1 /* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
2
3 /* Authors: Bernard Metzler <bmt@zurich.ibm.com> */
4 /* Copyright (c) 2008-2019, IBM Corporation */
5
6 #ifndef _SIW_H
7 #define _SIW_H
8
9 #include <rdma/ib_verbs.h>
10 #include <rdma/restrack.h>
11 #include <linux/socket.h>
12 #include <linux/skbuff.h>
13 #include <crypto/hash.h>
14 #include <linux/crc32.h>
15 #include <linux/crc32c.h>
16
17 #include <rdma/siw-abi.h>
18 #include "iwarp.h"
19
20 #define SIW_VENDOR_ID 0x626d74 /* ascii 'bmt' for now */
21 #define SIW_VENDORT_PART_ID 0
22 #define SIW_MAX_QP (1024 * 100)
23 #define SIW_MAX_QP_WR (1024 * 32)
24 #define SIW_MAX_ORD_QP 128
25 #define SIW_MAX_IRD_QP 128
26 #define SIW_MAX_SGE_PBL 256 /* max num sge's for PBL */
27 #define SIW_MAX_SGE_RD 1 /* iwarp limitation. we could relax */
28 #define SIW_MAX_CQ (1024 * 100)
29 #define SIW_MAX_CQE (SIW_MAX_QP_WR * 100)
30 #define SIW_MAX_MR (SIW_MAX_QP * 10)
31 #define SIW_MAX_PD SIW_MAX_QP
32 #define SIW_MAX_MW 0 /* to be set if MW's are supported */
33 #define SIW_MAX_SRQ SIW_MAX_QP
34 #define SIW_MAX_SRQ_WR (SIW_MAX_QP_WR * 10)
35 #define SIW_MAX_CONTEXT SIW_MAX_PD
36
37 /* Min number of bytes for using zero copy transmit */
38 #define SENDPAGE_THRESH PAGE_SIZE
39
40 /* Maximum number of frames which can be send in one SQ processing */
41 #define SQ_USER_MAXBURST 100
42
43 /* Maximum number of consecutive IRQ elements which get served
44 * if SQ has pending work. Prevents starving local SQ processing
45 * by serving peer Read Requests.
46 */
47 #define SIW_IRQ_MAXBURST_SQ_ACTIVE 4
48
49 struct siw_dev_cap {
50 int max_qp;
51 int max_qp_wr;
52 int max_ord; /* max. outbound read queue depth */
53 int max_ird; /* max. inbound read queue depth */
54 int max_sge;
55 int max_sge_rd;
56 int max_cq;
57 int max_cqe;
58 int max_mr;
59 int max_pd;
60 int max_mw;
61 int max_srq;
62 int max_srq_wr;
63 int max_srq_sge;
64 };
65
66 struct siw_pd {
67 struct ib_pd base_pd;
68 };
69
70 struct siw_device {
71 struct ib_device base_dev;
72 struct net_device *netdev;
73 struct siw_dev_cap attrs;
74
75 u32 vendor_part_id;
76 int numa_node;
77 char raw_gid[ETH_ALEN];
78
79 /* physical port state (only one port per device) */
80 enum ib_port_state state;
81
82 spinlock_t lock;
83
84 struct xarray qp_xa;
85 struct xarray mem_xa;
86
87 struct list_head cep_list;
88 struct list_head qp_list;
89
90 /* active objects statistics to enforce limits */
91 atomic_t num_qp;
92 atomic_t num_cq;
93 atomic_t num_pd;
94 atomic_t num_mr;
95 atomic_t num_srq;
96 atomic_t num_ctx;
97
98 struct work_struct netdev_down;
99 };
100
101 struct siw_ucontext {
102 struct ib_ucontext base_ucontext;
103 struct siw_device *sdev;
104 };
105
106 /*
107 * The RDMA core does not define LOCAL_READ access, which is always
108 * enabled implictely.
109 */
110 #define IWARP_ACCESS_MASK \
111 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | \
112 IB_ACCESS_REMOTE_READ)
113
114 /*
115 * siw presentation of user memory registered as source
116 * or target of RDMA operations.
117 */
118
119 struct siw_page_chunk {
120 struct page **plist;
121 };
122
123 struct siw_umem {
124 struct siw_page_chunk *page_chunk;
125 int num_pages;
126 bool writable;
127 u64 fp_addr; /* First page base address */
128 struct mm_struct *owning_mm;
129 };
130
131 struct siw_pble {
132 dma_addr_t addr; /* Address of assigned buffer */
133 unsigned int size; /* Size of this entry */
134 unsigned long pbl_off; /* Total offset from start of PBL */
135 };
136
137 struct siw_pbl {
138 unsigned int num_buf;
139 unsigned int max_buf;
140 struct siw_pble pbe[];
141 };
142
143 /*
144 * Generic memory representation for registered siw memory.
145 * Memory lookup always via higher 24 bit of STag (STag index).
146 */
147 struct siw_mem {
148 struct siw_device *sdev;
149 struct kref ref;
150 u64 va; /* VA of memory */
151 u64 len; /* length of the memory buffer in bytes */
152 u32 stag; /* iWarp memory access steering tag */
153 u8 stag_valid; /* VALID or INVALID */
154 u8 is_pbl; /* PBL or user space mem */
155 u8 is_mw; /* Memory Region or Memory Window */
156 enum ib_access_flags perms; /* local/remote READ & WRITE */
157 union {
158 struct siw_umem *umem;
159 struct siw_pbl *pbl;
160 void *mem_obj;
161 };
162 struct ib_pd *pd;
163 };
164
165 struct siw_mr {
166 struct ib_mr base_mr;
167 struct siw_mem *mem;
168 struct rcu_head rcu;
169 };
170
171 /*
172 * Error codes for local or remote
173 * access to registered memory
174 */
175 enum siw_access_state {
176 E_ACCESS_OK,
177 E_STAG_INVALID,
178 E_BASE_BOUNDS,
179 E_ACCESS_PERM,
180 E_PD_MISMATCH
181 };
182
183 enum siw_wr_state {
184 SIW_WR_IDLE,
185 SIW_WR_QUEUED, /* processing has not started yet */
186 SIW_WR_INPROGRESS /* initiated processing of the WR */
187 };
188
189 /* The WQE currently being processed (RX or TX) */
190 struct siw_wqe {
191 /* Copy of applications SQE or RQE */
192 union {
193 struct siw_sqe sqe;
194 struct siw_rqe rqe;
195 };
196 struct siw_mem *mem[SIW_MAX_SGE]; /* per sge's resolved mem */
197 enum siw_wr_state wr_status;
198 enum siw_wc_status wc_status;
199 u32 bytes; /* total bytes to process */
200 u32 processed; /* bytes processed */
201 };
202
203 struct siw_cq {
204 struct ib_cq base_cq;
205 spinlock_t lock;
206 struct siw_cq_ctrl *notify;
207 struct siw_cqe *queue;
208 u32 cq_put;
209 u32 cq_get;
210 u32 num_cqe;
211 struct rdma_user_mmap_entry *cq_entry; /* mmap info for CQE array */
212 u32 id; /* For debugging only */
213 };
214
215 enum siw_qp_state {
216 SIW_QP_STATE_IDLE,
217 SIW_QP_STATE_RTR,
218 SIW_QP_STATE_RTS,
219 SIW_QP_STATE_CLOSING,
220 SIW_QP_STATE_TERMINATE,
221 SIW_QP_STATE_ERROR,
222 SIW_QP_STATE_COUNT
223 };
224
225 enum siw_qp_flags {
226 SIW_RDMA_BIND_ENABLED = (1 << 0),
227 SIW_RDMA_WRITE_ENABLED = (1 << 1),
228 SIW_RDMA_READ_ENABLED = (1 << 2),
229 SIW_SIGNAL_ALL_WR = (1 << 3),
230 SIW_MPA_CRC = (1 << 4),
231 SIW_QP_IN_DESTROY = (1 << 5)
232 };
233
234 enum siw_qp_attr_mask {
235 SIW_QP_ATTR_STATE = (1 << 0),
236 SIW_QP_ATTR_ACCESS_FLAGS = (1 << 1),
237 SIW_QP_ATTR_LLP_HANDLE = (1 << 2),
238 SIW_QP_ATTR_ORD = (1 << 3),
239 SIW_QP_ATTR_IRD = (1 << 4),
240 SIW_QP_ATTR_SQ_SIZE = (1 << 5),
241 SIW_QP_ATTR_RQ_SIZE = (1 << 6),
242 SIW_QP_ATTR_MPA = (1 << 7)
243 };
244
245 struct siw_srq {
246 struct ib_srq base_srq;
247 spinlock_t lock;
248 u32 max_sge;
249 u32 limit; /* low watermark for async event */
250 struct siw_rqe *recvq;
251 u32 rq_put;
252 u32 rq_get;
253 u32 num_rqe; /* max # of wqe's allowed */
254 struct rdma_user_mmap_entry *srq_entry; /* mmap info for SRQ array */
255 bool armed:1; /* inform user if limit hit */
256 bool is_kernel_res:1; /* true if kernel client */
257 };
258
259 struct siw_qp_attrs {
260 enum siw_qp_state state;
261 u32 sq_size;
262 u32 rq_size;
263 u32 orq_size;
264 u32 irq_size;
265 u32 sq_max_sges;
266 u32 rq_max_sges;
267 enum siw_qp_flags flags;
268
269 struct socket *sk;
270 };
271
272 enum siw_tx_ctx {
273 SIW_SEND_HDR, /* start or continue sending HDR */
274 SIW_SEND_DATA, /* start or continue sending DDP payload */
275 SIW_SEND_TRAILER, /* start or continue sending TRAILER */
276 SIW_SEND_SHORT_FPDU/* send whole FPDU hdr|data|trailer at once */
277 };
278
279 enum siw_rx_state {
280 SIW_GET_HDR, /* await new hdr or within hdr */
281 SIW_GET_DATA_START, /* start of inbound DDP payload */
282 SIW_GET_DATA_MORE, /* continuation of (misaligned) DDP payload */
283 SIW_GET_TRAILER/* await new trailer or within trailer */
284 };
285
286 struct siw_rx_stream {
287 struct sk_buff *skb;
288 int skb_new; /* pending unread bytes in skb */
289 int skb_offset; /* offset in skb */
290 int skb_copied; /* processed bytes in skb */
291
292 union iwarp_hdr hdr;
293 struct mpa_trailer trailer;
294
295 enum siw_rx_state state;
296
297 /*
298 * For each FPDU, main RX loop runs through 3 stages:
299 * Receiving protocol headers, placing DDP payload and receiving
300 * trailer information (CRC + possibly padding).
301 * Next two variables keep state on receive status of the
302 * current FPDU part (hdr, data, trailer).
303 */
304 int fpdu_part_rcvd; /* bytes in pkt part copied */
305 int fpdu_part_rem; /* bytes in pkt part not seen */
306
307 /*
308 * Next expected DDP MSN for each QN +
309 * expected steering tag +
310 * expected DDP tagget offset (all HBO)
311 */
312 u32 ddp_msn[RDMAP_UNTAGGED_QN_COUNT];
313 u32 ddp_stag;
314 u64 ddp_to;
315 u32 inval_stag; /* Stag to be invalidated */
316
317 struct shash_desc *mpa_crc_hd;
318 u8 rx_suspend : 1;
319 u8 pad : 2; /* # of pad bytes expected */
320 u8 rdmap_op : 4; /* opcode of current frame */
321 };
322
323 struct siw_rx_fpdu {
324 /*
325 * Local destination memory of inbound RDMA operation.
326 * Valid, according to wqe->wr_status
327 */
328 struct siw_wqe wqe_active;
329
330 unsigned int pbl_idx; /* Index into current PBL */
331 unsigned int sge_idx; /* current sge in rx */
332 unsigned int sge_off; /* already rcvd in curr. sge */
333
334 char first_ddp_seg; /* this is the first DDP seg */
335 char more_ddp_segs; /* more DDP segs expected */
336 u8 prev_rdmap_op : 4; /* opcode of prev frame */
337 };
338
339 /*
340 * Shorthands for short packets w/o payload
341 * to be transmitted more efficient.
342 */
343 struct siw_send_pkt {
344 struct iwarp_send send;
345 __be32 crc;
346 };
347
348 struct siw_write_pkt {
349 struct iwarp_rdma_write write;
350 __be32 crc;
351 };
352
353 struct siw_rreq_pkt {
354 struct iwarp_rdma_rreq rreq;
355 __be32 crc;
356 };
357
358 struct siw_rresp_pkt {
359 struct iwarp_rdma_rresp rresp;
360 __be32 crc;
361 };
362
363 struct siw_iwarp_tx {
364 union {
365 union iwarp_hdr hdr;
366
367 /* Generic part of FPDU header */
368 struct iwarp_ctrl ctrl;
369 struct iwarp_ctrl_untagged c_untagged;
370 struct iwarp_ctrl_tagged c_tagged;
371
372 /* FPDU headers */
373 struct iwarp_rdma_write rwrite;
374 struct iwarp_rdma_rreq rreq;
375 struct iwarp_rdma_rresp rresp;
376 struct iwarp_terminate terminate;
377 struct iwarp_send send;
378 struct iwarp_send_inv send_inv;
379
380 /* complete short FPDUs */
381 struct siw_send_pkt send_pkt;
382 struct siw_write_pkt write_pkt;
383 struct siw_rreq_pkt rreq_pkt;
384 struct siw_rresp_pkt rresp_pkt;
385 } pkt;
386
387 struct mpa_trailer trailer;
388 /* DDP MSN for untagged messages */
389 u32 ddp_msn[RDMAP_UNTAGGED_QN_COUNT];
390
391 enum siw_tx_ctx state;
392 u16 ctrl_len; /* ddp+rdmap hdr */
393 u16 ctrl_sent;
394 int burst;
395 int bytes_unsent; /* ddp payload bytes */
396
397 struct shash_desc *mpa_crc_hd;
398
399 u8 do_crc : 1; /* do crc for segment */
400 u8 use_sendpage : 1; /* send w/o copy */
401 u8 tx_suspend : 1; /* stop sending DDP segs. */
402 u8 pad : 2; /* # pad in current fpdu */
403 u8 orq_fence : 1; /* ORQ full or Send fenced */
404 u8 in_syscall : 1; /* TX out of user context */
405 u8 zcopy_tx : 1; /* Use TCP_SENDPAGE if possible */
406 u8 gso_seg_limit; /* Maximum segments for GSO, 0 = unbound */
407
408 u16 fpdu_len; /* len of FPDU to tx */
409 unsigned int tcp_seglen; /* remaining tcp seg space */
410
411 struct siw_wqe wqe_active;
412
413 int pbl_idx; /* Index into current PBL */
414 int sge_idx; /* current sge in tx */
415 u32 sge_off; /* already sent in curr. sge */
416 };
417
418 struct siw_qp {
419 struct ib_qp base_qp;
420 struct siw_device *sdev;
421 struct kref ref;
422 struct completion qp_free;
423 struct list_head devq;
424 int tx_cpu;
425 struct siw_qp_attrs attrs;
426
427 struct siw_cep *cep;
428 struct rw_semaphore state_lock;
429
430 struct ib_pd *pd;
431 struct siw_cq *scq;
432 struct siw_cq *rcq;
433 struct siw_srq *srq;
434
435 struct siw_iwarp_tx tx_ctx; /* Transmit context */
436 spinlock_t sq_lock;
437 struct siw_sqe *sendq; /* send queue element array */
438 uint32_t sq_get; /* consumer index into sq array */
439 uint32_t sq_put; /* kernel prod. index into sq array */
440 struct llist_node tx_list;
441
442 struct siw_sqe *orq; /* outbound read queue element array */
443 spinlock_t orq_lock;
444 uint32_t orq_get; /* consumer index into orq array */
445 uint32_t orq_put; /* shared producer index for ORQ */
446
447 struct siw_rx_stream rx_stream;
448 struct siw_rx_fpdu *rx_fpdu;
449 struct siw_rx_fpdu rx_tagged;
450 struct siw_rx_fpdu rx_untagged;
451 spinlock_t rq_lock;
452 struct siw_rqe *recvq; /* recv queue element array */
453 uint32_t rq_get; /* consumer index into rq array */
454 uint32_t rq_put; /* kernel prod. index into rq array */
455
456 struct siw_sqe *irq; /* inbound read queue element array */
457 uint32_t irq_get; /* consumer index into irq array */
458 uint32_t irq_put; /* producer index into irq array */
459 int irq_burst;
460
461 struct { /* information to be carried in TERMINATE pkt, if valid */
462 u8 valid;
463 u8 in_tx;
464 u8 layer : 4, etype : 4;
465 u8 ecode;
466 } term_info;
467 struct rdma_user_mmap_entry *sq_entry; /* mmap info for SQE array */
468 struct rdma_user_mmap_entry *rq_entry; /* mmap info for RQE array */
469 struct rcu_head rcu;
470 };
471
472 /* helper macros */
473 #define rx_qp(rx) container_of(rx, struct siw_qp, rx_stream)
474 #define tx_qp(tx) container_of(tx, struct siw_qp, tx_ctx)
475 #define tx_wqe(qp) (&(qp)->tx_ctx.wqe_active)
476 #define rx_wqe(rctx) (&(rctx)->wqe_active)
477 #define rx_mem(rctx) ((rctx)->wqe_active.mem[0])
478 #define tx_type(wqe) ((wqe)->sqe.opcode)
479 #define rx_type(wqe) ((wqe)->rqe.opcode)
480 #define tx_flags(wqe) ((wqe)->sqe.flags)
481
482 struct iwarp_msg_info {
483 int hdr_len;
484 struct iwarp_ctrl ctrl;
485 int (*rx_data)(struct siw_qp *qp);
486 };
487
488 struct siw_user_mmap_entry {
489 struct rdma_user_mmap_entry rdma_entry;
490 void *address;
491 };
492
493 /* Global siw parameters. Currently set in siw_main.c */
494 extern const bool zcopy_tx;
495 extern const bool try_gso;
496 extern const bool loopback_enabled;
497 extern const bool mpa_crc_required;
498 extern const bool mpa_crc_strict;
499 extern const bool siw_tcp_nagle;
500 extern u_char mpa_version;
501 extern const bool peer_to_peer;
502 extern struct task_struct *siw_tx_thread[];
503
504 extern struct crypto_shash *siw_crypto_shash;
505 extern struct iwarp_msg_info iwarp_pktinfo[RDMAP_TERMINATE + 1];
506
507 /* QP general functions */
508 int siw_qp_modify(struct siw_qp *qp, struct siw_qp_attrs *attr,
509 enum siw_qp_attr_mask mask);
510 int siw_qp_mpa_rts(struct siw_qp *qp, enum mpa_v2_ctrl ctrl);
511 void siw_qp_llp_close(struct siw_qp *qp);
512 void siw_qp_cm_drop(struct siw_qp *qp, int schedule);
513 void siw_send_terminate(struct siw_qp *qp);
514
515 void siw_qp_get_ref(struct ib_qp *qp);
516 void siw_qp_put_ref(struct ib_qp *qp);
517 int siw_qp_add(struct siw_device *sdev, struct siw_qp *qp);
518 void siw_free_qp(struct kref *ref);
519
520 void siw_init_terminate(struct siw_qp *qp, enum term_elayer layer,
521 u8 etype, u8 ecode, int in_tx);
522 enum ddp_ecode siw_tagged_error(enum siw_access_state state);
523 enum rdmap_ecode siw_rdmap_error(enum siw_access_state state);
524
525 void siw_read_to_orq(struct siw_sqe *rreq, struct siw_sqe *sqe);
526 int siw_sqe_complete(struct siw_qp *qp, struct siw_sqe *sqe, u32 bytes,
527 enum siw_wc_status status);
528 int siw_rqe_complete(struct siw_qp *qp, struct siw_rqe *rqe, u32 bytes,
529 u32 inval_stag, enum siw_wc_status status);
530 void siw_qp_llp_data_ready(struct sock *sk);
531 void siw_qp_llp_write_space(struct sock *sk);
532
533 /* QP TX path functions */
534 int siw_run_sq(void *arg);
535 int siw_qp_sq_process(struct siw_qp *qp);
536 int siw_sq_start(struct siw_qp *qp);
537 int siw_activate_tx(struct siw_qp *qp);
538 void siw_stop_tx_thread(int nr_cpu);
539 int siw_get_tx_cpu(struct siw_device *sdev);
540 void siw_put_tx_cpu(int cpu);
541
542 /* QP RX path functions */
543 int siw_proc_send(struct siw_qp *qp);
544 int siw_proc_rreq(struct siw_qp *qp);
545 int siw_proc_rresp(struct siw_qp *qp);
546 int siw_proc_write(struct siw_qp *qp);
547 int siw_proc_terminate(struct siw_qp *qp);
548
549 int siw_tcp_rx_data(read_descriptor_t *rd_desc, struct sk_buff *skb,
550 unsigned int off, size_t len);
551
set_rx_fpdu_context(struct siw_qp * qp,u8 opcode)552 static inline void set_rx_fpdu_context(struct siw_qp *qp, u8 opcode)
553 {
554 if (opcode == RDMAP_RDMA_WRITE || opcode == RDMAP_RDMA_READ_RESP)
555 qp->rx_fpdu = &qp->rx_tagged;
556 else
557 qp->rx_fpdu = &qp->rx_untagged;
558
559 qp->rx_stream.rdmap_op = opcode;
560 }
561
to_siw_ctx(struct ib_ucontext * base_ctx)562 static inline struct siw_ucontext *to_siw_ctx(struct ib_ucontext *base_ctx)
563 {
564 return container_of(base_ctx, struct siw_ucontext, base_ucontext);
565 }
566
to_siw_qp(struct ib_qp * base_qp)567 static inline struct siw_qp *to_siw_qp(struct ib_qp *base_qp)
568 {
569 return container_of(base_qp, struct siw_qp, base_qp);
570 }
571
to_siw_cq(struct ib_cq * base_cq)572 static inline struct siw_cq *to_siw_cq(struct ib_cq *base_cq)
573 {
574 return container_of(base_cq, struct siw_cq, base_cq);
575 }
576
to_siw_srq(struct ib_srq * base_srq)577 static inline struct siw_srq *to_siw_srq(struct ib_srq *base_srq)
578 {
579 return container_of(base_srq, struct siw_srq, base_srq);
580 }
581
to_siw_dev(struct ib_device * base_dev)582 static inline struct siw_device *to_siw_dev(struct ib_device *base_dev)
583 {
584 return container_of(base_dev, struct siw_device, base_dev);
585 }
586
to_siw_mr(struct ib_mr * base_mr)587 static inline struct siw_mr *to_siw_mr(struct ib_mr *base_mr)
588 {
589 return container_of(base_mr, struct siw_mr, base_mr);
590 }
591
592 static inline struct siw_user_mmap_entry *
to_siw_mmap_entry(struct rdma_user_mmap_entry * rdma_mmap)593 to_siw_mmap_entry(struct rdma_user_mmap_entry *rdma_mmap)
594 {
595 return container_of(rdma_mmap, struct siw_user_mmap_entry, rdma_entry);
596 }
597
siw_qp_id2obj(struct siw_device * sdev,int id)598 static inline struct siw_qp *siw_qp_id2obj(struct siw_device *sdev, int id)
599 {
600 struct siw_qp *qp;
601
602 rcu_read_lock();
603 qp = xa_load(&sdev->qp_xa, id);
604 if (likely(qp && kref_get_unless_zero(&qp->ref))) {
605 rcu_read_unlock();
606 return qp;
607 }
608 rcu_read_unlock();
609 return NULL;
610 }
611
qp_id(struct siw_qp * qp)612 static inline u32 qp_id(struct siw_qp *qp)
613 {
614 return qp->base_qp.qp_num;
615 }
616
siw_qp_get(struct siw_qp * qp)617 static inline void siw_qp_get(struct siw_qp *qp)
618 {
619 kref_get(&qp->ref);
620 }
621
siw_qp_put(struct siw_qp * qp)622 static inline void siw_qp_put(struct siw_qp *qp)
623 {
624 kref_put(&qp->ref, siw_free_qp);
625 }
626
siw_sq_empty(struct siw_qp * qp)627 static inline int siw_sq_empty(struct siw_qp *qp)
628 {
629 struct siw_sqe *sqe = &qp->sendq[qp->sq_get % qp->attrs.sq_size];
630
631 return READ_ONCE(sqe->flags) == 0;
632 }
633
sq_get_next(struct siw_qp * qp)634 static inline struct siw_sqe *sq_get_next(struct siw_qp *qp)
635 {
636 struct siw_sqe *sqe = &qp->sendq[qp->sq_get % qp->attrs.sq_size];
637
638 if (READ_ONCE(sqe->flags) & SIW_WQE_VALID)
639 return sqe;
640
641 return NULL;
642 }
643
orq_get_current(struct siw_qp * qp)644 static inline struct siw_sqe *orq_get_current(struct siw_qp *qp)
645 {
646 return &qp->orq[qp->orq_get % qp->attrs.orq_size];
647 }
648
orq_get_free(struct siw_qp * qp)649 static inline struct siw_sqe *orq_get_free(struct siw_qp *qp)
650 {
651 struct siw_sqe *orq_e = &qp->orq[qp->orq_put % qp->attrs.orq_size];
652
653 if (READ_ONCE(orq_e->flags) == 0)
654 return orq_e;
655
656 return NULL;
657 }
658
siw_orq_empty(struct siw_qp * qp)659 static inline int siw_orq_empty(struct siw_qp *qp)
660 {
661 return qp->orq[qp->orq_get % qp->attrs.orq_size].flags == 0 ? 1 : 0;
662 }
663
irq_alloc_free(struct siw_qp * qp)664 static inline struct siw_sqe *irq_alloc_free(struct siw_qp *qp)
665 {
666 struct siw_sqe *irq_e = &qp->irq[qp->irq_put % qp->attrs.irq_size];
667
668 if (READ_ONCE(irq_e->flags) == 0) {
669 qp->irq_put++;
670 return irq_e;
671 }
672 return NULL;
673 }
674
siw_csum_update(const void * buff,int len,__wsum sum)675 static inline __wsum siw_csum_update(const void *buff, int len, __wsum sum)
676 {
677 return (__force __wsum)crc32c((__force __u32)sum, buff, len);
678 }
679
siw_csum_combine(__wsum csum,__wsum csum2,int offset,int len)680 static inline __wsum siw_csum_combine(__wsum csum, __wsum csum2, int offset,
681 int len)
682 {
683 return (__force __wsum)__crc32c_le_combine((__force __u32)csum,
684 (__force __u32)csum2, len);
685 }
686
siw_crc_skb(struct siw_rx_stream * srx,unsigned int len)687 static inline void siw_crc_skb(struct siw_rx_stream *srx, unsigned int len)
688 {
689 const struct skb_checksum_ops siw_cs_ops = {
690 .update = siw_csum_update,
691 .combine = siw_csum_combine,
692 };
693 __wsum crc = *(u32 *)shash_desc_ctx(srx->mpa_crc_hd);
694
695 crc = __skb_checksum(srx->skb, srx->skb_offset, len, crc,
696 &siw_cs_ops);
697 *(u32 *)shash_desc_ctx(srx->mpa_crc_hd) = crc;
698 }
699
700 #define siw_dbg(ibdev, fmt, ...) \
701 ibdev_dbg(ibdev, "%s: " fmt, __func__, ##__VA_ARGS__)
702
703 #define siw_dbg_qp(qp, fmt, ...) \
704 ibdev_dbg(&qp->sdev->base_dev, "QP[%u] %s: " fmt, qp_id(qp), __func__, \
705 ##__VA_ARGS__)
706
707 #define siw_dbg_cq(cq, fmt, ...) \
708 ibdev_dbg(cq->base_cq.device, "CQ[%u] %s: " fmt, cq->id, __func__, \
709 ##__VA_ARGS__)
710
711 #define siw_dbg_pd(pd, fmt, ...) \
712 ibdev_dbg(pd->device, "PD[%u] %s: " fmt, pd->res.id, __func__, \
713 ##__VA_ARGS__)
714
715 #define siw_dbg_mem(mem, fmt, ...) \
716 ibdev_dbg(&mem->sdev->base_dev, \
717 "MEM[0x%08x] %s: " fmt, mem->stag, __func__, ##__VA_ARGS__)
718
719 #define siw_dbg_cep(cep, fmt, ...) \
720 ibdev_dbg(&cep->sdev->base_dev, "CEP[0x%pK] %s: " fmt, \
721 cep, __func__, ##__VA_ARGS__)
722
723 void siw_cq_flush(struct siw_cq *cq);
724 void siw_sq_flush(struct siw_qp *qp);
725 void siw_rq_flush(struct siw_qp *qp);
726 int siw_reap_cqe(struct siw_cq *cq, struct ib_wc *wc);
727
728 #endif
729