)
3415 {
3416     struct port_info *pi = netdev_priv(dev);
3417     struct adapter *adap = pi->adapter;
3418     u32 param, val;
3419     int ret = 0;
3420 
3421     if (vf >= adap->num_vfs)
3422         return -EINVAL;
3423 
3424     switch (link) {
3425     case IFLA_VF_LINK_STATE_AUTO:
3426         val = FW_VF_LINK_STATE_AUTO;
3427         break;
3428 
3429     case IFLA_VF_LINK_STATE_ENABLE:
3430         val = FW_VF_LINK_STATE_ENABLE;
3431         break;
3432 
3433     case IFLA_VF_LINK_STATE_DISABLE:
3434         val = FW_VF_LINK_STATE_DISABLE;
3435         break;
3436 
3437     default:
3438         return -EINVAL;
3439     }
3440 
3441     param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
3442          FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_LINK_STATE));
3443     ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1,
3444                 &param, &val);
3445     if (ret) {
3446         dev_err(adap->pdev_dev,
3447             "Error %d in setting PF %d VF %d link state\n",
3448             ret, adap->pf, vf);
3449         return -EINVAL;
3450     }
3451 
3452     adap->vfinfo[vf].link_state = link;
3453     return ret;
3454 }
3455 #endif /* CONFIG_PCI_IOV */
3456 
3457 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
3458 {
3459     int ret;
3460     struct sockaddr *addr = p;
3461     struct port_info *pi = netdev_priv(dev);
3462 
3463     if (!is_valid_ether_addr(addr->sa_data))
3464         return -EADDRNOTAVAIL;
3465 
3466     ret = cxgb4_update_mac_filt(pi, pi->viid, &pi->xact_addr_filt,
3467                     addr->sa_data, true, &pi->smt_idx);
3468     if (ret < 0)
3469         return ret;
3470 
3471     eth_hw_addr_set(dev, addr->sa_data);
3472     return 0;
3473 }
3474 
3475 #ifdef CONFIG_NET_POLL_CONTROLLER
3476 static void cxgb_netpoll(struct net_device *dev)
3477 {
3478     struct port_info *pi = netdev_priv(dev);
3479     struct adapter *adap = pi->adapter;
3480 
3481     if (adap->flags & CXGB4_USING_MSIX) {
3482         int i;
3483         struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
3484 
3485         for (i = pi->nqsets; i; i--, rx++)
3486             t4_sge_intr_msix(0, &rx->rspq);
3487     } else
3488         t4_intr_handler(adap)(0, adap);
3489 }
3490 #endif
3491 
3492 static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
3493 {
3494     struct port_info *pi = netdev_priv(dev);
3495     struct adapter *adap = pi->adapter;
3496     struct ch_sched_queue qe = { 0 };
3497     struct ch_sched_params p = { 0 };
3498     struct sched_class *e;
3499     u32 req_rate;
3500     int err = 0;
3501 
3502     if (!can_sched(dev))
3503         return -ENOTSUPP;
3504 
3505     if (index < 0 || index > pi->nqsets - 1)
3506         return -EINVAL;
3507 
3508     if (!(adap->flags & CXGB4_FULL_INIT_DONE)) {
3509         dev_err(adap->pdev_dev,
3510             "Failed to rate limit on queue %d. Link Down?\n",
3511             index);
3512         return -EINVAL;
3513     }
3514 
3515     qe.queue = index;
3516     e = cxgb4_sched_queue_lookup(dev, &qe);
3517     if (e && e->info.u.params.level != SCHED_CLASS_LEVEL_CL_RL) {
3518         dev_err(adap->pdev_dev,
3519             "Queue %u already bound to class %u of type: %u\n",
3520             index, e->idx, e->info.u.params.level);
3521         return -EBUSY;
3522     }
3523 
3524     /* Convert from Mbps to Kbps */
3525     req_rate = rate * 1000;
3526 
3527     /* Max rate is 100 Gbps */
3528     if (req_rate > SCHED_MAX_RATE_KBPS) {
3529         dev_err(adap->pdev_dev,
3530             "Invalid rate %u Mbps, Max rate is %u Mbps\n",
3531             rate, SCHED_MAX_RATE_KBPS / 1000);
3532         return -ERANGE;
3533     }
3534 
3535     /* First unbind the queue from any existing class */
3536     memset(&qe, 0, sizeof(qe));
3537     qe.queue = index;
3538     qe.class = SCHED_CLS_NONE;
3539 
3540     err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
3541     if (err) {
3542         dev_err(adap->pdev_dev,
3543             "Unbinding Queue %d on port %d fail. Err: %d\n",
3544             index, pi->port_id, err);
3545         return err;
3546     }
3547 
3548     /* Queue already unbound */
3549     if (!req_rate)
3550         return 0;
3551 
3552     /* Fetch any available unused or matching scheduling class */
3553     p.type = SCHED_CLASS_TYPE_PACKET;
3554     p.u.params.level    = SCHED_CLASS_LEVEL_CL_RL;
3555     p.u.params.mode     = SCHED_CLASS_MODE_CLASS;
3556     p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
3557     p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
3558     p.u.params.channel  = pi->tx_chan;
3559     p.u.params.class    = SCHED_CLS_NONE;
3560     p.u.params.minrate  = 0;
3561     p.u.params.maxrate  = req_rate;
3562     p.u.params.weight   = 0;
3563     p.u.params.pktsize  = dev->mtu;
3564 
3565     e = cxgb4_sched_class_alloc(dev, &p);
3566     if (!e)
3567         return -ENOMEM;
3568 
3569     /* Bind the queue to a scheduling class */
3570     memset(&qe, 0, sizeof(qe));
3571     qe.queue = index;
3572     qe.class = e->idx;
3573 
3574     err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
3575     if (err)
3576         dev_err(adap->pdev_dev,
3577             "Queue rate limiting failed. Err: %d\n", err);
3578     return err;
3579 }
3580 
3581 static int cxgb_setup_tc_flower(struct net_device *dev,
3582                 struct flow_cls_offload *cls_flower)
3583 {
3584     switch (cls_flower->command) {
3585     case FLOW_CLS_REPLACE:
3586         return cxgb4_tc_flower_replace(dev, cls_flower);
3587     case FLOW_CLS_DESTROY:
3588         return cxgb4_tc_flower_destroy(dev, cls_flower);
3589     case FLOW_CLS_STATS:
3590         return cxgb4_tc_flower_stats(dev, cls_flower);
3591     default:
3592         return -EOPNOTSUPP;
3593     }
3594 }
3595 
3596 static int cxgb_setup_tc_cls_u32(struct net_device *dev,
3597                  struct tc_cls_u32_offload *cls_u32)
3598 {
3599     switch (cls_u32->command) {
3600     case TC_CLSU32_NEW_KNODE:
3601     case TC_CLSU32_REPLACE_KNODE:
3602         return cxgb4_config_knode(dev, cls_u32);
3603     case TC_CLSU32_DELETE_KNODE:
3604         return cxgb4_delete_knode(dev, cls_u32);
3605     default:
3606         return -EOPNOTSUPP;
3607     }
3608 }
3609 
3610 static int cxgb_setup_tc_matchall(struct net_device *dev,
3611                   struct tc_cls_matchall_offload *cls_matchall,
3612                   bool ingress)
3613 {
3614     struct adapter *adap = netdev2adap(dev);
3615 
3616     if (!adap->tc_matchall)
3617         return -ENOMEM;
3618 
3619     switch (cls_matchall->command) {
3620     case TC_CLSMATCHALL_REPLACE:
3621         return cxgb4_tc_matchall_replace(dev, cls_matchall, ingress);
3622     case TC_CLSMATCHALL_DESTROY:
3623         return cxgb4_tc_matchall_destroy(dev, cls_matchall, ingress);
3624     case TC_CLSMATCHALL_STATS:
3625         if (ingress)
3626             return cxgb4_tc_matchall_stats(dev, cls_matchall);
3627         break;
3628     default:
3629         break;
3630     }
3631 
3632     return -EOPNOTSUPP;
3633 }
3634 
3635 static int cxgb_setup_tc_block_ingress_cb(enum tc_setup_type type,
3636                       void *type_data, void *cb_priv)
3637 {
3638     struct net_device *dev = cb_priv;
3639     struct port_info *pi = netdev2pinfo(dev);
3640     struct adapter *adap = netdev2adap(dev);
3641 
3642     if (!(adap->flags & CXGB4_FULL_INIT_DONE)) {
3643         dev_err(adap->pdev_dev,
3644             "Failed to setup tc on port %d. Link Down?\n",
3645             pi->port_id);
3646         return -EINVAL;
3647     }
3648 
3649     if (!tc_cls_can_offload_and_chain0(dev, type_data))
3650         return -EOPNOTSUPP;
3651 
3652     switch (type) {
3653     case TC_SETUP_CLSU32:
3654         return cxgb_setup_tc_cls_u32(dev, type_data);
3655     case TC_SETUP_CLSFLOWER:
3656         return cxgb_setup_tc_flower(dev, type_data);
3657     case TC_SETUP_CLSMATCHALL:
3658         return cxgb_setup_tc_matchall(dev, type_data, true);
3659     default:
3660         return -EOPNOTSUPP;
3661     }
3662 }
3663 
3664 static int cxgb_setup_tc_block_egress_cb(enum tc_setup_type type,
3665                      void *type_data, void *cb_priv)
3666 {
3667     struct net_device *dev = cb_priv;
3668     struct port_info *pi = netdev2pinfo(dev);
3669     struct adapter *adap = netdev2adap(dev);
3670 
3671     if (!(adap->flags & CXGB4_FULL_INIT_DONE)) {
3672         dev_err(adap->pdev_dev,
3673             "Failed to setup tc on port %d. Link Down?\n",
3674             pi->port_id);
3675         return -EINVAL;
3676     }
3677 
3678     if (!tc_cls_can_offload_and_chain0(dev, type_data))
3679         return -EOPNOTSUPP;
3680 
3681     switch (type) {
3682     case TC_SETUP_CLSMATCHALL:
3683         return cxgb_setup_tc_matchall(dev, type_data, false);
3684     default:
3685         break;
3686     }
3687 
3688     return -EOPNOTSUPP;
3689 }
3690 
3691 static int cxgb_setup_tc_mqprio(struct net_device *dev,
3692                 struct tc_mqprio_qopt_offload *mqprio)
3693 {
3694     struct adapter *adap = netdev2adap(dev);
3695 
3696     if (!is_ethofld(adap) || !adap->tc_mqprio)
3697         return -ENOMEM;
3698 
3699     return cxgb4_setup_tc_mqprio(dev, mqprio);
3700 }
3701 
3702 static LIST_HEAD(cxgb_block_cb_list);
3703 
3704 static int cxgb_setup_tc_block(struct net_device *dev,
3705                    struct flow_block_offload *f)
3706 {
3707     struct port_info *pi = netdev_priv(dev);
3708     flow_setup_cb_t *cb;
3709     bool ingress_only;
3710 
3711     pi->tc_block_shared = f->block_shared;
3712     if (f->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS) {
3713         cb = cxgb_setup_tc_block_egress_cb;
3714         ingress_only = false;
3715     } else {
3716         cb = cxgb_setup_tc_block_ingress_cb;
3717         ingress_only = true;
3718     }
3719 
3720     return flow_block_cb_setup_simple(f, &cxgb_block_cb_list,
3721                       cb, pi, dev, ingress_only);
3722 }
3723 
3724 static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
3725              void *type_data)
3726 {
3727     switch (type) {
3728     case TC_SETUP_QDISC_MQPRIO:
3729         return cxgb_setup_tc_mqprio(dev, type_data);
3730     case TC_SETUP_BLOCK:
3731         return cxgb_setup_tc_block(dev, type_data);
3732     default:
3733         return -EOPNOTSUPP;
3734     }
3735 }
3736 
3737 static int cxgb_udp_tunnel_unset_port(struct net_device *netdev,
3738                       unsigned int table, unsigned int entry,
3739                       struct udp_tunnel_info *ti)
3740 {
3741     struct port_info *pi = netdev_priv(netdev);
3742     struct adapter *adapter = pi->adapter;
3743     u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3744     int ret = 0, i;
3745 
3746     switch (ti->type) {
3747     case UDP_TUNNEL_TYPE_VXLAN:
3748         adapter->vxlan_port = 0;
3749         t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A, 0);
3750         break;
3751     case UDP_TUNNEL_TYPE_GENEVE:
3752         adapter->geneve_port = 0;
3753         t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A, 0);
3754         break;
3755     default:
3756         return -EINVAL;
3757     }
3758 
3759     /* Matchall mac entries can be deleted only after all tunnel ports
3760      * are brought down or removed.
3761      */
3762     if (!adapter->rawf_cnt)
3763         return 0;
3764     for_each_port(adapter, i) {
3765         pi = adap2pinfo(adapter, i);
3766         ret = t4_free_raw_mac_filt(adapter, pi->viid,
3767                        match_all_mac, match_all_mac,
3768                        adapter->rawf_start + pi->port_id,
3769                        1, pi->port_id, false);
3770         if (ret < 0) {
3771             netdev_info(netdev, "Failed to free mac filter entry, for port %d\n",
3772                     i);
3773             return ret;
3774         }
3775     }
3776 
3777     return 0;
3778 }
3779 
3780 static int cxgb_udp_tunnel_set_port(struct net_device *netdev,
3781                     unsigned int table, unsigned int entry,
3782                     struct udp_tunnel_info *ti)
3783 {
3784     struct port_info *pi = netdev_priv(netdev);
3785     struct adapter *adapter = pi->adapter;
3786     u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3787     int i, ret;
3788 
3789     switch (ti->type) {
3790     case UDP_TUNNEL_TYPE_VXLAN:
3791         adapter->vxlan_port = ti->port;
3792         t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A,
3793                  VXLAN_V(be16_to_cpu(ti->port)) | VXLAN_EN_F);
3794         break;
3795     case UDP_TUNNEL_TYPE_GENEVE:
3796         adapter->geneve_port = ti->port;
3797         t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A,
3798                  GENEVE_V(be16_to_cpu(ti->port)) | GENEVE_EN_F);
3799         break;
3800     default:
3801         return -EINVAL;
3802     }
3803 
3804     /* Create a 'match all' mac filter entry for inner mac,
3805      * if raw mac interface is supported. Once the linux kernel provides
3806      * driver entry points for adding/deleting the inner mac addresses,
3807      * we will remove this 'match all' entry and fallback to adding
3808      * exact match filters.
3809      */
3810     for_each_port(adapter, i) {
3811         pi = adap2pinfo(adapter, i);
3812 
3813         ret = t4_alloc_raw_mac_filt(adapter, pi->viid,
3814                         match_all_mac,
3815                         match_all_mac,
3816                         adapter->rawf_start + pi->port_id,
3817                         1, pi->port_id, false);
3818         if (ret < 0) {
3819             netdev_info(netdev, "Failed to allocate a mac filter entry, not adding port %d\n",
3820                     be16_to_cpu(ti->port));
3821             return ret;
3822         }
3823     }
3824 
3825     return 0;
3826 }
3827 
3828 static const struct udp_tunnel_nic_info cxgb_udp_tunnels = {
3829     .set_port   = cxgb_udp_tunnel_set_port,
3830     .unset_port = cxgb_udp_tunnel_unset_port,
3831     .tables     = {
3832         { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
3833         { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_GENEVE, },
3834     },
3835 };
3836 
3837 static netdev_features_t cxgb_features_check(struct sk_buff *skb,
3838                          struct net_device *dev,
3839                          netdev_features_t features)
3840 {
3841     struct port_info *pi = netdev_priv(dev);
3842     struct adapter *adapter = pi->adapter;
3843 
3844     if (CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
3845         return features;
3846 
3847     /* Check if hw supports offload for this packet */
3848     if (!skb->encapsulation || cxgb_encap_offload_supported(skb))
3849         return features;
3850 
3851     /* Offload is not supported for this encapsulated packet */
3852     return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3853 }
3854 
3855 static netdev_features_t cxgb_fix_features(struct net_device *dev,
3856                        netdev_features_t features)
3857 {
3858     /* Disable GRO, if RX_CSUM is disabled */
3859     if (!(features & NETIF_F_RXCSUM))
3860         features &= ~NETIF_F_GRO;
3861 
3862     return features;
3863 }
3864 
3865 static const struct net_device_ops cxgb4_netdev_ops = {
3866     .ndo_open             = cxgb_open,
3867     .ndo_stop             = cxgb_close,
3868     .ndo_start_xmit       = t4_start_xmit,
3869     .ndo_select_queue     = cxgb_select_queue,
3870     .ndo_get_stats64      = cxgb_get_stats,
3871     .ndo_set_rx_mode      = cxgb_set_rxmode,
3872     .ndo_set_mac_address  = cxgb_set_mac_addr,
3873     .ndo_set_features     = cxgb_set_features,
3874     .ndo_validate_addr    = eth_validate_addr,
3875     .ndo_eth_ioctl         = cxgb_ioctl,
3876     .ndo_change_mtu       = cxgb_change_mtu,
3877 #ifdef CONFIG_NET_POLL_CONTROLLER
3878     .ndo_poll_controller  = cxgb_netpoll,
3879 #endif
3880 #ifdef CONFIG_CHELSIO_T4_FCOE
3881     .ndo_fcoe_enable      = cxgb_fcoe_enable,
3882     .ndo_fcoe_disable     = cxgb_fcoe_disable,
3883 #endif /* CONFIG_CHELSIO_T4_FCOE */
3884     .ndo_set_tx_maxrate   = cxgb_set_tx_maxrate,
3885     .ndo_setup_tc         = cxgb_setup_tc,
3886     .ndo_features_check   = cxgb_features_check,
3887     .ndo_fix_features     = cxgb_fix_features,
3888 };
3889 
3890 #ifdef CONFIG_PCI_IOV
3891 static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
3892     .ndo_open               = cxgb4_mgmt_open,
3893     .ndo_set_vf_mac         = cxgb4_mgmt_set_vf_mac,
3894     .ndo_get_vf_config      = cxgb4_mgmt_get_vf_config,
3895     .ndo_set_vf_rate        = cxgb4_mgmt_set_vf_rate,
3896     .ndo_get_phys_port_id   = cxgb4_mgmt_get_phys_port_id,
3897     .ndo_set_vf_vlan        = cxgb4_mgmt_set_vf_vlan,
3898     .ndo_set_vf_link_state  = cxgb4_mgmt_set_vf_link_state,
3899 };
3900 
3901 static void cxgb4_mgmt_get_drvinfo(struct net_device *dev,
3902                    struct ethtool_drvinfo *info)
3903 {
3904     struct adapter *adapter = netdev2adap(dev);
3905 
3906     strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
3907     strlcpy(info->bus_info, pci_name(adapter->pdev),
3908         sizeof(info->bus_info));
3909 }
3910 
3911 static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
3912     .get_drvinfo       = cxgb4_mgmt_get_drvinfo,
3913 };
3914 #endif
3915 
3916 static void notify_fatal_err(struct work_struct *work)
3917 {
3918     struct adapter *adap;
3919 
3920     adap = container_of(work, struct adapter, fatal_err_notify_task);
3921     notify_ulds(adap, CXGB4_STATE_FATAL_ERROR);
3922 }
3923 
3924 void t4_fatal_err(struct adapter *adap)
3925 {
3926     int port;
3927 
3928     if (pci_channel_offline(adap->pdev))
3929         return;
3930 
3931     /* Disable the SGE since ULDs are going to free resources that
3932      * could be exposed to the adapter.  RDMA MWs for example...
3933      */
3934     t4_shutdown_adapter(adap);
3935     for_each_port(adap, port) {
3936         struct net_device *dev = adap->port[port];
3937 
3938         /* If we get here in very early initialization the network
3939          * devices may not have been set up yet.
3940          */
3941         if (!dev)
3942             continue;
3943 
3944         netif_tx_stop_all_queues(dev);
3945         netif_carrier_off(dev);
3946     }
3947     dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
3948     queue_work(adap->workq, &adap->fatal_err_notify_task);
3949 }
3950 
3951 static void setup_memwin(struct adapter *adap)
3952 {
3953     u32 nic_win_base = t4_get_util_window(adap);
3954 
3955     t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
3956 }
3957 
3958 static void setup_memwin_rdma(struct adapter *adap)
3959 {
3960     if (adap->vres.ocq.size) {
3961         u32 start;
3962         unsigned int sz_kb;
3963 
3964         start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
3965         start &= PCI_BASE_ADDRESS_MEM_MASK;
3966         start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
3967         sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
3968         t4_write_reg(adap,
3969                  PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
3970                  start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
3971         t4_write_reg(adap,
3972                  PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
3973                  adap->vres.ocq.start);
3974         t4_read_reg(adap,
3975                 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
3976     }
3977 }
3978 
3979 /* HMA Definitions */
3980 
3981 /* The maximum number of address that can be send in a single FW cmd */
3982 #define HMA_MAX_ADDR_IN_CMD 5
3983 
3984 #define HMA_PAGE_SIZE       PAGE_SIZE
3985 
3986 #define HMA_MAX_NO_FW_ADDRESS   (16 << 10)  /* FW supports 16K addresses */
3987 
3988 #define HMA_PAGE_ORDER                  \
3989     ((HMA_PAGE_SIZE < HMA_MAX_NO_FW_ADDRESS) ?  \
3990     ilog2(HMA_MAX_NO_FW_ADDRESS / HMA_PAGE_SIZE) : 0)
3991 
3992 /* The minimum and maximum possible HMA sizes that can be specified in the FW
3993  * configuration(in units of MB).
3994  */
3995 #define HMA_MIN_TOTAL_SIZE  1
3996 #define HMA_MAX_TOTAL_SIZE              \
3997     (((HMA_PAGE_SIZE << HMA_PAGE_ORDER) *       \
3998       HMA_MAX_NO_FW_ADDRESS) >> 20)
3999 
4000 static void adap_free_hma_mem(struct adapter *adapter)
4001 {
4002     struct scatterlist *iter;
4003     struct page *page;
4004     int i;
4005 
4006     if (!adapter->hma.sgt)
4007         return;
4008 
4009     if (adapter->hma.flags & HMA_DMA_MAPPED_FLAG) {
4010         dma_unmap_sg(adapter->pdev_dev, adapter->hma.sgt->sgl,
4011                  adapter->hma.sgt->nents, DMA_BIDIRECTIONAL);
4012         adapter->hma.flags &= ~HMA_DMA_MAPPED_FLAG;
4013     }
4014 
4015     for_each_sg(adapter->hma.sgt->sgl, iter,
4016             adapter->hma.sgt->orig_nents, i) {
4017         page = sg_page(iter);
4018         if (page)
4019             __free_pages(page, HMA_PAGE_ORDER);
4020     }
4021 
4022     kfree(adapter->hma.phy_addr);
4023     sg_free_table(adapter->hma.sgt);
4024     kfree(adapter->hma.sgt);
4025     adapter->hma.sgt = NULL;
4026 }
4027 
4028 static int adap_config_hma(struct adapter *adapter)
4029 {
4030     struct scatterlist *sgl, *iter;
4031     struct sg_table *sgt;
4032     struct page *newpage;
4033     unsigned int i, j, k;
4034     u32 param, hma_size;
4035     unsigned int ncmds;
4036     size_t page_size;
4037     u32 page_order;
4038     int node, ret;
4039 
4040     /* HMA is supported only for T6+ cards.
4041      * Avoid initializing HMA in kdump kernels.
4042      */
4043     if (is_kdump_kernel() ||
4044         CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
4045         return 0;
4046 
4047     /* Get the HMA region size required by fw */
4048     param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4049          FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_HMA_SIZE));
4050     ret = t4_query_params(adapter, adapter->mbox, adapter->pf, 0,
4051                   1, &param, &hma_size);
4052     /* An error means card has its own memory or HMA is not supported by
4053      * the firmware. Return without any errors.
4054      */
4055     if (ret || !hma_size)
4056         return 0;
4057 
4058     if (hma_size < HMA_MIN_TOTAL_SIZE ||
4059         hma_size > HMA_MAX_TOTAL_SIZE) {
4060         dev_err(adapter->pdev_dev,
4061             "HMA size %uMB beyond bounds(%u-%lu)MB\n",
4062             hma_size, HMA_MIN_TOTAL_SIZE, HMA_MAX_TOTAL_SIZE);
4063         return -EINVAL;
4064     }
4065 
4066     page_size = HMA_PAGE_SIZE;
4067     page_order = HMA_PAGE_ORDER;
4068     adapter->hma.sgt = kzalloc(sizeof(*adapter->hma.sgt), GFP_KERNEL);
4069     if (unlikely(!adapter->hma.sgt)) {
4070         dev_err(adapter->pdev_dev, "HMA SG table allocation failed\n");
4071         return -ENOMEM;
4072     }
4073     sgt = adapter->hma.sgt;
4074     /* FW returned value will be in MB's
4075      */
4076     sgt->orig_nents = (hma_size << 20) / (page_size << page_order);
4077     if (sg_alloc_table(sgt, sgt->orig_nents, GFP_KERNEL)) {
4078         dev_err(adapter->pdev_dev, "HMA SGL allocation failed\n");
4079         kfree(adapter->hma.sgt);
4080         adapter->hma.sgt = NULL;
4081         return -ENOMEM;
4082     }
4083 
4084     sgl = adapter->hma.sgt->sgl;
4085     node = dev_to_node(adapter->pdev_dev);
4086     for_each_sg(sgl, iter, sgt->orig_nents, i) {
4087         newpage = alloc_pages_node(node, __GFP_NOWARN | GFP_KERNEL |
4088                        __GFP_ZERO, page_order);
4089         if (!newpage) {
4090             dev_err(adapter->pdev_dev,
4091                 "Not enough memory for HMA page allocation\n");
4092             ret = -ENOMEM;
4093             goto free_hma;
4094         }
4095         sg_set_page(iter, newpage, page_size << page_order, 0);
4096     }
4097 
4098     sgt->nents = dma_map_sg(adapter->pdev_dev, sgl, sgt->orig_nents,
4099                 DMA_BIDIRECTIONAL);
4100     if (!sgt->nents) {
4101         dev_err(adapter->pdev_dev,
4102             "Not enough memory for HMA DMA mapping");
4103         ret = -ENOMEM;
4104         goto free_hma;
4105     }
4106     adapter->hma.flags |= HMA_DMA_MAPPED_FLAG;
4107 
4108     adapter->hma.phy_addr = kcalloc(sgt->nents, sizeof(dma_addr_t),
4109                     GFP_KERNEL);
4110     if (unlikely(!adapter->hma.phy_addr))
4111         goto free_hma;
4112 
4113     for_each_sg(sgl, iter, sgt->nents, i) {
4114         newpage = sg_page(iter);
4115         adapter->hma.phy_addr[i] = sg_dma_address(iter);
4116     }
4117 
4118     ncmds = DIV_ROUND_UP(sgt->nents, HMA_MAX_ADDR_IN_CMD);
4119     /* Pass on the addresses to firmware */
4120     for (i = 0, k = 0; i < ncmds; i++, k += HMA_MAX_ADDR_IN_CMD) {
4121         struct fw_hma_cmd hma_cmd;
4122         u8 naddr = HMA_MAX_ADDR_IN_CMD;
4123         u8 soc = 0, eoc = 0;
4124         u8 hma_mode = 1; /* Presently we support only Page table mode */
4125 
4126         soc = (i == 0) ? 1 : 0;
4127         eoc = (i == ncmds - 1) ? 1 : 0;
4128 
4129         /* For last cmd, set naddr corresponding to remaining
4130          * addresses
4131          */
4132         if (i == ncmds - 1) {
4133             naddr = sgt->nents % HMA_MAX_ADDR_IN_CMD;
4134             naddr = naddr ? naddr : HMA_MAX_ADDR_IN_CMD;
4135         }
4136         memset(&hma_cmd, 0, sizeof(hma_cmd));
4137         hma_cmd.op_pkd = htonl(FW_CMD_OP_V(FW_HMA_CMD) |
4138                        FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
4139         hma_cmd.retval_len16 = htonl(FW_LEN16(hma_cmd));
4140 
4141         hma_cmd.mode_to_pcie_params =
4142             htonl(FW_HMA_CMD_MODE_V(hma_mode) |
4143                   FW_HMA_CMD_SOC_V(soc) | FW_HMA_CMD_EOC_V(eoc));
4144 
4145         /* HMA cmd size specified in MB's */
4146         hma_cmd.naddr_size =
4147             htonl(FW_HMA_CMD_SIZE_V(hma_size) |
4148                   FW_HMA_CMD_NADDR_V(naddr));
4149 
4150         /* Total Page size specified in units of 4K */
4151         hma_cmd.addr_size_pkd =
4152             htonl(FW_HMA_CMD_ADDR_SIZE_V
4153                 ((page_size << page_order) >> 12));
4154 
4155         /* Fill the 5 addresses */
4156         for (j = 0; j < naddr; j++) {
4157             hma_cmd.phy_address[j] =
4158                 cpu_to_be64(adapter->hma.phy_addr[j + k]);
4159         }
4160         ret = t4_wr_mbox(adapter, adapter->mbox, &hma_cmd,
4161                  sizeof(hma_cmd), &hma_cmd);
4162         if (ret) {
4163             dev_err(adapter->pdev_dev,
4164                 "HMA FW command failed with err %d\n", ret);
4165             goto free_hma;
4166         }
4167     }
4168 
4169     if (!ret)
4170         dev_info(adapter->pdev_dev,
4171              "Reserved %uMB host memory for HMA\n", hma_size);
4172     return ret;
4173 
4174 free_hma:
4175     adap_free_hma_mem(adapter);
4176     return ret;
4177 }
4178 
4179 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
4180 {
4181     u32 v;
4182     int ret;
4183 
4184     /* Now that we've successfully configured and initialized the adapter
4185      * can ask the Firmware what resources it has provisioned for us.
4186      */
4187     ret = t4_get_pfres(adap);
4188     if (ret) {
4189         dev_err(adap->pdev_dev,
4190             "Unable to retrieve resource provisioning information\n");
4191         return ret;
4192     }
4193 
4194     /* get device capabilities */
4195     memset(c, 0, sizeof(*c));
4196     c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4197                    FW_CMD_REQUEST_F | FW_CMD_READ_F);
4198     c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
4199     ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
4200     if (ret < 0)
4201         return ret;
4202 
4203     c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4204                    FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
4205     ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
4206     if (ret < 0)
4207         return ret;
4208 
4209     ret = t4_config_glbl_rss(adap, adap->pf,
4210                  FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4211                  FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
4212                  FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
4213     if (ret < 0)
4214         return ret;
4215 
4216     ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
4217               MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
4218               FW_CMD_CAP_PF);
4219     if (ret < 0)
4220         return ret;
4221 
4222     t4_sge_init(adap);
4223 
4224     /* tweak some settings */
4225     t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
4226     t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
4227     t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
4228     v = t4_read_reg(adap, TP_PIO_DATA_A);
4229     t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
4230 
4231     /* first 4 Tx modulation queues point to consecutive Tx channels */
4232     adap->params.tp.tx_modq_map = 0xE4;
4233     t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
4234              TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
4235 
4236     /* associate each Tx modulation queue with consecutive Tx channels */
4237     v = 0x84218421;
4238     t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
4239               &v, 1, TP_TX_SCHED_HDR_A);
4240     t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
4241               &v, 1, TP_TX_SCHED_FIFO_A);
4242     t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
4243               &v, 1, TP_TX_SCHED_PCMD_A);
4244 
4245 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
4246     if (is_offload(adap)) {
4247         t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
4248                  TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4249                  TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4250                  TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4251                  TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4252         t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
4253                  TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4254                  TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4255                  TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4256                  TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4257     }
4258 
4259     /* get basic stuff going */
4260     return t4_early_init(adap, adap->pf);
4261 }
4262 
4263 /*
4264  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
4265  */
4266 #define MAX_ATIDS 8192U
4267 
4268 /*
4269  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4270  *
4271  * If the firmware we're dealing with has Configuration File support, then
4272  * we use that to perform all configuration
4273  */
4274 
4275 /*
4276  * Tweak configuration based on module parameters, etc.  Most of these have
4277  * defaults assigned to them by Firmware Configuration Files (if we're using
4278  * them) but need to be explicitly set if we're using hard-coded
4279  * initialization.  But even in the case of using Firmware Configuration
4280  * Files, we'd like to expose the ability to change these via module
4281  * parameters so these are essentially common tweaks/settings for
4282  * Configuration Files and hard-coded initialization ...
4283  */
4284 static int adap_init0_tweaks(struct adapter *adapter)
4285 {
4286     /*
4287      * Fix up various Host-Dependent Parameters like Page Size, Cache
4288      * Line Size, etc.  The firmware default is for a 4KB Page Size and
4289      * 64B Cache Line Size ...
4290      */
4291     t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
4292 
4293     /*
4294      * Process module parameters which affect early initialization.
4295      */
4296     if (rx_dma_offset != 2 && rx_dma_offset != 0) {
4297         dev_err(&adapter->pdev->dev,
4298             "Ignoring illegal rx_dma_offset=%d, using 2\n",
4299             rx_dma_offset);
4300         rx_dma_offset = 2;
4301     }
4302     t4_set_reg_field(adapter, SGE_CONTROL_A,
4303              PKTSHIFT_V(PKTSHIFT_M),
4304              PKTSHIFT_V(rx_dma_offset));
4305 
4306     /*
4307      * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
4308      * adds the pseudo header itself.
4309      */
4310     t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
4311                    CSUM_HAS_PSEUDO_HDR_F, 0);
4312 
4313     return 0;
4314 }
4315 
4316 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
4317  * unto themselves and they contain their own firmware to perform their
4318  * tasks ...
4319  */
4320 static int phy_aq1202_version(const u8 *phy_fw_data,
4321                   size_t phy_fw_size)
4322 {
4323     int offset;
4324 
4325     /* At offset 0x8 you're looking for the primary image's
4326      * starting offset which is 3 Bytes wide
4327      *
4328      * At offset 0xa of the primary image, you look for the offset
4329      * of the DRAM segment which is 3 Bytes wide.
4330      *
4331      * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
4332      * wide
4333      */
4334     #define be16(__p) (((__p)[0] << 8) | (__p)[1])
4335     #define le16(__p) ((__p)[0] | ((__p)[1] << 8))
4336     #define le24(__p) (le16(__p) | ((__p)[2] << 16))
4337 
4338     offset = le24(phy_fw_data + 0x8) << 12;
4339     offset = le24(phy_fw_data + offset + 0xa);
4340     return be16(phy_fw_data + offset + 0x27e);
4341 
4342     #undef be16
4343     #undef le16
4344     #undef le24
4345 }
4346 
4347 static struct info_10gbt_phy_fw {
4348     unsigned int phy_fw_id;     /* PCI Device ID */
4349     char *phy_fw_file;      /* /lib/firmware/ PHY Firmware file */
4350     int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
4351     int phy_flash;          /* Has FLASH for PHY Firmware */
4352 } phy_info_array[] = {
4353     {
4354         PHY_AQ1202_DEVICEID,
4355         PHY_AQ1202_FIRMWARE,
4356         phy_aq1202_version,
4357         1,
4358     },
4359     {
4360         PHY_BCM84834_DEVICEID,
4361         PHY_BCM84834_FIRMWARE,
4362         NULL,
4363         0,
4364     },
4365     { 0, NULL, NULL },
4366 };
4367 
4368 static struct info_10gbt_phy_fw *find_phy_info(int devid)
4369 {
4370     int i;
4371 
4372     for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
4373         if (phy_info_array[i].phy_fw_id == devid)
4374             return &phy_info_array[i];
4375     }
4376     return NULL;
4377 }
4378 
4379 /* Handle updating of chip-external 10Gb/s-BT PHY firmware.  This needs to
4380  * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD.  On error
4381  * we return a negative error number.  If we transfer new firmware we return 1
4382  * (from t4_load_phy_fw()).  If we don't do anything we return 0.
4383  */
4384 static int adap_init0_phy(struct adapter *adap)
4385 {
4386     const struct firmware *phyf;
4387     int ret;
4388     struct info_10gbt_phy_fw *phy_info;
4389 
4390     /* Use the device ID to determine which PHY file to flash.
4391      */
4392     phy_info = find_phy_info(adap->pdev->device);
4393     if (!phy_info) {
4394         dev_warn(adap->pdev_dev,
4395              "No PHY Firmware file found for this PHY\n");
4396         return -EOPNOTSUPP;
4397     }
4398 
4399     /* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
4400      * use that. The adapter firmware provides us with a memory buffer
4401      * where we can load a PHY firmware file from the host if we want to
4402      * override the PHY firmware File in flash.
4403      */
4404     ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
4405                       adap->pdev_dev);
4406     if (ret < 0) {
4407         /* For adapters without FLASH attached to PHY for their
4408          * firmware, it's obviously a fatal error if we can't get the
4409          * firmware to the adapter.  For adapters with PHY firmware
4410          * FLASH storage, it's worth a warning if we can't find the
4411          * PHY Firmware but we'll neuter the error ...
4412          */
4413         dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
4414             "/lib/firmware/%s, error %d\n",
4415             phy_info->phy_fw_file, -ret);
4416         if (phy_info->phy_flash) {
4417             int cur_phy_fw_ver = 0;
4418 
4419             t4_phy_fw_ver(adap, &cur_phy_fw_ver);
4420             dev_warn(adap->pdev_dev, "continuing with, on-adapter "
4421                  "FLASH copy, version %#x\n", cur_phy_fw_ver);
4422             ret = 0;
4423         }
4424 
4425         return ret;
4426     }
4427 
4428     /* Load PHY Firmware onto adapter.
4429      */
4430     ret = t4_load_phy_fw(adap, MEMWIN_NIC, phy_info->phy_fw_version,
4431                  (u8 *)phyf->data, phyf->size);
4432     if (ret < 0)
4433         dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
4434             -ret);
4435     else if (ret > 0) {
4436         int new_phy_fw_ver = 0;
4437 
4438         if (phy_info->phy_fw_version)
4439             new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
4440                                   phyf->size);
4441         dev_info(adap->pdev_dev, "Successfully transferred PHY "
4442              "Firmware /lib/firmware/%s, version %#x\n",
4443              phy_info->phy_fw_file, new_phy_fw_ver);
4444     }
4445 
4446     release_firmware(phyf);
4447 
4448     return ret;
4449 }
4450 
4451 /*
4452  * Attempt to initialize the adapter via a Firmware Configuration File.
4453  */
4454 static int adap_init0_config(struct adapter *adapter, int reset)
4455 {
4456     char *fw_config_file, fw_config_file_path[256];
4457     u32 finiver, finicsum, cfcsum, param, val;
4458     struct fw_caps_config_cmd caps_cmd;
4459     unsigned long mtype = 0, maddr = 0;
4460     const struct firmware *cf;
4461     char *config_name = NULL;
4462     int config_issued = 0;
4463     int ret;
4464 
4465     /*
4466      * Reset device if necessary.
4467      */
4468     if (reset) {
4469         ret = t4_fw_reset(adapter, adapter->mbox,
4470                   PIORSTMODE_F | PIORST_F);
4471         if (ret < 0)
4472             goto bye;
4473     }
4474 
4475     /* If this is a 10Gb/s-BT adapter make sure the chip-external
4476      * 10Gb/s-BT PHYs have up-to-date firmware.  Note that this step needs
4477      * to be performed after any global adapter RESET above since some
4478      * PHYs only have local RAM copies of the PHY firmware.
4479      */
4480     if (is_10gbt_device(adapter->pdev->device)) {
4481         ret = adap_init0_phy(adapter);
4482         if (ret < 0)
4483             goto bye;
4484     }
4485     /*
4486      * If we have a T4 configuration file under /lib/firmware/cxgb4/,
4487      * then use that.  Otherwise, use the configuration file stored
4488      * in the adapter flash ...
4489      */
4490     switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
4491     case CHELSIO_T4:
4492         fw_config_file = FW4_CFNAME;
4493         break;
4494     case CHELSIO_T5:
4495         fw_config_file = FW5_CFNAME;
4496         break;
4497     case CHELSIO_T6:
4498         fw_config_file = FW6_CFNAME;
4499         break;
4500     default:
4501         dev_err(adapter->pdev_dev, "Device %d is not supported\n",
4502                adapter->pdev->device);
4503         ret = -EINVAL;
4504         goto bye;
4505     }
4506 
4507     ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
4508     if (ret < 0) {
4509         config_name = "On FLASH";
4510         mtype = FW_MEMTYPE_CF_FLASH;
4511         maddr = t4_flash_cfg_addr(adapter);
4512     } else {
4513         u32 params[7], val[7];
4514 
4515         sprintf(fw_config_file_path,
4516             "/lib/firmware/%s", fw_config_file);
4517         config_name = fw_config_file_path;
4518 
4519         if (cf->size >= FLASH_CFG_MAX_SIZE)
4520             ret = -ENOMEM;
4521         else {
4522             params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4523                  FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
4524             ret = t4_query_params(adapter, adapter->mbox,
4525                           adapter->pf, 0, 1, params, val);
4526             if (ret == 0) {
4527                 /*
4528                  * For t4_memory_rw() below addresses and
4529                  * sizes have to be in terms of multiples of 4
4530                  * bytes.  So, if the Configuration File isn't
4531                  * a multiple of 4 bytes in length we'll have
4532                  * to write that out separately since we can't
4533                  * guarantee that the bytes following the
4534                  * residual byte in the buffer returned by
4535                  * request_firmware() are zeroed out ...
4536                  */
4537                 size_t resid = cf->size & 0x3;
4538                 size_t size = cf->size & ~0x3;
4539                 __be32 *data = (__be32 *)cf->data;
4540 
4541                 mtype = FW_PARAMS_PARAM_Y_G(val[0]);
4542                 maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
4543 
4544                 spin_lock(&adapter->win0_lock);
4545                 ret = t4_memory_rw(adapter, 0, mtype, maddr,
4546                            size, data, T4_MEMORY_WRITE);
4547                 if (ret == 0 && resid != 0) {
4548                     union {
4549                         __be32 word;
4550                         char buf[4];
4551                     } last;
4552                     int i;
4553 
4554                     last.word = data[size >> 2];
4555                     for (i = resid; i < 4; i++)
4556                         last.buf[i] = 0;
4557                     ret = t4_memory_rw(adapter, 0, mtype,
4558                                maddr + size,
4559                                4, &last.word,
4560                                T4_MEMORY_WRITE);
4561                 }
4562                 spin_unlock(&adapter->win0_lock);
4563             }
4564         }
4565 
4566         release_firmware(cf);
4567         if (ret)
4568             goto bye;
4569     }
4570 
4571     val = 0;
4572 
4573     /* Ofld + Hash filter is supported. Older fw will fail this request and
4574      * it is fine.
4575      */
4576     param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4577          FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_HASHFILTER_WITH_OFLD));
4578     ret = t4_set_params(adapter, adapter->mbox, adapter->pf, 0,
4579                 1, &param, &val);
4580 
4581     /* FW doesn't know about Hash filter + ofld support,
4582      * it's not a problem, don't return an error.
4583      */
4584     if (ret < 0) {
4585         dev_warn(adapter->pdev_dev,
4586              "Hash filter with ofld is not supported by FW\n");
4587     }
4588 
4589     /*
4590      * Issue a Capability Configuration command to the firmware to get it
4591      * to parse the Configuration File.  We don't use t4_fw_config_file()
4592      * because we want the ability to modify various features after we've
4593      * processed the configuration file ...
4594      */
4595     memset(&caps_cmd, 0, sizeof(caps_cmd));
4596     caps_cmd.op_to_write =
4597         htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4598               FW_CMD_REQUEST_F |
4599               FW_CMD_READ_F);
4600     caps_cmd.cfvalid_to_len16 =
4601         htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
4602               FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
4603               FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
4604               FW_LEN16(caps_cmd));
4605     ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4606              &caps_cmd);
4607 
4608     /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
4609      * Configuration File in FLASH), our last gasp effort is to use the
4610      * Firmware Configuration File which is embedded in the firmware.  A
4611      * very few early versions of the firmware didn't have one embedded
4612      * but we can ignore those.
4613      */
4614     if (ret == -ENOENT) {
4615         memset(&caps_cmd, 0, sizeof(caps_cmd));
4616         caps_cmd.op_to_write =
4617             htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4618                     FW_CMD_REQUEST_F |
4619                     FW_CMD_READ_F);
4620         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4621         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
4622                 sizeof(caps_cmd), &caps_cmd);
4623         config_name = "Firmware Default";
4624     }
4625 
4626     config_issued = 1;
4627     if (ret < 0)
4628         goto bye;
4629 
4630     finiver = ntohl(caps_cmd.finiver);
4631     finicsum = ntohl(caps_cmd.finicsum);
4632     cfcsum = ntohl(caps_cmd.cfcsum);
4633     if (finicsum != cfcsum)
4634         dev_warn(adapter->pdev_dev, "Configuration File checksum "\
4635              "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4636              finicsum, cfcsum);
4637 
4638     /*
4639      * And now tell the firmware to use the configuration we just loaded.
4640      */
4641     caps_cmd.op_to_write =
4642         htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4643               FW_CMD_REQUEST_F |
4644               FW_CMD_WRITE_F);
4645     caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4646     ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4647              NULL);
4648     if (ret < 0)
4649         goto bye;
4650 
4651     /*
4652      * Tweak configuration based on system architecture, module
4653      * parameters, etc.
4654      */
4655     ret = adap_init0_tweaks(adapter);
4656     if (ret < 0)
4657         goto bye;
4658 
4659     /* We will proceed even if HMA init fails. */
4660     ret = adap_config_hma(adapter);
4661     if (ret)
4662         dev_err(adapter->pdev_dev,
4663             "HMA configuration failed with error %d\n", ret);
4664 
4665     if (is_t6(adapter->params.chip)) {
4666         adap_config_hpfilter(adapter);
4667         ret = setup_ppod_edram(adapter);
4668         if (!ret)
4669             dev_info(adapter->pdev_dev, "Successfully enabled "
4670                  "ppod edram feature\n");
4671     }
4672 
4673     /*
4674      * And finally tell the firmware to initialize itself using the
4675      * parameters from the Configuration File.
4676      */
4677     ret = t4_fw_initialize(adapter, adapter->mbox);
4678     if (ret < 0)
4679         goto bye;
4680 
4681     /* Emit Firmware Configuration File information and return
4682      * successfully.
4683      */
4684     dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4685          "Configuration File \"%s\", version %#x, computed checksum %#x\n",
4686          config_name, finiver, cfcsum);
4687     return 0;
4688 
4689     /*
4690      * Something bad happened.  Return the error ...  (If the "error"
4691      * is that there's no Configuration File on the adapter we don't
4692      * want to issue a warning since this is fairly common.)
4693      */
4694 bye:
4695     if (config_issued && ret != -ENOENT)
4696         dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
4697              config_name, -ret);
4698     return ret;
4699 }
4700 
4701 static struct fw_info fw_info_array[] = {
4702     {
4703         .chip = CHELSIO_T4,
4704         .fs_name = FW4_CFNAME,
4705         .fw_mod_name = FW4_FNAME,
4706         .fw_hdr = {
4707             .chip = FW_HDR_CHIP_T4,
4708             .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
4709             .intfver_nic = FW_INTFVER(T4, NIC),
4710             .intfver_vnic = FW_INTFVER(T4, VNIC),
4711             .intfver_ri = FW_INTFVER(T4, RI),
4712             .intfver_iscsi = FW_INTFVER(T4, ISCSI),
4713             .intfver_fcoe = FW_INTFVER(T4, FCOE),
4714         },
4715     }, {
4716         .chip = CHELSIO_T5,
4717         .fs_name = FW5_CFNAME,
4718         .fw_mod_name = FW5_FNAME,
4719         .fw_hdr = {
4720             .chip = FW_HDR_CHIP_T5,
4721             .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
4722             .intfver_nic = FW_INTFVER(T5, NIC),
4723             .intfver_vnic = FW_INTFVER(T5, VNIC),
4724             .intfver_ri = FW_INTFVER(T5, RI),
4725             .intfver_iscsi = FW_INTFVER(T5, ISCSI),
4726             .intfver_fcoe = FW_INTFVER(T5, FCOE),
4727         },
4728     }, {
4729         .chip = CHELSIO_T6,
4730         .fs_name = FW6_CFNAME,
4731         .fw_mod_name = FW6_FNAME,
4732         .fw_hdr = {
4733             .chip = FW_HDR_CHIP_T6,
4734             .fw_ver = __cpu_to_be32(FW_VERSION(T6)),
4735             .intfver_nic = FW_INTFVER(T6, NIC),
4736             .intfver_vnic = FW_INTFVER(T6, VNIC),
4737             .intfver_ofld = FW_INTFVER(T6, OFLD),
4738             .intfver_ri = FW_INTFVER(T6, RI),
4739             .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
4740             .intfver_iscsi = FW_INTFVER(T6, ISCSI),
4741             .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
4742             .intfver_fcoe = FW_INTFVER(T6, FCOE),
4743         },
4744     }
4745 
4746 };
4747 
4748 static struct fw_info *find_fw_info(int chip)
4749 {
4750     int i;
4751 
4752     for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
4753         if (fw_info_array[i].chip == chip)
4754             return &fw_info_array[i];
4755     }
4756     return NULL;
4757 }
4758 
4759 /*
4760  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4761  */
4762 static int adap_init0(struct adapter *adap, int vpd_skip)
4763 {
4764     struct fw_caps_config_cmd caps_cmd;
4765     u32 params[7], val[7];
4766     enum dev_state state;
4767     u32 v, port_vec;
4768     int reset = 1;
4769     int ret;
4770 
4771     /* Grab Firmware Device Log parameters as early as possible so we have
4772      * access to it for debugging, etc.
4773      */
4774     ret = t4_init_devlog_params(adap);
4775     if (ret < 0)
4776         return ret;
4777 
4778     /* Contact FW, advertising Master capability */
4779     ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
4780               is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
4781     if (ret < 0) {
4782         dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
4783             ret);
4784         return ret;
4785     }
4786     if (ret == adap->mbox)
4787         adap->flags |= CXGB4_MASTER_PF;
4788 
4789     /*
4790      * If we're the Master PF Driver and the device is uninitialized,
4791      * then let's consider upgrading the firmware ...  (We always want
4792      * to check the firmware version number in order to A. get it for
4793      * later reporting and B. to warn if the currently loaded firmware
4794      * is excessively mismatched relative to the driver.)
4795      */
4796 
4797     t4_get_version_info(adap);
4798     ret = t4_check_fw_version(adap);
4799     /* If firmware is too old (not supported by driver) force an update. */
4800     if (ret)
4801         state = DEV_STATE_UNINIT;
4802     if ((adap->flags & CXGB4_MASTER_PF) && state != DEV_STATE_INIT) {
4803         struct fw_info *fw_info;
4804         struct fw_hdr *card_fw;
4805         const struct firmware *fw;
4806         const u8 *fw_data = NULL;
4807         unsigned int fw_size = 0;
4808 
4809         /* This is the firmware whose headers the driver was compiled
4810          * against
4811          */
4812         fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
4813         if (fw_info == NULL) {
4814             dev_err(adap->pdev_dev,
4815                 "unable to get firmware info for chip %d.\n",
4816                 CHELSIO_CHIP_VERSION(adap->params.chip));
4817             return -EINVAL;
4818         }
4819 
4820         /* allocate memory to read the header of the firmware on the
4821          * card
4822          */
4823         card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
4824         if (!card_fw) {
4825             ret = -ENOMEM;
4826             goto bye;
4827         }
4828 
4829         /* Get FW from from /lib/firmware/ */
4830         ret = request_firmware(&fw, fw_info->fw_mod_name,
4831                        adap->pdev_dev);
4832         if (ret < 0) {
4833             dev_err(adap->pdev_dev,
4834                 "unable to load firmware image %s, error %d\n",
4835                 fw_info->fw_mod_name, ret);
4836         } else {
4837             fw_data = fw->data;
4838             fw_size = fw->size;
4839         }
4840 
4841         /* upgrade FW logic */
4842         ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
4843                  state, &reset);
4844 
4845         /* Cleaning up */
4846         release_firmware(fw);
4847         kvfree(card_fw);
4848 
4849         if (ret < 0)
4850             goto bye;
4851     }
4852 
4853     /* If the firmware is initialized already, emit a simply note to that
4854      * effect. Otherwise, it's time to try initializing the adapter.
4855      */
4856     if (state == DEV_STATE_INIT) {
4857         ret = adap_config_hma(adap);
4858         if (ret)
4859             dev_err(adap->pdev_dev,
4860                 "HMA configuration failed with error %d\n",
4861                 ret);
4862         dev_info(adap->pdev_dev, "Coming up as %s: "\
4863              "Adapter already initialized\n",
4864              adap->flags & CXGB4_MASTER_PF ? "MASTER" : "SLAVE");
4865     } else {
4866         dev_info(adap->pdev_dev, "Coming up as MASTER: "\
4867              "Initializing adapter\n");
4868 
4869         /* Find out whether we're dealing with a version of the
4870          * firmware which has configuration file support.
4871          */
4872         params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4873                  FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
4874         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
4875                       params, val);
4876 
4877         /* If the firmware doesn't support Configuration Files,
4878          * return an error.
4879          */
4880         if (ret < 0) {
4881             dev_err(adap->pdev_dev, "firmware doesn't support "
4882                 "Firmware Configuration Files\n");
4883             goto bye;
4884         }
4885 
4886         /* The firmware provides us with a memory buffer where we can
4887          * load a Configuration File from the host if we want to
4888          * override the Configuration File in flash.
4889          */
4890         ret = adap_init0_config(adap, reset);
4891         if (ret == -ENOENT) {
4892             dev_err(adap->pdev_dev, "no Configuration File "
4893                 "present on adapter.\n");
4894             goto bye;
4895         }
4896         if (ret < 0) {
4897             dev_err(adap->pdev_dev, "could not initialize "
4898                 "adapter, error %d\n", -ret);
4899             goto bye;
4900         }
4901     }
4902 
4903     /* Now that we've successfully configured and initialized the adapter
4904      * (or found it already initialized), we can ask the Firmware what
4905      * resources it has provisioned for us.
4906      */
4907     ret = t4_get_pfres(adap);
4908     if (ret) {
4909         dev_err(adap->pdev_dev,
4910             "Unable to retrieve resource provisioning information\n");
4911         goto bye;
4912     }
4913 
4914     /* Grab VPD parameters.  This should be done after we establish a
4915      * connection to the firmware since some of the VPD parameters
4916      * (notably the Core Clock frequency) are retrieved via requests to
4917      * the firmware.  On the other hand, we need these fairly early on
4918      * so we do this right after getting ahold of the firmware.
4919      *
4920      * We need to do this after initializing the adapter because someone
4921      * could have FLASHed a new VPD which won't be read by the firmware
4922      * until we do the RESET ...
4923      */
4924     if (!vpd_skip) {
4925         ret = t4_get_vpd_params(adap, &adap->params.vpd);
4926         if (ret < 0)
4927             goto bye;
4928     }
4929 
4930     /* Find out what ports are available to us.  Note that we need to do
4931      * this before calling adap_init0_no_config() since it needs nports
4932      * and portvec ...
4933      */
4934     v =
4935         FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4936         FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
4937     ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
4938     if (ret < 0)
4939         goto bye;
4940 
4941     adap->params.nports = hweight32(port_vec);
4942     adap->params.portvec = port_vec;
4943 
4944     /* Give the SGE code a chance to pull in anything that it needs ...
4945      * Note that this must be called after we retrieve our VPD parameters
4946      * in order to know how to convert core ticks to seconds, etc.
4947      */
4948     ret = t4_sge_init(adap);
4949     if (ret < 0)
4950         goto bye;
4951 
4952     /* Grab the SGE Doorbell Queue Timer values.  If successful, that
4953      * indicates that the Firmware and Hardware support this.
4954      */
4955     params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4956             FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_DBQ_TIMERTICK));
4957     ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4958                   1, params, val);
4959 
4960     if (!ret) {
4961         adap->sge.dbqtimer_tick = val[0];
4962         ret = t4_read_sge_dbqtimers(adap,
4963                         ARRAY_SIZE(adap->sge.dbqtimer_val),
4964                         adap->sge.dbqtimer_val);
4965     }
4966 
4967     if (!ret)
4968         adap->flags |= CXGB4_SGE_DBQ_TIMER;
4969 
4970     if (is_bypass_device(adap->pdev->device))
4971         adap->params.bypass = 1;
4972 
4973     /*
4974      * Grab some of our basic fundamental operating parameters.
4975      */
4976     params[0] = FW_PARAM_PFVF(EQ_START);
4977     params[1] = FW_PARAM_PFVF(L2T_START);
4978     params[2] = FW_PARAM_PFVF(L2T_END);
4979     params[3] = FW_PARAM_PFVF(FILTER_START);
4980     params[4] = FW_PARAM_PFVF(FILTER_END);
4981     params[5] = FW_PARAM_PFVF(IQFLINT_START);
4982     ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
4983     if (ret < 0)
4984         goto bye;
4985     adap->sge.egr_start = val[0];
4986     adap->l2t_start = val[1];
4987     adap->l2t_end = val[2];
4988     adap->tids.ftid_base = val[3];
4989     adap->tids.nftids = val[4] - val[3] + 1;
4990     adap->sge.ingr_start = val[5];
4991 
4992     if (CHELSIO_CHIP_VERSION(adap->params.chip) > CHELSIO_T5) {
4993         params[0] = FW_PARAM_PFVF(HPFILTER_START);
4994         params[1] = FW_PARAM_PFVF(HPFILTER_END);
4995         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4996                       params, val);
4997         if (ret < 0)
4998             goto bye;
4999 
5000         adap->tids.hpftid_base = val[0];
5001         adap->tids.nhpftids = val[1] - val[0] + 1;
5002 
5003         /* Read the raw mps entries. In T6, the last 2 tcam entries
5004          * are reserved for raw mac addresses (rawf = 2, one per port).
5005          */
5006         params[0] = FW_PARAM_PFVF(RAWF_START);
5007         params[1] = FW_PARAM_PFVF(RAWF_END);
5008         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
5009                       params, val);
5010         if (ret == 0) {
5011             adap->rawf_start = val[0];
5012             adap->rawf_cnt = val[1] - val[0] + 1;
5013         }
5014 
5015         adap->tids.tid_base =
5016             t4_read_reg(adap, LE_DB_ACTIVE_TABLE_START_INDEX_A);
5017     }
5018 
5019     /* qids (ingress/egress) returned from firmware can be anywhere
5020      * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
5021      * Hence driver needs to allocate memory for this range to
5022      * store the queue info. Get the highest IQFLINT/EQ index returned
5023      * in FW_EQ_*_CMD.alloc command.
5024      */
5025     params[0] = FW_PARAM_PFVF(EQ_END);
5026     params[1] = FW_PARAM_PFVF(IQFLINT_END);
5027     ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
5028     if (ret < 0)
5029         goto bye;
5030     adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
5031     adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
5032 
5033     adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
5034                     sizeof(*adap->sge.egr_map), GFP_KERNEL);
5035     if (!adap->sge.egr_map) {
5036         ret = -ENOMEM;
5037         goto bye;
5038     }
5039 
5040     adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
5041                      sizeof(*adap->sge.ingr_map), GFP_KERNEL);
5042     if (!adap->sge.ingr_map) {
5043         ret = -ENOMEM;
5044         goto bye;
5045     }
5046 
5047     /* Allocate the memory for the vaious egress queue bitmaps
5048      * ie starving_fl, txq_maperr and blocked_fl.
5049      */
5050     adap->sge.starving_fl = bitmap_zalloc(adap->sge.egr_sz, GFP_KERNEL);
5051     if (!adap->sge.starving_fl) {
5052         ret = -ENOMEM;
5053         goto bye;
5054     }
5055 
5056     adap->sge.txq_maperr = bitmap_zalloc(adap->sge.egr_sz, GFP_KERNEL);
5057     if (!adap->sge.txq_maperr) {
5058         ret = -ENOMEM;
5059         goto bye;
5060     }
5061 
5062 #ifdef CONFIG_DEBUG_FS
5063     adap->sge.blocked_fl = bitmap_zalloc(adap->sge.egr_sz, GFP_KERNEL);
5064     if (!adap->sge.blocked_fl) {
5065         ret = -ENOMEM;
5066         goto bye;
5067     }
5068 #endif
5069 
5070     params[0] = FW_PARAM_PFVF(CLIP_START);
5071     params[1] = FW_PARAM_PFVF(CLIP_END);
5072     ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
5073     if (ret < 0)
5074         goto bye;
5075     adap->clipt_start = val[0];
5076     adap->clipt_end = val[1];
5077 
5078     /* Get the supported number of traffic classes */
5079     params[0] = FW_PARAM_DEV(NUM_TM_CLASS);
5080     ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
5081     if (ret < 0) {
5082         /* We couldn't retrieve the number of Traffic Classes
5083          * supported by the hardware/firmware. So we hard
5084          * code it here.
5085          */
5086         adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
5087     } else {
5088         adap->params.nsched_cls = val[0];
5089     }
5090 
5091     /* query params related to active filter region */
5092     params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
5093     params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
5094     ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
5095     /* If Active filter size is set we enable establishing
5096      * offload connection through firmware work request
5097      */
5098     if ((val[0] != val[1]) && (ret >= 0)) {
5099         adap->flags |= CXGB4_FW_OFLD_CONN;
5100         adap->tids.aftid_base = val[0];
5101         adap->tids.aftid_end = val[1];
5102     }
5103 
5104     /* If we're running on newer firmware, let it know that we're
5105      * prepared to deal with encapsulated CPL messages.  Older
5106      * firmware won't understand this and we'll just get
5107      * unencapsulated messages ...
5108      */
5109     params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
5110     val[0] = 1;
5111     (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
5112 
5113     /*
5114      * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
5115      * capability.  Earlier versions of the firmware didn't have the
5116      * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
5117      * permission to use ULPTX MEMWRITE DSGL.
5118      */
5119     if (is_t4(adap->params.chip)) {
5120         adap->params.ulptx_memwrite_dsgl = false;
5121     } else {
5122         params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
5123         ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
5124                       1, params, val);
5125         adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
5126     }
5127 
5128     /* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
5129     params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
5130     ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
5131                   1, params, val);
5132     adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
5133 
5134     /* See if FW supports FW_FILTER2 work request */
5135     if (is_t4(adap->params.chip)) {
5136         adap->params.filter2_wr_support = false;
5137     } else {
5138         params[0] = FW_PARAM_DEV(FILTER2_WR);
5139         ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
5140                       1, params, val);
5141         adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
5142     }
5143 
5144     /* Check if FW supports returning vin and smt index.
5145      * If this is not supported, driver will interpret
5146      * these values from viid.
5147      */
5148     params[0] = FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
5149     ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
5150                   1, params, val);
5151     adap->params.viid_smt_extn_support = (ret == 0 && val[0] != 0);
5152 
5153     /*
5154      * Get device capabilities so we can determine what resources we need
5155      * to manage.
5156      */
5157     memset(&caps_cmd, 0, sizeof(caps_cmd));
5158     caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
5159                      FW_CMD_REQUEST_F | FW_CMD_READ_F);
5160     caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
5161     ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
5162              &caps_cmd);
5163     if (ret < 0)
5164         goto bye;
5165 
5166     /* hash filter has some mandatory register settings to be tested and for
5167      * that it needs to test whether offload is enabled or not, hence
5168      * checking and setting it here.
5169      */
5170     if (caps_cmd.ofldcaps)
5171         adap->params.offload = 1;
5172 
5173     if (caps_cmd.ofldcaps ||
5174         (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER)) ||
5175         (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_ETHOFLD))) {
5176         /* query offload-related parameters */
5177         params[0] = FW_PARAM_DEV(NTID);
5178         params[1] = FW_PARAM_PFVF(SERVER_START);
5179         params[2] = FW_PARAM_PFVF(SERVER_END);
5180         params[3] = FW_PARAM_PFVF(TDDP_START);
5181         params[4] = FW_PARAM_PFVF(TDDP_END);
5182         params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
5183         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
5184                       params, val);
5185         if (ret < 0)
5186             goto bye;
5187         adap->tids.ntids = val[0];
5188         adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
5189         adap->tids.stid_base = val[1];
5190         adap->tids.nstids = val[2] - val[1] + 1;
5191         /*
5192          * Setup server filter region. Divide the available filter
5193          * region into two parts. Regular filters get 1/3rd and server
5194          * filters get 2/3rd part. This is only enabled if workarond
5195          * path is enabled.
5196          * 1. For regular filters.
5197          * 2. Server filter: This are special filters which are used
5198          * to redirect SYN packets to offload queue.
5199          */
5200         if (adap->flags & CXGB4_FW_OFLD_CONN && !is_bypass(adap)) {
5201             adap->tids.sftid_base = adap->tids.ftid_base +
5202                     DIV_ROUND_UP(adap->tids.nftids, 3);
5203             adap->tids.nsftids = adap->tids.nftids -
5204                      DIV_ROUND_UP(adap->tids.nftids, 3);
5205             adap->tids.nftids = adap->tids.sftid_base -
5206                         adap->tids.ftid_base;
5207         }
5208         adap->vres.ddp.start = val[3];
5209         adap->vres.ddp.size = val[4] - val[3] + 1;
5210         adap->params.ofldq_wr_cred = val[5];
5211 
5212         if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
5213             init_hash_filter(adap);
5214         } else {
5215             adap->num_ofld_uld += 1;
5216         }
5217 
5218         if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_ETHOFLD)) {
5219             params[0] = FW_PARAM_PFVF(ETHOFLD_START);
5220             params[1] = FW_PARAM_PFVF(ETHOFLD_END);
5221             ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
5222                           params, val);
5223             if (!ret) {
5224                 adap->tids.eotid_base = val[0];
5225                 adap->tids.neotids = min_t(u32, MAX_ATIDS,
5226                                val[1] - val[0] + 1);
5227                 adap->params.ethofld = 1;
5228             }
5229         }
5230     }
5231     if (caps_cmd.rdmacaps) {
5232         params[0] = FW_PARAM_PFVF(STAG_START);
5233         params[1] = FW_PARAM_PFVF(STAG_END);
5234         params[2] = FW_PARAM_PFVF(RQ_START);
5235         params[3] = FW_PARAM_PFVF(RQ_END);
5236         params[4] = FW_PARAM_PFVF(PBL_START);
5237         params[5] = FW_PARAM_PFVF(PBL_END);
5238         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
5239                       params, val);
5240         if (ret < 0)
5241             goto bye;
5242         adap->vres.stag.start = val[0];
5243         adap->vres.stag.size = val[1] - val[0] + 1;
5244         adap->vres.rq.start = val[2];
5245         adap->vres.rq.size = val[3] - val[2] + 1;
5246         adap->vres.pbl.start = val[4];
5247         adap->vres.pbl.size = val[5] - val[4] + 1;
5248 
5249         params[0] = FW_PARAM_PFVF(SRQ_START);
5250         params[1] = FW_PARAM_PFVF(SRQ_END);
5251         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
5252                       params, val);
5253         if (!ret) {
5254             adap->vres.srq.start = val[0];
5255             adap->vres.srq.size = val[1] - val[0] + 1;
5256         }
5257         if (adap->vres.srq.size) {
5258             adap->srq = t4_init_srq(adap->vres.srq.size);
5259             if (!adap->srq)
5260                 dev_warn(&adap->pdev->dev, "could not allocate SRQ, continuing\n");
5261         }
5262 
5263         params[0] = FW_PARAM_PFVF(SQRQ_START);
5264         params[1] = FW_PARAM_PFVF(SQRQ_END);
5265         params[2] = FW_PARAM_PFVF(CQ_START);
5266         params[3] = FW_PARAM_PFVF(CQ_END);
5267         params[4] = FW_PARAM_PFVF(OCQ_START);
5268         params[5] = FW_PARAM_PFVF(OCQ_END);
5269         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
5270                       val);
5271         if (ret < 0)
5272             goto bye;
5273         adap->vres.qp.start = val[0];
5274         adap->vres.qp.size = val[1] - val[0] + 1;
5275         adap->vres.cq.start = val[2];
5276         adap->vres.cq.size = val[3] - val[2] + 1;
5277         adap->vres.ocq.start = val[4];
5278         adap->vres.ocq.size = val[5] - val[4] + 1;
5279 
5280         params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
5281         params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
5282         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
5283                       val);
5284         if (ret < 0) {
5285             adap->params.max_ordird_qp = 8;
5286             adap->params.max_ird_adapter = 32 * adap->tids.ntids;
5287             ret = 0;
5288         } else {
5289             adap->params.max_ordird_qp = val[0];
5290             adap->params.max_ird_adapter = val[1];
5291         }
5292         dev_info(adap->pdev_dev,
5293              "max_ordird_qp %d max_ird_adapter %d\n",
5294              adap->params.max_ordird_qp,
5295              adap->params.max_ird_adapter);
5296 
5297         /* Enable write_with_immediate if FW supports it */
5298         params[0] = FW_PARAM_DEV(RDMA_WRITE_WITH_IMM);
5299         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params,
5300                       val);
5301         adap->params.write_w_imm_support = (ret == 0 && val[0] != 0);
5302 
5303         /* Enable write_cmpl if FW supports it */
5304         params[0] = FW_PARAM_DEV(RI_WRITE_CMPL_WR);
5305         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params,
5306                       val);
5307         adap->params.write_cmpl_support = (ret == 0 && val[0] != 0);
5308         adap->num_ofld_uld += 2;
5309     }
5310     if (caps_cmd.iscsicaps) {
5311         params[0] = FW_PARAM_PFVF(ISCSI_START);
5312         params[1] = FW_PARAM_PFVF(ISCSI_END);
5313         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
5314                       params, val);
5315         if (ret < 0)
5316             goto bye;
5317         adap->vres.iscsi.start = val[0];
5318         adap->vres.iscsi.size = val[1] - val[0] + 1;
5319         if (is_t6(adap->params.chip)) {
5320             params[0] = FW_PARAM_PFVF(PPOD_EDRAM_START);
5321             params[1] = FW_PARAM_PFVF(PPOD_EDRAM_END);
5322             ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
5323                           params, val);
5324             if (!ret) {
5325                 adap->vres.ppod_edram.start = val[0];
5326                 adap->vres.ppod_edram.size =
5327                     val[1] - val[0] + 1;
5328 
5329                 dev_info(adap->pdev_dev,
5330                      "ppod edram start 0x%x end 0x%x size 0x%x\n",
5331                      val[0], val[1],
5332                      adap->vres.ppod_edram.size);
5333             }
5334         }
5335         /* LIO target and cxgb4i initiaitor */
5336         adap->num_ofld_uld += 2;
5337     }
5338     if (caps_cmd.cryptocaps) {
5339         if (ntohs(caps_cmd.cryptocaps) &
5340             FW_CAPS_CONFIG_CRYPTO_LOOKASIDE) {
5341             params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
5342             ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
5343                           2, params, val);
5344             if (ret < 0) {
5345                 if (ret != -EINVAL)
5346                     goto bye;
5347             } else {
5348                 adap->vres.ncrypto_fc = val[0];
5349             }
5350             adap->num_ofld_uld += 1;
5351         }
5352         if (ntohs(caps_cmd.cryptocaps) &
5353             FW_CAPS_CONFIG_TLS_INLINE) {
5354             params[0] = FW_PARAM_PFVF(TLS_START);
5355             params[1] = FW_PARAM_PFVF(TLS_END);
5356             ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
5357                           2, params, val);
5358             if (ret < 0)
5359                 goto bye;
5360             adap->vres.key.start = val[0];
5361             adap->vres.key.size = val[1] - val[0] + 1;
5362             adap->num_uld += 1;
5363         }
5364         adap->params.crypto = ntohs(caps_cmd.cryptocaps);
5365     }
5366 
5367     /* The MTU/MSS Table is initialized by now, so load their values.  If
5368      * we're initializing the adapter, then we'll make any modifications
5369      * we want to the MTU/MSS Table and also initialize the congestion
5370      * parameters.
5371      */
5372     t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
5373     if (state != DEV_STATE_INIT) {
5374         int i;
5375 
5376         /* The default MTU Table contains values 1492 and 1500.
5377          * However, for TCP, it's better to have two values which are
5378          * a multiple of 8 +/- 4 bytes apart near this popular MTU.
5379          * This allows us to have a TCP Data Payload which is a
5380          * multiple of 8 regardless of what combination of TCP Options
5381          * are in use (always a multiple of 4 bytes) which is
5382          * important for performance reasons.  For instance, if no
5383          * options are in use, then we have a 20-byte IP header and a
5384          * 20-byte TCP header.  In this case, a 1500-byte MSS would
5385          * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
5386          * which is not a multiple of 8.  So using an MSS of 1488 in
5387          * this case results in a TCP Data Payload of 1448 bytes which
5388          * is a multiple of 8.  On the other hand, if 12-byte TCP Time
5389          * Stamps have been negotiated, then an MTU of 1500 bytes
5390          * results in a TCP Data Payload of 1448 bytes which, as
5391          * above, is a multiple of 8 bytes ...
5392          */
5393         for (i = 0; i < NMTUS; i++)
5394             if (adap->params.mtus[i] == 1492) {
5395                 adap->params.mtus[i] = 1488;
5396                 break;
5397             }
5398 
5399         t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5400                  adap->params.b_wnd);
5401     }
5402     t4_init_sge_params(adap);
5403     adap->flags |= CXGB4_FW_OK;
5404     t4_init_tp_params(adap, true);
5405     return 0;
5406 
5407     /*
5408      * Something bad happened.  If a command timed out or failed with EIO
5409      * FW does not operate within its spec or something catastrophic
5410      * happened to HW/FW, stop issuing commands.
5411      */
5412 bye:
5413     adap_free_hma_mem(adap);
5414     kfree(adap->sge.egr_map);
5415     kfree(adap->sge.ingr_map);
5416     bitmap_free(adap->sge.starving_fl);
5417     bitmap_free(adap->sge.txq_maperr);
5418 #ifdef CONFIG_DEBUG_FS
5419     bitmap_free(adap->sge.blocked_fl);
5420 #endif
5421     if (ret != -ETIMEDOUT && ret != -EIO)
5422         t4_fw_bye(adap, adap->mbox);
5423     return ret;
5424 }
5425 
5426 /* EEH callbacks */
5427 
5428 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
5429                      pci_channel_state_t state)
5430 {
5431     int i;
5432     struct adapter *adap = pci_get_drvdata(pdev);
5433 
5434     if (!adap)
5435         goto out;
5436 
5437     rtnl_lock();
5438     adap->flags &= ~CXGB4_FW_OK;
5439     notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
5440     spin_lock(&adap->stats_lock);
5441     for_each_port(adap, i) {
5442         struct net_device *dev = adap->port[i];
5443         if (dev) {
5444             netif_device_detach(dev);
5445             netif_carrier_off(dev);
5446         }
5447     }
5448     spin_unlock(&adap->stats_lock);
5449     disable_interrupts(adap);
5450     if (adap->flags & CXGB4_FULL_INIT_DONE)
5451         cxgb_down(adap);
5452     rtnl_unlock();
5453     if ((adap->flags & CXGB4_DEV_ENABLED)) {
5454         pci_disable_device(pdev);
5455         adap->flags &= ~CXGB4_DEV_ENABLED;
5456     }
5457 out:    return state == pci_channel_io_perm_failure ?
5458         PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
5459 }
5460 
5461 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
5462 {
5463     int i, ret;
5464     struct fw_caps_config_cmd c;
5465     struct adapter *adap = pci_get_drvdata(pdev);
5466 
5467     if (!adap) {
5468         pci_restore_state(pdev);
5469         pci_save_state(pdev);
5470         return PCI_ERS_RESULT_RECOVERED;
5471     }
5472 
5473     if (!(adap->flags & CXGB4_DEV_ENABLED)) {
5474         if (pci_enable_device(pdev)) {
5475             dev_err(&pdev->dev, "Cannot reenable PCI "
5476                         "device after reset\n");
5477             return PCI_ERS_RESULT_DISCONNECT;
5478         }
5479         adap->flags |= CXGB4_DEV_ENABLED;
5480     }
5481 
5482     pci_set_master(pdev);
5483     pci_restore_state(pdev);
5484     pci_save_state(pdev);
5485 
5486     if (t4_wait_dev_ready(adap->regs) < 0)
5487         return PCI_ERS_RESULT_DISCONNECT;
5488     if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
5489         return PCI_ERS_RESULT_DISCONNECT;
5490     adap->flags |= CXGB4_FW_OK;
5491     if (adap_init1(adap, &c))
5492         return PCI_ERS_RESULT_DISCONNECT;
5493 
5494     for_each_port(adap, i) {
5495         struct port_info *pi = adap2pinfo(adap, i);
5496         u8 vivld = 0, vin = 0;
5497 
5498         ret = t4_alloc_vi(adap, adap->mbox, pi->tx_chan, adap->pf, 0, 1,
5499                   NULL, NULL, &vivld, &vin);
5500         if (ret < 0)
5501             return PCI_ERS_RESULT_DISCONNECT;
5502         pi->viid = ret;
5503         pi->xact_addr_filt = -1;
5504         /* If fw supports returning the VIN as part of FW_VI_CMD,
5505          * save the returned values.
5506          */
5507         if (adap->params.viid_smt_extn_support) {
5508             pi->vivld = vivld;
5509             pi->vin = vin;
5510         } else {
5511             /* Retrieve the values from VIID */
5512             pi->vivld = FW_VIID_VIVLD_G(pi->viid);
5513             pi->vin = FW_VIID_VIN_G(pi->viid);
5514         }
5515     }
5516 
5517     t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5518              adap->params.b_wnd);
5519     setup_memwin(adap);
5520     if (cxgb_up(adap))
5521         return PCI_ERS_RESULT_DISCONNECT;
5522     return PCI_ERS_RESULT_RECOVERED;
5523 }
5524 
5525 static void eeh_resume(struct pci_dev *pdev)
5526 {
5527     int i;
5528     struct adapter *adap = pci_get_drvdata(pdev);
5529 
5530     if (!adap)
5531         return;
5532 
5533     rtnl_lock();
5534     for_each_port(adap, i) {
5535         struct net_device *dev = adap->port[i];
5536         if (dev) {
5537             if (netif_running(dev)) {
5538                 link_start(dev);
5539                 cxgb_set_rxmode(dev);
5540             }
5541             netif_device_attach(dev);
5542         }
5543     }
5544     rtnl_unlock();
5545 }
5546 
5547 static void eeh_reset_prepare(struct pci_dev *pdev)
5548 {
5549     struct adapter *adapter = pci_get_drvdata(pdev);
5550     int i;
5551 
5552     if (adapter->pf != 4)
5553         return;
5554 
5555     adapter->flags &= ~CXGB4_FW_OK;
5556 
5557     notify_ulds(adapter, CXGB4_STATE_DOWN);
5558 
5559     for_each_port(adapter, i)
5560         if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5561             cxgb_close(adapter->port[i]);
5562 
5563     disable_interrupts(adapter);
5564     cxgb4_free_mps_ref_entries(adapter);
5565 
5566     adap_free_hma_mem(adapter);
5567 
5568     if (adapter->flags & CXGB4_FULL_INIT_DONE)
5569         cxgb_down(adapter);
5570 }
5571 
5572 static void eeh_reset_done(struct pci_dev *pdev)
5573 {
5574     struct adapter *adapter = pci_get_drvdata(pdev);
5575     int err, i;
5576 
5577     if (adapter->pf != 4)
5578         return;
5579 
5580     err = t4_wait_dev_ready(adapter->regs);
5581     if (err < 0) {
5582         dev_err(adapter->pdev_dev,
5583             "Device not ready, err %d", err);
5584         return;
5585     }
5586 
5587     setup_memwin(adapter);
5588 
5589     err = adap_init0(adapter, 1);
5590     if (err) {
5591         dev_err(adapter->pdev_dev,
5592             "Adapter init failed, err %d", err);
5593         return;
5594     }
5595 
5596     setup_memwin_rdma(adapter);
5597 
5598     if (adapter->flags & CXGB4_FW_OK) {
5599         err = t4_port_init(adapter, adapter->pf, adapter->pf, 0);
5600         if (err) {
5601             dev_err(adapter->pdev_dev,
5602                 "Port init failed, err %d", err);
5603             return;
5604         }
5605     }
5606 
5607     err = cfg_queues(adapter);
5608     if (err) {
5609         dev_err(adapter->pdev_dev,
5610             "Config queues failed, err %d", err);
5611         return;
5612     }
5613 
5614     cxgb4_init_mps_ref_entries(adapter);
5615 
5616     err = setup_fw_sge_queues(adapter);
5617     if (err) {
5618         dev_err(adapter->pdev_dev,
5619             "FW sge queue allocation failed, err %d", err);
5620         return;
5621     }
5622 
5623     for_each_port(adapter, i)
5624         if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5625             cxgb_open(adapter->port[i]);
5626 }
5627 
5628 static const struct pci_error_handlers cxgb4_eeh = {
5629     .error_detected = eeh_err_detected,
5630     .slot_reset     = eeh_slot_reset,
5631     .resume         = eeh_resume,
5632     .reset_prepare  = eeh_reset_prepare,
5633     .reset_done     = eeh_reset_done,
5634 };
5635 
5636 /* Return true if the Link Configuration supports "High Speeds" (those greater
5637  * than 1Gb/s).
5638  */
5639 static inline bool is_x_10g_port(const struct link_config *lc)
5640 {
5641     unsigned int speeds, high_speeds;
5642 
5643     speeds = FW_PORT_CAP32_SPEED_V(FW_PORT_CAP32_SPEED_G(lc->pcaps));
5644     high_speeds = speeds &
5645             ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
5646 
5647     return high_speeds != 0;
5648 }
5649 
5650 /* Perform default configuration of DMA queues depending on the number and type
5651  * of ports we found and the number of available CPUs.  Most settings can be
5652  * modified by the admin prior to actual use.
5653  */
5654 static int cfg_queues(struct adapter *adap)
5655 {
5656     u32 avail_qsets, avail_eth_qsets, avail_uld_qsets;
5657     u32 ncpus = num_online_cpus();
5658     u32 niqflint, neq, num_ulds;
5659     struct sge *s = &adap->sge;
5660     u32 i, n10g = 0, qidx = 0;
5661     u32 q10g = 0, q1g;
5662 
5663     /* Reduce memory usage in kdump environment, disable all offload. */
5664     if (is_kdump_kernel() || (is_uld(adap) && t4_uld_mem_alloc(adap))) {
5665         adap->params.offload = 0;
5666         adap->params.crypto = 0;
5667         adap->params.ethofld = 0;
5668     }
5669 
5670     /* Calculate the number of Ethernet Queue Sets available based on
5671      * resources provisioned for us.  We always have an Asynchronous
5672      * Firmware Event Ingress Queue.  If we're operating in MSI or Legacy
5673      * IRQ Pin Interrupt mode, then we'll also have a Forwarded Interrupt
5674      * Ingress Queue.  Meanwhile, we need two Egress Queues for each
5675      * Queue Set: one for the Free List and one for the Ethernet TX Queue.
5676      *
5677      * Note that we should also take into account all of the various
5678      * Offload Queues.  But, in any situation where we're operating in
5679      * a Resource Constrained Provisioning environment, doing any Offload
5680      * at all is problematic ...
5681      */
5682     niqflint = adap->params.pfres.niqflint - 1;
5683     if (!(adap->flags & CXGB4_USING_MSIX))
5684         niqflint--;
5685     neq = adap->params.pfres.neq / 2;
5686     avail_qsets = min(niqflint, neq);
5687 
5688     if (avail_qsets < adap->params.nports) {
5689         dev_err(adap->pdev_dev, "avail_eth_qsets=%d < nports=%d\n",
5690             avail_qsets, adap->params.nports);
5691         return -ENOMEM;
5692     }
5693 
5694     /* Count the number of 10Gb/s or better ports */
5695     for_each_port(adap, i)
5696         n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
5697 
5698     avail_eth_qsets = min_t(u32, avail_qsets, MAX_ETH_QSETS);
5699 
5700     /* We default to 1 queue per non-10G port and up to # of cores queues
5701      * per 10G port.
5702      */
5703     if (n10g)
5704         q10g = (avail_eth_qsets - (adap->params.nports - n10g)) / n10g;
5705 
5706 #ifdef CONFIG_CHELSIO_T4_DCB
5707     /* For Data Center Bridging support we need to be able to support up
5708      * to 8 Traffic Priorities; each of which will be assigned to its
5709      * own TX Queue in order to prevent Head-Of-Line Blocking.
5710      */
5711     q1g = 8;
5712     if (adap->params.nports * 8 > avail_eth_qsets) {
5713         dev_err(adap->pdev_dev, "DCB avail_eth_qsets=%d < %d!\n",
5714             avail_eth_qsets, adap->params.nports * 8);
5715         return -ENOMEM;
5716     }
5717 
5718     if (adap->params.nports * ncpus < avail_eth_qsets)
5719         q10g = max(8U, ncpus);
5720     else
5721         q10g = max(8U, q10g);
5722 
5723     while ((q10g * n10g) >
5724            (avail_eth_qsets - (adap->params.nports - n10g) * q1g))
5725         q10g--;
5726 
5727 #else /* !CONFIG_CHELSIO_T4_DCB */
5728     q1g = 1;
5729     q10g = min(q10g, ncpus);
5730 #endif /* !CONFIG_CHELSIO_T4_DCB */
5731     if (is_kdump_kernel()) {
5732         q10g = 1;
5733         q1g = 1;
5734     }
5735 
5736     for_each_port(adap, i) {
5737         struct port_info *pi = adap2pinfo(adap, i);
5738 
5739         pi->first_qset = qidx;
5740         pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : q1g;
5741         qidx += pi->nqsets;
5742     }
5743 
5744     s->ethqsets = qidx;
5745     s->max_ethqsets = qidx;   /* MSI-X may lower it later */
5746     avail_qsets -= qidx;
5747 
5748     if (is_uld(adap)) {
5749         /* For offload we use 1 queue/channel if all ports are up to 1G,
5750          * otherwise we divide all available queues amongst the channels
5751          * capped by the number of available cores.
5752          */
5753         num_ulds = adap->num_uld + adap->num_ofld_uld;
5754         i = min_t(u32, MAX_OFLD_QSETS, ncpus);
5755         avail_uld_qsets = roundup(i, adap->params.nports);
5756         if (avail_qsets < num_ulds * adap->params.nports) {
5757             adap->params.offload = 0;
5758             adap->params.crypto = 0;
5759             s->ofldqsets = 0;
5760         } else if (avail_qsets < num_ulds * avail_uld_qsets || !n10g) {
5761             s->ofldqsets = adap->params.nports;
5762         } else {
5763             s->ofldqsets = avail_uld_qsets;
5764         }
5765 
5766         avail_qsets -= num_ulds * s->ofldqsets;
5767     }
5768 
5769     /* ETHOFLD Queues used for QoS offload should follow same
5770      * allocation scheme as normal Ethernet Queues.
5771      */
5772     if (is_ethofld(adap)) {
5773         if (avail_qsets < s->max_ethqsets) {
5774             adap->params.ethofld = 0;
5775             s->eoqsets = 0;
5776         } else {
5777             s->eoqsets = s->max_ethqsets;
5778         }
5779         avail_qsets -= s->eoqsets;
5780     }
5781 
5782     /* Mirror queues must follow same scheme as normal Ethernet
5783      * Queues, when there are enough queues available. Otherwise,
5784      * allocate at least 1 queue per port. If even 1 queue is not
5785      * available, then disable mirror queues support.
5786      */
5787     if (avail_qsets >= s->max_ethqsets)
5788         s->mirrorqsets = s->max_ethqsets;
5789     else if (avail_qsets >= adap->params.nports)
5790         s->mirrorqsets = adap->params.nports;
5791     else
5792         s->mirrorqsets = 0;
5793     avail_qsets -= s->mirrorqsets;
5794 
5795     for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
5796         struct sge_eth_rxq *r = &s->ethrxq[i];
5797 
5798         init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
5799         r->fl.size = 72;
5800     }
5801 
5802     for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
5803         s->ethtxq[i].q.size = 1024;
5804 
5805     for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
5806         s->ctrlq[i].q.size = 512;
5807 
5808     if (!is_t4(adap->params.chip))
5809         s->ptptxq.q.size = 8;
5810 
5811     init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
5812     init_rspq(adap, &s->intrq, 0, 1, 512, 64);
5813 
5814     return 0;
5815 }
5816 
5817 /*
5818  * Reduce the number of Ethernet queues across all ports to at most n.
5819  * n provides at least one queue per port.
5820  */
5821 static void reduce_ethqs(struct adapter *adap, int n)
5822 {
5823     int i;
5824     struct port_info *pi;
5825 
5826     while (n < adap->sge.ethqsets)
5827         for_each_port(adap, i) {
5828             pi = adap2pinfo(adap, i);
5829             if (pi->nqsets > 1) {
5830                 pi->nqsets--;
5831                 adap->sge.ethqsets--;
5832                 if (adap->sge.ethqsets <= n)
5833                     break;
5834             }
5835         }
5836 
5837     n = 0;
5838     for_each_port(adap, i) {
5839         pi = adap2pinfo(adap, i);
5840         pi->first_qset = n;
5841         n += pi->nqsets;
5842     }
5843 }
5844 
5845 static int alloc_msix_info(struct adapter *adap, u32 num_vec)
5846 {
5847     struct msix_info *msix_info;
5848 
5849     msix_info = kcalloc(num_vec, sizeof(*msix_info), GFP_KERNEL);
5850     if (!msix_info)
5851         return -ENOMEM;
5852 
5853     adap->msix_bmap.msix_bmap = bitmap_zalloc(num_vec, GFP_KERNEL);
5854     if (!adap->msix_bmap.msix_bmap) {
5855         kfree(msix_info);
5856         return -ENOMEM;
5857     }
5858 
5859     spin_lock_init(&adap->msix_bmap.lock);
5860     adap->msix_bmap.mapsize = num_vec;
5861 
5862     adap->msix_info = msix_info;
5863     return 0;
5864 }
5865 
5866 static void free_msix_info(struct adapter *adap)
5867 {
5868     bitmap_free(adap->msix_bmap.msix_bmap);
5869     kfree(adap->msix_info);
5870 }
5871 
5872 int cxgb4_get_msix_idx_from_bmap(struct adapter *adap)
5873 {
5874     struct msix_bmap *bmap = &adap->msix_bmap;
5875     unsigned int msix_idx;
5876     unsigned long flags;
5877 
5878     spin_lock_irqsave(&bmap->lock, flags);
5879     msix_idx = find_first_zero_bit(bmap->msix_bmap, bmap->mapsize);
5880     if (msix_idx < bmap->mapsize) {
5881         __set_bit(msix_idx, bmap->msix_bmap);
5882     } else {
5883         spin_unlock_irqrestore(&bmap->lock, flags);
5884         return -ENOSPC;
5885     }
5886 
5887     spin_unlock_irqrestore(&bmap->lock, flags);
5888     return msix_idx;
5889 }
5890 
5891 void cxgb4_free_msix_idx_in_bmap(struct adapter *adap,
5892                  unsigned int msix_idx)
5893 {
5894     struct msix_bmap *bmap = &adap->msix_bmap;
5895     unsigned long flags;
5896 
5897     spin_lock_irqsave(&bmap->lock, flags);
5898     __clear_bit(msix_idx, bmap->msix_bmap);
5899     spin_unlock_irqrestore(&bmap->lock, flags);
5900 }
5901 
5902 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5903 #define EXTRA_VECS 2
5904 
5905 static int enable_msix(struct adapter *adap)
5906 {
5907     u32 eth_need, uld_need = 0, ethofld_need = 0, mirror_need = 0;
5908     u32 ethqsets = 0, ofldqsets = 0, eoqsets = 0, mirrorqsets = 0;
5909     u8 num_uld = 0, nchan = adap->params.nports;
5910     u32 i, want, need, num_vec;
5911     struct sge *s = &adap->sge;
5912     struct msix_entry *entries;
5913     struct port_info *pi;
5914     int allocated, ret;
5915 
5916     want = s->max_ethqsets;
5917 #ifdef CONFIG_CHELSIO_T4_DCB
5918     /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
5919      * each port.
5920      */
5921     need = 8 * nchan;
5922 #else
5923     need = nchan;
5924 #endif
5925     eth_need = need;
5926     if (is_uld(adap)) {
5927         num_uld = adap->num_ofld_uld + adap->num_uld;
5928         want += num_uld * s->ofldqsets;
5929         uld_need = num_uld * nchan;
5930         need += uld_need;
5931     }
5932 
5933     if (is_ethofld(adap)) {
5934         want += s->eoqsets;
5935         ethofld_need = eth_need;
5936         need += ethofld_need;
5937     }
5938 
5939     if (s->mirrorqsets) {
5940         want += s->mirrorqsets;
5941         mirror_need = nchan;
5942         need += mirror_need;
5943     }
5944 
5945     want += EXTRA_VECS;
5946     need += EXTRA_VECS;
5947 
5948     entries = kmalloc_array(want, sizeof(*entries), GFP_KERNEL);
5949     if (!entries)
5950         return -ENOMEM;
5951 
5952     for (i = 0; i < want; i++)
5953         entries[i].entry = i;
5954 
5955     allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
5956     if (allocated < 0) {
5957         /* Disable offload and attempt to get vectors for NIC
5958          * only mode.
5959          */
5960         want = s->max_ethqsets + EXTRA_VECS;
5961         need = eth_need + EXTRA_VECS;
5962         allocated = pci_enable_msix_range(adap->pdev, entries,
5963                           need, want);
5964         if (allocated < 0) {
5965             dev_info(adap->pdev_dev,
5966                  "Disabling MSI-X due to insufficient MSI-X vectors\n");
5967             ret = allocated;
5968             goto out_free;
5969         }
5970 
5971         dev_info(adap->pdev_dev,
5972              "Disabling offload due to insufficient MSI-X vectors\n");
5973         adap->params.offload = 0;
5974         adap->params.crypto = 0;
5975         adap->params.ethofld = 0;
5976         s->ofldqsets = 0;
5977         s->eoqsets = 0;
5978         s->mirrorqsets = 0;
5979         uld_need = 0;
5980         ethofld_need = 0;
5981         mirror_need = 0;
5982     }
5983 
5984     num_vec = allocated;
5985     if (num_vec < want) {
5986         /* Distribute available vectors to the various queue groups.
5987          * Every group gets its minimum requirement and NIC gets top
5988          * priority for leftovers.
5989          */
5990         ethqsets = eth_need;
5991         if (is_uld(adap))
5992             ofldqsets = nchan;
5993         if (is_ethofld(adap))
5994             eoqsets = ethofld_need;
5995         if (s->mirrorqsets)
5996             mirrorqsets = mirror_need;
5997 
5998         num_vec -= need;
5999         while (num_vec) {
6000             if (num_vec < eth_need + ethofld_need ||
6001                 ethqsets > s->max_ethqsets)
6002                 break;
6003 
6004             for_each_port(adap, i) {
6005                 pi = adap2pinfo(adap, i);
6006                 if (pi->nqsets < 2)
6007                     continue;
6008 
6009                 ethqsets++;
6010                 num_vec--;
6011                 if (ethofld_need) {
6012                     eoqsets++;
6013                     num_vec--;
6014                 }
6015             }
6016         }
6017 
6018         if (is_uld(adap)) {
6019             while (num_vec) {
6020                 if (num_vec < uld_need ||
6021                     ofldqsets > s->ofldqsets)
6022                     break;
6023 
6024                 ofldqsets++;
6025                 num_vec -= uld_need;
6026             }
6027         }
6028 
6029         if (s->mirrorqsets) {
6030             while (num_vec) {
6031                 if (num_vec < mirror_need ||
6032                     mirrorqsets > s->mirrorqsets)
6033                     break;
6034 
6035                 mirrorqsets++;
6036                 num_vec -= mirror_need;
6037             }
6038         }
6039     } else {
6040         ethqsets = s->max_ethqsets;
6041         if (is_uld(adap))
6042             ofldqsets = s->ofldqsets;
6043         if (is_ethofld(adap))
6044             eoqsets = s->eoqsets;
6045         if (s->mirrorqsets)
6046             mirrorqsets = s->mirrorqsets;
6047     }
6048 
6049     if (ethqsets < s->max_ethqsets) {
6050         s->max_ethqsets = ethqsets;
6051         reduce_ethqs(adap, ethqsets);
6052     }
6053 
6054     if (is_uld(adap)) {
6055         s->ofldqsets = ofldqsets;
6056         s->nqs_per_uld = s->ofldqsets;
6057     }
6058 
6059     if (is_ethofld(adap))
6060         s->eoqsets = eoqsets;
6061 
6062     if (s->mirrorqsets) {
6063         s->mirrorqsets = mirrorqsets;
6064         for_each_port(adap, i) {
6065             pi = adap2pinfo(adap, i);
6066             pi->nmirrorqsets = s->mirrorqsets / nchan;
6067             mutex_init(&pi->vi_mirror_mutex);
6068         }
6069     }
6070 
6071     /* map for msix */
6072     ret = alloc_msix_info(adap, allocated);
6073     if (ret)
6074         goto out_disable_msix;
6075 
6076     for (i = 0; i < allocated; i++) {
6077         adap->msix_info[i].vec = entries[i].vector;
6078         adap->msix_info[i].idx = i;
6079     }
6080 
6081     dev_info(adap->pdev_dev,
6082          "%d MSI-X vectors allocated, nic %d eoqsets %d per uld %d mirrorqsets %d\n",
6083          allocated, s->max_ethqsets, s->eoqsets, s->nqs_per_uld,
6084          s->mirrorqsets);
6085 
6086     kfree(entries);
6087     return 0;
6088 
6089 out_disable_msix:
6090     pci_disable_msix(adap->pdev);
6091 
6092 out_free:
6093     kfree(entries);
6094     return ret;
6095 }
6096 
6097 #undef EXTRA_VECS
6098 
6099 static int init_rss(struct adapter *adap)
6100 {
6101     unsigned int i;
6102     int err;
6103 
6104     err = t4_init_rss_mode(adap, adap->mbox);
6105     if (err)
6106         return err;
6107 
6108     for_each_port(adap, i) {
6109         struct port_info *pi = adap2pinfo(adap, i);
6110 
6111         pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
6112         if (!pi->rss)
6113             return -ENOMEM;
6114     }
6115     return 0;
6116 }
6117 
6118 /* Dump basic information about the adapter */
6119 static void print_adapter_info(struct adapter *adapter)
6120 {
6121     /* Hardware/Firmware/etc. Version/Revision IDs */
6122     t4_dump_version_info(adapter);
6123 
6124     /* Software/Hardware configuration */
6125     dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
6126          is_offload(adapter) ? "R" : "",
6127          ((adapter->flags & CXGB4_USING_MSIX) ? "MSI-X" :
6128           (adapter->flags & CXGB4_USING_MSI) ? "MSI" : ""),
6129          is_offload(adapter) ? "Offload" : "non-Offload");
6130 }
6131 
6132 static void print_port_info(const struct net_device *dev)
6133 {
6134     char buf[80];
6135     char *bufp = buf;
6136     const struct port_info *pi = netdev_priv(dev);
6137     const struct adapter *adap = pi->adapter;
6138 
6139     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
6140         bufp += sprintf(bufp, "100M/");
6141     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
6142         bufp += sprintf(bufp, "1G/");
6143     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
6144         bufp += sprintf(bufp, "10G/");
6145     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
6146         bufp += sprintf(bufp, "25G/");
6147     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
6148         bufp += sprintf(bufp, "40G/");
6149     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
6150         bufp += sprintf(bufp, "50G/");
6151     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
6152         bufp += sprintf(bufp, "100G/");
6153     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_200G)
6154         bufp += sprintf(bufp, "200G/");
6155     if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_400G)
6156         bufp += sprintf(bufp, "400G/");
6157     if (bufp != buf)
6158         --bufp;
6159     sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
6160 
6161     netdev_info(dev, "Chelsio %s %s\n", adap->params.vpd.id, buf);
6162 }
6163 
6164 /*
6165  * Free the following resources:
6166  * - memory used for tables
6167  * - MSI/MSI-X
6168  * - net devices
6169  * - resources FW is holding for us
6170  */
6171 static void free_some_resources(struct adapter *adapter)
6172 {
6173     unsigned int i;
6174 
6175     kvfree(adapter->smt);
6176     kvfree(adapter->l2t);
6177     kvfree(adapter->srq);
6178     t4_cleanup_sched(adapter);
6179     kvfree(adapter->tids.tid_tab);
6180     cxgb4_cleanup_tc_matchall(adapter);
6181     cxgb4_cleanup_tc_mqprio(adapter);
6182     cxgb4_cleanup_tc_flower(adapter);
6183     cxgb4_cleanup_tc_u32(adapter);
6184     cxgb4_cleanup_ethtool_filters(adapter);
6185     kfree(adapter->sge.egr_map);
6186     kfree(adapter->sge.ingr_map);
6187     bitmap_free(adapter->sge.starving_fl);
6188     bitmap_free(adapter->sge.txq_maperr);
6189 #ifdef CONFIG_DEBUG_FS
6190     bitmap_free(adapter->sge.blocked_fl);
6191 #endif
6192     disable_msi(adapter);
6193 
6194     for_each_port(adapter, i)
6195         if (adapter->port[i]) {
6196             struct port_info *pi = adap2pinfo(adapter, i);
6197 
6198             if (pi->viid != 0)
6199                 t4_free_vi(adapter, adapter->mbox, adapter->pf,
6200                        0, pi->viid);
6201             kfree(adap2pinfo(adapter, i)->rss);
6202             free_netdev(adapter->port[i]);
6203         }
6204     if (adapter->flags & CXGB4_FW_OK)
6205         t4_fw_bye(adapter, adapter->pf);
6206 }
6207 
6208 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN | \
6209            NETIF_F_GSO_UDP_L4)
6210 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
6211            NETIF_F_GRO | NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
6212 #define SEGMENT_SIZE 128
6213 
6214 static int t4_get_chip_type(struct adapter *adap, int ver)
6215 {
6216     u32 pl_rev = REV_G(t4_read_reg(adap, PL_REV_A));
6217 
6218     switch (ver) {
6219     case CHELSIO_T4:
6220         return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
6221     case CHELSIO_T5:
6222         return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
6223     case CHELSIO_T6:
6224         return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
6225     default:
6226         break;
6227     }
6228     return -EINVAL;
6229 }
6230 
6231 #ifdef CONFIG_PCI_IOV
6232 static void cxgb4_mgmt_setup(struct net_device *dev)
6233 {
6234     dev->type = ARPHRD_NONE;
6235     dev->mtu = 0;
6236     dev->hard_header_len = 0;
6237     dev->addr_len = 0;
6238     dev->tx_queue_len = 0;
6239     dev->flags |= IFF_NOARP;
6240     dev->priv_flags |= IFF_NO_QUEUE;
6241 
6242     /* Initialize the device structure. */
6243     dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
6244     dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
6245 }
6246 
6247 static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
6248 {
6249     struct adapter *adap = pci_get_drvdata(pdev);
6250     int err = 0;
6251     int current_vfs = pci_num_vf(pdev);
6252     u32 pcie_fw;
6253 
6254     pcie_fw = readl(adap->regs + PCIE_FW_A);
6255     /* Check if fw is initialized */
6256     if (!(pcie_fw & PCIE_FW_INIT_F)) {
6257         dev_warn(&pdev->dev, "Device not initialized\n");
6258         return -EOPNOTSUPP;
6259     }
6260 
6261     /* If any of the VF's is already assigned to Guest OS, then
6262      * SRIOV for the same cannot be modified
6263      */
6264     if (current_vfs && pci_vfs_assigned(pdev)) {
6265         dev_err(&pdev->dev,
6266             "Cannot modify SR-IOV while VFs are assigned\n");
6267         return current_vfs;
6268     }
6269     /* Note that the upper-level code ensures that we're never called with
6270      * a non-zero "num_vfs" when we already have VFs instantiated.  But
6271      * it never hurts to code defensively.
6272      */
6273     if (num_vfs != 0 && current_vfs != 0)
6274         return -EBUSY;
6275 
6276     /* Nothing to do for no change. */
6277     if (num_vfs == current_vfs)
6278         return num_vfs;
6279 
6280     /* Disable SRIOV when zero is passed. */
6281     if (!num_vfs) {
6282         pci_disable_sriov(pdev);
6283         /* free VF Management Interface */
6284         unregister_netdev(adap->port[0]);
6285         free_netdev(adap->port[0]);
6286         adap->port[0] = NULL;
6287 
6288         /* free VF resources */
6289         adap->num_vfs = 0;
6290         kfree(adap->vfinfo);
6291         adap->vfinfo = NULL;
6292         return 0;
6293     }
6294 
6295     if (!current_vfs) {
6296         struct fw_pfvf_cmd port_cmd, port_rpl;
6297         struct net_device *netdev;
6298         unsigned int pmask, port;
6299         struct pci_dev *pbridge;
6300         struct port_info *pi;
6301         char name[IFNAMSIZ];
6302         u32 devcap2;
6303         u16 flags;
6304 
6305         /* If we want to instantiate Virtual Functions, then our
6306          * parent bridge's PCI-E needs to support Alternative Routing
6307          * ID (ARI) because our VFs will show up at function offset 8
6308          * and above.
6309          */
6310         pbridge = pdev->bus->self;
6311         pcie_capability_read_word(pbridge, PCI_EXP_FLAGS, &flags);
6312         pcie_capability_read_dword(pbridge, PCI_EXP_DEVCAP2, &devcap2);
6313 
6314         if ((flags & PCI_EXP_FLAGS_VERS) < 2 ||
6315             !(devcap2 & PCI_EXP_DEVCAP2_ARI)) {
6316             /* Our parent bridge does not support ARI so issue a
6317              * warning and skip instantiating the VFs.  They
6318              * won't be reachable.
6319              */
6320             dev_warn(&pdev->dev, "Parent bridge %02x:%02x.%x doesn't support ARI; can't instantiate Virtual Functions\n",
6321                  pbridge->bus->number, PCI_SLOT(pbridge->devfn),
6322                  PCI_FUNC(pbridge->devfn));
6323             return -ENOTSUPP;
6324         }
6325         memset(&port_cmd, 0, sizeof(port_cmd));
6326         port_cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_PFVF_CMD) |
6327                          FW_CMD_REQUEST_F |
6328                          FW_CMD_READ_F |
6329                          FW_PFVF_CMD_PFN_V(adap->pf) |
6330                          FW_PFVF_CMD_VFN_V(0));
6331         port_cmd.retval_len16 = cpu_to_be32(FW_LEN16(port_cmd));
6332         err = t4_wr_mbox(adap, adap->mbox, &port_cmd, sizeof(port_cmd),
6333                  &port_rpl);
6334         if (err)
6335             return err;
6336         pmask = FW_PFVF_CMD_PMASK_G(be32_to_cpu(port_rpl.type_to_neq));
6337         port = ffs(pmask) - 1;
6338         /* Allocate VF Management Interface. */
6339         snprintf(name, IFNAMSIZ, "mgmtpf%d,%d", adap->adap_idx,
6340              adap->pf);
6341         netdev = alloc_netdev(sizeof(struct port_info),
6342                       name, NET_NAME_UNKNOWN, cxgb4_mgmt_setup);
6343         if (!netdev)
6344             return -ENOMEM;
6345 
6346         pi = netdev_priv(netdev);
6347         pi->adapter = adap;
6348         pi->lport = port;
6349         pi->tx_chan = port;
6350         SET_NETDEV_DEV(netdev, &pdev->dev);
6351 
6352         adap->port[0] = netdev;
6353         pi->port_id = 0;
6354 
6355         err = register_netdev(adap->port[0]);
6356         if (err) {
6357             pr_info("Unable to register VF mgmt netdev %s\n", name);
6358             free_netdev(adap->port[0]);
6359             adap->port[0] = NULL;
6360             return err;
6361         }
6362         /* Allocate and set up VF Information. */
6363         adap->vfinfo = kcalloc(pci_sriov_get_totalvfs(pdev),
6364                        sizeof(struct vf_info), GFP_KERNEL);
6365         if (!adap->vfinfo) {
6366             unregister_netdev(adap->port[0]);
6367             free_netdev(adap->port[0]);
6368             adap->port[0] = NULL;
6369             return -ENOMEM;
6370         }
6371         cxgb4_mgmt_fill_vf_station_mac_addr(adap);
6372     }
6373     /* Instantiate the requested number of VFs. */
6374     err = pci_enable_sriov(pdev, num_vfs);
6375     if (err) {
6376         pr_info("Unable to instantiate %d VFs\n", num_vfs);
6377         if (!current_vfs) {
6378             unregister_netdev(adap->port[0]);
6379             free_netdev(adap->port[0]);
6380             adap->port[0] = NULL;
6381             kfree(adap->vfinfo);
6382             adap->vfinfo = NULL;
6383         }
6384         return err;
6385     }
6386 
6387     adap->num_vfs = num_vfs;
6388     return num_vfs;
6389 }
6390 #endif /* CONFIG_PCI_IOV */
6391 
6392 #if IS_ENABLED(CONFIG_CHELSIO_TLS_DEVICE) || IS_ENABLED(CONFIG_CHELSIO_IPSEC_INLINE)
6393 
6394 static int chcr_offload_state(struct adapter *adap,
6395                   enum cxgb4_netdev_tls_ops op_val)
6396 {
6397     switch (op_val) {
6398 #if IS_ENABLED(CONFIG_CHELSIO_TLS_DEVICE)
6399     case CXGB4_TLSDEV_OPS:
6400         if (!adap->uld[CXGB4_ULD_KTLS].handle) {
6401             dev_dbg(adap->pdev_dev, "ch_ktls driver is not loaded\n");
6402             return -EOPNOTSUPP;
6403         }
6404         if (!adap->uld[CXGB4_ULD_KTLS].tlsdev_ops) {
6405             dev_dbg(adap->pdev_dev,
6406                 "ch_ktls driver has no registered tlsdev_ops\n");
6407             return -EOPNOTSUPP;
6408         }
6409         break;
6410 #endif /* CONFIG_CHELSIO_TLS_DEVICE */
6411 #if IS_ENABLED(CONFIG_CHELSIO_IPSEC_INLINE)
6412     case CXGB4_XFRMDEV_OPS:
6413         if (!adap->uld[CXGB4_ULD_IPSEC].handle) {
6414             dev_dbg(adap->pdev_dev, "chipsec driver is not loaded\n");
6415             return -EOPNOTSUPP;
6416         }
6417         if (!adap->uld[CXGB4_ULD_IPSEC].xfrmdev_ops) {
6418             dev_dbg(adap->pdev_dev,
6419                 "chipsec driver has no registered xfrmdev_ops\n");
6420             return -EOPNOTSUPP;
6421         }
6422         break;
6423 #endif /* CONFIG_CHELSIO_IPSEC_INLINE */
6424     default:
6425         dev_dbg(adap->pdev_dev,
6426             "driver has no support for offload %d\n", op_val);
6427         return -EOPNOTSUPP;
6428     }
6429 
6430     return 0;
6431 }
6432 
6433 #endif /* CONFIG_CHELSIO_TLS_DEVICE || CONFIG_CHELSIO_IPSEC_INLINE */
6434 
6435 #if IS_ENABLED(CONFIG_CHELSIO_TLS_DEVICE)
6436 
6437 static int cxgb4_ktls_dev_add(struct net_device *netdev, struct sock *sk,
6438                   enum tls_offload_ctx_dir direction,
6439                   struct tls_crypto_info *crypto_info,
6440                   u32 tcp_sn)
6441 {
6442     struct adapter *adap = netdev2adap(netdev);
6443     int ret;
6444 
6445     mutex_lock(&uld_mutex);
6446     ret = chcr_offload_state(adap, CXGB4_TLSDEV_OPS);
6447     if (ret)
6448         goto out_unlock;
6449 
6450     ret = cxgb4_set_ktls_feature(adap, FW_PARAMS_PARAM_DEV_KTLS_HW_ENABLE);
6451     if (ret)
6452         goto out_unlock;
6453 
6454     ret = adap->uld[CXGB4_ULD_KTLS].tlsdev_ops->tls_dev_add(netdev, sk,
6455                                 direction,
6456                                 crypto_info,
6457                                 tcp_sn);
6458     /* if there is a failure, clear the refcount */
6459     if (ret)
6460         cxgb4_set_ktls_feature(adap,
6461                        FW_PARAMS_PARAM_DEV_KTLS_HW_DISABLE);
6462 out_unlock:
6463     mutex_unlock(&uld_mutex);
6464     return ret;
6465 }
6466 
6467 static void cxgb4_ktls_dev_del(struct net_device *netdev,
6468                    struct tls_context *tls_ctx,
6469                    enum tls_offload_ctx_dir direction)
6470 {
6471     struct adapter *adap = netdev2adap(netdev);
6472 
6473     mutex_lock(&uld_mutex);
6474     if (chcr_offload_state(adap, CXGB4_TLSDEV_OPS))
6475         goto out_unlock;
6476 
6477     adap->uld[CXGB4_ULD_KTLS].tlsdev_ops->tls_dev_del(netdev, tls_ctx,
6478                               direction);
6479 
6480 out_unlock:
6481     cxgb4_set_ktls_feature(adap, FW_PARAMS_PARAM_DEV_KTLS_HW_DISABLE);
6482     mutex_unlock(&uld_mutex);
6483 }
6484 
6485 static const struct tlsdev_ops cxgb4_ktls_ops = {
6486     .tls_dev_add = cxgb4_ktls_dev_add,
6487     .tls_dev_del = cxgb4_ktls_dev_del,
6488 };
6489 #endif /* CONFIG_CHELSIO_TLS_DEVICE */
6490 
6491 #if IS_ENABLED(CONFIG_CHELSIO_IPSEC_INLINE)
6492 
6493 static int cxgb4_xfrm_add_state(struct xfrm_state *x)
6494 {
6495     struct adapter *adap = netdev2adap(x->xso.dev);
6496     int ret;
6497 
6498     if (!mutex_trylock(&uld_mutex)) {
6499         dev_dbg(adap->pdev_dev,
6500             "crypto uld critical resource is under use\n");
6501         return -EBUSY;
6502     }
6503     ret = chcr_offload_state(adap, CXGB4_XFRMDEV_OPS);
6504     if (ret)
6505         goto out_unlock;
6506 
6507     ret = adap->uld[CXGB4_ULD_IPSEC].xfrmdev_ops->xdo_dev_state_add(x);
6508 
6509 out_unlock:
6510     mutex_unlock(&uld_mutex);
6511 
6512     return ret;
6513 }
6514 
6515 static void cxgb4_xfrm_del_state(struct xfrm_state *x)
6516 {
6517     struct adapter *adap = netdev2adap(x->xso.dev);
6518 
6519     if (!mutex_trylock(&uld_mutex)) {
6520         dev_dbg(adap->pdev_dev,
6521             "crypto uld critical resource is under use\n");
6522         return;
6523     }
6524     if (chcr_offload_state(adap, CXGB4_XFRMDEV_OPS))
6525         goto out_unlock;
6526 
6527     adap->uld[CXGB4_ULD_IPSEC].xfrmdev_ops->xdo_dev_state_delete(x);
6528 
6529 out_unlock:
6530     mutex_unlock(&uld_mutex);
6531 }
6532 
6533 static void cxgb4_xfrm_free_state(struct xfrm_state *x)
6534 {
6535     struct adapter *adap = netdev2adap(x->xso.dev);
6536 
6537     if (!mutex_trylock(&uld_mutex)) {
6538         dev_dbg(adap->pdev_dev,
6539             "crypto uld critical resource is under use\n");
6540         return;
6541     }
6542     if (chcr_offload_state(adap, CXGB4_XFRMDEV_OPS))
6543         goto out_unlock;
6544 
6545     adap->uld[CXGB4_ULD_IPSEC].xfrmdev_ops->xdo_dev_state_free(x);
6546 
6547 out_unlock:
6548     mutex_unlock(&uld_mutex);
6549 }
6550 
6551 static bool cxgb4_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
6552 {
6553     struct adapter *adap = netdev2adap(x->xso.dev);
6554     bool ret = false;
6555 
6556     if (!mutex_trylock(&uld_mutex)) {
6557         dev_dbg(adap->pdev_dev,
6558             "crypto uld critical resource is under use\n");
6559         return ret;
6560     }
6561     if (chcr_offload_state(adap, CXGB4_XFRMDEV_OPS))
6562         goto out_unlock;
6563 
6564     ret = adap->uld[CXGB4_ULD_IPSEC].xfrmdev_ops->xdo_dev_offload_ok(skb, x);
6565 
6566 out_unlock:
6567     mutex_unlock(&uld_mutex);
6568     return ret;
6569 }
6570 
6571 static void cxgb4_advance_esn_state(struct xfrm_state *x)
6572 {
6573     struct adapter *adap = netdev2adap(x->xso.dev);
6574 
6575     if (!mutex_trylock(&uld_mutex)) {
6576         dev_dbg(adap->pdev_dev,
6577             "crypto uld critical resource is under use\n");
6578         return;
6579     }
6580     if (chcr_offload_state(adap, CXGB4_XFRMDEV_OPS))
6581         goto out_unlock;
6582 
6583     adap->uld[CXGB4_ULD_IPSEC].xfrmdev_ops->xdo_dev_state_advance_esn(x);
6584 
6585 out_unlock:
6586     mutex_unlock(&uld_mutex);
6587 }
6588 
6589 static const struct xfrmdev_ops cxgb4_xfrmdev_ops = {
6590     .xdo_dev_state_add      = cxgb4_xfrm_add_state,
6591     .xdo_dev_state_delete   = cxgb4_xfrm_del_state,
6592     .xdo_dev_state_free     = cxgb4_xfrm_free_state,
6593     .xdo_dev_offload_ok     = cxgb4_ipsec_offload_ok,
6594     .xdo_dev_state_advance_esn = cxgb4_advance_esn_state,
6595 };
6596 
6597 #endif /* CONFIG_CHELSIO_IPSEC_INLINE */
6598 
6599 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
6600 {
6601     struct net_device *netdev;
6602     struct adapter *adapter;
6603     static int adap_idx = 1;
6604     int s_qpp, qpp, num_seg;
6605     struct port_info *pi;
6606     enum chip_type chip;
6607     void __iomem *regs;
6608     int func, chip_ver;
6609     u16 device_id;
6610     int i, err;
6611     u32 whoami;
6612 
6613     err = pci_request_regions(pdev, KBUILD_MODNAME);
6614     if (err) {
6615         /* Just info, some other driver may have claimed the device. */
6616         dev_info(&pdev->dev, "cannot obtain PCI resources\n");
6617         return err;
6618     }
6619 
6620     err = pci_enable_device(pdev);
6621     if (err) {
6622         dev_err(&pdev->dev, "cannot enable PCI device\n");
6623         goto out_release_regions;
6624     }
6625 
6626     regs = pci_ioremap_bar(pdev, 0);
6627     if (!regs) {
6628         dev_err(&pdev->dev, "cannot map device registers\n");
6629         err = -ENOMEM;
6630         goto out_disable_device;
6631     }
6632 
6633     adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
6634     if (!adapter) {
6635         err = -ENOMEM;
6636         goto out_unmap_bar0;
6637     }
6638 
6639     adapter->regs = regs;
6640     err = t4_wait_dev_ready(regs);
6641     if (err < 0)
6642         goto out_free_adapter;
6643 
6644     /* We control everything through one PF */
6645     whoami = t4_read_reg(adapter, PL_WHOAMI_A);
6646     pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
6647     chip = t4_get_chip_type(adapter, CHELSIO_PCI_ID_VER(device_id));
6648     if ((int)chip < 0) {
6649         dev_err(&pdev->dev, "Device %d is not supported\n", device_id);
6650         err = chip;
6651         goto out_free_adapter;
6652     }
6653     chip_ver = CHELSIO_CHIP_VERSION(chip);
6654     func = chip_ver <= CHELSIO_T5 ?
6655            SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
6656 
6657     adapter->pdev = pdev;
6658     adapter->pdev_dev = &pdev->dev;
6659     adapter->name = pci_name(pdev);
6660     adapter->mbox = func;
6661     adapter->pf = func;
6662     adapter->params.chip = chip;
6663     adapter->adap_idx = adap_idx;
6664     adapter->msg_enable = DFLT_MSG_ENABLE;
6665     adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
6666                     (sizeof(struct mbox_cmd) *
6667                      T4_OS_LOG_MBOX_CMDS),
6668                     GFP_KERNEL);
6669     if (!adapter->mbox_log) {
6670         err = -ENOMEM;
6671         goto out_free_adapter;
6672     }
6673     spin_lock_init(&adapter->mbox_lock);
6674     INIT_LIST_HEAD(&adapter->mlist.list);
6675     adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
6676     pci_set_drvdata(pdev, adapter);
6677 
6678     if (func != ent->driver_data) {
6679         pci_disable_device(pdev);
6680         pci_save_state(pdev);        /* to restore SR-IOV later */
6681         return 0;
6682     }
6683 
6684     err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
6685     if (err) {
6686         dev_err(&pdev->dev, "no usable DMA configuration\n");
6687         goto out_free_adapter;
6688     }
6689 
6690     pci_enable_pcie_error_reporting(pdev);
6691     pci_set_master(pdev);
6692     pci_save_state(pdev);
6693     adap_idx++;
6694     adapter->workq = create_singlethread_workqueue("cxgb4");
6695     if (!adapter->workq) {
6696         err = -ENOMEM;
6697         goto out_free_adapter;
6698     }
6699 
6700     /* PCI device has been enabled */
6701     adapter->flags |= CXGB4_DEV_ENABLED;
6702     memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
6703 
6704     /* If possible, we use PCIe Relaxed Ordering Attribute to deliver
6705      * Ingress Packet Data to Free List Buffers in order to allow for
6706      * chipset performance optimizations between the Root Complex and
6707      * Memory Controllers.  (Messages to the associated Ingress Queue
6708      * notifying new Packet Placement in the Free Lists Buffers will be
6709      * send without the Relaxed Ordering Attribute thus guaranteeing that
6710      * all preceding PCIe Transaction Layer Packets will be processed
6711      * first.)  But some Root Complexes have various issues with Upstream
6712      * Transaction Layer Packets with the Relaxed Ordering Attribute set.
6713      * The PCIe devices which under the Root Complexes will be cleared the
6714      * Relaxed Ordering bit in the configuration space, So we check our
6715      * PCIe configuration space to see if it's flagged with advice against
6716      * using Relaxed Ordering.
6717      */
6718     if (!pcie_relaxed_ordering_enabled(pdev))
6719         adapter->flags |= CXGB4_ROOT_NO_RELAXED_ORDERING;
6720 
6721     spin_lock_init(&adapter->stats_lock);
6722     spin_lock_init(&adapter->tid_release_lock);
6723     spin_lock_init(&adapter->win0_lock);
6724 
6725     INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
6726     INIT_WORK(&adapter->db_full_task, process_db_full);
6727     INIT_WORK(&adapter->db_drop_task, process_db_drop);
6728     INIT_WORK(&adapter->fatal_err_notify_task, notify_fatal_err);
6729 
6730     err = t4_prep_adapter(adapter);
6731     if (err)
6732         goto out_free_adapter;
6733 
6734     if (is_kdump_kernel()) {
6735         /* Collect hardware state and append to /proc/vmcore */
6736         err = cxgb4_cudbg_vmcore_add_dump(adapter);
6737         if (err) {
6738             dev_warn(adapter->pdev_dev,
6739                  "Fail collecting vmcore device dump, err: %d. Continuing\n",
6740                  err);
6741             err = 0;
6742         }
6743     }
6744 
6745     if (!is_t4(adapter->params.chip)) {
6746         s_qpp = (QUEUESPERPAGEPF0_S +
6747             (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
6748             adapter->pf);
6749         qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
6750               SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
6751         num_seg = PAGE_SIZE / SEGMENT_SIZE;
6752 
6753         /* Each segment size is 128B. Write coalescing is enabled only
6754          * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
6755          * queue is less no of segments that can be accommodated in
6756          * a page size.
6757          */
6758         if (qpp > num_seg) {
6759             dev_err(&pdev->dev,
6760                 "Incorrect number of egress queues per page\n");
6761             err = -EINVAL;
6762             goto out_free_adapter;
6763         }
6764         adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
6765         pci_resource_len(pdev, 2));
6766         if (!adapter->bar2) {
6767             dev_err(&pdev->dev, "cannot map device bar2 region\n");
6768             err = -ENOMEM;
6769             goto out_free_adapter;
6770         }
6771     }
6772 
6773     setup_memwin(adapter);
6774     err = adap_init0(adapter, 0);
6775     if (err)
6776         goto out_unmap_bar;
6777 
6778     setup_memwin_rdma(adapter);
6779 
6780     /* configure SGE_STAT_CFG_A to read WC stats */
6781     if (!is_t4(adapter->params.chip))
6782         t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
6783                  (is_t5(adapter->params.chip) ? STATMODE_V(0) :
6784                   T6_STATMODE_V(0)));
6785 
6786     /* Initialize hash mac addr list */
6787     INIT_LIST_HEAD(&adapter->mac_hlist);
6788 
6789     for_each_port(adapter, i) {
6790         /* For supporting MQPRIO Offload, need some extra
6791          * queues for each ETHOFLD TIDs. Keep it equal to
6792          * MAX_ATIDs for now. Once we connect to firmware
6793          * later and query the EOTID params, we'll come to
6794          * know the actual # of EOTIDs supported.
6795          */
6796         netdev = alloc_etherdev_mq(sizeof(struct port_info),
6797                        MAX_ETH_QSETS + MAX_ATIDS);
6798         if (!netdev) {
6799             err = -ENOMEM;
6800             goto out_free_dev;
6801         }
6802 
6803         SET_NETDEV_DEV(netdev, &pdev->dev);
6804 
6805         adapter->port[i] = netdev;
6806         pi = netdev_priv(netdev);
6807         pi->adapter = adapter;
6808         pi->xact_addr_filt = -1;
6809         pi->port_id = i;
6810         netdev->irq = pdev->irq;
6811 
6812         netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
6813             NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
6814             NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_GRO |
6815             NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
6816             NETIF_F_HW_TC | NETIF_F_NTUPLE | NETIF_F_HIGHDMA;
6817 
6818         if (chip_ver > CHELSIO_T5) {
6819             netdev->hw_enc_features |= NETIF_F_IP_CSUM |
6820                            NETIF_F_IPV6_CSUM |
6821                            NETIF_F_RXCSUM |
6822                            NETIF_F_GSO_UDP_TUNNEL |
6823                            NETIF_F_GSO_UDP_TUNNEL_CSUM |
6824                            NETIF_F_TSO | NETIF_F_TSO6;
6825 
6826             netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
6827                            NETIF_F_GSO_UDP_TUNNEL_CSUM |
6828                            NETIF_F_HW_TLS_RECORD;
6829 
6830             if (adapter->rawf_cnt)
6831                 netdev->udp_tunnel_nic_info = &cxgb_udp_tunnels;
6832         }
6833 
6834         netdev->features |= netdev->hw_features;
6835         netdev->vlan_features = netdev->features & VLAN_FEAT;
6836 #if IS_ENABLED(CONFIG_CHELSIO_TLS_DEVICE)
6837         if (pi->adapter->params.crypto & FW_CAPS_CONFIG_TLS_HW) {
6838             netdev->hw_features |= NETIF_F_HW_TLS_TX;
6839             netdev->tlsdev_ops = &cxgb4_ktls_ops;
6840             /* initialize the refcount */
6841             refcount_set(&pi->adapter->chcr_ktls.ktls_refcount, 0);
6842         }
6843 #endif /* CONFIG_CHELSIO_TLS_DEVICE */
6844 #if IS_ENABLED(CONFIG_CHELSIO_IPSEC_INLINE)
6845         if (pi->adapter->params.crypto & FW_CAPS_CONFIG_IPSEC_INLINE) {
6846             netdev->hw_enc_features |= NETIF_F_HW_ESP;
6847             netdev->features |= NETIF_F_HW_ESP;
6848             netdev->xfrmdev_ops = &cxgb4_xfrmdev_ops;
6849         }
6850 #endif /* CONFIG_CHELSIO_IPSEC_INLINE */
6851 
6852         netdev->priv_flags |= IFF_UNICAST_FLT;
6853 
6854         /* MTU range: 81 - 9600 */
6855         netdev->min_mtu = 81;              /* accommodate SACK */
6856         netdev->max_mtu = MAX_MTU;
6857 
6858         netdev->netdev_ops = &cxgb4_netdev_ops;
6859 #ifdef CONFIG_CHELSIO_T4_DCB
6860         netdev->dcbnl_ops = &cxgb4_dcb_ops;
6861         cxgb4_dcb_state_init(netdev);
6862         cxgb4_dcb_version_init(netdev);
6863 #endif
6864         cxgb4_set_ethtool_ops(netdev);
6865     }
6866 
6867     cxgb4_init_ethtool_dump(adapter);
6868 
6869     pci_set_drvdata(pdev, adapter);
6870 
6871     if (adapter->flags & CXGB4_FW_OK) {
6872         err = t4_port_init(adapter, func, func, 0);
6873         if (err)
6874             goto out_free_dev;
6875     } else if (adapter->params.nports == 1) {
6876         /* If we don't have a connection to the firmware -- possibly
6877          * because of an error -- grab the raw VPD parameters so we
6878          * can set the proper MAC Address on the debug network
6879          * interface that we've created.
6880          */
6881         u8 hw_addr[ETH_ALEN];
6882         u8 *na = adapter->params.vpd.na;
6883 
6884         err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
6885         if (!err) {
6886             for (i = 0; i < ETH_ALEN; i++)
6887                 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
6888                           hex2val(na[2 * i + 1]));
6889             t4_set_hw_addr(adapter, 0, hw_addr);
6890         }
6891     }
6892 
6893     if (!(adapter->flags & CXGB4_FW_OK))
6894         goto fw_attach_fail;
6895 
6896     /* Configure queues and allocate tables now, they can be needed as
6897      * soon as the first register_netdev completes.
6898      */
6899     err = cfg_queues(adapter);
6900     if (err)
6901         goto out_free_dev;
6902 
6903     adapter->smt = t4_init_smt();
6904     if (!adapter->smt) {
6905         /* We tolerate a lack of SMT, giving up some functionality */
6906         dev_warn(&pdev->dev, "could not allocate SMT, continuing\n");
6907     }
6908 
6909     adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
6910     if (!adapter->l2t) {
6911         /* We tolerate a lack of L2T, giving up some functionality */
6912         dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
6913         adapter->params.offload = 0;
6914     }
6915 
6916 #if IS_ENABLED(CONFIG_IPV6)
6917     if (chip_ver <= CHELSIO_T5 &&
6918         (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
6919         /* CLIP functionality is not present in hardware,
6920          * hence disable all offload features
6921          */
6922         dev_warn(&pdev->dev,
6923              "CLIP not enabled in hardware, continuing\n");
6924         adapter->params.offload = 0;
6925     } else {
6926         adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
6927                           adapter->clipt_end);
6928         if (!adapter->clipt) {
6929             /* We tolerate a lack of clip_table, giving up
6930              * some functionality
6931              */
6932             dev_warn(&pdev->dev,
6933                  "could not allocate Clip table, continuing\n");
6934             adapter->params.offload = 0;
6935         }
6936     }
6937 #endif
6938 
6939     for_each_port(adapter, i) {
6940         pi = adap2pinfo(adapter, i);
6941         pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
6942         if (!pi->sched_tbl)
6943             dev_warn(&pdev->dev,
6944                  "could not activate scheduling on port %d\n",
6945                  i);
6946     }
6947 
6948     if (is_offload(adapter) || is_hashfilter(adapter)) {
6949         if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
6950             u32 v;
6951 
6952             v = t4_read_reg(adapter, LE_DB_HASH_CONFIG_A);
6953             if (chip_ver <= CHELSIO_T5) {
6954                 adapter->tids.nhash = 1 << HASHTIDSIZE_G(v);
6955                 v = t4_read_reg(adapter, LE_DB_TID_HASHBASE_A);
6956                 adapter->tids.hash_base = v / 4;
6957             } else {
6958                 adapter->tids.nhash = HASHTBLSIZE_G(v) << 3;
6959                 v = t4_read_reg(adapter,
6960                         T6_LE_DB_HASH_TID_BASE_A);
6961                 adapter->tids.hash_base = v;
6962             }
6963         }
6964     }
6965 
6966     if (tid_init(&adapter->tids) < 0) {
6967         dev_warn(&pdev->dev, "could not allocate TID table, "
6968              "continuing\n");
6969         adapter->params.offload = 0;
6970     } else {
6971         adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
6972         if (!adapter->tc_u32)
6973             dev_warn(&pdev->dev,
6974                  "could not offload tc u32, continuing\n");
6975 
6976         if (cxgb4_init_tc_flower(adapter))
6977             dev_warn(&pdev->dev,
6978                  "could not offload tc flower, continuing\n");
6979 
6980         if (cxgb4_init_tc_mqprio(adapter))
6981             dev_warn(&pdev->dev,
6982                  "could not offload tc mqprio, continuing\n");
6983 
6984         if (cxgb4_init_tc_matchall(adapter))
6985             dev_warn(&pdev->dev,
6986                  "could not offload tc matchall, continuing\n");
6987         if (cxgb4_init_ethtool_filters(adapter))
6988             dev_warn(&pdev->dev,
6989                  "could not initialize ethtool filters, continuing\n");
6990     }
6991 
6992     /* See what interrupts we'll be using */
6993     if (msi > 1 && enable_msix(adapter) == 0)
6994         adapter->flags |= CXGB4_USING_MSIX;
6995     else if (msi > 0 && pci_enable_msi(pdev) == 0) {
6996         adapter->flags |= CXGB4_USING_MSI;
6997         if (msi > 1)
6998             free_msix_info(adapter);
6999     }
7000 
7001     /* check for PCI Express bandwidth capabiltites */
7002     pcie_print_link_status(pdev);
7003 
7004     cxgb4_init_mps_ref_entries(adapter);
7005 
7006     err = init_rss(adapter);
7007     if (err)
7008         goto out_free_dev;
7009 
7010     err = setup_non_data_intr(adapter);
7011     if (err) {
7012         dev_err(adapter->pdev_dev,
7013             "Non Data interrupt allocation failed, err: %d\n", err);
7014         goto out_free_dev;
7015     }
7016 
7017     err = setup_fw_sge_queues(adapter);
7018     if (err) {
7019         dev_err(adapter->pdev_dev,
7020             "FW sge queue allocation failed, err %d", err);
7021         goto out_free_dev;
7022     }
7023 
7024 fw_attach_fail:
7025     /*
7026      * The card is now ready to go.  If any errors occur during device
7027      * registration we do not fail the whole card but rather proceed only
7028      * with the ports we manage to register successfully.  However we must
7029      * register at least one net device.
7030      */
7031     for_each_port(adapter, i) {
7032         pi = adap2pinfo(adapter, i);
7033         adapter->port[i]->dev_port = pi->lport;
7034         netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
7035         netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
7036 
7037         netif_carrier_off(adapter->port[i]);
7038 
7039         err = register_netdev(adapter->port[i]);
7040         if (err)
7041             break;
7042         adapter->chan_map[pi->tx_chan] = i;
7043         print_port_info(adapter->port[i]);
7044     }
7045     if (i == 0) {
7046         dev_err(&pdev->dev, "could not register any net devices\n");
7047         goto out_free_dev;
7048     }
7049     if (err) {
7050         dev_warn(&pdev->dev, "only %d net devices registered\n", i);
7051         err = 0;
7052     }
7053 
7054     if (cxgb4_debugfs_root) {
7055         adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
7056                                cxgb4_debugfs_root);
7057         setup_debugfs(adapter);
7058     }
7059 
7060     /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
7061     pdev->needs_freset = 1;
7062 
7063     if (is_uld(adapter))
7064         cxgb4_uld_enable(adapter);
7065 
7066     if (!is_t4(adapter->params.chip))
7067         cxgb4_ptp_init(adapter);
7068 
7069     if (IS_REACHABLE(CONFIG_THERMAL) &&
7070         !is_t4(adapter->params.chip) && (adapter->flags & CXGB4_FW_OK))
7071         cxgb4_thermal_init(adapter);
7072 
7073     print_adapter_info(adapter);
7074     return 0;
7075 
7076  out_free_dev:
7077     t4_free_sge_resources(adapter);
7078     free_some_resources(adapter);
7079     if (adapter->flags & CXGB4_USING_MSIX)
7080         free_msix_info(adapter);
7081     if (adapter->num_uld || adapter->num_ofld_uld)
7082         t4_uld_mem_free(adapter);
7083  out_unmap_bar:
7084     if (!is_t4(adapter->params.chip))
7085         iounmap(adapter->bar2);
7086  out_free_adapter:
7087     if (adapter->workq)
7088         destroy_workqueue(adapter->workq);
7089 
7090     kfree(adapter->mbox_log);
7091     kfree(adapter);
7092  out_unmap_bar0:
7093     iounmap(regs);
7094  out_disable_device:
7095     pci_disable_pcie_error_reporting(pdev);
7096     pci_disable_device(pdev);
7097  out_release_regions:
7098     pci_release_regions(pdev);
7099     return err;
7100 }
7101 
7102 static void remove_one(struct pci_dev *pdev)
7103 {
7104     struct adapter *adapter = pci_get_drvdata(pdev);
7105     struct hash_mac_addr *entry, *tmp;
7106 
7107     if (!adapter) {
7108         pci_release_regions(pdev);
7109         return;
7110     }
7111 
7112     /* If we allocated filters, free up state associated with any
7113      * valid filters ...
7114      */
7115     clear_all_filters(adapter);
7116 
7117     adapter->flags |= CXGB4_SHUTTING_DOWN;
7118 
7119     if (adapter->pf == 4) {
7120         int i;
7121 
7122         /* Tear down per-adapter Work Queue first since it can contain
7123          * references to our adapter data structure.
7124          */
7125         destroy_workqueue(adapter->workq);
7126 
7127         detach_ulds(adapter);
7128 
7129         for_each_port(adapter, i)
7130             if (adapter->port[i]->reg_state == NETREG_REGISTERED)
7131                 unregister_netdev(adapter->port[i]);
7132 
7133         t4_uld_clean_up(adapter);
7134 
7135         adap_free_hma_mem(adapter);
7136 
7137         disable_interrupts(adapter);
7138 
7139         cxgb4_free_mps_ref_entries(adapter);
7140 
7141         debugfs_remove_recursive(adapter->debugfs_root);
7142 
7143         if (!is_t4(adapter->params.chip))
7144             cxgb4_ptp_stop(adapter);
7145         if (IS_REACHABLE(CONFIG_THERMAL))
7146             cxgb4_thermal_remove(adapter);
7147 
7148         if (adapter->flags & CXGB4_FULL_INIT_DONE)
7149             cxgb_down(adapter);
7150 
7151         if (adapter->flags & CXGB4_USING_MSIX)
7152             free_msix_info(adapter);
7153         if (adapter->num_uld || adapter->num_ofld_uld)
7154             t4_uld_mem_free(adapter);
7155         free_some_resources(adapter);
7156         list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist,
7157                      list) {
7158             list_del(&entry->list);
7159             kfree(entry);
7160         }
7161 
7162 #if IS_ENABLED(CONFIG_IPV6)
7163         t4_cleanup_clip_tbl(adapter);
7164 #endif
7165         if (!is_t4(adapter->params.chip))
7166             iounmap(adapter->bar2);
7167     }
7168 #ifdef CONFIG_PCI_IOV
7169     else {
7170         cxgb4_iov_configure(adapter->pdev, 0);
7171     }
7172 #endif
7173     iounmap(adapter->regs);
7174     pci_disable_pcie_error_reporting(pdev);
7175     if ((adapter->flags & CXGB4_DEV_ENABLED)) {
7176         pci_disable_device(pdev);
7177         adapter->flags &= ~CXGB4_DEV_ENABLED;
7178     }
7179     pci_release_regions(pdev);
7180     kfree(adapter->mbox_log);
7181     synchronize_rcu();
7182     kfree(adapter);
7183 }
7184 
7185 /* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
7186  * delivery.  This is essentially a stripped down version of the PCI remove()
7187  * function where we do the minimal amount of work necessary to shutdown any
7188  * further activity.
7189  */
7190 static void shutdown_one(struct pci_dev *pdev)
7191 {
7192     struct adapter *adapter = pci_get_drvdata(pdev);
7193 
7194     /* As with remove_one() above (see extended comment), we only want do
7195      * do cleanup on PCI Devices which went all the way through init_one()
7196      * ...
7197      */
7198     if (!adapter) {
7199         pci_release_regions(pdev);
7200         return;
7201     }
7202 
7203     adapter->flags |= CXGB4_SHUTTING_DOWN;
7204 
7205     if (adapter->pf == 4) {
7206         int i;
7207 
7208         for_each_port(adapter, i)
7209             if (adapter->port[i]->reg_state == NETREG_REGISTERED)
7210                 cxgb_close(adapter->port[i]);
7211 
7212         rtnl_lock();
7213         cxgb4_mqprio_stop_offload(adapter);
7214         rtnl_unlock();
7215 
7216         if (is_uld(adapter)) {
7217             detach_ulds(adapter);
7218             t4_uld_clean_up(adapter);
7219         }
7220 
7221         disable_interrupts(adapter);
7222         disable_msi(adapter);
7223 
7224         t4_sge_stop(adapter);
7225         if (adapter->flags & CXGB4_FW_OK)
7226             t4_fw_bye(adapter, adapter->mbox);
7227     }
7228 }
7229 
7230 static struct pci_driver cxgb4_driver = {
7231     .name     = KBUILD_MODNAME,
7232     .id_table = cxgb4_pci_tbl,
7233     .probe    = init_one,
7234     .remove   = remove_one,
7235     .shutdown = shutdown_one,
7236 #ifdef CONFIG_PCI_IOV
7237     .sriov_configure = cxgb4_iov_configure,
7238 #endif
7239     .err_handler = &cxgb4_eeh,
7240 };
7241 
7242 static int __init cxgb4_init_module(void)
7243 {
7244     int ret;
7245 
7246     cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
7247 
7248     ret = pci_register_driver(&cxgb4_driver);
7249     if (ret < 0)
7250         goto err_pci;
7251 
7252 #if IS_ENABLED(CONFIG_IPV6)
7253     if (!inet6addr_registered) {
7254         ret = register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
7255         if (ret)
7256             pci_unregister_driver(&cxgb4_driver);
7257         else
7258             inet6addr_registered = true;
7259     }
7260 #endif
7261 
7262     if (ret == 0)
7263         return ret;
7264 
7265 err_pci:
7266     debugfs_remove(cxgb4_debugfs_root);
7267 
7268     return ret;
7269 }
7270 
7271 static void __exit cxgb4_cleanup_module(void)
7272 {
7273 #if IS_ENABLED(CONFIG_IPV6)
7274     if (inet6addr_registered) {
7275         unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
7276         inet6addr_registered = false;
7277     }
7278 #endif
7279     pci_unregister_driver(&cxgb4_driver);
7280     debugfs_remove(cxgb4_debugfs_root);  /* NULL ok */
7281 }
7282 
7283 module_init(cxgb4_init_module);
7284 module_exit(cxgb4_cleanup_module);