hw/hfi1/verbs.h

0001 /* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
0002 /*
0003  * Copyright(c) 2015 - 2018 Intel Corporation.
0004  */
0005
0006 #ifndef HFI1_VERBS_H
0007 #define HFI1_VERBS_H
0008
0009 #include <linux/types.h>
0010 #include <linux/seqlock.h>
0011 #include <linux/kernel.h>
0012 #include <linux/interrupt.h>
0013 #include <linux/kref.h>
0014 #include <linux/workqueue.h>
0015 #include <linux/kthread.h>
0016 #include <linux/completion.h>
0017 #include <linux/slab.h>
0018 #include <rdma/ib_pack.h>
0019 #include <rdma/ib_user_verbs.h>
0020 #include <rdma/ib_mad.h>
0021 #include <rdma/ib_hdrs.h>
0022 #include <rdma/rdma_vt.h>
0023 #include <rdma/rdmavt_qp.h>
0024 #include <rdma/rdmavt_cq.h>
0025
0026 struct hfi1_ctxtdata;
0027 struct hfi1_pportdata;
0028 struct hfi1_devdata;
0029 struct hfi1_packet;
0030
0031 #include "iowait.h"
0032 #include "tid_rdma.h"
0033 #include "opfn.h"
0034
0035 #define HFI1_MAX_RDMA_ATOMIC     16
0036
0037 /*
0038  * Increment this value if any changes that break userspace ABI
0039  * compatibility are made.
0040  */
0041 #define HFI1_UVERBS_ABI_VERSION       2
0042
0043 /* IB Performance Manager status values */
0044 #define IB_PMA_SAMPLE_STATUS_DONE       0x00
0045 #define IB_PMA_SAMPLE_STATUS_STARTED    0x01
0046 #define IB_PMA_SAMPLE_STATUS_RUNNING    0x02
0047
0048 /* Mandatory IB performance counter select values. */
0049 #define IB_PMA_PORT_XMIT_DATA   cpu_to_be16(0x0001)
0050 #define IB_PMA_PORT_RCV_DATA    cpu_to_be16(0x0002)
0051 #define IB_PMA_PORT_XMIT_PKTS   cpu_to_be16(0x0003)
0052 #define IB_PMA_PORT_RCV_PKTS    cpu_to_be16(0x0004)
0053 #define IB_PMA_PORT_XMIT_WAIT   cpu_to_be16(0x0005)
0054
0055 #define HFI1_VENDOR_IPG     cpu_to_be16(0xFFA0)
0056
0057 #define IB_DEFAULT_GID_PREFIX   cpu_to_be64(0xfe80000000000000ULL)
0058 #define OPA_BTH_MIG_REQ     BIT(31)
0059
0060 #define RC_OP(x) IB_OPCODE_RC_##x
0061 #define UC_OP(x) IB_OPCODE_UC_##x
0062
0063 /* flags passed by hfi1_ib_rcv() */
0064 enum {
0065     HFI1_HAS_GRH = (1 << 0),
0066 };
0067
0068 #define LRH_16B_BYTES (sizeof_field(struct hfi1_16b_header, lrh))
0069 #define LRH_16B_DWORDS (LRH_16B_BYTES / sizeof(u32))
0070 #define LRH_9B_BYTES (sizeof_field(struct ib_header, lrh))
0071 #define LRH_9B_DWORDS (LRH_9B_BYTES / sizeof(u32))
0072
0073 /* 24Bits for qpn, upper 8Bits reserved */
0074 struct opa_16b_mgmt {
0075     __be32 dest_qpn;
0076     __be32 src_qpn;
0077 };
0078
0079 struct hfi1_16b_header {
0080     u32 lrh[4];
0081     union {
0082         struct {
0083             struct ib_grh grh;
0084             struct ib_other_headers oth;
0085         } l;
0086         struct ib_other_headers oth;
0087         struct opa_16b_mgmt mgmt;
0088     } u;
0089 } __packed;
0090
0091 struct hfi1_opa_header {
0092     union {
0093         struct ib_header ibh; /* 9B header */
0094         struct hfi1_16b_header opah; /* 16B header */
0095     };
0096     u8 hdr_type; /* 9B or 16B */
0097 } __packed;
0098
0099 struct hfi1_ahg_info {
0100     u32 ahgdesc[2];
0101     u16 tx_flags;
0102     u8 ahgcount;
0103     u8 ahgidx;
0104 };
0105
0106 struct hfi1_sdma_header {
0107     __le64 pbc;
0108     struct hfi1_opa_header hdr;
0109 } __packed;
0110
0111 /*
0112  * hfi1 specific data structures that will be hidden from rvt after the queue
0113  * pair is made common
0114  */
0115 struct hfi1_qp_priv {
0116     struct hfi1_ahg_info *s_ahg;              /* ahg info for next header */
0117     struct sdma_engine *s_sde;                /* current sde */
0118     struct send_context *s_sendcontext;       /* current sendcontext */
0119     struct hfi1_ctxtdata *rcd;                /* QP's receive context */
0120     struct page **pages;                      /* for TID page scan */
0121     u32 tid_enqueue;                          /* saved when tid waited */
0122     u8 s_sc;                          /* SC[0..4] for next packet */
0123     struct iowait s_iowait;
0124     struct timer_list s_tid_timer;            /* for timing tid wait */
0125     struct timer_list s_tid_retry_timer;      /* for timing tid ack */
0126     struct list_head tid_wait;                /* for queueing tid space */
0127     struct hfi1_opfn_data opfn;
0128     struct tid_flow_state flow_state;
0129     struct tid_rdma_qp_params tid_rdma;
0130     struct rvt_qp *owner;
0131     u16 s_running_pkt_size;
0132     u8 hdr_type; /* 9B or 16B */
0133     struct rvt_sge_state tid_ss;       /* SGE state pointer for 2nd leg */
0134     atomic_t n_requests;               /* # of TID RDMA requests in the */
0135                        /* queue */
0136     atomic_t n_tid_requests;            /* # of sent TID RDMA requests */
0137     unsigned long tid_timer_timeout_jiffies;
0138     unsigned long tid_retry_timeout_jiffies;
0139
0140     /* variables for the TID RDMA SE state machine */
0141     u8 s_state;
0142     u8 s_retry;
0143     u8 rnr_nak_state;       /* RNR NAK state */
0144     u8 s_nak_state;
0145     u32 s_nak_psn;
0146     u32 s_flags;
0147     u32 s_tid_cur;
0148     u32 s_tid_head;
0149     u32 s_tid_tail;
0150     u32 r_tid_head;     /* Most recently added TID RDMA request */
0151     u32 r_tid_tail;     /* the last completed TID RDMA request */
0152     u32 r_tid_ack;      /* the TID RDMA request to be ACK'ed */
0153     u32 r_tid_alloc;    /* Request for which we are allocating resources */
0154     u32 pending_tid_w_segs; /* Num of pending tid write segments */
0155     u32 pending_tid_w_resp; /* Num of pending tid write responses */
0156     u32 alloc_w_segs;       /* Number of segments for which write */
0157                    /* resources have been allocated for this QP */
0158
0159     /* For TID RDMA READ */
0160     u32 tid_r_reqs;         /* Num of tid reads requested */
0161     u32 tid_r_comp;         /* Num of tid reads completed */
0162     u32 pending_tid_r_segs; /* Num of pending tid read segments */
0163     u16 pkts_ps;            /* packets per segment */
0164     u8 timeout_shift;       /* account for number of packets per segment */
0165
0166     u32 r_next_psn_kdeth;
0167     u32 r_next_psn_kdeth_save;
0168     u32 s_resync_psn;
0169     u8 sync_pt;           /* Set when QP reaches sync point */
0170     u8 resync;
0171 };
0172
0173 #define HFI1_QP_WQE_INVALID   ((u32)-1)
0174
0175 struct hfi1_swqe_priv {
0176     struct tid_rdma_request tid_req;
0177     struct rvt_sge_state ss;  /* Used for TID RDMA READ Request */
0178 };
0179
0180 struct hfi1_ack_priv {
0181     struct rvt_sge_state ss;               /* used for TID WRITE RESP */
0182     struct tid_rdma_request tid_req;
0183 };
0184
0185 /*
0186  * This structure is used to hold commonly lookedup and computed values during
0187  * the send engine progress.
0188  */
0189 struct iowait_work;
0190 struct hfi1_pkt_state {
0191     struct hfi1_ibdev *dev;
0192     struct hfi1_ibport *ibp;
0193     struct hfi1_pportdata *ppd;
0194     struct verbs_txreq *s_txreq;
0195     struct iowait_work *wait;
0196     unsigned long flags;
0197     unsigned long timeout;
0198     unsigned long timeout_int;
0199     int cpu;
0200     u8 opcode;
0201     bool in_thread;
0202     bool pkts_sent;
0203 };
0204
0205 #define HFI1_PSN_CREDIT  16
0206
0207 struct hfi1_opcode_stats {
0208     u64 n_packets;          /* number of packets */
0209     u64 n_bytes;            /* total number of bytes */
0210 };
0211
0212 struct hfi1_opcode_stats_perctx {
0213     struct hfi1_opcode_stats stats[256];
0214 };
0215
0216 static inline void inc_opstats(
0217     u32 tlen,
0218     struct hfi1_opcode_stats *stats)
0219 {
0220 #ifdef CONFIG_DEBUG_FS
0221     stats->n_bytes += tlen;
0222     stats->n_packets++;
0223 #endif
0224 }
0225
0226 struct hfi1_ibport {
0227     struct rvt_qp __rcu *qp[2];
0228     struct rvt_ibport rvp;
0229
0230     /* the first 16 entries are sl_to_vl for !OPA */
0231     u8 sl_to_sc[32];
0232     u8 sc_to_sl[32];
0233 };
0234
0235 struct hfi1_ibdev {
0236     struct rvt_dev_info rdi; /* Must be first */
0237
0238     /* QP numbers are shared by all IB ports */
0239     /* protect txwait list */
0240     seqlock_t txwait_lock ____cacheline_aligned_in_smp;
0241     struct list_head txwait;        /* list for wait verbs_txreq */
0242     struct list_head memwait;       /* list for wait kernel memory */
0243     struct kmem_cache *verbs_txreq_cache;
0244     u64 n_txwait;
0245     u64 n_kmem_wait;
0246     u64 n_tidwait;
0247
0248     /* protect iowait lists */
0249     seqlock_t iowait_lock ____cacheline_aligned_in_smp;
0250     u64 n_piowait;
0251     u64 n_piodrain;
0252     struct timer_list mem_timer;
0253
0254 #ifdef CONFIG_DEBUG_FS
0255     /* per HFI debugfs */
0256     struct dentry *hfi1_ibdev_dbg;
0257     /* per HFI symlinks to above */
0258     struct dentry *hfi1_ibdev_link;
0259 #ifdef CONFIG_FAULT_INJECTION
0260     struct fault *fault;
0261 #endif
0262 #endif
0263 };
0264
0265 static inline struct hfi1_ibdev *to_idev(struct ib_device *ibdev)
0266 {
0267     struct rvt_dev_info *rdi;
0268
0269     rdi = container_of(ibdev, struct rvt_dev_info, ibdev);
0270     return container_of(rdi, struct hfi1_ibdev, rdi);
0271 }
0272
0273 static inline struct rvt_qp *iowait_to_qp(struct iowait *s_iowait)
0274 {
0275     struct hfi1_qp_priv *priv;
0276
0277     priv = container_of(s_iowait, struct hfi1_qp_priv, s_iowait);
0278     return priv->owner;
0279 }
0280
0281 /*
0282  * This must be called with s_lock held.
0283  */
0284 void hfi1_bad_pkey(struct hfi1_ibport *ibp, u32 key, u32 sl,
0285            u32 qp1, u32 qp2, u32 lid1, u32 lid2);
0286 void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u32 port_num);
0287 void hfi1_sys_guid_chg(struct hfi1_ibport *ibp);
0288 void hfi1_node_desc_chg(struct hfi1_ibport *ibp);
0289 int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u32 port,
0290              const struct ib_wc *in_wc, const struct ib_grh *in_grh,
0291              const struct ib_mad *in_mad, struct ib_mad *out_mad,
0292              size_t *out_mad_size, u16 *out_mad_pkey_index);
0293
0294 /*
0295  * The PSN_MASK and PSN_SHIFT allow for
0296  * 1) comparing two PSNs
0297  * 2) returning the PSN with any upper bits masked
0298  * 3) returning the difference between to PSNs
0299  *
0300  * The number of significant bits in the PSN must
0301  * necessarily be at least one bit less than
0302  * the container holding the PSN.
0303  */
0304 #define PSN_MASK 0x7FFFFFFF
0305 #define PSN_SHIFT 1
0306 #define PSN_MODIFY_MASK 0xFFFFFF
0307
0308 /*
0309  * Compare two PSNs
0310  * Returns an integer <, ==, or > than zero.
0311  */
0312 static inline int cmp_psn(u32 a, u32 b)
0313 {
0314     return (((int)a) - ((int)b)) << PSN_SHIFT;
0315 }
0316
0317 /*
0318  * Return masked PSN
0319  */
0320 static inline u32 mask_psn(u32 a)
0321 {
0322     return a & PSN_MASK;
0323 }
0324
0325 /*
0326  * Return delta between two PSNs
0327  */
0328 static inline u32 delta_psn(u32 a, u32 b)
0329 {
0330     return (((int)a - (int)b) << PSN_SHIFT) >> PSN_SHIFT;
0331 }
0332
0333 static inline struct tid_rdma_request *wqe_to_tid_req(struct rvt_swqe *wqe)
0334 {
0335     return &((struct hfi1_swqe_priv *)wqe->priv)->tid_req;
0336 }
0337
0338 static inline struct tid_rdma_request *ack_to_tid_req(struct rvt_ack_entry *e)
0339 {
0340     return &((struct hfi1_ack_priv *)e->priv)->tid_req;
0341 }
0342
0343 /*
0344  * Look through all the active flows for a TID RDMA request and find
0345  * the one (if it exists) that contains the specified PSN.
0346  */
0347 static inline u32 __full_flow_psn(struct flow_state *state, u32 psn)
0348 {
0349     return mask_psn((state->generation << HFI1_KDETH_BTH_SEQ_SHIFT) |
0350             (psn & HFI1_KDETH_BTH_SEQ_MASK));
0351 }
0352
0353 static inline u32 full_flow_psn(struct tid_rdma_flow *flow, u32 psn)
0354 {
0355     return __full_flow_psn(&flow->flow_state, psn);
0356 }
0357
0358 struct verbs_txreq;
0359 void hfi1_put_txreq(struct verbs_txreq *tx);
0360
0361 int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
0362
0363 void hfi1_cnp_rcv(struct hfi1_packet *packet);
0364
0365 void hfi1_uc_rcv(struct hfi1_packet *packet);
0366
0367 void hfi1_rc_rcv(struct hfi1_packet *packet);
0368
0369 void hfi1_rc_hdrerr(
0370     struct hfi1_ctxtdata *rcd,
0371     struct hfi1_packet *packet,
0372     struct rvt_qp *qp);
0373
0374 u8 ah_to_sc(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr);
0375
0376 void hfi1_rc_verbs_aborted(struct rvt_qp *qp, struct hfi1_opa_header *opah);
0377 void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah);
0378
0379 void hfi1_ud_rcv(struct hfi1_packet *packet);
0380
0381 int hfi1_lookup_pkey_idx(struct hfi1_ibport *ibp, u16 pkey);
0382
0383 void hfi1_migrate_qp(struct rvt_qp *qp);
0384
0385 int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
0386              int attr_mask, struct ib_udata *udata);
0387
0388 void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
0389             int attr_mask, struct ib_udata *udata);
0390 void hfi1_restart_rc(struct rvt_qp *qp, u32 psn, int wait);
0391 int hfi1_setup_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe,
0392            bool *call_send);
0393
0394 extern const u32 rc_only_opcode;
0395 extern const u32 uc_only_opcode;
0396
0397 int hfi1_ruc_check_hdr(struct hfi1_ibport *ibp, struct hfi1_packet *packet);
0398
0399 u32 hfi1_make_grh(struct hfi1_ibport *ibp, struct ib_grh *hdr,
0400           const struct ib_global_route *grh, u32 hwords, u32 nwords);
0401
0402 void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
0403               u32 bth0, u32 bth1, u32 bth2, int middle,
0404               struct hfi1_pkt_state *ps);
0405
0406 bool hfi1_schedule_send_yield(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
0407                   bool tid);
0408
0409 void _hfi1_do_send(struct work_struct *work);
0410
0411 void hfi1_do_send_from_rvt(struct rvt_qp *qp);
0412
0413 void hfi1_do_send(struct rvt_qp *qp, bool in_thread);
0414
0415 void hfi1_send_rc_ack(struct hfi1_packet *packet, bool is_fecn);
0416
0417 int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
0418
0419 int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
0420
0421 int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
0422
0423 int hfi1_register_ib_device(struct hfi1_devdata *);
0424
0425 void hfi1_unregister_ib_device(struct hfi1_devdata *);
0426
0427 void hfi1_kdeth_eager_rcv(struct hfi1_packet *packet);
0428
0429 void hfi1_kdeth_expected_rcv(struct hfi1_packet *packet);
0430
0431 void hfi1_ib_rcv(struct hfi1_packet *packet);
0432
0433 void hfi1_16B_rcv(struct hfi1_packet *packet);
0434
0435 unsigned hfi1_get_npkeys(struct hfi1_devdata *);
0436
0437 int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
0438             u64 pbc);
0439
0440 int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
0441             u64 pbc);
0442
0443 static inline bool opa_bth_is_migration(struct ib_other_headers *ohdr)
0444 {
0445     return ohdr->bth[1] & cpu_to_be32(OPA_BTH_MIG_REQ);
0446 }
0447
0448 void hfi1_wait_kmem(struct rvt_qp *qp);
0449
0450 static inline void hfi1_trdma_send_complete(struct rvt_qp *qp,
0451                         struct rvt_swqe *wqe,
0452                         enum ib_wc_status status)
0453 {
0454     trdma_clean_swqe(qp, wqe);
0455     rvt_send_complete(qp, wqe, status);
0456 }
0457
0458 extern const enum ib_wc_opcode ib_hfi1_wc_opcode[];
0459
0460 extern const u8 hdr_len_by_opcode[];
0461
0462 extern const int ib_rvt_state_ops[];
0463
0464 extern __be64 ib_hfi1_sys_image_guid;    /* in network order */
0465
0466 extern unsigned int hfi1_max_cqes;
0467
0468 extern unsigned int hfi1_max_cqs;
0469
0470 extern unsigned int hfi1_max_qp_wrs;
0471
0472 extern unsigned int hfi1_max_qps;
0473
0474 extern unsigned int hfi1_max_sges;
0475
0476 extern unsigned int hfi1_max_mcast_grps;
0477
0478 extern unsigned int hfi1_max_mcast_qp_attached;
0479
0480 extern unsigned int hfi1_max_srqs;
0481
0482 extern unsigned int hfi1_max_srq_sges;
0483
0484 extern unsigned int hfi1_max_srq_wrs;
0485
0486 extern unsigned short piothreshold;
0487
0488 extern const u32 ib_hfi1_rnr_table[];
0489
0490 #endif                          /* HFI1_VERBS_H */