Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * Copyright (C) 2015 Cavium, Inc.
0004  */
0005 
0006 #include <linux/module.h>
0007 #include <linux/interrupt.h>
0008 #include <linux/pci.h>
0009 #include <linux/netdevice.h>
0010 #include <linux/if_vlan.h>
0011 #include <linux/etherdevice.h>
0012 #include <linux/ethtool.h>
0013 #include <linux/log2.h>
0014 #include <linux/prefetch.h>
0015 #include <linux/irq.h>
0016 #include <linux/iommu.h>
0017 #include <linux/bpf.h>
0018 #include <linux/bpf_trace.h>
0019 #include <linux/filter.h>
0020 #include <linux/net_tstamp.h>
0021 #include <linux/workqueue.h>
0022 
0023 #include "nic_reg.h"
0024 #include "nic.h"
0025 #include "nicvf_queues.h"
0026 #include "thunder_bgx.h"
0027 #include "../common/cavium_ptp.h"
0028 
0029 #define DRV_NAME    "nicvf"
0030 #define DRV_VERSION "1.0"
0031 
0032 /* NOTE: Packets bigger than 1530 are split across multiple pages and XDP needs
0033  * the buffer to be contiguous. Allow XDP to be set up only if we don't exceed
0034  * this value, keeping headroom for the 14 byte Ethernet header and two
0035  * VLAN tags (for QinQ)
0036  */
0037 #define MAX_XDP_MTU (1530 - ETH_HLEN - VLAN_HLEN * 2)
0038 
0039 /* Supported devices */
0040 static const struct pci_device_id nicvf_id_table[] = {
0041     { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
0042              PCI_DEVICE_ID_THUNDER_NIC_VF,
0043              PCI_VENDOR_ID_CAVIUM,
0044              PCI_SUBSYS_DEVID_88XX_NIC_VF) },
0045     { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
0046              PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF,
0047              PCI_VENDOR_ID_CAVIUM,
0048              PCI_SUBSYS_DEVID_88XX_PASS1_NIC_VF) },
0049     { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
0050              PCI_DEVICE_ID_THUNDER_NIC_VF,
0051              PCI_VENDOR_ID_CAVIUM,
0052              PCI_SUBSYS_DEVID_81XX_NIC_VF) },
0053     { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM,
0054              PCI_DEVICE_ID_THUNDER_NIC_VF,
0055              PCI_VENDOR_ID_CAVIUM,
0056              PCI_SUBSYS_DEVID_83XX_NIC_VF) },
0057     { 0, }  /* end of table */
0058 };
0059 
0060 MODULE_AUTHOR("Sunil Goutham");
0061 MODULE_DESCRIPTION("Cavium Thunder NIC Virtual Function Driver");
0062 MODULE_LICENSE("GPL v2");
0063 MODULE_VERSION(DRV_VERSION);
0064 MODULE_DEVICE_TABLE(pci, nicvf_id_table);
0065 
0066 static int debug = 0x00;
0067 module_param(debug, int, 0644);
0068 MODULE_PARM_DESC(debug, "Debug message level bitmap");
0069 
0070 static int cpi_alg = CPI_ALG_NONE;
0071 module_param(cpi_alg, int, 0444);
0072 MODULE_PARM_DESC(cpi_alg,
0073          "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)");
0074 
0075 static inline u8 nicvf_netdev_qidx(struct nicvf *nic, u8 qidx)
0076 {
0077     if (nic->sqs_mode)
0078         return qidx + ((nic->sqs_id + 1) * MAX_CMP_QUEUES_PER_QS);
0079     else
0080         return qidx;
0081 }
0082 
0083 /* The Cavium ThunderX network controller can *only* be found in SoCs
0084  * containing the ThunderX ARM64 CPU implementation.  All accesses to the device
0085  * registers on this platform are implicitly strongly ordered with respect
0086  * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use
0087  * with no memory barriers in this driver.  The readq()/writeq() functions add
0088  * explicit ordering operation which in this case are redundant, and only
0089  * add overhead.
0090  */
0091 
0092 /* Register read/write APIs */
0093 void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val)
0094 {
0095     writeq_relaxed(val, nic->reg_base + offset);
0096 }
0097 
0098 u64 nicvf_reg_read(struct nicvf *nic, u64 offset)
0099 {
0100     return readq_relaxed(nic->reg_base + offset);
0101 }
0102 
0103 void nicvf_queue_reg_write(struct nicvf *nic, u64 offset,
0104                u64 qidx, u64 val)
0105 {
0106     void __iomem *addr = nic->reg_base + offset;
0107 
0108     writeq_relaxed(val, addr + (qidx << NIC_Q_NUM_SHIFT));
0109 }
0110 
0111 u64 nicvf_queue_reg_read(struct nicvf *nic, u64 offset, u64 qidx)
0112 {
0113     void __iomem *addr = nic->reg_base + offset;
0114 
0115     return readq_relaxed(addr + (qidx << NIC_Q_NUM_SHIFT));
0116 }
0117 
0118 /* VF -> PF mailbox communication */
0119 static void nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx)
0120 {
0121     u64 *msg = (u64 *)mbx;
0122 
0123     nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]);
0124     nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]);
0125 }
0126 
0127 int nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx)
0128 {
0129     unsigned long timeout;
0130     int ret = 0;
0131 
0132     mutex_lock(&nic->rx_mode_mtx);
0133 
0134     nic->pf_acked = false;
0135     nic->pf_nacked = false;
0136 
0137     nicvf_write_to_mbx(nic, mbx);
0138 
0139     timeout = jiffies + msecs_to_jiffies(NIC_MBOX_MSG_TIMEOUT);
0140     /* Wait for previous message to be acked, timeout 2sec */
0141     while (!nic->pf_acked) {
0142         if (nic->pf_nacked) {
0143             netdev_err(nic->netdev,
0144                    "PF NACK to mbox msg 0x%02x from VF%d\n",
0145                    (mbx->msg.msg & 0xFF), nic->vf_id);
0146             ret = -EINVAL;
0147             break;
0148         }
0149         usleep_range(8000, 10000);
0150         if (nic->pf_acked)
0151             break;
0152         if (time_after(jiffies, timeout)) {
0153             netdev_err(nic->netdev,
0154                    "PF didn't ACK to mbox msg 0x%02x from VF%d\n",
0155                    (mbx->msg.msg & 0xFF), nic->vf_id);
0156             ret = -EBUSY;
0157             break;
0158         }
0159     }
0160     mutex_unlock(&nic->rx_mode_mtx);
0161     return ret;
0162 }
0163 
0164 /* Checks if VF is able to comminicate with PF
0165 * and also gets the VNIC number this VF is associated to.
0166 */
0167 static int nicvf_check_pf_ready(struct nicvf *nic)
0168 {
0169     union nic_mbx mbx = {};
0170 
0171     mbx.msg.msg = NIC_MBOX_MSG_READY;
0172     if (nicvf_send_msg_to_pf(nic, &mbx)) {
0173         netdev_err(nic->netdev,
0174                "PF didn't respond to READY msg\n");
0175         return 0;
0176     }
0177 
0178     return 1;
0179 }
0180 
0181 static void nicvf_send_cfg_done(struct nicvf *nic)
0182 {
0183     union nic_mbx mbx = {};
0184 
0185     mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE;
0186     if (nicvf_send_msg_to_pf(nic, &mbx)) {
0187         netdev_err(nic->netdev,
0188                "PF didn't respond to CFG DONE msg\n");
0189     }
0190 }
0191 
0192 static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx)
0193 {
0194     if (bgx->rx)
0195         nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats;
0196     else
0197         nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats;
0198 }
0199 
0200 static void  nicvf_handle_mbx_intr(struct nicvf *nic)
0201 {
0202     union nic_mbx mbx = {};
0203     u64 *mbx_data;
0204     u64 mbx_addr;
0205     int i;
0206 
0207     mbx_addr = NIC_VF_PF_MAILBOX_0_1;
0208     mbx_data = (u64 *)&mbx;
0209 
0210     for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) {
0211         *mbx_data = nicvf_reg_read(nic, mbx_addr);
0212         mbx_data++;
0213         mbx_addr += sizeof(u64);
0214     }
0215 
0216     netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg);
0217     switch (mbx.msg.msg) {
0218     case NIC_MBOX_MSG_READY:
0219         nic->pf_acked = true;
0220         nic->vf_id = mbx.nic_cfg.vf_id & 0x7F;
0221         nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F;
0222         nic->node = mbx.nic_cfg.node_id;
0223         if (!nic->set_mac_pending)
0224             eth_hw_addr_set(nic->netdev, mbx.nic_cfg.mac_addr);
0225         nic->sqs_mode = mbx.nic_cfg.sqs_mode;
0226         nic->loopback_supported = mbx.nic_cfg.loopback_supported;
0227         nic->link_up = false;
0228         nic->duplex = 0;
0229         nic->speed = 0;
0230         break;
0231     case NIC_MBOX_MSG_ACK:
0232         nic->pf_acked = true;
0233         break;
0234     case NIC_MBOX_MSG_NACK:
0235         nic->pf_nacked = true;
0236         break;
0237     case NIC_MBOX_MSG_RSS_SIZE:
0238         nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size;
0239         nic->pf_acked = true;
0240         break;
0241     case NIC_MBOX_MSG_BGX_STATS:
0242         nicvf_read_bgx_stats(nic, &mbx.bgx_stats);
0243         nic->pf_acked = true;
0244         break;
0245     case NIC_MBOX_MSG_BGX_LINK_CHANGE:
0246         nic->pf_acked = true;
0247         if (nic->link_up != mbx.link_status.link_up) {
0248             nic->link_up = mbx.link_status.link_up;
0249             nic->duplex = mbx.link_status.duplex;
0250             nic->speed = mbx.link_status.speed;
0251             nic->mac_type = mbx.link_status.mac_type;
0252             if (nic->link_up) {
0253                 netdev_info(nic->netdev,
0254                         "Link is Up %d Mbps %s duplex\n",
0255                         nic->speed,
0256                         nic->duplex == DUPLEX_FULL ?
0257                         "Full" : "Half");
0258                 netif_carrier_on(nic->netdev);
0259                 netif_tx_start_all_queues(nic->netdev);
0260             } else {
0261                 netdev_info(nic->netdev, "Link is Down\n");
0262                 netif_carrier_off(nic->netdev);
0263                 netif_tx_stop_all_queues(nic->netdev);
0264             }
0265         }
0266         break;
0267     case NIC_MBOX_MSG_ALLOC_SQS:
0268         nic->sqs_count = mbx.sqs_alloc.qs_count;
0269         nic->pf_acked = true;
0270         break;
0271     case NIC_MBOX_MSG_SNICVF_PTR:
0272         /* Primary VF: make note of secondary VF's pointer
0273          * to be used while packet transmission.
0274          */
0275         nic->snicvf[mbx.nicvf.sqs_id] =
0276             (struct nicvf *)mbx.nicvf.nicvf;
0277         nic->pf_acked = true;
0278         break;
0279     case NIC_MBOX_MSG_PNICVF_PTR:
0280         /* Secondary VF/Qset: make note of primary VF's pointer
0281          * to be used while packet reception, to handover packet
0282          * to primary VF's netdev.
0283          */
0284         nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf;
0285         nic->pf_acked = true;
0286         break;
0287     case NIC_MBOX_MSG_PFC:
0288         nic->pfc.autoneg = mbx.pfc.autoneg;
0289         nic->pfc.fc_rx = mbx.pfc.fc_rx;
0290         nic->pfc.fc_tx = mbx.pfc.fc_tx;
0291         nic->pf_acked = true;
0292         break;
0293     default:
0294         netdev_err(nic->netdev,
0295                "Invalid message from PF, msg 0x%x\n", mbx.msg.msg);
0296         break;
0297     }
0298     nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0);
0299 }
0300 
0301 static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev)
0302 {
0303     union nic_mbx mbx = {};
0304 
0305     mbx.mac.msg = NIC_MBOX_MSG_SET_MAC;
0306     mbx.mac.vf_id = nic->vf_id;
0307     ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr);
0308 
0309     return nicvf_send_msg_to_pf(nic, &mbx);
0310 }
0311 
0312 static void nicvf_config_cpi(struct nicvf *nic)
0313 {
0314     union nic_mbx mbx = {};
0315 
0316     mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG;
0317     mbx.cpi_cfg.vf_id = nic->vf_id;
0318     mbx.cpi_cfg.cpi_alg = nic->cpi_alg;
0319     mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt;
0320 
0321     nicvf_send_msg_to_pf(nic, &mbx);
0322 }
0323 
0324 static void nicvf_get_rss_size(struct nicvf *nic)
0325 {
0326     union nic_mbx mbx = {};
0327 
0328     mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE;
0329     mbx.rss_size.vf_id = nic->vf_id;
0330     nicvf_send_msg_to_pf(nic, &mbx);
0331 }
0332 
0333 void nicvf_config_rss(struct nicvf *nic)
0334 {
0335     union nic_mbx mbx = {};
0336     struct nicvf_rss_info *rss = &nic->rss_info;
0337     int ind_tbl_len = rss->rss_size;
0338     int i, nextq = 0;
0339 
0340     mbx.rss_cfg.vf_id = nic->vf_id;
0341     mbx.rss_cfg.hash_bits = rss->hash_bits;
0342     while (ind_tbl_len) {
0343         mbx.rss_cfg.tbl_offset = nextq;
0344         mbx.rss_cfg.tbl_len = min(ind_tbl_len,
0345                            RSS_IND_TBL_LEN_PER_MBX_MSG);
0346         mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ?
0347               NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG;
0348 
0349         for (i = 0; i < mbx.rss_cfg.tbl_len; i++)
0350             mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++];
0351 
0352         nicvf_send_msg_to_pf(nic, &mbx);
0353 
0354         ind_tbl_len -= mbx.rss_cfg.tbl_len;
0355     }
0356 }
0357 
0358 void nicvf_set_rss_key(struct nicvf *nic)
0359 {
0360     struct nicvf_rss_info *rss = &nic->rss_info;
0361     u64 key_addr = NIC_VNIC_RSS_KEY_0_4;
0362     int idx;
0363 
0364     for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) {
0365         nicvf_reg_write(nic, key_addr, rss->key[idx]);
0366         key_addr += sizeof(u64);
0367     }
0368 }
0369 
0370 static int nicvf_rss_init(struct nicvf *nic)
0371 {
0372     struct nicvf_rss_info *rss = &nic->rss_info;
0373     int idx;
0374 
0375     nicvf_get_rss_size(nic);
0376 
0377     if (cpi_alg != CPI_ALG_NONE) {
0378         rss->enable = false;
0379         rss->hash_bits = 0;
0380         return 0;
0381     }
0382 
0383     rss->enable = true;
0384 
0385     netdev_rss_key_fill(rss->key, RSS_HASH_KEY_SIZE * sizeof(u64));
0386     nicvf_set_rss_key(nic);
0387 
0388     rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA;
0389     nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg);
0390 
0391     rss->hash_bits =  ilog2(rounddown_pow_of_two(rss->rss_size));
0392 
0393     for (idx = 0; idx < rss->rss_size; idx++)
0394         rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx,
0395                                    nic->rx_queues);
0396     nicvf_config_rss(nic);
0397     return 1;
0398 }
0399 
0400 /* Request PF to allocate additional Qsets */
0401 static void nicvf_request_sqs(struct nicvf *nic)
0402 {
0403     union nic_mbx mbx = {};
0404     int sqs;
0405     int sqs_count = nic->sqs_count;
0406     int rx_queues = 0, tx_queues = 0;
0407 
0408     /* Only primary VF should request */
0409     if (nic->sqs_mode ||  !nic->sqs_count)
0410         return;
0411 
0412     mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS;
0413     mbx.sqs_alloc.vf_id = nic->vf_id;
0414     mbx.sqs_alloc.qs_count = nic->sqs_count;
0415     if (nicvf_send_msg_to_pf(nic, &mbx)) {
0416         /* No response from PF */
0417         nic->sqs_count = 0;
0418         return;
0419     }
0420 
0421     /* Return if no Secondary Qsets available */
0422     if (!nic->sqs_count)
0423         return;
0424 
0425     if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS)
0426         rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS;
0427 
0428     tx_queues = nic->tx_queues + nic->xdp_tx_queues;
0429     if (tx_queues > MAX_SND_QUEUES_PER_QS)
0430         tx_queues = tx_queues - MAX_SND_QUEUES_PER_QS;
0431 
0432     /* Set no of Rx/Tx queues in each of the SQsets */
0433     for (sqs = 0; sqs < nic->sqs_count; sqs++) {
0434         mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR;
0435         mbx.nicvf.vf_id = nic->vf_id;
0436         mbx.nicvf.sqs_id = sqs;
0437         nicvf_send_msg_to_pf(nic, &mbx);
0438 
0439         nic->snicvf[sqs]->sqs_id = sqs;
0440         if (rx_queues > MAX_RCV_QUEUES_PER_QS) {
0441             nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS;
0442             rx_queues -= MAX_RCV_QUEUES_PER_QS;
0443         } else {
0444             nic->snicvf[sqs]->qs->rq_cnt = rx_queues;
0445             rx_queues = 0;
0446         }
0447 
0448         if (tx_queues > MAX_SND_QUEUES_PER_QS) {
0449             nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS;
0450             tx_queues -= MAX_SND_QUEUES_PER_QS;
0451         } else {
0452             nic->snicvf[sqs]->qs->sq_cnt = tx_queues;
0453             tx_queues = 0;
0454         }
0455 
0456         nic->snicvf[sqs]->qs->cq_cnt =
0457         max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt);
0458 
0459         /* Initialize secondary Qset's queues and its interrupts */
0460         nicvf_open(nic->snicvf[sqs]->netdev);
0461     }
0462 
0463     /* Update stack with actual Rx/Tx queue count allocated */
0464     if (sqs_count != nic->sqs_count)
0465         nicvf_set_real_num_queues(nic->netdev,
0466                       nic->tx_queues, nic->rx_queues);
0467 }
0468 
0469 /* Send this Qset's nicvf pointer to PF.
0470  * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs
0471  * so that packets received by these Qsets can use primary VF's netdev
0472  */
0473 static void nicvf_send_vf_struct(struct nicvf *nic)
0474 {
0475     union nic_mbx mbx = {};
0476 
0477     mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR;
0478     mbx.nicvf.sqs_mode = nic->sqs_mode;
0479     mbx.nicvf.nicvf = (u64)nic;
0480     nicvf_send_msg_to_pf(nic, &mbx);
0481 }
0482 
0483 static void nicvf_get_primary_vf_struct(struct nicvf *nic)
0484 {
0485     union nic_mbx mbx = {};
0486 
0487     mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR;
0488     nicvf_send_msg_to_pf(nic, &mbx);
0489 }
0490 
0491 int nicvf_set_real_num_queues(struct net_device *netdev,
0492                   int tx_queues, int rx_queues)
0493 {
0494     int err = 0;
0495 
0496     err = netif_set_real_num_tx_queues(netdev, tx_queues);
0497     if (err) {
0498         netdev_err(netdev,
0499                "Failed to set no of Tx queues: %d\n", tx_queues);
0500         return err;
0501     }
0502 
0503     err = netif_set_real_num_rx_queues(netdev, rx_queues);
0504     if (err)
0505         netdev_err(netdev,
0506                "Failed to set no of Rx queues: %d\n", rx_queues);
0507     return err;
0508 }
0509 
0510 static int nicvf_init_resources(struct nicvf *nic)
0511 {
0512     int err;
0513 
0514     /* Enable Qset */
0515     nicvf_qset_config(nic, true);
0516 
0517     /* Initialize queues and HW for data transfer */
0518     err = nicvf_config_data_transfer(nic, true);
0519     if (err) {
0520         netdev_err(nic->netdev,
0521                "Failed to alloc/config VF's QSet resources\n");
0522         return err;
0523     }
0524 
0525     return 0;
0526 }
0527 
0528 static inline bool nicvf_xdp_rx(struct nicvf *nic, struct bpf_prog *prog,
0529                 struct cqe_rx_t *cqe_rx, struct snd_queue *sq,
0530                 struct rcv_queue *rq, struct sk_buff **skb)
0531 {
0532     unsigned char *hard_start, *data;
0533     struct xdp_buff xdp;
0534     struct page *page;
0535     u32 action;
0536     u16 len, offset = 0;
0537     u64 dma_addr, cpu_addr;
0538     void *orig_data;
0539 
0540     /* Retrieve packet buffer's DMA address and length */
0541     len = *((u16 *)((void *)cqe_rx + (3 * sizeof(u64))));
0542     dma_addr = *((u64 *)((void *)cqe_rx + (7 * sizeof(u64))));
0543 
0544     cpu_addr = nicvf_iova_to_phys(nic, dma_addr);
0545     if (!cpu_addr)
0546         return false;
0547     cpu_addr = (u64)phys_to_virt(cpu_addr);
0548     page = virt_to_page((void *)cpu_addr);
0549 
0550     xdp_init_buff(&xdp, RCV_FRAG_LEN + XDP_PACKET_HEADROOM,
0551               &rq->xdp_rxq);
0552     hard_start = page_address(page);
0553     data = (unsigned char *)cpu_addr;
0554     xdp_prepare_buff(&xdp, hard_start, data - hard_start, len, false);
0555     orig_data = xdp.data;
0556 
0557     action = bpf_prog_run_xdp(prog, &xdp);
0558 
0559     len = xdp.data_end - xdp.data;
0560     /* Check if XDP program has changed headers */
0561     if (orig_data != xdp.data) {
0562         offset = orig_data - xdp.data;
0563         dma_addr -= offset;
0564     }
0565 
0566     switch (action) {
0567     case XDP_PASS:
0568         /* Check if it's a recycled page, if not
0569          * unmap the DMA mapping.
0570          *
0571          * Recycled page holds an extra reference.
0572          */
0573         if (page_ref_count(page) == 1) {
0574             dma_addr &= PAGE_MASK;
0575             dma_unmap_page_attrs(&nic->pdev->dev, dma_addr,
0576                          RCV_FRAG_LEN + XDP_PACKET_HEADROOM,
0577                          DMA_FROM_DEVICE,
0578                          DMA_ATTR_SKIP_CPU_SYNC);
0579         }
0580 
0581         /* Build SKB and pass on packet to network stack */
0582         *skb = build_skb(xdp.data,
0583                  RCV_FRAG_LEN - cqe_rx->align_pad + offset);
0584         if (!*skb)
0585             put_page(page);
0586         else
0587             skb_put(*skb, len);
0588         return false;
0589     case XDP_TX:
0590         nicvf_xdp_sq_append_pkt(nic, sq, (u64)xdp.data, dma_addr, len);
0591         return true;
0592     default:
0593         bpf_warn_invalid_xdp_action(nic->netdev, prog, action);
0594         fallthrough;
0595     case XDP_ABORTED:
0596         trace_xdp_exception(nic->netdev, prog, action);
0597         fallthrough;
0598     case XDP_DROP:
0599         /* Check if it's a recycled page, if not
0600          * unmap the DMA mapping.
0601          *
0602          * Recycled page holds an extra reference.
0603          */
0604         if (page_ref_count(page) == 1) {
0605             dma_addr &= PAGE_MASK;
0606             dma_unmap_page_attrs(&nic->pdev->dev, dma_addr,
0607                          RCV_FRAG_LEN + XDP_PACKET_HEADROOM,
0608                          DMA_FROM_DEVICE,
0609                          DMA_ATTR_SKIP_CPU_SYNC);
0610         }
0611         put_page(page);
0612         return true;
0613     }
0614     return false;
0615 }
0616 
0617 static void nicvf_snd_ptp_handler(struct net_device *netdev,
0618                   struct cqe_send_t *cqe_tx)
0619 {
0620     struct nicvf *nic = netdev_priv(netdev);
0621     struct skb_shared_hwtstamps ts;
0622     u64 ns;
0623 
0624     nic = nic->pnicvf;
0625 
0626     /* Sync for 'ptp_skb' */
0627     smp_rmb();
0628 
0629     /* New timestamp request can be queued now */
0630     atomic_set(&nic->tx_ptp_skbs, 0);
0631 
0632     /* Check for timestamp requested skb */
0633     if (!nic->ptp_skb)
0634         return;
0635 
0636     /* Check if timestamping is timedout, which is set to 10us */
0637     if (cqe_tx->send_status == CQ_TX_ERROP_TSTMP_TIMEOUT ||
0638         cqe_tx->send_status == CQ_TX_ERROP_TSTMP_CONFLICT)
0639         goto no_tstamp;
0640 
0641     /* Get the timestamp */
0642     memset(&ts, 0, sizeof(ts));
0643     ns = cavium_ptp_tstamp2time(nic->ptp_clock, cqe_tx->ptp_timestamp);
0644     ts.hwtstamp = ns_to_ktime(ns);
0645     skb_tstamp_tx(nic->ptp_skb, &ts);
0646 
0647 no_tstamp:
0648     /* Free the original skb */
0649     dev_kfree_skb_any(nic->ptp_skb);
0650     nic->ptp_skb = NULL;
0651     /* Sync 'ptp_skb' */
0652     smp_wmb();
0653 }
0654 
0655 static void nicvf_snd_pkt_handler(struct net_device *netdev,
0656                   struct cqe_send_t *cqe_tx,
0657                   int budget, int *subdesc_cnt,
0658                   unsigned int *tx_pkts, unsigned int *tx_bytes)
0659 {
0660     struct sk_buff *skb = NULL;
0661     struct page *page;
0662     struct nicvf *nic = netdev_priv(netdev);
0663     struct snd_queue *sq;
0664     struct sq_hdr_subdesc *hdr;
0665     struct sq_hdr_subdesc *tso_sqe;
0666 
0667     sq = &nic->qs->sq[cqe_tx->sq_idx];
0668 
0669     hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr);
0670     if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER)
0671         return;
0672 
0673     /* Check for errors */
0674     if (cqe_tx->send_status)
0675         nicvf_check_cqe_tx_errs(nic->pnicvf, cqe_tx);
0676 
0677     /* Is this a XDP designated Tx queue */
0678     if (sq->is_xdp) {
0679         page = (struct page *)sq->xdp_page[cqe_tx->sqe_ptr];
0680         /* Check if it's recycled page or else unmap DMA mapping */
0681         if (page && (page_ref_count(page) == 1))
0682             nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr,
0683                          hdr->subdesc_cnt);
0684 
0685         /* Release page reference for recycling */
0686         if (page)
0687             put_page(page);
0688         sq->xdp_page[cqe_tx->sqe_ptr] = (u64)NULL;
0689         *subdesc_cnt += hdr->subdesc_cnt + 1;
0690         return;
0691     }
0692 
0693     skb = (struct sk_buff *)sq->skbuff[cqe_tx->sqe_ptr];
0694     if (skb) {
0695         /* Check for dummy descriptor used for HW TSO offload on 88xx */
0696         if (hdr->dont_send) {
0697             /* Get actual TSO descriptors and free them */
0698             tso_sqe =
0699              (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, hdr->rsvd2);
0700             nicvf_unmap_sndq_buffers(nic, sq, hdr->rsvd2,
0701                          tso_sqe->subdesc_cnt);
0702             *subdesc_cnt += tso_sqe->subdesc_cnt + 1;
0703         } else {
0704             nicvf_unmap_sndq_buffers(nic, sq, cqe_tx->sqe_ptr,
0705                          hdr->subdesc_cnt);
0706         }
0707         *subdesc_cnt += hdr->subdesc_cnt + 1;
0708         prefetch(skb);
0709         (*tx_pkts)++;
0710         *tx_bytes += skb->len;
0711         /* If timestamp is requested for this skb, don't free it */
0712         if (skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS &&
0713             !nic->pnicvf->ptp_skb)
0714             nic->pnicvf->ptp_skb = skb;
0715         else
0716             napi_consume_skb(skb, budget);
0717         sq->skbuff[cqe_tx->sqe_ptr] = (u64)NULL;
0718     } else {
0719         /* In case of SW TSO on 88xx, only last segment will have
0720          * a SKB attached, so just free SQEs here.
0721          */
0722         if (!nic->hw_tso)
0723             *subdesc_cnt += hdr->subdesc_cnt + 1;
0724     }
0725 }
0726 
0727 static inline void nicvf_set_rxhash(struct net_device *netdev,
0728                     struct cqe_rx_t *cqe_rx,
0729                     struct sk_buff *skb)
0730 {
0731     u8 hash_type;
0732     u32 hash;
0733 
0734     if (!(netdev->features & NETIF_F_RXHASH))
0735         return;
0736 
0737     switch (cqe_rx->rss_alg) {
0738     case RSS_ALG_TCP_IP:
0739     case RSS_ALG_UDP_IP:
0740         hash_type = PKT_HASH_TYPE_L4;
0741         hash = cqe_rx->rss_tag;
0742         break;
0743     case RSS_ALG_IP:
0744         hash_type = PKT_HASH_TYPE_L3;
0745         hash = cqe_rx->rss_tag;
0746         break;
0747     default:
0748         hash_type = PKT_HASH_TYPE_NONE;
0749         hash = 0;
0750     }
0751 
0752     skb_set_hash(skb, hash, hash_type);
0753 }
0754 
0755 static inline void nicvf_set_rxtstamp(struct nicvf *nic, struct sk_buff *skb)
0756 {
0757     u64 ns;
0758 
0759     if (!nic->ptp_clock || !nic->hw_rx_tstamp)
0760         return;
0761 
0762     /* The first 8 bytes is the timestamp */
0763     ns = cavium_ptp_tstamp2time(nic->ptp_clock,
0764                     be64_to_cpu(*(__be64 *)skb->data));
0765     skb_hwtstamps(skb)->hwtstamp = ns_to_ktime(ns);
0766 
0767     __skb_pull(skb, 8);
0768 }
0769 
0770 static void nicvf_rcv_pkt_handler(struct net_device *netdev,
0771                   struct napi_struct *napi,
0772                   struct cqe_rx_t *cqe_rx,
0773                   struct snd_queue *sq, struct rcv_queue *rq)
0774 {
0775     struct sk_buff *skb = NULL;
0776     struct nicvf *nic = netdev_priv(netdev);
0777     struct nicvf *snic = nic;
0778     int err = 0;
0779     int rq_idx;
0780 
0781     rq_idx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx);
0782 
0783     if (nic->sqs_mode) {
0784         /* Use primary VF's 'nicvf' struct */
0785         nic = nic->pnicvf;
0786         netdev = nic->netdev;
0787     }
0788 
0789     /* Check for errors */
0790     if (cqe_rx->err_level || cqe_rx->err_opcode) {
0791         err = nicvf_check_cqe_rx_errs(nic, cqe_rx);
0792         if (err && !cqe_rx->rb_cnt)
0793             return;
0794     }
0795 
0796     /* For XDP, ignore pkts spanning multiple pages */
0797     if (nic->xdp_prog && (cqe_rx->rb_cnt == 1)) {
0798         /* Packet consumed by XDP */
0799         if (nicvf_xdp_rx(snic, nic->xdp_prog, cqe_rx, sq, rq, &skb))
0800             return;
0801     } else {
0802         skb = nicvf_get_rcv_skb(snic, cqe_rx,
0803                     nic->xdp_prog ? true : false);
0804     }
0805 
0806     if (!skb)
0807         return;
0808 
0809     if (netif_msg_pktdata(nic)) {
0810         netdev_info(nic->netdev, "skb 0x%p, len=%d\n", skb, skb->len);
0811         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
0812                    skb->data, skb->len, true);
0813     }
0814 
0815     /* If error packet, drop it here */
0816     if (err) {
0817         dev_kfree_skb_any(skb);
0818         return;
0819     }
0820 
0821     nicvf_set_rxtstamp(nic, skb);
0822     nicvf_set_rxhash(netdev, cqe_rx, skb);
0823 
0824     skb_record_rx_queue(skb, rq_idx);
0825     if (netdev->hw_features & NETIF_F_RXCSUM) {
0826         /* HW by default verifies TCP/UDP/SCTP checksums */
0827         skb->ip_summed = CHECKSUM_UNNECESSARY;
0828     } else {
0829         skb_checksum_none_assert(skb);
0830     }
0831 
0832     skb->protocol = eth_type_trans(skb, netdev);
0833 
0834     /* Check for stripped VLAN */
0835     if (cqe_rx->vlan_found && cqe_rx->vlan_stripped)
0836         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
0837                        ntohs((__force __be16)cqe_rx->vlan_tci));
0838 
0839     if (napi && (netdev->features & NETIF_F_GRO))
0840         napi_gro_receive(napi, skb);
0841     else
0842         netif_receive_skb(skb);
0843 }
0844 
0845 static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx,
0846                  struct napi_struct *napi, int budget)
0847 {
0848     int processed_cqe, work_done = 0, tx_done = 0;
0849     int cqe_count, cqe_head;
0850     int subdesc_cnt = 0;
0851     struct nicvf *nic = netdev_priv(netdev);
0852     struct queue_set *qs = nic->qs;
0853     struct cmp_queue *cq = &qs->cq[cq_idx];
0854     struct cqe_rx_t *cq_desc;
0855     struct netdev_queue *txq;
0856     struct snd_queue *sq = &qs->sq[cq_idx];
0857     struct rcv_queue *rq = &qs->rq[cq_idx];
0858     unsigned int tx_pkts = 0, tx_bytes = 0, txq_idx;
0859 
0860     spin_lock_bh(&cq->lock);
0861 loop:
0862     processed_cqe = 0;
0863     /* Get no of valid CQ entries to process */
0864     cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx);
0865     cqe_count &= CQ_CQE_COUNT;
0866     if (!cqe_count)
0867         goto done;
0868 
0869     /* Get head of the valid CQ entries */
0870     cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9;
0871     cqe_head &= 0xFFFF;
0872 
0873     while (processed_cqe < cqe_count) {
0874         /* Get the CQ descriptor */
0875         cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head);
0876         cqe_head++;
0877         cqe_head &= (cq->dmem.q_len - 1);
0878         /* Initiate prefetch for next descriptor */
0879         prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head));
0880 
0881         if ((work_done >= budget) && napi &&
0882             (cq_desc->cqe_type != CQE_TYPE_SEND)) {
0883             break;
0884         }
0885 
0886         switch (cq_desc->cqe_type) {
0887         case CQE_TYPE_RX:
0888             nicvf_rcv_pkt_handler(netdev, napi, cq_desc, sq, rq);
0889             work_done++;
0890         break;
0891         case CQE_TYPE_SEND:
0892             nicvf_snd_pkt_handler(netdev, (void *)cq_desc,
0893                           budget, &subdesc_cnt,
0894                           &tx_pkts, &tx_bytes);
0895             tx_done++;
0896         break;
0897         case CQE_TYPE_SEND_PTP:
0898             nicvf_snd_ptp_handler(netdev, (void *)cq_desc);
0899         break;
0900         case CQE_TYPE_INVALID:
0901         case CQE_TYPE_RX_SPLIT:
0902         case CQE_TYPE_RX_TCP:
0903             /* Ignore for now */
0904         break;
0905         }
0906         processed_cqe++;
0907     }
0908 
0909     /* Ring doorbell to inform H/W to reuse processed CQEs */
0910     nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR,
0911                   cq_idx, processed_cqe);
0912 
0913     if ((work_done < budget) && napi)
0914         goto loop;
0915 
0916 done:
0917     /* Update SQ's descriptor free count */
0918     if (subdesc_cnt)
0919         nicvf_put_sq_desc(sq, subdesc_cnt);
0920 
0921     txq_idx = nicvf_netdev_qidx(nic, cq_idx);
0922     /* Handle XDP TX queues */
0923     if (nic->pnicvf->xdp_prog) {
0924         if (txq_idx < nic->pnicvf->xdp_tx_queues) {
0925             nicvf_xdp_sq_doorbell(nic, sq, cq_idx);
0926             goto out;
0927         }
0928         nic = nic->pnicvf;
0929         txq_idx -= nic->pnicvf->xdp_tx_queues;
0930     }
0931 
0932     /* Wakeup TXQ if its stopped earlier due to SQ full */
0933     if (tx_done ||
0934         (atomic_read(&sq->free_cnt) >= MIN_SQ_DESC_PER_PKT_XMIT)) {
0935         netdev = nic->pnicvf->netdev;
0936         txq = netdev_get_tx_queue(netdev, txq_idx);
0937         if (tx_pkts)
0938             netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
0939 
0940         /* To read updated queue and carrier status */
0941         smp_mb();
0942         if (netif_tx_queue_stopped(txq) && netif_carrier_ok(netdev)) {
0943             netif_tx_wake_queue(txq);
0944             nic = nic->pnicvf;
0945             this_cpu_inc(nic->drv_stats->txq_wake);
0946             netif_warn(nic, tx_err, netdev,
0947                    "Transmit queue wakeup SQ%d\n", txq_idx);
0948         }
0949     }
0950 
0951 out:
0952     spin_unlock_bh(&cq->lock);
0953     return work_done;
0954 }
0955 
0956 static int nicvf_poll(struct napi_struct *napi, int budget)
0957 {
0958     u64  cq_head;
0959     int  work_done = 0;
0960     struct net_device *netdev = napi->dev;
0961     struct nicvf *nic = netdev_priv(netdev);
0962     struct nicvf_cq_poll *cq;
0963 
0964     cq = container_of(napi, struct nicvf_cq_poll, napi);
0965     work_done = nicvf_cq_intr_handler(netdev, cq->cq_idx, napi, budget);
0966 
0967     if (work_done < budget) {
0968         /* Slow packet rate, exit polling */
0969         napi_complete_done(napi, work_done);
0970         /* Re-enable interrupts */
0971         cq_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD,
0972                            cq->cq_idx);
0973         nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
0974         nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_HEAD,
0975                       cq->cq_idx, cq_head);
0976         nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->cq_idx);
0977     }
0978     return work_done;
0979 }
0980 
0981 /* Qset error interrupt handler
0982  *
0983  * As of now only CQ errors are handled
0984  */
0985 static void nicvf_handle_qs_err(struct tasklet_struct *t)
0986 {
0987     struct nicvf *nic = from_tasklet(nic, t, qs_err_task);
0988     struct queue_set *qs = nic->qs;
0989     int qidx;
0990     u64 status;
0991 
0992     netif_tx_disable(nic->netdev);
0993 
0994     /* Check if it is CQ err */
0995     for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
0996         status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS,
0997                           qidx);
0998         if (!(status & CQ_ERR_MASK))
0999             continue;
1000         /* Process already queued CQEs and reconfig CQ */
1001         nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1002         nicvf_sq_disable(nic, qidx);
1003         nicvf_cq_intr_handler(nic->netdev, qidx, NULL, 0);
1004         nicvf_cmp_queue_config(nic, qs, qidx, true);
1005         nicvf_sq_free_used_descs(nic->netdev, &qs->sq[qidx], qidx);
1006         nicvf_sq_enable(nic, &qs->sq[qidx], qidx);
1007 
1008         nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1009     }
1010 
1011     netif_tx_start_all_queues(nic->netdev);
1012     /* Re-enable Qset error interrupt */
1013     nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1014 }
1015 
1016 static void nicvf_dump_intr_status(struct nicvf *nic)
1017 {
1018     netif_info(nic, intr, nic->netdev, "interrupt status 0x%llx\n",
1019            nicvf_reg_read(nic, NIC_VF_INT));
1020 }
1021 
1022 static irqreturn_t nicvf_misc_intr_handler(int irq, void *nicvf_irq)
1023 {
1024     struct nicvf *nic = (struct nicvf *)nicvf_irq;
1025     u64 intr;
1026 
1027     nicvf_dump_intr_status(nic);
1028 
1029     intr = nicvf_reg_read(nic, NIC_VF_INT);
1030     /* Check for spurious interrupt */
1031     if (!(intr & NICVF_INTR_MBOX_MASK))
1032         return IRQ_HANDLED;
1033 
1034     nicvf_handle_mbx_intr(nic);
1035 
1036     return IRQ_HANDLED;
1037 }
1038 
1039 static irqreturn_t nicvf_intr_handler(int irq, void *cq_irq)
1040 {
1041     struct nicvf_cq_poll *cq_poll = (struct nicvf_cq_poll *)cq_irq;
1042     struct nicvf *nic = cq_poll->nicvf;
1043     int qidx = cq_poll->cq_idx;
1044 
1045     nicvf_dump_intr_status(nic);
1046 
1047     /* Disable interrupts */
1048     nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1049 
1050     /* Schedule NAPI */
1051     napi_schedule_irqoff(&cq_poll->napi);
1052 
1053     /* Clear interrupt */
1054     nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1055 
1056     return IRQ_HANDLED;
1057 }
1058 
1059 static irqreturn_t nicvf_rbdr_intr_handler(int irq, void *nicvf_irq)
1060 {
1061     struct nicvf *nic = (struct nicvf *)nicvf_irq;
1062     u8 qidx;
1063 
1064 
1065     nicvf_dump_intr_status(nic);
1066 
1067     /* Disable RBDR interrupt and schedule softirq */
1068     for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) {
1069         if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx))
1070             continue;
1071         nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1072         tasklet_hi_schedule(&nic->rbdr_task);
1073         /* Clear interrupt */
1074         nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1075     }
1076 
1077     return IRQ_HANDLED;
1078 }
1079 
1080 static irqreturn_t nicvf_qs_err_intr_handler(int irq, void *nicvf_irq)
1081 {
1082     struct nicvf *nic = (struct nicvf *)nicvf_irq;
1083 
1084     nicvf_dump_intr_status(nic);
1085 
1086     /* Disable Qset err interrupt and schedule softirq */
1087     nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1088     tasklet_hi_schedule(&nic->qs_err_task);
1089     nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1090 
1091     return IRQ_HANDLED;
1092 }
1093 
1094 static void nicvf_set_irq_affinity(struct nicvf *nic)
1095 {
1096     int vec, cpu;
1097 
1098     for (vec = 0; vec < nic->num_vec; vec++) {
1099         if (!nic->irq_allocated[vec])
1100             continue;
1101 
1102         if (!zalloc_cpumask_var(&nic->affinity_mask[vec], GFP_KERNEL))
1103             return;
1104          /* CQ interrupts */
1105         if (vec < NICVF_INTR_ID_SQ)
1106             /* Leave CPU0 for RBDR and other interrupts */
1107             cpu = nicvf_netdev_qidx(nic, vec) + 1;
1108         else
1109             cpu = 0;
1110 
1111         cpumask_set_cpu(cpumask_local_spread(cpu, nic->node),
1112                 nic->affinity_mask[vec]);
1113         irq_set_affinity_hint(pci_irq_vector(nic->pdev, vec),
1114                       nic->affinity_mask[vec]);
1115     }
1116 }
1117 
1118 static int nicvf_register_interrupts(struct nicvf *nic)
1119 {
1120     int irq, ret = 0;
1121 
1122     for_each_cq_irq(irq)
1123         sprintf(nic->irq_name[irq], "%s-rxtx-%d",
1124             nic->pnicvf->netdev->name,
1125             nicvf_netdev_qidx(nic, irq));
1126 
1127     for_each_sq_irq(irq)
1128         sprintf(nic->irq_name[irq], "%s-sq-%d",
1129             nic->pnicvf->netdev->name,
1130             nicvf_netdev_qidx(nic, irq - NICVF_INTR_ID_SQ));
1131 
1132     for_each_rbdr_irq(irq)
1133         sprintf(nic->irq_name[irq], "%s-rbdr-%d",
1134             nic->pnicvf->netdev->name,
1135             nic->sqs_mode ? (nic->sqs_id + 1) : 0);
1136 
1137     /* Register CQ interrupts */
1138     for (irq = 0; irq < nic->qs->cq_cnt; irq++) {
1139         ret = request_irq(pci_irq_vector(nic->pdev, irq),
1140                   nicvf_intr_handler,
1141                   0, nic->irq_name[irq], nic->napi[irq]);
1142         if (ret)
1143             goto err;
1144         nic->irq_allocated[irq] = true;
1145     }
1146 
1147     /* Register RBDR interrupt */
1148     for (irq = NICVF_INTR_ID_RBDR;
1149          irq < (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt); irq++) {
1150         ret = request_irq(pci_irq_vector(nic->pdev, irq),
1151                   nicvf_rbdr_intr_handler,
1152                   0, nic->irq_name[irq], nic);
1153         if (ret)
1154             goto err;
1155         nic->irq_allocated[irq] = true;
1156     }
1157 
1158     /* Register QS error interrupt */
1159     sprintf(nic->irq_name[NICVF_INTR_ID_QS_ERR], "%s-qset-err-%d",
1160         nic->pnicvf->netdev->name,
1161         nic->sqs_mode ? (nic->sqs_id + 1) : 0);
1162     irq = NICVF_INTR_ID_QS_ERR;
1163     ret = request_irq(pci_irq_vector(nic->pdev, irq),
1164               nicvf_qs_err_intr_handler,
1165               0, nic->irq_name[irq], nic);
1166     if (ret)
1167         goto err;
1168 
1169     nic->irq_allocated[irq] = true;
1170 
1171     /* Set IRQ affinities */
1172     nicvf_set_irq_affinity(nic);
1173 
1174 err:
1175     if (ret)
1176         netdev_err(nic->netdev, "request_irq failed, vector %d\n", irq);
1177 
1178     return ret;
1179 }
1180 
1181 static void nicvf_unregister_interrupts(struct nicvf *nic)
1182 {
1183     struct pci_dev *pdev = nic->pdev;
1184     int irq;
1185 
1186     /* Free registered interrupts */
1187     for (irq = 0; irq < nic->num_vec; irq++) {
1188         if (!nic->irq_allocated[irq])
1189             continue;
1190 
1191         irq_set_affinity_hint(pci_irq_vector(pdev, irq), NULL);
1192         free_cpumask_var(nic->affinity_mask[irq]);
1193 
1194         if (irq < NICVF_INTR_ID_SQ)
1195             free_irq(pci_irq_vector(pdev, irq), nic->napi[irq]);
1196         else
1197             free_irq(pci_irq_vector(pdev, irq), nic);
1198 
1199         nic->irq_allocated[irq] = false;
1200     }
1201 
1202     /* Disable MSI-X */
1203     pci_free_irq_vectors(pdev);
1204     nic->num_vec = 0;
1205 }
1206 
1207 /* Initialize MSIX vectors and register MISC interrupt.
1208  * Send READY message to PF to check if its alive
1209  */
1210 static int nicvf_register_misc_interrupt(struct nicvf *nic)
1211 {
1212     int ret = 0;
1213     int irq = NICVF_INTR_ID_MISC;
1214 
1215     /* Return if mailbox interrupt is already registered */
1216     if (nic->pdev->msix_enabled)
1217         return 0;
1218 
1219     /* Enable MSI-X */
1220     nic->num_vec = pci_msix_vec_count(nic->pdev);
1221     ret = pci_alloc_irq_vectors(nic->pdev, nic->num_vec, nic->num_vec,
1222                     PCI_IRQ_MSIX);
1223     if (ret < 0) {
1224         netdev_err(nic->netdev,
1225                "Req for #%d msix vectors failed\n", nic->num_vec);
1226         return ret;
1227     }
1228 
1229     sprintf(nic->irq_name[irq], "%s Mbox", "NICVF");
1230     /* Register Misc interrupt */
1231     ret = request_irq(pci_irq_vector(nic->pdev, irq),
1232               nicvf_misc_intr_handler, 0, nic->irq_name[irq], nic);
1233 
1234     if (ret)
1235         return ret;
1236     nic->irq_allocated[irq] = true;
1237 
1238     /* Enable mailbox interrupt */
1239     nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0);
1240 
1241     /* Check if VF is able to communicate with PF */
1242     if (!nicvf_check_pf_ready(nic)) {
1243         nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1244         nicvf_unregister_interrupts(nic);
1245         return -EIO;
1246     }
1247 
1248     return 0;
1249 }
1250 
1251 static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev)
1252 {
1253     struct nicvf *nic = netdev_priv(netdev);
1254     int qid = skb_get_queue_mapping(skb);
1255     struct netdev_queue *txq = netdev_get_tx_queue(netdev, qid);
1256     struct nicvf *snic;
1257     struct snd_queue *sq;
1258     int tmp;
1259 
1260     /* Check for minimum packet length */
1261     if (skb->len <= ETH_HLEN) {
1262         dev_kfree_skb(skb);
1263         return NETDEV_TX_OK;
1264     }
1265 
1266     /* In XDP case, initial HW tx queues are used for XDP,
1267      * but stack's queue mapping starts at '0', so skip the
1268      * Tx queues attached to Rx queues for XDP.
1269      */
1270     if (nic->xdp_prog)
1271         qid += nic->xdp_tx_queues;
1272 
1273     snic = nic;
1274     /* Get secondary Qset's SQ structure */
1275     if (qid >= MAX_SND_QUEUES_PER_QS) {
1276         tmp = qid / MAX_SND_QUEUES_PER_QS;
1277         snic = (struct nicvf *)nic->snicvf[tmp - 1];
1278         if (!snic) {
1279             netdev_warn(nic->netdev,
1280                     "Secondary Qset#%d's ptr not initialized\n",
1281                     tmp - 1);
1282             dev_kfree_skb(skb);
1283             return NETDEV_TX_OK;
1284         }
1285         qid = qid % MAX_SND_QUEUES_PER_QS;
1286     }
1287 
1288     sq = &snic->qs->sq[qid];
1289     if (!netif_tx_queue_stopped(txq) &&
1290         !nicvf_sq_append_skb(snic, sq, skb, qid)) {
1291         netif_tx_stop_queue(txq);
1292 
1293         /* Barrier, so that stop_queue visible to other cpus */
1294         smp_mb();
1295 
1296         /* Check again, incase another cpu freed descriptors */
1297         if (atomic_read(&sq->free_cnt) > MIN_SQ_DESC_PER_PKT_XMIT) {
1298             netif_tx_wake_queue(txq);
1299         } else {
1300             this_cpu_inc(nic->drv_stats->txq_stop);
1301             netif_warn(nic, tx_err, netdev,
1302                    "Transmit ring full, stopping SQ%d\n", qid);
1303         }
1304         return NETDEV_TX_BUSY;
1305     }
1306 
1307     return NETDEV_TX_OK;
1308 }
1309 
1310 static inline void nicvf_free_cq_poll(struct nicvf *nic)
1311 {
1312     struct nicvf_cq_poll *cq_poll;
1313     int qidx;
1314 
1315     for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1316         cq_poll = nic->napi[qidx];
1317         if (!cq_poll)
1318             continue;
1319         nic->napi[qidx] = NULL;
1320         kfree(cq_poll);
1321     }
1322 }
1323 
1324 int nicvf_stop(struct net_device *netdev)
1325 {
1326     int irq, qidx;
1327     struct nicvf *nic = netdev_priv(netdev);
1328     struct queue_set *qs = nic->qs;
1329     struct nicvf_cq_poll *cq_poll = NULL;
1330     union nic_mbx mbx = {};
1331 
1332     /* wait till all queued set_rx_mode tasks completes */
1333     if (nic->nicvf_rx_mode_wq) {
1334         cancel_delayed_work_sync(&nic->link_change_work);
1335         drain_workqueue(nic->nicvf_rx_mode_wq);
1336     }
1337 
1338     mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN;
1339     nicvf_send_msg_to_pf(nic, &mbx);
1340 
1341     netif_carrier_off(netdev);
1342     netif_tx_stop_all_queues(nic->netdev);
1343     nic->link_up = false;
1344 
1345     /* Teardown secondary qsets first */
1346     if (!nic->sqs_mode) {
1347         for (qidx = 0; qidx < nic->sqs_count; qidx++) {
1348             if (!nic->snicvf[qidx])
1349                 continue;
1350             nicvf_stop(nic->snicvf[qidx]->netdev);
1351             nic->snicvf[qidx] = NULL;
1352         }
1353     }
1354 
1355     /* Disable RBDR & QS error interrupts */
1356     for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1357         nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx);
1358         nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx);
1359     }
1360     nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0);
1361     nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0);
1362 
1363     /* Wait for pending IRQ handlers to finish */
1364     for (irq = 0; irq < nic->num_vec; irq++)
1365         synchronize_irq(pci_irq_vector(nic->pdev, irq));
1366 
1367     tasklet_kill(&nic->rbdr_task);
1368     tasklet_kill(&nic->qs_err_task);
1369     if (nic->rb_work_scheduled)
1370         cancel_delayed_work_sync(&nic->rbdr_work);
1371 
1372     for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) {
1373         cq_poll = nic->napi[qidx];
1374         if (!cq_poll)
1375             continue;
1376         napi_synchronize(&cq_poll->napi);
1377         /* CQ intr is enabled while napi_complete,
1378          * so disable it now
1379          */
1380         nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
1381         nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
1382         napi_disable(&cq_poll->napi);
1383         netif_napi_del(&cq_poll->napi);
1384     }
1385 
1386     netif_tx_disable(netdev);
1387 
1388     for (qidx = 0; qidx < netdev->num_tx_queues; qidx++)
1389         netdev_tx_reset_queue(netdev_get_tx_queue(netdev, qidx));
1390 
1391     /* Free resources */
1392     nicvf_config_data_transfer(nic, false);
1393 
1394     /* Disable HW Qset */
1395     nicvf_qset_config(nic, false);
1396 
1397     /* disable mailbox interrupt */
1398     nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1399 
1400     nicvf_unregister_interrupts(nic);
1401 
1402     nicvf_free_cq_poll(nic);
1403 
1404     /* Free any pending SKB saved to receive timestamp */
1405     if (nic->ptp_skb) {
1406         dev_kfree_skb_any(nic->ptp_skb);
1407         nic->ptp_skb = NULL;
1408     }
1409 
1410     /* Clear multiqset info */
1411     nic->pnicvf = nic;
1412 
1413     return 0;
1414 }
1415 
1416 static int nicvf_config_hw_rx_tstamp(struct nicvf *nic, bool enable)
1417 {
1418     union nic_mbx mbx = {};
1419 
1420     mbx.ptp.msg = NIC_MBOX_MSG_PTP_CFG;
1421     mbx.ptp.enable = enable;
1422 
1423     return nicvf_send_msg_to_pf(nic, &mbx);
1424 }
1425 
1426 static int nicvf_update_hw_max_frs(struct nicvf *nic, int mtu)
1427 {
1428     union nic_mbx mbx = {};
1429 
1430     mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS;
1431     mbx.frs.max_frs = mtu;
1432     mbx.frs.vf_id = nic->vf_id;
1433 
1434     return nicvf_send_msg_to_pf(nic, &mbx);
1435 }
1436 
1437 static void nicvf_link_status_check_task(struct work_struct *work_arg)
1438 {
1439     struct nicvf *nic = container_of(work_arg,
1440                      struct nicvf,
1441                      link_change_work.work);
1442     union nic_mbx mbx = {};
1443     mbx.msg.msg = NIC_MBOX_MSG_BGX_LINK_CHANGE;
1444     nicvf_send_msg_to_pf(nic, &mbx);
1445     queue_delayed_work(nic->nicvf_rx_mode_wq,
1446                &nic->link_change_work, 2 * HZ);
1447 }
1448 
1449 int nicvf_open(struct net_device *netdev)
1450 {
1451     int cpu, err, qidx;
1452     struct nicvf *nic = netdev_priv(netdev);
1453     struct queue_set *qs = nic->qs;
1454     struct nicvf_cq_poll *cq_poll = NULL;
1455 
1456     /* wait till all queued set_rx_mode tasks completes if any */
1457     if (nic->nicvf_rx_mode_wq)
1458         drain_workqueue(nic->nicvf_rx_mode_wq);
1459 
1460     netif_carrier_off(netdev);
1461 
1462     err = nicvf_register_misc_interrupt(nic);
1463     if (err)
1464         return err;
1465 
1466     /* Register NAPI handler for processing CQEs */
1467     for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1468         cq_poll = kzalloc(sizeof(*cq_poll), GFP_KERNEL);
1469         if (!cq_poll) {
1470             err = -ENOMEM;
1471             goto napi_del;
1472         }
1473         cq_poll->cq_idx = qidx;
1474         cq_poll->nicvf = nic;
1475         netif_napi_add(netdev, &cq_poll->napi, nicvf_poll,
1476                    NAPI_POLL_WEIGHT);
1477         napi_enable(&cq_poll->napi);
1478         nic->napi[qidx] = cq_poll;
1479     }
1480 
1481     /* Check if we got MAC address from PF or else generate a radom MAC */
1482     if (!nic->sqs_mode && is_zero_ether_addr(netdev->dev_addr)) {
1483         eth_hw_addr_random(netdev);
1484         nicvf_hw_set_mac_addr(nic, netdev);
1485     }
1486 
1487     if (nic->set_mac_pending) {
1488         nic->set_mac_pending = false;
1489         nicvf_hw_set_mac_addr(nic, netdev);
1490     }
1491 
1492     /* Init tasklet for handling Qset err interrupt */
1493     tasklet_setup(&nic->qs_err_task, nicvf_handle_qs_err);
1494 
1495     /* Init RBDR tasklet which will refill RBDR */
1496     tasklet_setup(&nic->rbdr_task, nicvf_rbdr_task);
1497     INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work);
1498 
1499     /* Configure CPI alorithm */
1500     nic->cpi_alg = cpi_alg;
1501     if (!nic->sqs_mode)
1502         nicvf_config_cpi(nic);
1503 
1504     nicvf_request_sqs(nic);
1505     if (nic->sqs_mode)
1506         nicvf_get_primary_vf_struct(nic);
1507 
1508     /* Configure PTP timestamp */
1509     if (nic->ptp_clock)
1510         nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1511     atomic_set(&nic->tx_ptp_skbs, 0);
1512     nic->ptp_skb = NULL;
1513 
1514     /* Configure receive side scaling and MTU */
1515     if (!nic->sqs_mode) {
1516         nicvf_rss_init(nic);
1517         err = nicvf_update_hw_max_frs(nic, netdev->mtu);
1518         if (err)
1519             goto cleanup;
1520 
1521         /* Clear percpu stats */
1522         for_each_possible_cpu(cpu)
1523             memset(per_cpu_ptr(nic->drv_stats, cpu), 0,
1524                    sizeof(struct nicvf_drv_stats));
1525     }
1526 
1527     err = nicvf_register_interrupts(nic);
1528     if (err)
1529         goto cleanup;
1530 
1531     /* Initialize the queues */
1532     err = nicvf_init_resources(nic);
1533     if (err)
1534         goto cleanup;
1535 
1536     /* Make sure queue initialization is written */
1537     wmb();
1538 
1539     nicvf_reg_write(nic, NIC_VF_INT, -1);
1540     /* Enable Qset err interrupt */
1541     nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
1542 
1543     /* Enable completion queue interrupt */
1544     for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1545         nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
1546 
1547     /* Enable RBDR threshold interrupt */
1548     for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1549         nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx);
1550 
1551     /* Send VF config done msg to PF */
1552     nicvf_send_cfg_done(nic);
1553 
1554     if (nic->nicvf_rx_mode_wq) {
1555         INIT_DELAYED_WORK(&nic->link_change_work,
1556                   nicvf_link_status_check_task);
1557         queue_delayed_work(nic->nicvf_rx_mode_wq,
1558                    &nic->link_change_work, 0);
1559     }
1560 
1561     return 0;
1562 cleanup:
1563     nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0);
1564     nicvf_unregister_interrupts(nic);
1565     tasklet_kill(&nic->qs_err_task);
1566     tasklet_kill(&nic->rbdr_task);
1567 napi_del:
1568     for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1569         cq_poll = nic->napi[qidx];
1570         if (!cq_poll)
1571             continue;
1572         napi_disable(&cq_poll->napi);
1573         netif_napi_del(&cq_poll->napi);
1574     }
1575     nicvf_free_cq_poll(nic);
1576     return err;
1577 }
1578 
1579 static int nicvf_change_mtu(struct net_device *netdev, int new_mtu)
1580 {
1581     struct nicvf *nic = netdev_priv(netdev);
1582     int orig_mtu = netdev->mtu;
1583 
1584     /* For now just support only the usual MTU sized frames,
1585      * plus some headroom for VLAN, QinQ.
1586      */
1587     if (nic->xdp_prog && new_mtu > MAX_XDP_MTU) {
1588         netdev_warn(netdev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1589                 netdev->mtu);
1590         return -EINVAL;
1591     }
1592 
1593     netdev->mtu = new_mtu;
1594 
1595     if (!netif_running(netdev))
1596         return 0;
1597 
1598     if (nicvf_update_hw_max_frs(nic, new_mtu)) {
1599         netdev->mtu = orig_mtu;
1600         return -EINVAL;
1601     }
1602 
1603     return 0;
1604 }
1605 
1606 static int nicvf_set_mac_address(struct net_device *netdev, void *p)
1607 {
1608     struct sockaddr *addr = p;
1609     struct nicvf *nic = netdev_priv(netdev);
1610 
1611     if (!is_valid_ether_addr(addr->sa_data))
1612         return -EADDRNOTAVAIL;
1613 
1614     eth_hw_addr_set(netdev, addr->sa_data);
1615 
1616     if (nic->pdev->msix_enabled) {
1617         if (nicvf_hw_set_mac_addr(nic, netdev))
1618             return -EBUSY;
1619     } else {
1620         nic->set_mac_pending = true;
1621     }
1622 
1623     return 0;
1624 }
1625 
1626 void nicvf_update_lmac_stats(struct nicvf *nic)
1627 {
1628     int stat = 0;
1629     union nic_mbx mbx = {};
1630 
1631     if (!netif_running(nic->netdev))
1632         return;
1633 
1634     mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS;
1635     mbx.bgx_stats.vf_id = nic->vf_id;
1636     /* Rx stats */
1637     mbx.bgx_stats.rx = 1;
1638     while (stat < BGX_RX_STATS_COUNT) {
1639         mbx.bgx_stats.idx = stat;
1640         if (nicvf_send_msg_to_pf(nic, &mbx))
1641             return;
1642         stat++;
1643     }
1644 
1645     stat = 0;
1646 
1647     /* Tx stats */
1648     mbx.bgx_stats.rx = 0;
1649     while (stat < BGX_TX_STATS_COUNT) {
1650         mbx.bgx_stats.idx = stat;
1651         if (nicvf_send_msg_to_pf(nic, &mbx))
1652             return;
1653         stat++;
1654     }
1655 }
1656 
1657 void nicvf_update_stats(struct nicvf *nic)
1658 {
1659     int qidx, cpu;
1660     u64 tmp_stats = 0;
1661     struct nicvf_hw_stats *stats = &nic->hw_stats;
1662     struct nicvf_drv_stats *drv_stats;
1663     struct queue_set *qs = nic->qs;
1664 
1665 #define GET_RX_STATS(reg) \
1666     nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3))
1667 #define GET_TX_STATS(reg) \
1668     nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3))
1669 
1670     stats->rx_bytes = GET_RX_STATS(RX_OCTS);
1671     stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST);
1672     stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST);
1673     stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST);
1674     stats->rx_fcs_errors = GET_RX_STATS(RX_FCS);
1675     stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR);
1676     stats->rx_drop_red = GET_RX_STATS(RX_RED);
1677     stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS);
1678     stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN);
1679     stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS);
1680     stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST);
1681     stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST);
1682     stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST);
1683     stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST);
1684 
1685     stats->tx_bytes = GET_TX_STATS(TX_OCTS);
1686     stats->tx_ucast_frames = GET_TX_STATS(TX_UCAST);
1687     stats->tx_bcast_frames = GET_TX_STATS(TX_BCAST);
1688     stats->tx_mcast_frames = GET_TX_STATS(TX_MCAST);
1689     stats->tx_drops = GET_TX_STATS(TX_DROP);
1690 
1691     /* On T88 pass 2.0, the dummy SQE added for TSO notification
1692      * via CQE has 'dont_send' set. Hence HW drops the pkt pointed
1693      * pointed by dummy SQE and results in tx_drops counter being
1694      * incremented. Subtracting it from tx_tso counter will give
1695      * exact tx_drops counter.
1696      */
1697     if (nic->t88 && nic->hw_tso) {
1698         for_each_possible_cpu(cpu) {
1699             drv_stats = per_cpu_ptr(nic->drv_stats, cpu);
1700             tmp_stats += drv_stats->tx_tso;
1701         }
1702         stats->tx_drops = tmp_stats - stats->tx_drops;
1703     }
1704     stats->tx_frames = stats->tx_ucast_frames +
1705                stats->tx_bcast_frames +
1706                stats->tx_mcast_frames;
1707     stats->rx_frames = stats->rx_ucast_frames +
1708                stats->rx_bcast_frames +
1709                stats->rx_mcast_frames;
1710     stats->rx_drops = stats->rx_drop_red +
1711               stats->rx_drop_overrun;
1712 
1713     /* Update RQ and SQ stats */
1714     for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1715         nicvf_update_rq_stats(nic, qidx);
1716     for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1717         nicvf_update_sq_stats(nic, qidx);
1718 }
1719 
1720 static void nicvf_get_stats64(struct net_device *netdev,
1721                   struct rtnl_link_stats64 *stats)
1722 {
1723     struct nicvf *nic = netdev_priv(netdev);
1724     struct nicvf_hw_stats *hw_stats = &nic->hw_stats;
1725 
1726     nicvf_update_stats(nic);
1727 
1728     stats->rx_bytes = hw_stats->rx_bytes;
1729     stats->rx_packets = hw_stats->rx_frames;
1730     stats->rx_dropped = hw_stats->rx_drops;
1731     stats->multicast = hw_stats->rx_mcast_frames;
1732 
1733     stats->tx_bytes = hw_stats->tx_bytes;
1734     stats->tx_packets = hw_stats->tx_frames;
1735     stats->tx_dropped = hw_stats->tx_drops;
1736 
1737 }
1738 
1739 static void nicvf_tx_timeout(struct net_device *dev, unsigned int txqueue)
1740 {
1741     struct nicvf *nic = netdev_priv(dev);
1742 
1743     netif_warn(nic, tx_err, dev, "Transmit timed out, resetting\n");
1744 
1745     this_cpu_inc(nic->drv_stats->tx_timeout);
1746     schedule_work(&nic->reset_task);
1747 }
1748 
1749 static void nicvf_reset_task(struct work_struct *work)
1750 {
1751     struct nicvf *nic;
1752 
1753     nic = container_of(work, struct nicvf, reset_task);
1754 
1755     if (!netif_running(nic->netdev))
1756         return;
1757 
1758     nicvf_stop(nic->netdev);
1759     nicvf_open(nic->netdev);
1760     netif_trans_update(nic->netdev);
1761 }
1762 
1763 static int nicvf_config_loopback(struct nicvf *nic,
1764                  netdev_features_t features)
1765 {
1766     union nic_mbx mbx = {};
1767 
1768     mbx.lbk.msg = NIC_MBOX_MSG_LOOPBACK;
1769     mbx.lbk.vf_id = nic->vf_id;
1770     mbx.lbk.enable = (features & NETIF_F_LOOPBACK) != 0;
1771 
1772     return nicvf_send_msg_to_pf(nic, &mbx);
1773 }
1774 
1775 static netdev_features_t nicvf_fix_features(struct net_device *netdev,
1776                         netdev_features_t features)
1777 {
1778     struct nicvf *nic = netdev_priv(netdev);
1779 
1780     if ((features & NETIF_F_LOOPBACK) &&
1781         netif_running(netdev) && !nic->loopback_supported)
1782         features &= ~NETIF_F_LOOPBACK;
1783 
1784     return features;
1785 }
1786 
1787 static int nicvf_set_features(struct net_device *netdev,
1788                   netdev_features_t features)
1789 {
1790     struct nicvf *nic = netdev_priv(netdev);
1791     netdev_features_t changed = features ^ netdev->features;
1792 
1793     if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1794         nicvf_config_vlan_stripping(nic, features);
1795 
1796     if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev))
1797         return nicvf_config_loopback(nic, features);
1798 
1799     return 0;
1800 }
1801 
1802 static void nicvf_set_xdp_queues(struct nicvf *nic, bool bpf_attached)
1803 {
1804     u8 cq_count, txq_count;
1805 
1806     /* Set XDP Tx queue count same as Rx queue count */
1807     if (!bpf_attached)
1808         nic->xdp_tx_queues = 0;
1809     else
1810         nic->xdp_tx_queues = nic->rx_queues;
1811 
1812     /* If queue count > MAX_CMP_QUEUES_PER_QS, then additional qsets
1813      * needs to be allocated, check how many.
1814      */
1815     txq_count = nic->xdp_tx_queues + nic->tx_queues;
1816     cq_count = max(nic->rx_queues, txq_count);
1817     if (cq_count > MAX_CMP_QUEUES_PER_QS) {
1818         nic->sqs_count = roundup(cq_count, MAX_CMP_QUEUES_PER_QS);
1819         nic->sqs_count = (nic->sqs_count / MAX_CMP_QUEUES_PER_QS) - 1;
1820     } else {
1821         nic->sqs_count = 0;
1822     }
1823 
1824     /* Set primary Qset's resources */
1825     nic->qs->rq_cnt = min_t(u8, nic->rx_queues, MAX_RCV_QUEUES_PER_QS);
1826     nic->qs->sq_cnt = min_t(u8, txq_count, MAX_SND_QUEUES_PER_QS);
1827     nic->qs->cq_cnt = max_t(u8, nic->qs->rq_cnt, nic->qs->sq_cnt);
1828 
1829     /* Update stack */
1830     nicvf_set_real_num_queues(nic->netdev, nic->tx_queues, nic->rx_queues);
1831 }
1832 
1833 static int nicvf_xdp_setup(struct nicvf *nic, struct bpf_prog *prog)
1834 {
1835     struct net_device *dev = nic->netdev;
1836     bool if_up = netif_running(nic->netdev);
1837     struct bpf_prog *old_prog;
1838     bool bpf_attached = false;
1839     int ret = 0;
1840 
1841     /* For now just support only the usual MTU sized frames,
1842      * plus some headroom for VLAN, QinQ.
1843      */
1844     if (prog && dev->mtu > MAX_XDP_MTU) {
1845         netdev_warn(dev, "Jumbo frames not yet supported with XDP, current MTU %d.\n",
1846                 dev->mtu);
1847         return -EOPNOTSUPP;
1848     }
1849 
1850     /* ALL SQs attached to CQs i.e same as RQs, are treated as
1851      * XDP Tx queues and more Tx queues are allocated for
1852      * network stack to send pkts out.
1853      *
1854      * No of Tx queues are either same as Rx queues or whatever
1855      * is left in max no of queues possible.
1856      */
1857     if ((nic->rx_queues + nic->tx_queues) > nic->max_queues) {
1858         netdev_warn(dev,
1859                 "Failed to attach BPF prog, RXQs + TXQs > Max %d\n",
1860                 nic->max_queues);
1861         return -ENOMEM;
1862     }
1863 
1864     if (if_up)
1865         nicvf_stop(nic->netdev);
1866 
1867     old_prog = xchg(&nic->xdp_prog, prog);
1868     /* Detach old prog, if any */
1869     if (old_prog)
1870         bpf_prog_put(old_prog);
1871 
1872     if (nic->xdp_prog) {
1873         /* Attach BPF program */
1874         bpf_prog_add(nic->xdp_prog, nic->rx_queues - 1);
1875         bpf_attached = true;
1876     }
1877 
1878     /* Calculate Tx queues needed for XDP and network stack */
1879     nicvf_set_xdp_queues(nic, bpf_attached);
1880 
1881     if (if_up) {
1882         /* Reinitialize interface, clean slate */
1883         nicvf_open(nic->netdev);
1884         netif_trans_update(nic->netdev);
1885     }
1886 
1887     return ret;
1888 }
1889 
1890 static int nicvf_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
1891 {
1892     struct nicvf *nic = netdev_priv(netdev);
1893 
1894     /* To avoid checks while retrieving buffer address from CQE_RX,
1895      * do not support XDP for T88 pass1.x silicons which are anyway
1896      * not in use widely.
1897      */
1898     if (pass1_silicon(nic->pdev))
1899         return -EOPNOTSUPP;
1900 
1901     switch (xdp->command) {
1902     case XDP_SETUP_PROG:
1903         return nicvf_xdp_setup(nic, xdp->prog);
1904     default:
1905         return -EINVAL;
1906     }
1907 }
1908 
1909 static int nicvf_config_hwtstamp(struct net_device *netdev, struct ifreq *ifr)
1910 {
1911     struct hwtstamp_config config;
1912     struct nicvf *nic = netdev_priv(netdev);
1913 
1914     if (!nic->ptp_clock)
1915         return -ENODEV;
1916 
1917     if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
1918         return -EFAULT;
1919 
1920     switch (config.tx_type) {
1921     case HWTSTAMP_TX_OFF:
1922     case HWTSTAMP_TX_ON:
1923         break;
1924     default:
1925         return -ERANGE;
1926     }
1927 
1928     switch (config.rx_filter) {
1929     case HWTSTAMP_FILTER_NONE:
1930         nic->hw_rx_tstamp = false;
1931         break;
1932     case HWTSTAMP_FILTER_ALL:
1933     case HWTSTAMP_FILTER_SOME:
1934     case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1935     case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1936     case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1937     case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1938     case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1939     case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1940     case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1941     case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1942     case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1943     case HWTSTAMP_FILTER_PTP_V2_EVENT:
1944     case HWTSTAMP_FILTER_PTP_V2_SYNC:
1945     case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1946         nic->hw_rx_tstamp = true;
1947         config.rx_filter = HWTSTAMP_FILTER_ALL;
1948         break;
1949     default:
1950         return -ERANGE;
1951     }
1952 
1953     if (netif_running(netdev))
1954         nicvf_config_hw_rx_tstamp(nic, nic->hw_rx_tstamp);
1955 
1956     if (copy_to_user(ifr->ifr_data, &config, sizeof(config)))
1957         return -EFAULT;
1958 
1959     return 0;
1960 }
1961 
1962 static int nicvf_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
1963 {
1964     switch (cmd) {
1965     case SIOCSHWTSTAMP:
1966         return nicvf_config_hwtstamp(netdev, req);
1967     default:
1968         return -EOPNOTSUPP;
1969     }
1970 }
1971 
1972 static void __nicvf_set_rx_mode_task(u8 mode, struct xcast_addr_list *mc_addrs,
1973                      struct nicvf *nic)
1974 {
1975     union nic_mbx mbx = {};
1976     int idx;
1977 
1978     /* From the inside of VM code flow we have only 128 bits memory
1979      * available to send message to host's PF, so send all mc addrs
1980      * one by one, starting from flush command in case if kernel
1981      * requests to configure specific MAC filtering
1982      */
1983 
1984     /* flush DMAC filters and reset RX mode */
1985     mbx.xcast.msg = NIC_MBOX_MSG_RESET_XCAST;
1986     if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
1987         goto free_mc;
1988 
1989     if (mode & BGX_XCAST_MCAST_FILTER) {
1990         /* once enabling filtering, we need to signal to PF to add
1991          * its' own LMAC to the filter to accept packets for it.
1992          */
1993         mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST;
1994         mbx.xcast.mac = 0;
1995         if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
1996             goto free_mc;
1997     }
1998 
1999     /* check if we have any specific MACs to be added to PF DMAC filter */
2000     if (mc_addrs) {
2001         /* now go through kernel list of MACs and add them one by one */
2002         for (idx = 0; idx < mc_addrs->count; idx++) {
2003             mbx.xcast.msg = NIC_MBOX_MSG_ADD_MCAST;
2004             mbx.xcast.mac = mc_addrs->mc[idx];
2005             if (nicvf_send_msg_to_pf(nic, &mbx) < 0)
2006                 goto free_mc;
2007         }
2008     }
2009 
2010     /* and finally set rx mode for PF accordingly */
2011     mbx.xcast.msg = NIC_MBOX_MSG_SET_XCAST;
2012     mbx.xcast.mode = mode;
2013 
2014     nicvf_send_msg_to_pf(nic, &mbx);
2015 free_mc:
2016     kfree(mc_addrs);
2017 }
2018 
2019 static void nicvf_set_rx_mode_task(struct work_struct *work_arg)
2020 {
2021     struct nicvf_work *vf_work = container_of(work_arg, struct nicvf_work,
2022                           work);
2023     struct nicvf *nic = container_of(vf_work, struct nicvf, rx_mode_work);
2024     u8 mode;
2025     struct xcast_addr_list *mc;
2026 
2027     /* Save message data locally to prevent them from
2028      * being overwritten by next ndo_set_rx_mode call().
2029      */
2030     spin_lock_bh(&nic->rx_mode_wq_lock);
2031     mode = vf_work->mode;
2032     mc = vf_work->mc;
2033     vf_work->mc = NULL;
2034     spin_unlock_bh(&nic->rx_mode_wq_lock);
2035 
2036     __nicvf_set_rx_mode_task(mode, mc, nic);
2037 }
2038 
2039 static void nicvf_set_rx_mode(struct net_device *netdev)
2040 {
2041     struct nicvf *nic = netdev_priv(netdev);
2042     struct netdev_hw_addr *ha;
2043     struct xcast_addr_list *mc_list = NULL;
2044     u8 mode = 0;
2045 
2046     if (netdev->flags & IFF_PROMISC) {
2047         mode = BGX_XCAST_BCAST_ACCEPT | BGX_XCAST_MCAST_ACCEPT;
2048     } else {
2049         if (netdev->flags & IFF_BROADCAST)
2050             mode |= BGX_XCAST_BCAST_ACCEPT;
2051 
2052         if (netdev->flags & IFF_ALLMULTI) {
2053             mode |= BGX_XCAST_MCAST_ACCEPT;
2054         } else if (netdev->flags & IFF_MULTICAST) {
2055             mode |= BGX_XCAST_MCAST_FILTER;
2056             /* here we need to copy mc addrs */
2057             if (netdev_mc_count(netdev)) {
2058                 mc_list = kmalloc(struct_size(mc_list, mc,
2059                                   netdev_mc_count(netdev)),
2060                           GFP_ATOMIC);
2061                 if (unlikely(!mc_list))
2062                     return;
2063                 mc_list->count = 0;
2064                 netdev_hw_addr_list_for_each(ha, &netdev->mc) {
2065                     mc_list->mc[mc_list->count] =
2066                         ether_addr_to_u64(ha->addr);
2067                     mc_list->count++;
2068                 }
2069             }
2070         }
2071     }
2072     spin_lock(&nic->rx_mode_wq_lock);
2073     kfree(nic->rx_mode_work.mc);
2074     nic->rx_mode_work.mc = mc_list;
2075     nic->rx_mode_work.mode = mode;
2076     queue_work(nic->nicvf_rx_mode_wq, &nic->rx_mode_work.work);
2077     spin_unlock(&nic->rx_mode_wq_lock);
2078 }
2079 
2080 static const struct net_device_ops nicvf_netdev_ops = {
2081     .ndo_open       = nicvf_open,
2082     .ndo_stop       = nicvf_stop,
2083     .ndo_start_xmit     = nicvf_xmit,
2084     .ndo_change_mtu     = nicvf_change_mtu,
2085     .ndo_set_mac_address    = nicvf_set_mac_address,
2086     .ndo_get_stats64    = nicvf_get_stats64,
2087     .ndo_tx_timeout         = nicvf_tx_timeout,
2088     .ndo_fix_features       = nicvf_fix_features,
2089     .ndo_set_features       = nicvf_set_features,
2090     .ndo_bpf        = nicvf_xdp,
2091     .ndo_eth_ioctl           = nicvf_ioctl,
2092     .ndo_set_rx_mode        = nicvf_set_rx_mode,
2093 };
2094 
2095 static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2096 {
2097     struct device *dev = &pdev->dev;
2098     struct net_device *netdev;
2099     struct nicvf *nic;
2100     int    err, qcount;
2101     u16    sdevid;
2102     struct cavium_ptp *ptp_clock;
2103 
2104     ptp_clock = cavium_ptp_get();
2105     if (IS_ERR(ptp_clock)) {
2106         if (PTR_ERR(ptp_clock) == -ENODEV)
2107             /* In virtualized environment we proceed without ptp */
2108             ptp_clock = NULL;
2109         else
2110             return PTR_ERR(ptp_clock);
2111     }
2112 
2113     err = pci_enable_device(pdev);
2114     if (err)
2115         return dev_err_probe(dev, err, "Failed to enable PCI device\n");
2116 
2117     err = pci_request_regions(pdev, DRV_NAME);
2118     if (err) {
2119         dev_err(dev, "PCI request regions failed 0x%x\n", err);
2120         goto err_disable_device;
2121     }
2122 
2123     err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48));
2124     if (err) {
2125         dev_err(dev, "Unable to get usable DMA configuration\n");
2126         goto err_release_regions;
2127     }
2128 
2129     qcount = netif_get_num_default_rss_queues();
2130 
2131     /* Restrict multiqset support only for host bound VFs */
2132     if (pdev->is_virtfn) {
2133         /* Set max number of queues per VF */
2134         qcount = min_t(int, num_online_cpus(),
2135                    (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS);
2136     }
2137 
2138     netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount);
2139     if (!netdev) {
2140         err = -ENOMEM;
2141         goto err_release_regions;
2142     }
2143 
2144     pci_set_drvdata(pdev, netdev);
2145 
2146     SET_NETDEV_DEV(netdev, &pdev->dev);
2147 
2148     nic = netdev_priv(netdev);
2149     nic->netdev = netdev;
2150     nic->pdev = pdev;
2151     nic->pnicvf = nic;
2152     nic->max_queues = qcount;
2153     /* If no of CPUs are too low, there won't be any queues left
2154      * for XDP_TX, hence double it.
2155      */
2156     if (!nic->t88)
2157         nic->max_queues *= 2;
2158     nic->ptp_clock = ptp_clock;
2159 
2160     /* Initialize mutex that serializes usage of VF's mailbox */
2161     mutex_init(&nic->rx_mode_mtx);
2162 
2163     /* MAP VF's configuration registers */
2164     nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0);
2165     if (!nic->reg_base) {
2166         dev_err(dev, "Cannot map config register space, aborting\n");
2167         err = -ENOMEM;
2168         goto err_free_netdev;
2169     }
2170 
2171     nic->drv_stats = netdev_alloc_pcpu_stats(struct nicvf_drv_stats);
2172     if (!nic->drv_stats) {
2173         err = -ENOMEM;
2174         goto err_free_netdev;
2175     }
2176 
2177     err = nicvf_set_qset_resources(nic);
2178     if (err)
2179         goto err_free_netdev;
2180 
2181     /* Check if PF is alive and get MAC address for this VF */
2182     err = nicvf_register_misc_interrupt(nic);
2183     if (err)
2184         goto err_free_netdev;
2185 
2186     nicvf_send_vf_struct(nic);
2187 
2188     if (!pass1_silicon(nic->pdev))
2189         nic->hw_tso = true;
2190 
2191     /* Get iommu domain for iova to physical addr conversion */
2192     nic->iommu_domain = iommu_get_domain_for_dev(dev);
2193 
2194     pci_read_config_word(nic->pdev, PCI_SUBSYSTEM_ID, &sdevid);
2195     if (sdevid == 0xA134)
2196         nic->t88 = true;
2197 
2198     /* Check if this VF is in QS only mode */
2199     if (nic->sqs_mode)
2200         return 0;
2201 
2202     err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues);
2203     if (err)
2204         goto err_unregister_interrupts;
2205 
2206     netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_SG |
2207                    NETIF_F_TSO | NETIF_F_GRO | NETIF_F_TSO6 |
2208                    NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2209                    NETIF_F_HW_VLAN_CTAG_RX);
2210 
2211     netdev->hw_features |= NETIF_F_RXHASH;
2212 
2213     netdev->features |= netdev->hw_features;
2214     netdev->hw_features |= NETIF_F_LOOPBACK;
2215 
2216     netdev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM |
2217                 NETIF_F_IPV6_CSUM | NETIF_F_TSO | NETIF_F_TSO6;
2218 
2219     netdev->netdev_ops = &nicvf_netdev_ops;
2220     netdev->watchdog_timeo = NICVF_TX_TIMEOUT;
2221 
2222     /* MTU range: 64 - 9200 */
2223     netdev->min_mtu = NIC_HW_MIN_FRS;
2224     netdev->max_mtu = NIC_HW_MAX_FRS;
2225 
2226     INIT_WORK(&nic->reset_task, nicvf_reset_task);
2227 
2228     nic->nicvf_rx_mode_wq = alloc_ordered_workqueue("nicvf_rx_mode_wq_VF%d",
2229                             WQ_MEM_RECLAIM,
2230                             nic->vf_id);
2231     if (!nic->nicvf_rx_mode_wq) {
2232         err = -ENOMEM;
2233         dev_err(dev, "Failed to allocate work queue\n");
2234         goto err_unregister_interrupts;
2235     }
2236 
2237     INIT_WORK(&nic->rx_mode_work.work, nicvf_set_rx_mode_task);
2238     spin_lock_init(&nic->rx_mode_wq_lock);
2239 
2240     err = register_netdev(netdev);
2241     if (err) {
2242         dev_err(dev, "Failed to register netdevice\n");
2243         goto err_unregister_interrupts;
2244     }
2245 
2246     nic->msg_enable = debug;
2247 
2248     nicvf_set_ethtool_ops(netdev);
2249 
2250     return 0;
2251 
2252 err_unregister_interrupts:
2253     nicvf_unregister_interrupts(nic);
2254 err_free_netdev:
2255     pci_set_drvdata(pdev, NULL);
2256     if (nic->drv_stats)
2257         free_percpu(nic->drv_stats);
2258     free_netdev(netdev);
2259 err_release_regions:
2260     pci_release_regions(pdev);
2261 err_disable_device:
2262     pci_disable_device(pdev);
2263     return err;
2264 }
2265 
2266 static void nicvf_remove(struct pci_dev *pdev)
2267 {
2268     struct net_device *netdev = pci_get_drvdata(pdev);
2269     struct nicvf *nic;
2270     struct net_device *pnetdev;
2271 
2272     if (!netdev)
2273         return;
2274 
2275     nic = netdev_priv(netdev);
2276     pnetdev = nic->pnicvf->netdev;
2277 
2278     /* Check if this Qset is assigned to different VF.
2279      * If yes, clean primary and all secondary Qsets.
2280      */
2281     if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED))
2282         unregister_netdev(pnetdev);
2283     if (nic->nicvf_rx_mode_wq) {
2284         destroy_workqueue(nic->nicvf_rx_mode_wq);
2285         nic->nicvf_rx_mode_wq = NULL;
2286     }
2287     nicvf_unregister_interrupts(nic);
2288     pci_set_drvdata(pdev, NULL);
2289     if (nic->drv_stats)
2290         free_percpu(nic->drv_stats);
2291     cavium_ptp_put(nic->ptp_clock);
2292     free_netdev(netdev);
2293     pci_release_regions(pdev);
2294     pci_disable_device(pdev);
2295 }
2296 
2297 static void nicvf_shutdown(struct pci_dev *pdev)
2298 {
2299     nicvf_remove(pdev);
2300 }
2301 
2302 static struct pci_driver nicvf_driver = {
2303     .name = DRV_NAME,
2304     .id_table = nicvf_id_table,
2305     .probe = nicvf_probe,
2306     .remove = nicvf_remove,
2307     .shutdown = nicvf_shutdown,
2308 };
2309 
2310 static int __init nicvf_init_module(void)
2311 {
2312     pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION);
2313     return pci_register_driver(&nicvf_driver);
2314 }
2315 
2316 static void __exit nicvf_cleanup_module(void)
2317 {
2318     pci_unregister_driver(&nicvf_driver);
2319 }
2320 
2321 module_init(nicvf_init_module);
2322 module_exit(nicvf_cleanup_module);
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!