1 #ifndef DEF_RDMAVT_INCQP_H
2 #define DEF_RDMAVT_INCQP_H
5 * Copyright(c) 2015 Intel Corporation.
7 * This file is provided under a dual BSD/GPLv2 license. When using or
8 * redistributing this file, you may do so under either license.
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of version 2 of the GNU General Public License as
14 * published by the Free Software Foundation.
16 * This program is distributed in the hope that it will be useful, but
17 * WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
27 * - Redistributions of source code must retain the above copyright
28 * notice, this list of conditions and the following disclaimer.
29 * - Redistributions in binary form must reproduce the above copyright
30 * notice, this list of conditions and the following disclaimer in
31 * the documentation and/or other materials provided with the
33 * - Neither the name of Intel Corporation nor the names of its
34 * contributors may be used to endorse or promote products derived
35 * from this software without specific prior written permission.
37 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
38 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
39 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
40 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
41 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
42 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
43 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
44 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
45 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
46 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
47 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
51 #include <rdma/rdma_vt.h>
52 #include <rdma/ib_pack.h>
54 * Atomic bit definitions for r_aflags.
56 #define RVT_R_WRID_VALID 0
57 #define RVT_R_REWIND_SGE 1
60 * Bit definitions for r_flags.
62 #define RVT_R_REUSE_SGE 0x01
63 #define RVT_R_RDMAR_SEQ 0x02
64 #define RVT_R_RSP_NAK 0x04
65 #define RVT_R_RSP_SEND 0x08
66 #define RVT_R_COMM_EST 0x10
69 * Bit definitions for s_flags.
71 * RVT_S_SIGNAL_REQ_WR - set if QP send WRs contain completion signaled
72 * RVT_S_BUSY - send tasklet is processing the QP
73 * RVT_S_TIMER - the RC retry timer is active
74 * RVT_S_ACK_PENDING - an ACK is waiting to be sent after RDMA read/atomics
75 * RVT_S_WAIT_FENCE - waiting for all prior RDMA read or atomic SWQEs
76 * before processing the next SWQE
77 * RVT_S_WAIT_RDMAR - waiting for a RDMA read or atomic SWQE to complete
78 * before processing the next SWQE
79 * RVT_S_WAIT_RNR - waiting for RNR timeout
80 * RVT_S_WAIT_SSN_CREDIT - waiting for RC credits to process next SWQE
81 * RVT_S_WAIT_DMA - waiting for send DMA queue to drain before generating
82 * next send completion entry not via send DMA
83 * RVT_S_WAIT_PIO - waiting for a send buffer to be available
84 * RVT_S_WAIT_TX - waiting for a struct verbs_txreq to be available
85 * RVT_S_WAIT_DMA_DESC - waiting for DMA descriptors to be available
86 * RVT_S_WAIT_KMEM - waiting for kernel memory to be available
87 * RVT_S_WAIT_PSN - waiting for a packet to exit the send DMA queue
88 * RVT_S_WAIT_ACK - waiting for an ACK packet before sending more requests
89 * RVT_S_SEND_ONE - send one packet, request ACK, then wait for ACK
90 * RVT_S_ECN - a BECN was queued to the send engine
92 #define RVT_S_SIGNAL_REQ_WR 0x0001
93 #define RVT_S_BUSY 0x0002
94 #define RVT_S_TIMER 0x0004
95 #define RVT_S_RESP_PENDING 0x0008
96 #define RVT_S_ACK_PENDING 0x0010
97 #define RVT_S_WAIT_FENCE 0x0020
98 #define RVT_S_WAIT_RDMAR 0x0040
99 #define RVT_S_WAIT_RNR 0x0080
100 #define RVT_S_WAIT_SSN_CREDIT 0x0100
101 #define RVT_S_WAIT_DMA 0x0200
102 #define RVT_S_WAIT_PIO 0x0400
103 #define RVT_S_WAIT_TX 0x0800
104 #define RVT_S_WAIT_DMA_DESC 0x1000
105 #define RVT_S_WAIT_KMEM 0x2000
106 #define RVT_S_WAIT_PSN 0x4000
107 #define RVT_S_WAIT_ACK 0x8000
108 #define RVT_S_SEND_ONE 0x10000
109 #define RVT_S_UNLIMITED_CREDIT 0x20000
110 #define RVT_S_AHG_VALID 0x40000
111 #define RVT_S_AHG_CLEAR 0x80000
112 #define RVT_S_ECN 0x100000
115 * Wait flags that would prevent any packet type from being sent.
117 #define RVT_S_ANY_WAIT_IO (RVT_S_WAIT_PIO | RVT_S_WAIT_TX | \
118 RVT_S_WAIT_DMA_DESC | RVT_S_WAIT_KMEM)
121 * Wait flags that would prevent send work requests from making progress.
123 #define RVT_S_ANY_WAIT_SEND (RVT_S_WAIT_FENCE | RVT_S_WAIT_RDMAR | \
124 RVT_S_WAIT_RNR | RVT_S_WAIT_SSN_CREDIT | RVT_S_WAIT_DMA | \
125 RVT_S_WAIT_PSN | RVT_S_WAIT_ACK)
127 #define RVT_S_ANY_WAIT (RVT_S_ANY_WAIT_IO | RVT_S_ANY_WAIT_SEND)
129 /* Number of bits to pay attention to in the opcode for checking qp type */
130 #define RVT_OPCODE_QP_MASK 0xE0
132 /* Flags for checking QP state (see ib_rvt_state_ops[]) */
133 #define RVT_POST_SEND_OK 0x01
134 #define RVT_POST_RECV_OK 0x02
135 #define RVT_PROCESS_RECV_OK 0x04
136 #define RVT_PROCESS_SEND_OK 0x08
137 #define RVT_PROCESS_NEXT_SEND_OK 0x10
138 #define RVT_FLUSH_SEND 0x20
139 #define RVT_FLUSH_RECV 0x40
140 #define RVT_PROCESS_OR_FLUSH_SEND \
141 (RVT_PROCESS_SEND_OK | RVT_FLUSH_SEND)
144 * Send work request queue entry.
145 * The size of the sg_list is determined when the QP is created and stored
150 struct ib_send_wr wr; /* don't use wr.sg_list */
151 struct ib_ud_wr ud_wr;
152 struct ib_reg_wr reg_wr;
153 struct ib_rdma_wr rdma_wr;
154 struct ib_atomic_wr atomic_wr;
156 u32 psn; /* first packet sequence number */
157 u32 lpsn; /* last packet sequence number */
158 u32 ssn; /* send sequence number */
159 u32 length; /* total length of data in sg_list */
160 struct rvt_sge sg_list[0];
164 * Receive work request queue entry.
165 * The size of the sg_list is determined when the QP (or SRQ) is created
166 * and stored in qp->r_rq.max_sge (or srq->rq.max_sge).
171 struct ib_sge sg_list[0];
175 * This structure is used to contain the head pointer, tail pointer,
176 * and receive work queue entries as a single memory allocation so
177 * it can be mmap'ed into user space.
178 * Note that the wq array elements are variable size so you can't
179 * just index into the array to get the N'th element;
180 * use get_rwqe_ptr() instead.
183 u32 head; /* new work requests posted to the head */
184 u32 tail; /* receives pull requests from here. */
185 struct rvt_rwqe wq[0];
190 u32 size; /* size of RWQE array */
192 /* protect changes in this struct */
193 spinlock_t lock ____cacheline_aligned_in_smp;
197 * This structure is used by rvt_mmap() to validate an offset
198 * when an mmap() request is made. The vm_area_struct then uses
199 * this as its vm_private_data.
201 struct rvt_mmap_info {
202 struct list_head pending_mmaps;
203 struct ib_ucontext *context;
210 #define RVT_MAX_RDMA_ATOMIC 16
213 * This structure holds the information that the send tasklet needs
214 * to send a RDMA read response or atomic operation.
216 struct rvt_ack_entry {
222 struct rvt_sge rdma_sge;
228 * Variables prefixed with s_ are for the requester (sender).
229 * Variables prefixed with r_ are for the responder (receiver).
230 * Variables prefixed with ack_ are for responder replies.
232 * Common variables are protected by both r_rq.lock and s_lock in that order
233 * which only happens in modify_qp() or changing the QP 'state'.
237 void *priv; /* Driver private data */
238 /* read mostly fields above and below */
239 struct ib_ah_attr remote_ah_attr;
240 struct ib_ah_attr alt_ah_attr;
241 struct rvt_qp __rcu *next; /* link list for QPN hash table */
242 struct rvt_swqe *s_wq; /* send work queue */
243 struct rvt_mmap_info *ip;
245 unsigned long timeout_jiffies; /* computed from timeout */
247 enum ib_mtu path_mtu;
248 int srate_mbps; /* s_srate (below) converted to Mbit/s */
250 u32 pmtu; /* decoded from path_mtu */
251 u32 qkey; /* QKEY for this QP (for UD or RD) */
252 u32 s_size; /* send work queue size */
253 u32 s_rnr_timeout; /* number of milliseconds for RNR timeout */
254 u32 s_ahgpsn; /* set to the psn in the copy of the header */
256 u8 state; /* QP state */
257 u8 allowed_ops; /* high order bits of allowed opcodes */
259 u8 alt_timeout; /* Alternate path timeout for this QP */
260 u8 timeout; /* Timeout for this QP */
264 u8 s_pkey_index; /* PKEY index to use */
265 u8 s_alt_pkey_index; /* Alternate path PKEY index to use */
266 u8 r_max_rd_atomic; /* max number of RDMA read/atomic to receive */
267 u8 s_max_rd_atomic; /* max number of RDMA read/atomic to send */
268 u8 s_retry_cnt; /* number of times to retry */
270 u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */
271 u8 s_max_sge; /* size of s_wq->sg_list */
274 /* start of read/write fields */
275 atomic_t refcount ____cacheline_aligned_in_smp;
276 wait_queue_head_t wait;
278 struct rvt_ack_entry s_ack_queue[RVT_MAX_RDMA_ATOMIC + 1]
279 ____cacheline_aligned_in_smp;
280 struct rvt_sge_state s_rdma_read_sge;
282 spinlock_t r_lock ____cacheline_aligned_in_smp; /* used for APM */
283 unsigned long r_aflags;
284 u64 r_wr_id; /* ID for current receive WQE */
285 u32 r_ack_psn; /* PSN for next ACK or atomic ACK */
286 u32 r_len; /* total length of r_sge */
287 u32 r_rcv_len; /* receive data len processed */
288 u32 r_psn; /* expected rcv packet sequence number */
289 u32 r_msn; /* message sequence number */
291 u8 r_state; /* opcode of last packet received */
293 u8 r_head_ack_queue; /* index into s_ack_queue[] */
295 struct list_head rspwait; /* link for waiting to respond */
297 struct rvt_sge_state r_sge; /* current receive data */
298 struct rvt_rq r_rq; /* receive work queue */
300 spinlock_t s_lock ____cacheline_aligned_in_smp;
301 struct rvt_sge_state *s_cur_sge;
303 struct rvt_swqe *s_wqe;
304 struct rvt_sge_state s_sge; /* current send request data */
305 struct rvt_mregion *s_rdma_mr;
306 struct sdma_engine *s_sde; /* current sde */
307 u32 s_cur_size; /* size of send packet in bytes */
308 u32 s_len; /* total length of s_sge */
309 u32 s_rdma_read_len; /* total length of s_rdma_read_sge */
310 u32 s_next_psn; /* PSN for next request */
311 u32 s_last_psn; /* last response PSN processed */
312 u32 s_sending_psn; /* lowest PSN that is being sent */
313 u32 s_sending_hpsn; /* highest PSN that is being sent */
314 u32 s_psn; /* current packet sequence number */
315 u32 s_ack_rdma_psn; /* PSN for sending RDMA read responses */
316 u32 s_ack_psn; /* PSN for acking sends and RDMA writes */
317 u32 s_head; /* new entries added here */
318 u32 s_tail; /* next entry to process */
319 u32 s_cur; /* current work queue entry */
320 u32 s_acked; /* last un-ACK'ed entry */
321 u32 s_last; /* last completed entry */
322 u32 s_ssn; /* SSN of tail entry */
323 u32 s_lsn; /* limit sequence number (credit) */
324 u16 s_hdrwords; /* size of s_hdr in 32 bit words */
327 u8 s_state; /* opcode of last packet sent */
328 u8 s_ack_state; /* opcode of packet to ACK */
329 u8 s_nak_state; /* non-zero if NAK is pending */
330 u8 r_nak_state; /* non-zero if NAK is pending */
331 u8 s_retry; /* requester retry counter */
332 u8 s_rnr_retry; /* requester RNR retry counter */
333 u8 s_num_rd_atomic; /* number of RDMA read/atomic pending */
334 u8 s_tail_ack_queue; /* index into s_ack_queue[] */
336 struct rvt_sge_state s_ack_rdma_sge;
337 struct timer_list s_timer;
340 * This sge list MUST be last. Do not add anything below here.
342 struct rvt_sge r_sg_list[0] /* verified SGEs */
343 ____cacheline_aligned_in_smp;
349 struct rvt_mmap_info *ip;
350 /* send signal when number of RWQEs < limit */
354 #define RVT_QPN_MAX BIT(24)
355 #define RVT_QPNMAP_ENTRIES (RVT_QPN_MAX / PAGE_SIZE / BITS_PER_BYTE)
356 #define RVT_BITS_PER_PAGE (PAGE_SIZE * BITS_PER_BYTE)
357 #define RVT_BITS_PER_PAGE_MASK (RVT_BITS_PER_PAGE - 1)
360 * QPN-map pages start out as NULL, they get allocated upon
361 * first use and are never deallocated. This way,
362 * large bitmaps are not allocated unless large numbers of QPs are used.
368 struct rvt_qpn_table {
369 spinlock_t lock; /* protect changes to the qp table */
370 unsigned flags; /* flags for QP0/1 allocated for each port */
371 u32 last; /* last QP number allocated */
372 u32 nmaps; /* size of the map table */
375 /* bit map of free QP numbers other than 0/1 */
376 struct rvt_qpn_map map[RVT_QPNMAP_ENTRIES];
379 struct rvt_qp_ibdev {
382 struct rvt_qp __rcu **qp_table;
383 spinlock_t qpt_lock; /* qptable lock */
384 struct rvt_qpn_table qpn_table;
388 * Since struct rvt_swqe is not a fixed size, we can't simply index into
389 * struct hfi1_qp.s_wq. This function does the array index computation.
391 static inline struct rvt_swqe *rvt_get_swqe_ptr(struct rvt_qp *qp,
394 return (struct rvt_swqe *)((char *)qp->s_wq +
395 (sizeof(struct rvt_swqe) +
397 sizeof(struct rvt_sge)) * n);
400 extern const int ib_rvt_state_ops[];
402 #endif /* DEF_RDMAVT_INCQP_H */