xref: /linux/drivers/infiniband/sw/siw/siw.h (revision 9a6b55ac)
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 <linux/socket.h>
11 #include <linux/skbuff.h>
12 #include <crypto/hash.h>
13 #include <linux/crc32.h>
14 #include <linux/crc32c.h>
15 
16 #include <rdma/siw-abi.h>
17 #include "iwarp.h"
18 
19 #define SIW_VENDOR_ID 0x626d74 /* ascii 'bmt' for now */
20 #define SIW_VENDORT_PART_ID 0
21 #define SIW_MAX_QP (1024 * 100)
22 #define SIW_MAX_QP_WR (1024 * 32)
23 #define SIW_MAX_ORD_QP 128
24 #define SIW_MAX_IRD_QP 128
25 #define SIW_MAX_SGE_PBL 256 /* max num sge's for PBL */
26 #define SIW_MAX_SGE_RD 1 /* iwarp limitation. we could relax */
27 #define SIW_MAX_CQ (1024 * 100)
28 #define SIW_MAX_CQE (SIW_MAX_QP_WR * 100)
29 #define SIW_MAX_MR (SIW_MAX_QP * 10)
30 #define SIW_MAX_PD SIW_MAX_QP
31 #define SIW_MAX_MW 0 /* to be set if MW's are supported */
32 #define SIW_MAX_FMR SIW_MAX_MR
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_fmr;
62 	int max_srq;
63 	int max_srq_wr;
64 	int max_srq_sge;
65 };
66 
67 struct siw_pd {
68 	struct ib_pd base_pd;
69 };
70 
71 struct siw_device {
72 	struct ib_device base_dev;
73 	struct device_dma_parameters dma_parms;
74 	struct net_device *netdev;
75 	struct siw_dev_cap attrs;
76 
77 	u32 vendor_part_id;
78 	int numa_node;
79 
80 	/* physical port state (only one port per device) */
81 	enum ib_port_state state;
82 
83 	spinlock_t lock;
84 
85 	struct xarray qp_xa;
86 	struct xarray mem_xa;
87 
88 	struct list_head cep_list;
89 	struct list_head qp_list;
90 
91 	/* active objects statistics to enforce limits */
92 	atomic_t num_qp;
93 	atomic_t num_cq;
94 	atomic_t num_pd;
95 	atomic_t num_mr;
96 	atomic_t num_srq;
97 	atomic_t num_ctx;
98 
99 	struct work_struct netdev_down;
100 };
101 
102 struct siw_ucontext {
103 	struct ib_ucontext base_ucontext;
104 	struct siw_device *sdev;
105 };
106 
107 /*
108  * The RDMA core does not define LOCAL_READ access, which is always
109  * enabled implictely.
110  */
111 #define IWARP_ACCESS_MASK					\
112 	(IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE	|	\
113 	 IB_ACCESS_REMOTE_READ)
114 
115 /*
116  * siw presentation of user memory registered as source
117  * or target of RDMA operations.
118  */
119 
120 struct siw_page_chunk {
121 	struct page **plist;
122 };
123 
124 struct siw_umem {
125 	struct siw_page_chunk *page_chunk;
126 	int num_pages;
127 	bool writable;
128 	u64 fp_addr; /* First page base address */
129 	struct mm_struct *owning_mm;
130 };
131 
132 struct siw_pble {
133 	dma_addr_t addr; /* Address of assigned buffer */
134 	unsigned int size; /* Size of this entry */
135 	unsigned long pbl_off; /* Total offset from start of PBL */
136 };
137 
138 struct siw_pbl {
139 	unsigned int num_buf;
140 	unsigned int max_buf;
141 	struct siw_pble pbe[1];
142 };
143 
144 /*
145  * Generic memory representation for registered siw memory.
146  * Memory lookup always via higher 24 bit of STag (STag index).
147  */
148 struct siw_mem {
149 	struct siw_device *sdev;
150 	struct kref ref;
151 	u64 va; /* VA of memory */
152 	u64 len; /* length of the memory buffer in bytes */
153 	u32 stag; /* iWarp memory access steering tag */
154 	u8 stag_valid; /* VALID or INVALID */
155 	u8 is_pbl; /* PBL or user space mem */
156 	u8 is_mw; /* Memory Region or Memory Window */
157 	enum ib_access_flags perms; /* local/remote READ & WRITE */
158 	union {
159 		struct siw_umem *umem;
160 		struct siw_pbl *pbl;
161 		void *mem_obj;
162 	};
163 	struct ib_pd *pd;
164 };
165 
166 struct siw_mr {
167 	struct ib_mr base_mr;
168 	struct siw_mem *mem;
169 	struct rcu_head rcu;
170 };
171 
172 /*
173  * Error codes for local or remote
174  * access to registered memory
175  */
176 enum siw_access_state {
177 	E_ACCESS_OK,
178 	E_STAG_INVALID,
179 	E_BASE_BOUNDS,
180 	E_ACCESS_PERM,
181 	E_PD_MISMATCH
182 };
183 
184 enum siw_wr_state {
185 	SIW_WR_IDLE,
186 	SIW_WR_QUEUED, /* processing has not started yet */
187 	SIW_WR_INPROGRESS /* initiated processing of the WR */
188 };
189 
190 /* The WQE currently being processed (RX or TX) */
191 struct siw_wqe {
192 	/* Copy of applications SQE or RQE */
193 	union {
194 		struct siw_sqe sqe;
195 		struct siw_rqe rqe;
196 	};
197 	struct siw_mem *mem[SIW_MAX_SGE]; /* per sge's resolved mem */
198 	enum siw_wr_state wr_status;
199 	enum siw_wc_status wc_status;
200 	u32 bytes; /* total bytes to process */
201 	u32 processed; /* bytes processed */
202 };
203 
204 struct siw_cq {
205 	struct ib_cq base_cq;
206 	spinlock_t lock;
207 	struct siw_cq_ctrl *notify;
208 	struct siw_cqe *queue;
209 	u32 cq_put;
210 	u32 cq_get;
211 	u32 num_cqe;
212 	bool kernel_verbs;
213 	struct rdma_user_mmap_entry *cq_entry; /* mmap info for CQE array */
214 	u32 id; /* For debugging only */
215 };
216 
217 enum siw_qp_state {
218 	SIW_QP_STATE_IDLE,
219 	SIW_QP_STATE_RTR,
220 	SIW_QP_STATE_RTS,
221 	SIW_QP_STATE_CLOSING,
222 	SIW_QP_STATE_TERMINATE,
223 	SIW_QP_STATE_ERROR,
224 	SIW_QP_STATE_COUNT
225 };
226 
227 enum siw_qp_flags {
228 	SIW_RDMA_BIND_ENABLED = (1 << 0),
229 	SIW_RDMA_WRITE_ENABLED = (1 << 1),
230 	SIW_RDMA_READ_ENABLED = (1 << 2),
231 	SIW_SIGNAL_ALL_WR = (1 << 3),
232 	SIW_MPA_CRC = (1 << 4),
233 	SIW_QP_IN_DESTROY = (1 << 5)
234 };
235 
236 enum siw_qp_attr_mask {
237 	SIW_QP_ATTR_STATE = (1 << 0),
238 	SIW_QP_ATTR_ACCESS_FLAGS = (1 << 1),
239 	SIW_QP_ATTR_LLP_HANDLE = (1 << 2),
240 	SIW_QP_ATTR_ORD = (1 << 3),
241 	SIW_QP_ATTR_IRD = (1 << 4),
242 	SIW_QP_ATTR_SQ_SIZE = (1 << 5),
243 	SIW_QP_ATTR_RQ_SIZE = (1 << 6),
244 	SIW_QP_ATTR_MPA = (1 << 7)
245 };
246 
247 struct siw_srq {
248 	struct ib_srq base_srq;
249 	spinlock_t lock;
250 	u32 max_sge;
251 	u32 limit; /* low watermark for async event */
252 	struct siw_rqe *recvq;
253 	u32 rq_put;
254 	u32 rq_get;
255 	u32 num_rqe; /* max # of wqe's allowed */
256 	struct rdma_user_mmap_entry *srq_entry; /* mmap info for SRQ array */
257 	char armed; /* inform user if limit hit */
258 	char kernel_verbs; /* '1' if kernel client */
259 };
260 
261 struct siw_qp_attrs {
262 	enum siw_qp_state state;
263 	u32 sq_size;
264 	u32 rq_size;
265 	u32 orq_size;
266 	u32 irq_size;
267 	u32 sq_max_sges;
268 	u32 rq_max_sges;
269 	enum siw_qp_flags flags;
270 
271 	struct socket *sk;
272 };
273 
274 enum siw_tx_ctx {
275 	SIW_SEND_HDR, /* start or continue sending HDR */
276 	SIW_SEND_DATA, /* start or continue sending DDP payload */
277 	SIW_SEND_TRAILER, /* start or continue sending TRAILER */
278 	SIW_SEND_SHORT_FPDU/* send whole FPDU hdr|data|trailer at once */
279 };
280 
281 enum siw_rx_state {
282 	SIW_GET_HDR, /* await new hdr or within hdr */
283 	SIW_GET_DATA_START, /* start of inbound DDP payload */
284 	SIW_GET_DATA_MORE, /* continuation of (misaligned) DDP payload */
285 	SIW_GET_TRAILER/* await new trailer or within trailer */
286 };
287 
288 struct siw_rx_stream {
289 	struct sk_buff *skb;
290 	int skb_new; /* pending unread bytes in skb */
291 	int skb_offset; /* offset in skb */
292 	int skb_copied; /* processed bytes in skb */
293 
294 	union iwarp_hdr hdr;
295 	struct mpa_trailer trailer;
296 
297 	enum siw_rx_state state;
298 
299 	/*
300 	 * For each FPDU, main RX loop runs through 3 stages:
301 	 * Receiving protocol headers, placing DDP payload and receiving
302 	 * trailer information (CRC + possibly padding).
303 	 * Next two variables keep state on receive status of the
304 	 * current FPDU part (hdr, data, trailer).
305 	 */
306 	int fpdu_part_rcvd; /* bytes in pkt part copied */
307 	int fpdu_part_rem; /* bytes in pkt part not seen */
308 
309 	/*
310 	 * Next expected DDP MSN for each QN +
311 	 * expected steering tag +
312 	 * expected DDP tagget offset (all HBO)
313 	 */
314 	u32 ddp_msn[RDMAP_UNTAGGED_QN_COUNT];
315 	u32 ddp_stag;
316 	u64 ddp_to;
317 	u32 inval_stag; /* Stag to be invalidated */
318 
319 	struct shash_desc *mpa_crc_hd;
320 	u8 rx_suspend : 1;
321 	u8 pad : 2; /* # of pad bytes expected */
322 	u8 rdmap_op : 4; /* opcode of current frame */
323 };
324 
325 struct siw_rx_fpdu {
326 	/*
327 	 * Local destination memory of inbound RDMA operation.
328 	 * Valid, according to wqe->wr_status
329 	 */
330 	struct siw_wqe wqe_active;
331 
332 	unsigned int pbl_idx; /* Index into current PBL */
333 	unsigned int sge_idx; /* current sge in rx */
334 	unsigned int sge_off; /* already rcvd in curr. sge */
335 
336 	char first_ddp_seg; /* this is the first DDP seg */
337 	char more_ddp_segs; /* more DDP segs expected */
338 	u8 prev_rdmap_op : 4; /* opcode of prev frame */
339 };
340 
341 /*
342  * Shorthands for short packets w/o payload
343  * to be transmitted more efficient.
344  */
345 struct siw_send_pkt {
346 	struct iwarp_send send;
347 	__be32 crc;
348 };
349 
350 struct siw_write_pkt {
351 	struct iwarp_rdma_write write;
352 	__be32 crc;
353 };
354 
355 struct siw_rreq_pkt {
356 	struct iwarp_rdma_rreq rreq;
357 	__be32 crc;
358 };
359 
360 struct siw_rresp_pkt {
361 	struct iwarp_rdma_rresp rresp;
362 	__be32 crc;
363 };
364 
365 struct siw_iwarp_tx {
366 	union {
367 		union iwarp_hdr hdr;
368 
369 		/* Generic part of FPDU header */
370 		struct iwarp_ctrl ctrl;
371 		struct iwarp_ctrl_untagged c_untagged;
372 		struct iwarp_ctrl_tagged c_tagged;
373 
374 		/* FPDU headers */
375 		struct iwarp_rdma_write rwrite;
376 		struct iwarp_rdma_rreq rreq;
377 		struct iwarp_rdma_rresp rresp;
378 		struct iwarp_terminate terminate;
379 		struct iwarp_send send;
380 		struct iwarp_send_inv send_inv;
381 
382 		/* complete short FPDUs */
383 		struct siw_send_pkt send_pkt;
384 		struct siw_write_pkt write_pkt;
385 		struct siw_rreq_pkt rreq_pkt;
386 		struct siw_rresp_pkt rresp_pkt;
387 	} pkt;
388 
389 	struct mpa_trailer trailer;
390 	/* DDP MSN for untagged messages */
391 	u32 ddp_msn[RDMAP_UNTAGGED_QN_COUNT];
392 
393 	enum siw_tx_ctx state;
394 	u16 ctrl_len; /* ddp+rdmap hdr */
395 	u16 ctrl_sent;
396 	int burst;
397 	int bytes_unsent; /* ddp payload bytes */
398 
399 	struct shash_desc *mpa_crc_hd;
400 
401 	u8 do_crc : 1; /* do crc for segment */
402 	u8 use_sendpage : 1; /* send w/o copy */
403 	u8 tx_suspend : 1; /* stop sending DDP segs. */
404 	u8 pad : 2; /* # pad in current fpdu */
405 	u8 orq_fence : 1; /* ORQ full or Send fenced */
406 	u8 in_syscall : 1; /* TX out of user context */
407 	u8 zcopy_tx : 1; /* Use TCP_SENDPAGE if possible */
408 	u8 gso_seg_limit; /* Maximum segments for GSO, 0 = unbound */
409 
410 	u16 fpdu_len; /* len of FPDU to tx */
411 	unsigned int tcp_seglen; /* remaining tcp seg space */
412 
413 	struct siw_wqe wqe_active;
414 
415 	int pbl_idx; /* Index into current PBL */
416 	int sge_idx; /* current sge in tx */
417 	u32 sge_off; /* already sent in curr. sge */
418 };
419 
420 struct siw_qp {
421 	struct siw_device *sdev;
422 	struct ib_qp *ib_qp;
423 	struct kref ref;
424 	u32 qp_num;
425 	struct list_head devq;
426 	int tx_cpu;
427 	bool kernel_verbs;
428 	struct siw_qp_attrs attrs;
429 
430 	struct siw_cep *cep;
431 	struct rw_semaphore state_lock;
432 
433 	struct ib_pd *pd;
434 	struct siw_cq *scq;
435 	struct siw_cq *rcq;
436 	struct siw_srq *srq;
437 
438 	struct siw_iwarp_tx tx_ctx; /* Transmit context */
439 	spinlock_t sq_lock;
440 	struct siw_sqe *sendq; /* send queue element array */
441 	uint32_t sq_get; /* consumer index into sq array */
442 	uint32_t sq_put; /* kernel prod. index into sq array */
443 	struct llist_node tx_list;
444 
445 	struct siw_sqe *orq; /* outbound read queue element array */
446 	spinlock_t orq_lock;
447 	uint32_t orq_get; /* consumer index into orq array */
448 	uint32_t orq_put; /* shared producer index for ORQ */
449 
450 	struct siw_rx_stream rx_stream;
451 	struct siw_rx_fpdu *rx_fpdu;
452 	struct siw_rx_fpdu rx_tagged;
453 	struct siw_rx_fpdu rx_untagged;
454 	spinlock_t rq_lock;
455 	struct siw_rqe *recvq; /* recv queue element array */
456 	uint32_t rq_get; /* consumer index into rq array */
457 	uint32_t rq_put; /* kernel prod. index into rq array */
458 
459 	struct siw_sqe *irq; /* inbound read queue element array */
460 	uint32_t irq_get; /* consumer index into irq array */
461 	uint32_t irq_put; /* producer index into irq array */
462 	int irq_burst;
463 
464 	struct { /* information to be carried in TERMINATE pkt, if valid */
465 		u8 valid;
466 		u8 in_tx;
467 		u8 layer : 4, etype : 4;
468 		u8 ecode;
469 	} term_info;
470 	struct rdma_user_mmap_entry *sq_entry; /* mmap info for SQE array */
471 	struct rdma_user_mmap_entry *rq_entry; /* mmap info for RQE array */
472 	struct rcu_head rcu;
473 };
474 
475 struct siw_base_qp {
476 	struct ib_qp base_qp;
477 	struct siw_qp *qp;
478 };
479 
480 /* helper macros */
481 #define rx_qp(rx) container_of(rx, struct siw_qp, rx_stream)
482 #define tx_qp(tx) container_of(tx, struct siw_qp, tx_ctx)
483 #define tx_wqe(qp) (&(qp)->tx_ctx.wqe_active)
484 #define rx_wqe(rctx) (&(rctx)->wqe_active)
485 #define rx_mem(rctx) ((rctx)->wqe_active.mem[0])
486 #define tx_type(wqe) ((wqe)->sqe.opcode)
487 #define rx_type(wqe) ((wqe)->rqe.opcode)
488 #define tx_flags(wqe) ((wqe)->sqe.flags)
489 
490 struct iwarp_msg_info {
491 	int hdr_len;
492 	struct iwarp_ctrl ctrl;
493 	int (*rx_data)(struct siw_qp *qp);
494 };
495 
496 struct siw_user_mmap_entry {
497 	struct rdma_user_mmap_entry rdma_entry;
498 	void *address;
499 };
500 
501 /* Global siw parameters. Currently set in siw_main.c */
502 extern const bool zcopy_tx;
503 extern const bool try_gso;
504 extern const bool loopback_enabled;
505 extern const bool mpa_crc_required;
506 extern const bool mpa_crc_strict;
507 extern const bool siw_tcp_nagle;
508 extern u_char mpa_version;
509 extern const bool peer_to_peer;
510 extern struct task_struct *siw_tx_thread[];
511 
512 extern struct crypto_shash *siw_crypto_shash;
513 extern struct iwarp_msg_info iwarp_pktinfo[RDMAP_TERMINATE + 1];
514 
515 /* QP general functions */
516 int siw_qp_modify(struct siw_qp *qp, struct siw_qp_attrs *attr,
517 		  enum siw_qp_attr_mask mask);
518 int siw_qp_mpa_rts(struct siw_qp *qp, enum mpa_v2_ctrl ctrl);
519 void siw_qp_llp_close(struct siw_qp *qp);
520 void siw_qp_cm_drop(struct siw_qp *qp, int schedule);
521 void siw_send_terminate(struct siw_qp *qp);
522 
523 void siw_qp_get_ref(struct ib_qp *qp);
524 void siw_qp_put_ref(struct ib_qp *qp);
525 int siw_qp_add(struct siw_device *sdev, struct siw_qp *qp);
526 void siw_free_qp(struct kref *ref);
527 
528 void siw_init_terminate(struct siw_qp *qp, enum term_elayer layer,
529 			u8 etype, u8 ecode, int in_tx);
530 enum ddp_ecode siw_tagged_error(enum siw_access_state state);
531 enum rdmap_ecode siw_rdmap_error(enum siw_access_state state);
532 
533 void siw_read_to_orq(struct siw_sqe *rreq, struct siw_sqe *sqe);
534 int siw_sqe_complete(struct siw_qp *qp, struct siw_sqe *sqe, u32 bytes,
535 		     enum siw_wc_status status);
536 int siw_rqe_complete(struct siw_qp *qp, struct siw_rqe *rqe, u32 bytes,
537 		     u32 inval_stag, enum siw_wc_status status);
538 void siw_qp_llp_data_ready(struct sock *sk);
539 void siw_qp_llp_write_space(struct sock *sk);
540 
541 /* QP TX path functions */
542 int siw_run_sq(void *arg);
543 int siw_qp_sq_process(struct siw_qp *qp);
544 int siw_sq_start(struct siw_qp *qp);
545 int siw_activate_tx(struct siw_qp *qp);
546 void siw_stop_tx_thread(int nr_cpu);
547 int siw_get_tx_cpu(struct siw_device *sdev);
548 void siw_put_tx_cpu(int cpu);
549 
550 /* QP RX path functions */
551 int siw_proc_send(struct siw_qp *qp);
552 int siw_proc_rreq(struct siw_qp *qp);
553 int siw_proc_rresp(struct siw_qp *qp);
554 int siw_proc_write(struct siw_qp *qp);
555 int siw_proc_terminate(struct siw_qp *qp);
556 
557 int siw_tcp_rx_data(read_descriptor_t *rd_desc, struct sk_buff *skb,
558 		    unsigned int off, size_t len);
559 
560 static inline void set_rx_fpdu_context(struct siw_qp *qp, u8 opcode)
561 {
562 	if (opcode == RDMAP_RDMA_WRITE || opcode == RDMAP_RDMA_READ_RESP)
563 		qp->rx_fpdu = &qp->rx_tagged;
564 	else
565 		qp->rx_fpdu = &qp->rx_untagged;
566 
567 	qp->rx_stream.rdmap_op = opcode;
568 }
569 
570 static inline struct siw_ucontext *to_siw_ctx(struct ib_ucontext *base_ctx)
571 {
572 	return container_of(base_ctx, struct siw_ucontext, base_ucontext);
573 }
574 
575 static inline struct siw_base_qp *to_siw_base_qp(struct ib_qp *base_qp)
576 {
577 	return container_of(base_qp, struct siw_base_qp, base_qp);
578 }
579 
580 static inline struct siw_qp *to_siw_qp(struct ib_qp *base_qp)
581 {
582 	return to_siw_base_qp(base_qp)->qp;
583 }
584 
585 static inline struct siw_cq *to_siw_cq(struct ib_cq *base_cq)
586 {
587 	return container_of(base_cq, struct siw_cq, base_cq);
588 }
589 
590 static inline struct siw_srq *to_siw_srq(struct ib_srq *base_srq)
591 {
592 	return container_of(base_srq, struct siw_srq, base_srq);
593 }
594 
595 static inline struct siw_device *to_siw_dev(struct ib_device *base_dev)
596 {
597 	return container_of(base_dev, struct siw_device, base_dev);
598 }
599 
600 static inline struct siw_mr *to_siw_mr(struct ib_mr *base_mr)
601 {
602 	return container_of(base_mr, struct siw_mr, base_mr);
603 }
604 
605 static inline struct siw_user_mmap_entry *
606 to_siw_mmap_entry(struct rdma_user_mmap_entry *rdma_mmap)
607 {
608 	return container_of(rdma_mmap, struct siw_user_mmap_entry, rdma_entry);
609 }
610 
611 static inline struct siw_qp *siw_qp_id2obj(struct siw_device *sdev, int id)
612 {
613 	struct siw_qp *qp;
614 
615 	rcu_read_lock();
616 	qp = xa_load(&sdev->qp_xa, id);
617 	if (likely(qp && kref_get_unless_zero(&qp->ref))) {
618 		rcu_read_unlock();
619 		return qp;
620 	}
621 	rcu_read_unlock();
622 	return NULL;
623 }
624 
625 static inline u32 qp_id(struct siw_qp *qp)
626 {
627 	return qp->qp_num;
628 }
629 
630 static inline void siw_qp_get(struct siw_qp *qp)
631 {
632 	kref_get(&qp->ref);
633 }
634 
635 static inline void siw_qp_put(struct siw_qp *qp)
636 {
637 	kref_put(&qp->ref, siw_free_qp);
638 }
639 
640 static inline int siw_sq_empty(struct siw_qp *qp)
641 {
642 	struct siw_sqe *sqe = &qp->sendq[qp->sq_get % qp->attrs.sq_size];
643 
644 	return READ_ONCE(sqe->flags) == 0;
645 }
646 
647 static inline struct siw_sqe *sq_get_next(struct siw_qp *qp)
648 {
649 	struct siw_sqe *sqe = &qp->sendq[qp->sq_get % qp->attrs.sq_size];
650 
651 	if (READ_ONCE(sqe->flags) & SIW_WQE_VALID)
652 		return sqe;
653 
654 	return NULL;
655 }
656 
657 static inline struct siw_sqe *orq_get_current(struct siw_qp *qp)
658 {
659 	return &qp->orq[qp->orq_get % qp->attrs.orq_size];
660 }
661 
662 static inline struct siw_sqe *orq_get_tail(struct siw_qp *qp)
663 {
664 	return &qp->orq[qp->orq_put % qp->attrs.orq_size];
665 }
666 
667 static inline struct siw_sqe *orq_get_free(struct siw_qp *qp)
668 {
669 	struct siw_sqe *orq_e = orq_get_tail(qp);
670 
671 	if (orq_e && READ_ONCE(orq_e->flags) == 0)
672 		return orq_e;
673 
674 	return NULL;
675 }
676 
677 static inline int siw_orq_empty(struct siw_qp *qp)
678 {
679 	return qp->orq[qp->orq_get % qp->attrs.orq_size].flags == 0 ? 1 : 0;
680 }
681 
682 static inline struct siw_sqe *irq_alloc_free(struct siw_qp *qp)
683 {
684 	struct siw_sqe *irq_e = &qp->irq[qp->irq_put % qp->attrs.irq_size];
685 
686 	if (READ_ONCE(irq_e->flags) == 0) {
687 		qp->irq_put++;
688 		return irq_e;
689 	}
690 	return NULL;
691 }
692 
693 static inline __wsum siw_csum_update(const void *buff, int len, __wsum sum)
694 {
695 	return (__force __wsum)crc32c((__force __u32)sum, buff, len);
696 }
697 
698 static inline __wsum siw_csum_combine(__wsum csum, __wsum csum2, int offset,
699 				      int len)
700 {
701 	return (__force __wsum)__crc32c_le_combine((__force __u32)csum,
702 						   (__force __u32)csum2, len);
703 }
704 
705 static inline void siw_crc_skb(struct siw_rx_stream *srx, unsigned int len)
706 {
707 	const struct skb_checksum_ops siw_cs_ops = {
708 		.update = siw_csum_update,
709 		.combine = siw_csum_combine,
710 	};
711 	__wsum crc = *(u32 *)shash_desc_ctx(srx->mpa_crc_hd);
712 
713 	crc = __skb_checksum(srx->skb, srx->skb_offset, len, crc,
714 			     &siw_cs_ops);
715 	*(u32 *)shash_desc_ctx(srx->mpa_crc_hd) = crc;
716 }
717 
718 #define siw_dbg(ibdev, fmt, ...)                                               \
719 	ibdev_dbg(ibdev, "%s: " fmt, __func__, ##__VA_ARGS__)
720 
721 #define siw_dbg_qp(qp, fmt, ...)                                               \
722 	ibdev_dbg(&qp->sdev->base_dev, "QP[%u] %s: " fmt, qp_id(qp), __func__, \
723 		  ##__VA_ARGS__)
724 
725 #define siw_dbg_cq(cq, fmt, ...)                                               \
726 	ibdev_dbg(cq->base_cq.device, "CQ[%u] %s: " fmt, cq->id, __func__,     \
727 		  ##__VA_ARGS__)
728 
729 #define siw_dbg_pd(pd, fmt, ...)                                               \
730 	ibdev_dbg(pd->device, "PD[%u] %s: " fmt, pd->res.id, __func__,         \
731 		  ##__VA_ARGS__)
732 
733 #define siw_dbg_mem(mem, fmt, ...)                                             \
734 	ibdev_dbg(&mem->sdev->base_dev,                                        \
735 		  "MEM[0x%08x] %s: " fmt, mem->stag, __func__, ##__VA_ARGS__)
736 
737 #define siw_dbg_cep(cep, fmt, ...)                                             \
738 	ibdev_dbg(&cep->sdev->base_dev, "CEP[0x%pK] %s: " fmt,                  \
739 		  cep, __func__, ##__VA_ARGS__)
740 
741 void siw_cq_flush(struct siw_cq *cq);
742 void siw_sq_flush(struct siw_qp *qp);
743 void siw_rq_flush(struct siw_qp *qp);
744 int siw_reap_cqe(struct siw_cq *cq, struct ib_wc *wc);
745 
746 #endif
747