2 * Copyright(c) 2015 - 2018 Intel Corporation.
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.
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.
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.
20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
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
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.
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.
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>
69 struct hfi1_pportdata
;
77 #define HFI1_MAX_RDMA_ATOMIC 16
80 * Increment this value if any changes that break userspace ABI
81 * compatibility are made.
83 #define HFI1_UVERBS_ABI_VERSION 2
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
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)
97 #define HFI1_VENDOR_IPG cpu_to_be16(0xFFA0)
99 #define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL)
100 #define OPA_BTH_MIG_REQ BIT(31)
102 #define RC_OP(x) IB_OPCODE_RC_##x
103 #define UC_OP(x) IB_OPCODE_UC_##x
105 /* flags passed by hfi1_ib_rcv() */
107 HFI1_HAS_GRH
= (1 << 0),
110 #define LRH_16B_BYTES (sizeof_field(struct hfi1_16b_header, lrh))
111 #define LRH_16B_DWORDS (LRH_16B_BYTES / sizeof(u32))
112 #define LRH_9B_BYTES (sizeof_field(struct ib_header, lrh))
113 #define LRH_9B_DWORDS (LRH_9B_BYTES / sizeof(u32))
115 /* 24Bits for qpn, upper 8Bits reserved */
116 struct opa_16b_mgmt
{
121 struct hfi1_16b_header
{
126 struct ib_other_headers oth
;
128 struct ib_other_headers oth
;
129 struct opa_16b_mgmt mgmt
;
133 struct hfi1_opa_header
{
135 struct ib_header ibh
; /* 9B header */
136 struct hfi1_16b_header opah
; /* 16B header */
138 u8 hdr_type
; /* 9B or 16B */
141 struct hfi1_ahg_info
{
148 struct hfi1_sdma_header
{
150 struct hfi1_opa_header hdr
;
154 * hfi1 specific data structures that will be hidden from rvt after the queue
155 * pair is made common
157 struct 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 */
161 struct hfi1_ctxtdata
*rcd
; /* QP's receive context */
162 struct page
**pages
; /* for TID page scan */
163 u32 tid_enqueue
; /* saved when tid waited */
164 u8 s_sc
; /* SC[0..4] for next packet */
165 struct iowait s_iowait
;
166 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
;
172 struct rvt_qp
*owner
;
173 u16 s_running_pkt_size
;
174 u8 hdr_type
; /* 9B or 16B */
175 struct rvt_sge_state tid_ss
; /* SGE state pointer for 2nd leg */
176 atomic_t n_requests
; /* # of TID RDMA requests in the */
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
;
182 /* variables for the TID RDMA SE state machine */
185 u8 rnr_nak_state
; /* RNR NAK state */
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 */
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 */
208 u32 r_next_psn_kdeth
;
209 u32 r_next_psn_kdeth_save
;
211 u8 sync_pt
; /* Set when QP reaches sync point */
215 #define HFI1_QP_WQE_INVALID ((u32)-1)
217 struct hfi1_swqe_priv
{
218 struct tid_rdma_request tid_req
;
219 struct rvt_sge_state ss
; /* Used for TID RDMA READ Request */
222 struct hfi1_ack_priv
{
223 struct rvt_sge_state ss
; /* used for TID WRITE RESP */
224 struct tid_rdma_request tid_req
;
228 * This structure is used to hold commonly lookedup and computed values during
229 * the send engine progress.
232 struct hfi1_pkt_state
{
233 struct hfi1_ibdev
*dev
;
234 struct hfi1_ibport
*ibp
;
235 struct hfi1_pportdata
*ppd
;
236 struct verbs_txreq
*s_txreq
;
237 struct iowait_work
*wait
;
239 unsigned long timeout
;
240 unsigned long timeout_int
;
247 #define HFI1_PSN_CREDIT 16
249 struct hfi1_opcode_stats
{
250 u64 n_packets
; /* number of packets */
251 u64 n_bytes
; /* total number of bytes */
254 struct hfi1_opcode_stats_perctx
{
255 struct hfi1_opcode_stats stats
[256];
258 static inline void inc_opstats(
260 struct hfi1_opcode_stats
*stats
)
262 #ifdef CONFIG_DEBUG_FS
263 stats
->n_bytes
+= tlen
;
269 struct rvt_qp __rcu
*qp
[2];
270 struct rvt_ibport rvp
;
272 /* the first 16 entries are sl_to_vl for !OPA */
278 struct rvt_dev_info rdi
; /* Must be first */
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
;
290 /* protect iowait lists */
291 seqlock_t iowait_lock ____cacheline_aligned_in_smp
;
294 struct timer_list mem_timer
;
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
307 static inline struct hfi1_ibdev
*to_idev(struct ib_device
*ibdev
)
309 struct rvt_dev_info
*rdi
;
311 rdi
= container_of(ibdev
, struct rvt_dev_info
, ibdev
);
312 return container_of(rdi
, struct hfi1_ibdev
, rdi
);
315 static inline struct rvt_qp
*iowait_to_qp(struct iowait
*s_iowait
)
317 struct hfi1_qp_priv
*priv
;
319 priv
= container_of(s_iowait
, struct hfi1_qp_priv
, s_iowait
);
324 * This must be called with s_lock held.
326 void hfi1_bad_pkey(struct hfi1_ibport
*ibp
, u32 key
, u32 sl
,
327 u32 qp1
, u32 qp2
, u32 lid1
, u32 lid2
);
328 void hfi1_cap_mask_chg(struct rvt_dev_info
*rdi
, u8 port_num
);
329 void hfi1_sys_guid_chg(struct hfi1_ibport
*ibp
);
330 void hfi1_node_desc_chg(struct hfi1_ibport
*ibp
);
331 int 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
*in_mad
, struct ib_mad
*out_mad
,
334 size_t *out_mad_size
, u16
*out_mad_pkey_index
);
337 * The PSN_MASK and PSN_SHIFT allow for
338 * 1) comparing two PSNs
339 * 2) returning the PSN with any upper bits masked
340 * 3) returning the difference between to PSNs
342 * The number of significant bits in the PSN must
343 * necessarily be at least one bit less than
344 * the container holding the PSN.
346 #define PSN_MASK 0x7FFFFFFF
348 #define PSN_MODIFY_MASK 0xFFFFFF
352 * Returns an integer <, ==, or > than zero.
354 static inline int cmp_psn(u32 a
, u32 b
)
356 return (((int)a
) - ((int)b
)) << PSN_SHIFT
;
362 static inline u32
mask_psn(u32 a
)
368 * Return delta between two PSNs
370 static inline u32
delta_psn(u32 a
, u32 b
)
372 return (((int)a
- (int)b
) << PSN_SHIFT
) >> PSN_SHIFT
;
375 static inline struct tid_rdma_request
*wqe_to_tid_req(struct rvt_swqe
*wqe
)
377 return &((struct hfi1_swqe_priv
*)wqe
->priv
)->tid_req
;
380 static inline struct tid_rdma_request
*ack_to_tid_req(struct rvt_ack_entry
*e
)
382 return &((struct hfi1_ack_priv
*)e
->priv
)->tid_req
;
386 * Look through all the active flows for a TID RDMA request and find
387 * the one (if it exists) that contains the specified PSN.
389 static inline u32
__full_flow_psn(struct flow_state
*state
, u32 psn
)
391 return mask_psn((state
->generation
<< HFI1_KDETH_BTH_SEQ_SHIFT
) |
392 (psn
& HFI1_KDETH_BTH_SEQ_MASK
));
395 static inline u32
full_flow_psn(struct tid_rdma_flow
*flow
, u32 psn
)
397 return __full_flow_psn(&flow
->flow_state
, psn
);
401 void hfi1_put_txreq(struct verbs_txreq
*tx
);
403 int hfi1_verbs_send(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
405 void hfi1_cnp_rcv(struct hfi1_packet
*packet
);
407 void hfi1_uc_rcv(struct hfi1_packet
*packet
);
409 void hfi1_rc_rcv(struct hfi1_packet
*packet
);
412 struct hfi1_ctxtdata
*rcd
,
413 struct hfi1_packet
*packet
,
416 u8
ah_to_sc(struct ib_device
*ibdev
, struct rdma_ah_attr
*ah_attr
);
418 void hfi1_rc_verbs_aborted(struct rvt_qp
*qp
, struct hfi1_opa_header
*opah
);
419 void hfi1_rc_send_complete(struct rvt_qp
*qp
, struct hfi1_opa_header
*opah
);
421 void hfi1_ud_rcv(struct hfi1_packet
*packet
);
423 int hfi1_lookup_pkey_idx(struct hfi1_ibport
*ibp
, u16 pkey
);
425 void hfi1_migrate_qp(struct rvt_qp
*qp
);
427 int hfi1_check_modify_qp(struct rvt_qp
*qp
, struct ib_qp_attr
*attr
,
428 int attr_mask
, struct ib_udata
*udata
);
430 void hfi1_modify_qp(struct rvt_qp
*qp
, struct ib_qp_attr
*attr
,
431 int attr_mask
, struct ib_udata
*udata
);
432 void hfi1_restart_rc(struct rvt_qp
*qp
, u32 psn
, int wait
);
433 int hfi1_setup_wqe(struct rvt_qp
*qp
, struct rvt_swqe
*wqe
,
436 extern const u32 rc_only_opcode
;
437 extern const u32 uc_only_opcode
;
439 int hfi1_ruc_check_hdr(struct hfi1_ibport
*ibp
, struct hfi1_packet
*packet
);
441 u32
hfi1_make_grh(struct hfi1_ibport
*ibp
, struct ib_grh
*hdr
,
442 const struct ib_global_route
*grh
, u32 hwords
, u32 nwords
);
444 void hfi1_make_ruc_header(struct rvt_qp
*qp
, struct ib_other_headers
*ohdr
,
445 u32 bth0
, u32 bth1
, u32 bth2
, int middle
,
446 struct hfi1_pkt_state
*ps
);
448 bool hfi1_schedule_send_yield(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
,
451 void _hfi1_do_send(struct work_struct
*work
);
453 void hfi1_do_send_from_rvt(struct rvt_qp
*qp
);
455 void hfi1_do_send(struct rvt_qp
*qp
, bool in_thread
);
457 void hfi1_send_rc_ack(struct hfi1_packet
*packet
, bool is_fecn
);
459 int hfi1_make_rc_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
461 int hfi1_make_uc_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
463 int hfi1_make_ud_req(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
);
465 int hfi1_register_ib_device(struct hfi1_devdata
*);
467 void hfi1_unregister_ib_device(struct hfi1_devdata
*);
469 void hfi1_kdeth_eager_rcv(struct hfi1_packet
*packet
);
471 void hfi1_kdeth_expected_rcv(struct hfi1_packet
*packet
);
473 void hfi1_ib_rcv(struct hfi1_packet
*packet
);
475 void hfi1_16B_rcv(struct hfi1_packet
*packet
);
477 unsigned hfi1_get_npkeys(struct hfi1_devdata
*);
479 int hfi1_verbs_send_dma(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
,
482 int hfi1_verbs_send_pio(struct rvt_qp
*qp
, struct hfi1_pkt_state
*ps
,
485 static inline bool opa_bth_is_migration(struct ib_other_headers
*ohdr
)
487 return ohdr
->bth
[1] & cpu_to_be32(OPA_BTH_MIG_REQ
);
490 void hfi1_wait_kmem(struct rvt_qp
*qp
);
492 static inline void hfi1_trdma_send_complete(struct rvt_qp
*qp
,
493 struct rvt_swqe
*wqe
,
494 enum ib_wc_status status
)
496 trdma_clean_swqe(qp
, wqe
);
497 rvt_send_complete(qp
, wqe
, status
);
500 extern const enum ib_wc_opcode ib_hfi1_wc_opcode
[];
502 extern const u8 hdr_len_by_opcode
[];
504 extern const int ib_rvt_state_ops
[];
506 extern __be64 ib_hfi1_sys_image_guid
; /* in network order */
508 extern unsigned int hfi1_max_cqes
;
510 extern unsigned int hfi1_max_cqs
;
512 extern unsigned int hfi1_max_qp_wrs
;
514 extern unsigned int hfi1_max_qps
;
516 extern unsigned int hfi1_max_sges
;
518 extern unsigned int hfi1_max_mcast_grps
;
520 extern unsigned int hfi1_max_mcast_qp_attached
;
522 extern unsigned int hfi1_max_srqs
;
524 extern unsigned int hfi1_max_srq_sges
;
526 extern unsigned int hfi1_max_srq_wrs
;
528 extern unsigned short piothreshold
;
530 extern const u32 ib_hfi1_rnr_table
[];
532 #endif /* HFI1_VERBS_H */