blob: ae9582ddbc8fbf038f56b4a2d7f8f5e76fe99180 [file] [log] [blame]
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001/*
2 * Copyright(c) 2015 - 2018 Intel Corporation.
3 *
4 * This file is provided under a dual BSD/GPLv2 license. When using or
5 * redistributing this file, you may do so under either license.
6 *
7 * GPL LICENSE SUMMARY
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
17 *
18 * BSD LICENSE
19 *
20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
22 * are met:
23 *
24 * - Redistributions of source code must retain the above copyright
25 * notice, this list of conditions and the following disclaimer.
26 * - Redistributions in binary form must reproduce the above copyright
27 * notice, this list of conditions and the following disclaimer in
28 * the documentation and/or other materials provided with the
29 * distribution.
30 * - Neither the name of Intel Corporation nor the names of its
31 * contributors may be used to endorse or promote products derived
32 * from this software without specific prior written permission.
33 *
34 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45 *
46 */
47
48#ifndef HFI1_VERBS_H
49#define HFI1_VERBS_H
50
51#include <linux/types.h>
52#include <linux/seqlock.h>
53#include <linux/kernel.h>
54#include <linux/interrupt.h>
55#include <linux/kref.h>
56#include <linux/workqueue.h>
57#include <linux/kthread.h>
58#include <linux/completion.h>
59#include <linux/slab.h>
60#include <rdma/ib_pack.h>
61#include <rdma/ib_user_verbs.h>
62#include <rdma/ib_mad.h>
63#include <rdma/ib_hdrs.h>
64#include <rdma/rdma_vt.h>
65#include <rdma/rdmavt_qp.h>
66#include <rdma/rdmavt_cq.h>
67
68struct hfi1_ctxtdata;
69struct hfi1_pportdata;
70struct hfi1_devdata;
71struct hfi1_packet;
72
73#include "iowait.h"
David Brazdil0f672f62019-12-10 10:32:29 +000074#include "tid_rdma.h"
75#include "opfn.h"
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000076
77#define HFI1_MAX_RDMA_ATOMIC 16
78
79/*
80 * Increment this value if any changes that break userspace ABI
81 * compatibility are made.
82 */
83#define HFI1_UVERBS_ABI_VERSION 2
84
85/* IB Performance Manager status values */
86#define IB_PMA_SAMPLE_STATUS_DONE 0x00
87#define IB_PMA_SAMPLE_STATUS_STARTED 0x01
88#define IB_PMA_SAMPLE_STATUS_RUNNING 0x02
89
90/* Mandatory IB performance counter select values. */
91#define IB_PMA_PORT_XMIT_DATA cpu_to_be16(0x0001)
92#define IB_PMA_PORT_RCV_DATA cpu_to_be16(0x0002)
93#define IB_PMA_PORT_XMIT_PKTS cpu_to_be16(0x0003)
94#define IB_PMA_PORT_RCV_PKTS cpu_to_be16(0x0004)
95#define IB_PMA_PORT_XMIT_WAIT cpu_to_be16(0x0005)
96
97#define HFI1_VENDOR_IPG cpu_to_be16(0xFFA0)
98
99#define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL)
100#define OPA_BTH_MIG_REQ BIT(31)
101
102#define RC_OP(x) IB_OPCODE_RC_##x
103#define UC_OP(x) IB_OPCODE_UC_##x
104
105/* flags passed by hfi1_ib_rcv() */
106enum {
107 HFI1_HAS_GRH = (1 << 0),
108};
109
110#define LRH_16B_BYTES (FIELD_SIZEOF(struct hfi1_16b_header, lrh))
111#define LRH_16B_DWORDS (LRH_16B_BYTES / sizeof(u32))
112#define LRH_9B_BYTES (FIELD_SIZEOF(struct ib_header, lrh))
113#define LRH_9B_DWORDS (LRH_9B_BYTES / sizeof(u32))
114
115/* 24Bits for qpn, upper 8Bits reserved */
116struct opa_16b_mgmt {
117 __be32 dest_qpn;
118 __be32 src_qpn;
119};
120
121struct hfi1_16b_header {
122 u32 lrh[4];
123 union {
124 struct {
125 struct ib_grh grh;
126 struct ib_other_headers oth;
127 } l;
128 struct ib_other_headers oth;
129 struct opa_16b_mgmt mgmt;
130 } u;
131} __packed;
132
133struct hfi1_opa_header {
134 union {
135 struct ib_header ibh; /* 9B header */
136 struct hfi1_16b_header opah; /* 16B header */
137 };
138 u8 hdr_type; /* 9B or 16B */
139} __packed;
140
141struct hfi1_ahg_info {
142 u32 ahgdesc[2];
143 u16 tx_flags;
144 u8 ahgcount;
145 u8 ahgidx;
146};
147
148struct hfi1_sdma_header {
149 __le64 pbc;
150 struct hfi1_opa_header hdr;
151} __packed;
152
153/*
154 * hfi1 specific data structures that will be hidden from rvt after the queue
155 * pair is made common
156 */
157struct hfi1_qp_priv {
158 struct hfi1_ahg_info *s_ahg; /* ahg info for next header */
159 struct sdma_engine *s_sde; /* current sde */
160 struct send_context *s_sendcontext; /* current sendcontext */
David Brazdil0f672f62019-12-10 10:32:29 +0000161 struct hfi1_ctxtdata *rcd; /* QP's receive context */
162 struct page **pages; /* for TID page scan */
163 u32 tid_enqueue; /* saved when tid waited */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000164 u8 s_sc; /* SC[0..4] for next packet */
165 struct iowait s_iowait;
David Brazdil0f672f62019-12-10 10:32:29 +0000166 struct timer_list s_tid_timer; /* for timing tid wait */
167 struct timer_list s_tid_retry_timer; /* for timing tid ack */
168 struct list_head tid_wait; /* for queueing tid space */
169 struct hfi1_opfn_data opfn;
170 struct tid_flow_state flow_state;
171 struct tid_rdma_qp_params tid_rdma;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000172 struct rvt_qp *owner;
David Brazdil0f672f62019-12-10 10:32:29 +0000173 u16 s_running_pkt_size;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000174 u8 hdr_type; /* 9B or 16B */
David Brazdil0f672f62019-12-10 10:32:29 +0000175 struct rvt_sge_state tid_ss; /* SGE state pointer for 2nd leg */
176 atomic_t n_requests; /* # of TID RDMA requests in the */
177 /* queue */
178 atomic_t n_tid_requests; /* # of sent TID RDMA requests */
179 unsigned long tid_timer_timeout_jiffies;
180 unsigned long tid_retry_timeout_jiffies;
181
182 /* variables for the TID RDMA SE state machine */
183 u8 s_state;
184 u8 s_retry;
185 u8 rnr_nak_state; /* RNR NAK state */
186 u8 s_nak_state;
187 u32 s_nak_psn;
188 u32 s_flags;
189 u32 s_tid_cur;
190 u32 s_tid_head;
191 u32 s_tid_tail;
192 u32 r_tid_head; /* Most recently added TID RDMA request */
193 u32 r_tid_tail; /* the last completed TID RDMA request */
194 u32 r_tid_ack; /* the TID RDMA request to be ACK'ed */
195 u32 r_tid_alloc; /* Request for which we are allocating resources */
196 u32 pending_tid_w_segs; /* Num of pending tid write segments */
197 u32 pending_tid_w_resp; /* Num of pending tid write responses */
198 u32 alloc_w_segs; /* Number of segments for which write */
199 /* resources have been allocated for this QP */
200
201 /* For TID RDMA READ */
202 u32 tid_r_reqs; /* Num of tid reads requested */
203 u32 tid_r_comp; /* Num of tid reads completed */
204 u32 pending_tid_r_segs; /* Num of pending tid read segments */
205 u16 pkts_ps; /* packets per segment */
206 u8 timeout_shift; /* account for number of packets per segment */
207
208 u32 r_next_psn_kdeth;
209 u32 r_next_psn_kdeth_save;
210 u32 s_resync_psn;
211 u8 sync_pt; /* Set when QP reaches sync point */
212 u8 resync;
213};
214
215#define HFI1_QP_WQE_INVALID ((u32)-1)
216
217struct hfi1_swqe_priv {
218 struct tid_rdma_request tid_req;
219 struct rvt_sge_state ss; /* Used for TID RDMA READ Request */
220};
221
222struct hfi1_ack_priv {
223 struct rvt_sge_state ss; /* used for TID WRITE RESP */
224 struct tid_rdma_request tid_req;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000225};
226
227/*
228 * This structure is used to hold commonly lookedup and computed values during
229 * the send engine progress.
230 */
David Brazdil0f672f62019-12-10 10:32:29 +0000231struct iowait_work;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000232struct hfi1_pkt_state {
233 struct hfi1_ibdev *dev;
234 struct hfi1_ibport *ibp;
235 struct hfi1_pportdata *ppd;
236 struct verbs_txreq *s_txreq;
David Brazdil0f672f62019-12-10 10:32:29 +0000237 struct iowait_work *wait;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000238 unsigned long flags;
239 unsigned long timeout;
240 unsigned long timeout_int;
241 int cpu;
242 u8 opcode;
243 bool in_thread;
244 bool pkts_sent;
245};
246
247#define HFI1_PSN_CREDIT 16
248
249struct hfi1_opcode_stats {
250 u64 n_packets; /* number of packets */
251 u64 n_bytes; /* total number of bytes */
252};
253
254struct hfi1_opcode_stats_perctx {
255 struct hfi1_opcode_stats stats[256];
256};
257
258static inline void inc_opstats(
259 u32 tlen,
260 struct hfi1_opcode_stats *stats)
261{
262#ifdef CONFIG_DEBUG_FS
263 stats->n_bytes += tlen;
264 stats->n_packets++;
265#endif
266}
267
268struct hfi1_ibport {
269 struct rvt_qp __rcu *qp[2];
270 struct rvt_ibport rvp;
271
272 /* the first 16 entries are sl_to_vl for !OPA */
273 u8 sl_to_sc[32];
274 u8 sc_to_sl[32];
275};
276
277struct hfi1_ibdev {
278 struct rvt_dev_info rdi; /* Must be first */
279
280 /* QP numbers are shared by all IB ports */
281 /* protect txwait list */
282 seqlock_t txwait_lock ____cacheline_aligned_in_smp;
283 struct list_head txwait; /* list for wait verbs_txreq */
284 struct list_head memwait; /* list for wait kernel memory */
285 struct kmem_cache *verbs_txreq_cache;
286 u64 n_txwait;
287 u64 n_kmem_wait;
David Brazdil0f672f62019-12-10 10:32:29 +0000288 u64 n_tidwait;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000289
290 /* protect iowait lists */
291 seqlock_t iowait_lock ____cacheline_aligned_in_smp;
292 u64 n_piowait;
293 u64 n_piodrain;
294 struct timer_list mem_timer;
295
296#ifdef CONFIG_DEBUG_FS
297 /* per HFI debugfs */
298 struct dentry *hfi1_ibdev_dbg;
299 /* per HFI symlinks to above */
300 struct dentry *hfi1_ibdev_link;
301#ifdef CONFIG_FAULT_INJECTION
302 struct fault *fault;
303#endif
304#endif
305};
306
307static inline struct hfi1_ibdev *to_idev(struct ib_device *ibdev)
308{
309 struct rvt_dev_info *rdi;
310
311 rdi = container_of(ibdev, struct rvt_dev_info, ibdev);
312 return container_of(rdi, struct hfi1_ibdev, rdi);
313}
314
David Brazdil0f672f62019-12-10 10:32:29 +0000315static inline struct rvt_qp *iowait_to_qp(struct iowait *s_iowait)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000316{
317 struct hfi1_qp_priv *priv;
318
319 priv = container_of(s_iowait, struct hfi1_qp_priv, s_iowait);
320 return priv->owner;
321}
322
323/*
324 * This must be called with s_lock held.
325 */
326void hfi1_bad_pkey(struct hfi1_ibport *ibp, u32 key, u32 sl,
327 u32 qp1, u32 qp2, u32 lid1, u32 lid2);
328void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num);
329void hfi1_sys_guid_chg(struct hfi1_ibport *ibp);
330void hfi1_node_desc_chg(struct hfi1_ibport *ibp);
331int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
332 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
333 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
334 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
335 u16 *out_mad_pkey_index);
336
337/*
338 * The PSN_MASK and PSN_SHIFT allow for
339 * 1) comparing two PSNs
340 * 2) returning the PSN with any upper bits masked
341 * 3) returning the difference between to PSNs
342 *
343 * The number of significant bits in the PSN must
344 * necessarily be at least one bit less than
345 * the container holding the PSN.
346 */
347#define PSN_MASK 0x7FFFFFFF
348#define PSN_SHIFT 1
349#define PSN_MODIFY_MASK 0xFFFFFF
350
351/*
352 * Compare two PSNs
353 * Returns an integer <, ==, or > than zero.
354 */
355static inline int cmp_psn(u32 a, u32 b)
356{
357 return (((int)a) - ((int)b)) << PSN_SHIFT;
358}
359
360/*
361 * Return masked PSN
362 */
363static inline u32 mask_psn(u32 a)
364{
365 return a & PSN_MASK;
366}
367
368/*
369 * Return delta between two PSNs
370 */
371static inline u32 delta_psn(u32 a, u32 b)
372{
373 return (((int)a - (int)b) << PSN_SHIFT) >> PSN_SHIFT;
374}
375
David Brazdil0f672f62019-12-10 10:32:29 +0000376static inline struct tid_rdma_request *wqe_to_tid_req(struct rvt_swqe *wqe)
377{
378 return &((struct hfi1_swqe_priv *)wqe->priv)->tid_req;
379}
380
381static inline struct tid_rdma_request *ack_to_tid_req(struct rvt_ack_entry *e)
382{
383 return &((struct hfi1_ack_priv *)e->priv)->tid_req;
384}
385
386/*
387 * Look through all the active flows for a TID RDMA request and find
388 * the one (if it exists) that contains the specified PSN.
389 */
390static inline u32 __full_flow_psn(struct flow_state *state, u32 psn)
391{
392 return mask_psn((state->generation << HFI1_KDETH_BTH_SEQ_SHIFT) |
393 (psn & HFI1_KDETH_BTH_SEQ_MASK));
394}
395
396static inline u32 full_flow_psn(struct tid_rdma_flow *flow, u32 psn)
397{
398 return __full_flow_psn(&flow->flow_state, psn);
399}
400
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000401struct verbs_txreq;
402void hfi1_put_txreq(struct verbs_txreq *tx);
403
404int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
405
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000406void hfi1_cnp_rcv(struct hfi1_packet *packet);
407
408void hfi1_uc_rcv(struct hfi1_packet *packet);
409
410void hfi1_rc_rcv(struct hfi1_packet *packet);
411
412void hfi1_rc_hdrerr(
413 struct hfi1_ctxtdata *rcd,
414 struct hfi1_packet *packet,
415 struct rvt_qp *qp);
416
417u8 ah_to_sc(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr);
418
David Brazdil0f672f62019-12-10 10:32:29 +0000419void hfi1_rc_verbs_aborted(struct rvt_qp *qp, struct hfi1_opa_header *opah);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000420void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah);
421
422void hfi1_ud_rcv(struct hfi1_packet *packet);
423
424int hfi1_lookup_pkey_idx(struct hfi1_ibport *ibp, u16 pkey);
425
426void hfi1_migrate_qp(struct rvt_qp *qp);
427
428int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
429 int attr_mask, struct ib_udata *udata);
430
431void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
432 int attr_mask, struct ib_udata *udata);
433void hfi1_restart_rc(struct rvt_qp *qp, u32 psn, int wait);
David Brazdil0f672f62019-12-10 10:32:29 +0000434int hfi1_setup_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe,
435 bool *call_send);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000436
437extern const u32 rc_only_opcode;
438extern const u32 uc_only_opcode;
439
440int hfi1_ruc_check_hdr(struct hfi1_ibport *ibp, struct hfi1_packet *packet);
441
442u32 hfi1_make_grh(struct hfi1_ibport *ibp, struct ib_grh *hdr,
443 const struct ib_global_route *grh, u32 hwords, u32 nwords);
444
445void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
David Brazdil0f672f62019-12-10 10:32:29 +0000446 u32 bth0, u32 bth1, u32 bth2, int middle,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000447 struct hfi1_pkt_state *ps);
448
David Brazdil0f672f62019-12-10 10:32:29 +0000449bool hfi1_schedule_send_yield(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
450 bool tid);
451
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000452void _hfi1_do_send(struct work_struct *work);
453
454void hfi1_do_send_from_rvt(struct rvt_qp *qp);
455
456void hfi1_do_send(struct rvt_qp *qp, bool in_thread);
457
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000458void hfi1_send_rc_ack(struct hfi1_packet *packet, bool is_fecn);
459
460int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
461
462int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
463
464int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
465
466int hfi1_register_ib_device(struct hfi1_devdata *);
467
468void hfi1_unregister_ib_device(struct hfi1_devdata *);
469
David Brazdil0f672f62019-12-10 10:32:29 +0000470void hfi1_kdeth_eager_rcv(struct hfi1_packet *packet);
471
472void hfi1_kdeth_expected_rcv(struct hfi1_packet *packet);
473
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000474void hfi1_ib_rcv(struct hfi1_packet *packet);
475
476void hfi1_16B_rcv(struct hfi1_packet *packet);
477
478unsigned hfi1_get_npkeys(struct hfi1_devdata *);
479
480int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
481 u64 pbc);
482
483int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
484 u64 pbc);
485
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000486static inline bool opa_bth_is_migration(struct ib_other_headers *ohdr)
487{
488 return ohdr->bth[1] & cpu_to_be32(OPA_BTH_MIG_REQ);
489}
490
David Brazdil0f672f62019-12-10 10:32:29 +0000491void hfi1_wait_kmem(struct rvt_qp *qp);
492
493static inline void hfi1_trdma_send_complete(struct rvt_qp *qp,
494 struct rvt_swqe *wqe,
495 enum ib_wc_status status)
496{
497 trdma_clean_swqe(qp, wqe);
498 rvt_send_complete(qp, wqe, status);
499}
500
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000501extern const enum ib_wc_opcode ib_hfi1_wc_opcode[];
502
503extern const u8 hdr_len_by_opcode[];
504
505extern const int ib_rvt_state_ops[];
506
507extern __be64 ib_hfi1_sys_image_guid; /* in network order */
508
509extern unsigned int hfi1_max_cqes;
510
511extern unsigned int hfi1_max_cqs;
512
513extern unsigned int hfi1_max_qp_wrs;
514
515extern unsigned int hfi1_max_qps;
516
517extern unsigned int hfi1_max_sges;
518
519extern unsigned int hfi1_max_mcast_grps;
520
521extern unsigned int hfi1_max_mcast_qp_attached;
522
523extern unsigned int hfi1_max_srqs;
524
525extern unsigned int hfi1_max_srq_sges;
526
527extern unsigned int hfi1_max_srq_wrs;
528
529extern unsigned short piothreshold;
530
531extern const u32 ib_hfi1_rnr_table[];
532
533#endif /* HFI1_VERBS_H */