xref: /freebsd/sys/dev/cxgbe/iw_cxgbe/t4.h (revision 148a8da8)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009-2013 Chelsio, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  *
33  * $FreeBSD$
34  */
35 #ifndef __T4_H__
36 #define __T4_H__
37 
38 #include "common/t4_regs_values.h"
39 #include "common/t4_regs.h"
40 /*
41  * Fixme: Adding missing defines
42  */
43 #define SGE_PF_KDOORBELL 0x0
44 #define  QID_MASK    0xffff8000U
45 #define  QID_SHIFT   15
46 #define  QID(x)      ((x) << QID_SHIFT)
47 #define  DBPRIO      0x00004000U
48 #define  PIDX_MASK   0x00003fffU
49 #define  PIDX_SHIFT  0
50 #define  PIDX(x)     ((x) << PIDX_SHIFT)
51 
52 #define SGE_PF_GTS 0x4
53 #define  INGRESSQID_MASK   0xffff0000U
54 #define  INGRESSQID_SHIFT  16
55 #define  INGRESSQID(x)     ((x) << INGRESSQID_SHIFT)
56 #define  TIMERREG_MASK     0x0000e000U
57 #define  TIMERREG_SHIFT    13
58 #define  TIMERREG(x)       ((x) << TIMERREG_SHIFT)
59 #define  SEINTARM_MASK     0x00001000U
60 #define  SEINTARM_SHIFT    12
61 #define  SEINTARM(x)       ((x) << SEINTARM_SHIFT)
62 #define  CIDXINC_MASK      0x00000fffU
63 #define  CIDXINC_SHIFT     0
64 #define  CIDXINC(x)        ((x) << CIDXINC_SHIFT)
65 
66 #define T4_MAX_NUM_PD 65536
67 #define T4_MAX_MR_SIZE (~0ULL)
68 #define T4_PAGESIZE_MASK 0xffffffff000 /* 4KB-8TB */
69 #define T4_STAG_UNSET 0xffffffff
70 #define T4_FW_MAJ 0
71 #define A_PCIE_MA_SYNC 0x30b4
72 
73 struct t4_status_page {
74 	__be32 rsvd1;	/* flit 0 - hw owns */
75 	__be16 rsvd2;
76 	__be16 qid;
77 	__be16 cidx;
78 	__be16 pidx;
79 	u8 qp_err;	/* flit 1 - sw owns */
80 	u8 db_off;
81 	u8 pad;
82 	u16 host_wq_pidx;
83 	u16 host_cidx;
84 	u16 host_pidx;
85 };
86 
87 #define T4_EQ_ENTRY_SIZE 64
88 
89 #define T4_SQ_NUM_SLOTS 5
90 #define T4_SQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_SQ_NUM_SLOTS)
91 #define T4_MAX_SEND_SGE ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_send_wr) - \
92 			sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
93 #define T4_MAX_SEND_INLINE ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_send_wr) - \
94 			sizeof(struct fw_ri_immd)))
95 #define T4_MAX_WRITE_INLINE ((T4_SQ_NUM_BYTES - \
96 			sizeof(struct fw_ri_rdma_write_wr) - \
97 			sizeof(struct fw_ri_immd)))
98 #define T4_MAX_WRITE_SGE ((T4_SQ_NUM_BYTES - \
99 			sizeof(struct fw_ri_rdma_write_wr) - \
100 			sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
101 #define T4_MAX_FR_IMMD ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_fr_nsmr_wr) - \
102 			sizeof(struct fw_ri_immd)) & ~31UL)
103 #define T4_MAX_FR_IMMD_DEPTH (T4_MAX_FR_IMMD / sizeof(u64))
104 #define T4_MAX_FR_DSGL 1024
105 #define T4_MAX_FR_DSGL_DEPTH (T4_MAX_FR_DSGL / sizeof(u64))
106 
107 static inline int t4_max_fr_depth(int use_dsgl)
108 {
109 	return use_dsgl ? T4_MAX_FR_DSGL_DEPTH : T4_MAX_FR_IMMD_DEPTH;
110 }
111 
112 #define T4_RQ_NUM_SLOTS 2
113 #define T4_RQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_RQ_NUM_SLOTS)
114 #define T4_MAX_RECV_SGE 4
115 
116 union t4_wr {
117 	struct fw_ri_res_wr res;
118 	struct fw_ri_wr ri;
119 	struct fw_ri_rdma_write_wr write;
120 	struct fw_ri_send_wr send;
121 	struct fw_ri_rdma_read_wr read;
122 	struct fw_ri_bind_mw_wr bind;
123 	struct fw_ri_fr_nsmr_wr fr;
124 	struct fw_ri_fr_nsmr_tpte_wr fr_tpte;
125 	struct fw_ri_inv_lstag_wr inv;
126 	struct t4_status_page status;
127 	__be64 flits[T4_EQ_ENTRY_SIZE / sizeof(__be64) * T4_SQ_NUM_SLOTS];
128 };
129 
130 union t4_recv_wr {
131 	struct fw_ri_recv_wr recv;
132 	struct t4_status_page status;
133 	__be64 flits[T4_EQ_ENTRY_SIZE / sizeof(__be64) * T4_RQ_NUM_SLOTS];
134 };
135 
136 static inline void init_wr_hdr(union t4_wr *wqe, u16 wrid,
137 			       enum fw_wr_opcodes opcode, u8 flags, u8 len16)
138 {
139 	wqe->send.opcode = (u8)opcode;
140 	wqe->send.flags = flags;
141 	wqe->send.wrid = wrid;
142 	wqe->send.r1[0] = 0;
143 	wqe->send.r1[1] = 0;
144 	wqe->send.r1[2] = 0;
145 	wqe->send.len16 = len16;
146 }
147 
148 /* CQE/AE status codes */
149 #define T4_ERR_SUCCESS                     0x0
150 #define T4_ERR_STAG                        0x1	/* STAG invalid: either the */
151 						/* STAG is offlimt, being 0, */
152 						/* or STAG_key mismatch */
153 #define T4_ERR_PDID                        0x2	/* PDID mismatch */
154 #define T4_ERR_QPID                        0x3	/* QPID mismatch */
155 #define T4_ERR_ACCESS                      0x4	/* Invalid access right */
156 #define T4_ERR_WRAP                        0x5	/* Wrap error */
157 #define T4_ERR_BOUND                       0x6	/* base and bounds voilation */
158 #define T4_ERR_INVALIDATE_SHARED_MR        0x7	/* attempt to invalidate a  */
159 						/* shared memory region */
160 #define T4_ERR_INVALIDATE_MR_WITH_MW_BOUND 0x8	/* attempt to invalidate a  */
161 						/* shared memory region */
162 #define T4_ERR_ECC                         0x9	/* ECC error detected */
163 #define T4_ERR_ECC_PSTAG                   0xA	/* ECC error detected when  */
164 						/* reading PSTAG for a MW  */
165 						/* Invalidate */
166 #define T4_ERR_PBL_ADDR_BOUND              0xB	/* pbl addr out of bounds:  */
167 						/* software error */
168 #define T4_ERR_SWFLUSH			   0xC	/* SW FLUSHED */
169 #define T4_ERR_CRC                         0x10 /* CRC error */
170 #define T4_ERR_MARKER                      0x11 /* Marker error */
171 #define T4_ERR_PDU_LEN_ERR                 0x12 /* invalid PDU length */
172 #define T4_ERR_OUT_OF_RQE                  0x13 /* out of RQE */
173 #define T4_ERR_DDP_VERSION                 0x14 /* wrong DDP version */
174 #define T4_ERR_RDMA_VERSION                0x15 /* wrong RDMA version */
175 #define T4_ERR_OPCODE                      0x16 /* invalid rdma opcode */
176 #define T4_ERR_DDP_QUEUE_NUM               0x17 /* invalid ddp queue number */
177 #define T4_ERR_MSN                         0x18 /* MSN error */
178 #define T4_ERR_TBIT                        0x19 /* tag bit not set correctly */
179 #define T4_ERR_MO                          0x1A /* MO not 0 for TERMINATE  */
180 						/* or READ_REQ */
181 #define T4_ERR_MSN_GAP                     0x1B
182 #define T4_ERR_MSN_RANGE                   0x1C
183 #define T4_ERR_IRD_OVERFLOW                0x1D
184 #define T4_ERR_RQE_ADDR_BOUND              0x1E /* RQE addr out of bounds:  */
185 						/* software error */
186 #define T4_ERR_INTERNAL_ERR                0x1F /* internal error (opcode  */
187 						/* mismatch) */
188 /*
189  * CQE defs
190  */
191 struct t4_cqe {
192 	__be32 header;
193 	__be32 len;
194 	union {
195 		struct {
196 			__be32 stag;
197 			__be32 msn;
198 		} rcqe;
199 		struct {
200 			u32 stag;
201 			u16 nada2;
202 			u16 cidx;
203 		} scqe;
204 		struct {
205 			__be32 wrid_hi;
206 			__be32 wrid_low;
207 		} gen;
208 		u64 drain_cookie;
209 	} u;
210 	__be64 reserved;
211 	__be64 bits_type_ts;
212 };
213 
214 /* macros for flit 0 of the cqe */
215 
216 #define S_CQE_QPID        12
217 #define M_CQE_QPID        0xFFFFF
218 #define G_CQE_QPID(x)     ((((x) >> S_CQE_QPID)) & M_CQE_QPID)
219 #define V_CQE_QPID(x)	  ((x)<<S_CQE_QPID)
220 
221 #define S_CQE_SWCQE       11
222 #define M_CQE_SWCQE       0x1
223 #define G_CQE_SWCQE(x)    ((((x) >> S_CQE_SWCQE)) & M_CQE_SWCQE)
224 #define V_CQE_SWCQE(x)	  ((x)<<S_CQE_SWCQE)
225 
226 #define S_CQE_STATUS      5
227 #define M_CQE_STATUS      0x1F
228 #define G_CQE_STATUS(x)   ((((x) >> S_CQE_STATUS)) & M_CQE_STATUS)
229 #define V_CQE_STATUS(x)   ((x)<<S_CQE_STATUS)
230 
231 #define S_CQE_TYPE        4
232 #define M_CQE_TYPE        0x1
233 #define G_CQE_TYPE(x)     ((((x) >> S_CQE_TYPE)) & M_CQE_TYPE)
234 #define V_CQE_TYPE(x)     ((x)<<S_CQE_TYPE)
235 
236 #define S_CQE_OPCODE      0
237 #define M_CQE_OPCODE      0xF
238 #define G_CQE_OPCODE(x)   ((((x) >> S_CQE_OPCODE)) & M_CQE_OPCODE)
239 #define V_CQE_OPCODE(x)   ((x)<<S_CQE_OPCODE)
240 
241 #define SW_CQE(x)         (G_CQE_SWCQE(be32_to_cpu((x)->header)))
242 #define CQE_QPID(x)       (G_CQE_QPID(be32_to_cpu((x)->header)))
243 #define CQE_TYPE(x)       (G_CQE_TYPE(be32_to_cpu((x)->header)))
244 #define SQ_TYPE(x)	  (CQE_TYPE((x)))
245 #define RQ_TYPE(x)	  (!CQE_TYPE((x)))
246 #define CQE_STATUS(x)     (G_CQE_STATUS(be32_to_cpu((x)->header)))
247 #define CQE_OPCODE(x)     (G_CQE_OPCODE(be32_to_cpu((x)->header)))
248 
249 #define CQE_SEND_OPCODE(x)(\
250 	(G_CQE_OPCODE(be32_to_cpu((x)->header)) == FW_RI_SEND) || \
251 	(G_CQE_OPCODE(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_SE) || \
252 	(G_CQE_OPCODE(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_INV) || \
253 	(G_CQE_OPCODE(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_SE_INV))
254 
255 #define CQE_LEN(x)        (be32_to_cpu((x)->len))
256 
257 /* used for RQ completion processing */
258 #define CQE_WRID_STAG(x)  (be32_to_cpu((x)->u.rcqe.stag))
259 #define CQE_WRID_MSN(x)   (be32_to_cpu((x)->u.rcqe.msn))
260 
261 /* used for SQ completion processing */
262 #define CQE_WRID_SQ_IDX(x)	((x)->u.scqe.cidx)
263 #define CQE_WRID_FR_STAG(x)     (be32_to_cpu((x)->u.scqe.stag))
264 
265 /* generic accessor macros */
266 #define CQE_WRID_HI(x)		((x)->u.gen.wrid_hi)
267 #define CQE_WRID_LOW(x)		((x)->u.gen.wrid_low)
268 #define CQE_DRAIN_COOKIE(x)	(x)->u.drain_cookie;
269 
270 /* macros for flit 3 of the cqe */
271 #define S_CQE_GENBIT	63
272 #define M_CQE_GENBIT	0x1
273 #define G_CQE_GENBIT(x)	(((x) >> S_CQE_GENBIT) & M_CQE_GENBIT)
274 #define V_CQE_GENBIT(x) ((x)<<S_CQE_GENBIT)
275 
276 #define S_CQE_OVFBIT	62
277 #define M_CQE_OVFBIT	0x1
278 #define G_CQE_OVFBIT(x)	((((x) >> S_CQE_OVFBIT)) & M_CQE_OVFBIT)
279 
280 #define S_CQE_IQTYPE	60
281 #define M_CQE_IQTYPE	0x3
282 #define G_CQE_IQTYPE(x)	((((x) >> S_CQE_IQTYPE)) & M_CQE_IQTYPE)
283 
284 #define M_CQE_TS	0x0fffffffffffffffULL
285 #define G_CQE_TS(x)	((x) & M_CQE_TS)
286 
287 #define CQE_OVFBIT(x)	((unsigned)G_CQE_OVFBIT(be64_to_cpu((x)->bits_type_ts)))
288 #define CQE_GENBIT(x)	((unsigned)G_CQE_GENBIT(be64_to_cpu((x)->bits_type_ts)))
289 #define CQE_TS(x)	(G_CQE_TS(be64_to_cpu((x)->bits_type_ts)))
290 
291 struct t4_swsqe {
292 	u64			wr_id;
293 	struct t4_cqe		cqe;
294 	int			read_len;
295 	int			opcode;
296 	int			complete;
297 	int			signaled;
298 	u16			idx;
299 	int                     flushed;
300 	struct timespec         host_ts;
301 	u64                     sge_ts;
302 };
303 
304 static inline pgprot_t t4_pgprot_wc(pgprot_t prot)
305 {
306 #if defined(__i386__) || defined(__x86_64__) || defined(CONFIG_PPC64)
307 	return pgprot_writecombine(prot);
308 #else
309 	return pgprot_noncached(prot);
310 #endif
311 }
312 
313 enum {
314 	T4_SQ_ONCHIP = (1<<0),
315 };
316 
317 struct t4_sq {
318 	union t4_wr *queue;
319 	bus_addr_t dma_addr;
320 	DEFINE_DMA_UNMAP_ADDR(mapping);
321 	unsigned long phys_addr;
322 	struct t4_swsqe *sw_sq;
323 	struct t4_swsqe *oldest_read;
324 	void __iomem *bar2_va;
325 	u64 bar2_pa;
326 	size_t memsize;
327 	u32 bar2_qid;
328 	u32 qid;
329 	u16 in_use;
330 	u16 size;
331 	u16 cidx;
332 	u16 pidx;
333 	u16 wq_pidx;
334 	u16 wq_pidx_inc;
335 	u16 flags;
336 	short flush_cidx;
337 };
338 
339 struct t4_swrqe {
340 	u64 wr_id;
341 };
342 
343 struct t4_rq {
344 	union  t4_recv_wr *queue;
345 	bus_addr_t dma_addr;
346 	DEFINE_DMA_UNMAP_ADDR(mapping);
347 	unsigned long phys_addr;
348 	struct t4_swrqe *sw_rq;
349 	void __iomem *bar2_va;
350 	u64 bar2_pa;
351 	size_t memsize;
352 	u32 bar2_qid;
353 	u32 qid;
354 	u32 msn;
355 	u32 rqt_hwaddr;
356 	u16 rqt_size;
357 	u16 in_use;
358 	u16 size;
359 	u16 cidx;
360 	u16 pidx;
361 	u16 wq_pidx;
362 	u16 wq_pidx_inc;
363 };
364 
365 struct t4_wq {
366 	struct t4_sq sq;
367 	struct t4_rq rq;
368 	struct c4iw_rdev *rdev;
369 	int flushed;
370 };
371 
372 static inline int t4_rqes_posted(struct t4_wq *wq)
373 {
374 	return wq->rq.in_use;
375 }
376 
377 static inline int t4_rq_empty(struct t4_wq *wq)
378 {
379 	return wq->rq.in_use == 0;
380 }
381 
382 static inline int t4_rq_full(struct t4_wq *wq)
383 {
384 	return wq->rq.in_use == (wq->rq.size - 1);
385 }
386 
387 static inline u32 t4_rq_avail(struct t4_wq *wq)
388 {
389 	return wq->rq.size - 1 - wq->rq.in_use;
390 }
391 
392 static inline void t4_rq_produce(struct t4_wq *wq, u8 len16)
393 {
394 	wq->rq.in_use++;
395 	if (++wq->rq.pidx == wq->rq.size)
396 		wq->rq.pidx = 0;
397 	wq->rq.wq_pidx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
398 	if (wq->rq.wq_pidx >= wq->rq.size * T4_RQ_NUM_SLOTS)
399 		wq->rq.wq_pidx %= wq->rq.size * T4_RQ_NUM_SLOTS;
400 }
401 
402 static inline void t4_rq_consume(struct t4_wq *wq)
403 {
404 	wq->rq.in_use--;
405 	wq->rq.msn++;
406 	if (++wq->rq.cidx == wq->rq.size)
407 		wq->rq.cidx = 0;
408 }
409 
410 static inline u16 t4_rq_host_wq_pidx(struct t4_wq *wq)
411 {
412 	return wq->rq.queue[wq->rq.size].status.host_wq_pidx;
413 }
414 
415 static inline u16 t4_rq_wq_size(struct t4_wq *wq)
416 {
417 	return wq->rq.size * T4_RQ_NUM_SLOTS;
418 }
419 
420 static inline int t4_sq_onchip(struct t4_sq *sq)
421 {
422 	return sq->flags & T4_SQ_ONCHIP;
423 }
424 
425 static inline int t4_sq_empty(struct t4_wq *wq)
426 {
427 	return wq->sq.in_use == 0;
428 }
429 
430 static inline int t4_sq_full(struct t4_wq *wq)
431 {
432 	return wq->sq.in_use == (wq->sq.size - 1);
433 }
434 
435 static inline u32 t4_sq_avail(struct t4_wq *wq)
436 {
437 	return wq->sq.size - 1 - wq->sq.in_use;
438 }
439 
440 static inline void t4_sq_produce(struct t4_wq *wq, u8 len16)
441 {
442 	wq->sq.in_use++;
443 	if (++wq->sq.pidx == wq->sq.size)
444 		wq->sq.pidx = 0;
445 	wq->sq.wq_pidx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
446 	if (wq->sq.wq_pidx >= wq->sq.size * T4_SQ_NUM_SLOTS)
447 		wq->sq.wq_pidx %= wq->sq.size * T4_SQ_NUM_SLOTS;
448 }
449 
450 static inline void t4_sq_consume(struct t4_wq *wq)
451 {
452 	BUG_ON(wq->sq.in_use < 1);
453 	if (wq->sq.cidx == wq->sq.flush_cidx)
454 		wq->sq.flush_cidx = -1;
455 	wq->sq.in_use--;
456 	if (++wq->sq.cidx == wq->sq.size)
457 		wq->sq.cidx = 0;
458 }
459 
460 static inline u16 t4_sq_host_wq_pidx(struct t4_wq *wq)
461 {
462 	return wq->sq.queue[wq->sq.size].status.host_wq_pidx;
463 }
464 
465 static inline u16 t4_sq_wq_size(struct t4_wq *wq)
466 {
467 		return wq->sq.size * T4_SQ_NUM_SLOTS;
468 }
469 
470 /* This function copies 64 byte coalesced work request to memory
471  * mapped BAR2 space. For coalesced WRs, the SGE fetches data
472  * from the FIFO instead of from Host.
473  */
474 static inline void pio_copy(u64 __iomem *dst, u64 *src)
475 {
476 	int count = 8;
477 
478 	while (count) {
479 		writeq(*src, dst);
480 		src++;
481 		dst++;
482 		count--;
483 	}
484 }
485 
486 static inline void
487 t4_ring_sq_db(struct t4_wq *wq, u16 inc, union t4_wr *wqe, u8 wc)
488 {
489 
490 	/* Flush host queue memory writes. */
491 	wmb();
492 	if (wc && inc == 1 && wq->sq.bar2_qid == 0 && wqe) {
493 		CTR2(KTR_IW_CXGBE, "%s: WC wq->sq.pidx = %d",
494 				__func__, wq->sq.pidx);
495 		pio_copy((u64 __iomem *)
496 				((u64)wq->sq.bar2_va + SGE_UDB_WCDOORBELL),
497 				(u64 *)wqe);
498 	} else {
499 		CTR2(KTR_IW_CXGBE, "%s: DB wq->sq.pidx = %d",
500 				__func__, wq->sq.pidx);
501 		writel(V_PIDX_T5(inc) | V_QID(wq->sq.bar2_qid),
502 				(void __iomem *)((u64)wq->sq.bar2_va +
503 					SGE_UDB_KDOORBELL));
504 	}
505 
506 	/* Flush user doorbell area writes. */
507 	wmb();
508 	return;
509 }
510 
511 static inline void
512 t4_ring_rq_db(struct t4_wq *wq, u16 inc, union t4_recv_wr *wqe, u8 wc)
513 {
514 
515 	/* Flush host queue memory writes. */
516 	wmb();
517 	if (wc && inc == 1 && wq->rq.bar2_qid == 0 && wqe) {
518 		CTR2(KTR_IW_CXGBE, "%s: WC wq->rq.pidx = %d",
519 				__func__, wq->rq.pidx);
520 		pio_copy((u64 __iomem *)((u64)wq->rq.bar2_va +
521 					SGE_UDB_WCDOORBELL), (u64 *)wqe);
522 	} else {
523 		CTR2(KTR_IW_CXGBE, "%s: DB wq->rq.pidx = %d",
524 				__func__, wq->rq.pidx);
525 		writel(V_PIDX_T5(inc) | V_QID(wq->rq.bar2_qid),
526 				(void __iomem *)((u64)wq->rq.bar2_va +
527 					SGE_UDB_KDOORBELL));
528 	}
529 
530 	/* Flush user doorbell area writes. */
531 	wmb();
532 	return;
533 }
534 
535 static inline int t4_wq_in_error(struct t4_wq *wq)
536 {
537 	return wq->rq.queue[wq->rq.size].status.qp_err;
538 }
539 
540 static inline void t4_set_wq_in_error(struct t4_wq *wq)
541 {
542 	wq->rq.queue[wq->rq.size].status.qp_err = 1;
543 }
544 
545 enum t4_cq_flags {
546 	CQ_ARMED	= 1,
547 };
548 
549 struct t4_cq {
550 	struct t4_cqe *queue;
551 	bus_addr_t dma_addr;
552 	DEFINE_DMA_UNMAP_ADDR(mapping);
553 	struct t4_cqe *sw_queue;
554 	void __iomem *bar2_va;
555 	u64 bar2_pa;
556 	u32 bar2_qid;
557 	struct c4iw_rdev *rdev;
558 	size_t memsize;
559 	__be64 bits_type_ts;
560 	u32 cqid;
561 	u32 qid_mask;
562 	int vector;
563 	u16 size; /* including status page */
564 	u16 cidx;
565 	u16 sw_pidx;
566 	u16 sw_cidx;
567 	u16 sw_in_use;
568 	u16 cidx_inc;
569 	u8 gen;
570 	u8 error;
571 	unsigned long flags;
572 };
573 
574 static inline void write_gts(struct t4_cq *cq, u32 val)
575 {
576 	writel(val | V_INGRESSQID(cq->bar2_qid),
577 		       (void __iomem *)((u64)cq->bar2_va + SGE_UDB_GTS));
578 }
579 
580 static inline int t4_clear_cq_armed(struct t4_cq *cq)
581 {
582 	return test_and_clear_bit(CQ_ARMED, &cq->flags);
583 }
584 
585 static inline int t4_arm_cq(struct t4_cq *cq, int se)
586 {
587 	u32 val;
588 
589 	set_bit(CQ_ARMED, &cq->flags);
590 	while (cq->cidx_inc > CIDXINC_MASK) {
591 		val = SEINTARM(0) | CIDXINC(CIDXINC_MASK) | TIMERREG(7);
592 		writel(val | V_INGRESSQID(cq->bar2_qid),
593 		       (void __iomem *)((u64)cq->bar2_va + SGE_UDB_GTS));
594 		cq->cidx_inc -= CIDXINC_MASK;
595 	}
596 	val = SEINTARM(se) | CIDXINC(cq->cidx_inc) | TIMERREG(6);
597 	writel(val | V_INGRESSQID(cq->bar2_qid),
598 		       (void __iomem *)((u64)cq->bar2_va + SGE_UDB_GTS));
599 	cq->cidx_inc = 0;
600 	return 0;
601 }
602 
603 static inline void t4_swcq_produce(struct t4_cq *cq)
604 {
605 	cq->sw_in_use++;
606 	if (cq->sw_in_use == cq->size) {
607 		CTR2(KTR_IW_CXGBE, "%s cxgb4 sw cq overflow cqid %u",
608 			 __func__, cq->cqid);
609 		cq->error = 1;
610 		BUG_ON(1);
611 	}
612 	if (++cq->sw_pidx == cq->size)
613 		cq->sw_pidx = 0;
614 }
615 
616 static inline void t4_swcq_consume(struct t4_cq *cq)
617 {
618 	BUG_ON(cq->sw_in_use < 1);
619 	cq->sw_in_use--;
620 	if (++cq->sw_cidx == cq->size)
621 		cq->sw_cidx = 0;
622 }
623 
624 static inline void t4_hwcq_consume(struct t4_cq *cq)
625 {
626 	cq->bits_type_ts = cq->queue[cq->cidx].bits_type_ts;
627 	if (++cq->cidx_inc == (cq->size >> 4) || cq->cidx_inc == M_CIDXINC) {
628 		u32 val;
629 
630 		val = SEINTARM(0) | CIDXINC(cq->cidx_inc) | TIMERREG(7);
631 		write_gts(cq, val);
632 		cq->cidx_inc = 0;
633 	}
634 	if (++cq->cidx == cq->size) {
635 		cq->cidx = 0;
636 		cq->gen ^= 1;
637 	}
638 }
639 
640 static inline int t4_valid_cqe(struct t4_cq *cq, struct t4_cqe *cqe)
641 {
642 	return (CQE_GENBIT(cqe) == cq->gen);
643 }
644 
645 static inline int t4_cq_notempty(struct t4_cq *cq)
646 {
647 	return cq->sw_in_use || t4_valid_cqe(cq, &cq->queue[cq->cidx]);
648 }
649 
650 static inline int t4_next_hw_cqe(struct t4_cq *cq, struct t4_cqe **cqe)
651 {
652 	int ret;
653 	u16 prev_cidx;
654 
655 	if (cq->cidx == 0)
656 		prev_cidx = cq->size - 1;
657 	else
658 		prev_cidx = cq->cidx - 1;
659 
660 	if (cq->queue[prev_cidx].bits_type_ts != cq->bits_type_ts) {
661 		ret = -EOVERFLOW;
662 		cq->error = 1;
663 		printk(KERN_ERR MOD "cq overflow cqid %u\n", cq->cqid);
664 		BUG_ON(1);
665 	} else if (t4_valid_cqe(cq, &cq->queue[cq->cidx])) {
666 
667 		/* Ensure CQE is flushed to memory */
668 		rmb();
669 		*cqe = &cq->queue[cq->cidx];
670 		ret = 0;
671 	} else
672 		ret = -ENODATA;
673 	return ret;
674 }
675 
676 static inline struct t4_cqe *t4_next_sw_cqe(struct t4_cq *cq)
677 {
678 	if (cq->sw_in_use == cq->size) {
679 		CTR2(KTR_IW_CXGBE, "%s cxgb4 sw cq overflow cqid %u",
680 			 __func__, cq->cqid);
681 		cq->error = 1;
682 		BUG_ON(1);
683 		return NULL;
684 	}
685 	if (cq->sw_in_use)
686 		return &cq->sw_queue[cq->sw_cidx];
687 	return NULL;
688 }
689 
690 static inline int t4_next_cqe(struct t4_cq *cq, struct t4_cqe **cqe)
691 {
692 	int ret = 0;
693 
694 	if (cq->error)
695 		ret = -ENODATA;
696 	else if (cq->sw_in_use)
697 		*cqe = &cq->sw_queue[cq->sw_cidx];
698 	else
699 		ret = t4_next_hw_cqe(cq, cqe);
700 	return ret;
701 }
702 
703 static inline int t4_cq_in_error(struct t4_cq *cq)
704 {
705 	return ((struct t4_status_page *)&cq->queue[cq->size])->qp_err;
706 }
707 
708 static inline void t4_set_cq_in_error(struct t4_cq *cq)
709 {
710 	((struct t4_status_page *)&cq->queue[cq->size])->qp_err = 1;
711 }
712 struct t4_dev_status_page {
713 	u8 db_off;
714 	u8 wc_supported;
715 	u16 pad2;
716 	u32 pad3;
717 	u64 qp_start;
718 	u64 qp_size;
719 	u64 cq_start;
720 	u64 cq_size;
721 };
722 #endif
723