2 * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
17 * - Redistributions in binary form must reproduce the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer in the documentation and/or other materials
20 * provided with the distribution.
22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
23 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
24 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
25 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
26 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
27 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
28 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #include "t4_values.h"
38 #include "t4fw_ri_api.h"
40 #define T4_MAX_NUM_PD 65536
41 #define T4_MAX_MR_SIZE (~0ULL)
42 #define T4_PAGESIZE_MASK 0xffff000 /* 4KB-128MB */
43 #define T4_STAG_UNSET 0xffffffff
45 #define PCIE_MA_SYNC_A 0x30b4
47 struct t4_status_page
{
48 __be32 rsvd1
; /* flit 0 - hw owns */
53 u8 qp_err
; /* flit 1 - sw owns */
61 #define T4_EQ_ENTRY_SIZE 64
63 #define T4_SQ_NUM_SLOTS 5
64 #define T4_SQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_SQ_NUM_SLOTS)
65 #define T4_MAX_SEND_SGE ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_send_wr) - \
66 sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
67 #define T4_MAX_SEND_INLINE ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_send_wr) - \
68 sizeof(struct fw_ri_immd)))
69 #define T4_MAX_WRITE_INLINE ((T4_SQ_NUM_BYTES - \
70 sizeof(struct fw_ri_rdma_write_wr) - \
71 sizeof(struct fw_ri_immd)))
72 #define T4_MAX_WRITE_SGE ((T4_SQ_NUM_BYTES - \
73 sizeof(struct fw_ri_rdma_write_wr) - \
74 sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
75 #define T4_MAX_FR_IMMD ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_fr_nsmr_wr) - \
76 sizeof(struct fw_ri_immd)) & ~31UL)
77 #define T4_MAX_FR_IMMD_DEPTH (T4_MAX_FR_IMMD / sizeof(u64))
78 #define T4_MAX_FR_DSGL 1024
79 #define T4_MAX_FR_DSGL_DEPTH (T4_MAX_FR_DSGL / sizeof(u64))
81 static inline int t4_max_fr_depth(int use_dsgl
)
83 return use_dsgl
? T4_MAX_FR_DSGL_DEPTH
: T4_MAX_FR_IMMD_DEPTH
;
86 #define T4_RQ_NUM_SLOTS 2
87 #define T4_RQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_RQ_NUM_SLOTS)
88 #define T4_MAX_RECV_SGE 4
91 struct fw_ri_res_wr res
;
93 struct fw_ri_rdma_write_wr write
;
94 struct fw_ri_send_wr send
;
95 struct fw_ri_rdma_read_wr read
;
96 struct fw_ri_bind_mw_wr bind
;
97 struct fw_ri_fr_nsmr_wr fr
;
98 struct fw_ri_inv_lstag_wr inv
;
99 struct t4_status_page status
;
100 __be64 flits
[T4_EQ_ENTRY_SIZE
/ sizeof(__be64
) * T4_SQ_NUM_SLOTS
];
104 struct fw_ri_recv_wr recv
;
105 struct t4_status_page status
;
106 __be64 flits
[T4_EQ_ENTRY_SIZE
/ sizeof(__be64
) * T4_RQ_NUM_SLOTS
];
109 static inline void init_wr_hdr(union t4_wr
*wqe
, u16 wrid
,
110 enum fw_wr_opcodes opcode
, u8 flags
, u8 len16
)
112 wqe
->send
.opcode
= (u8
)opcode
;
113 wqe
->send
.flags
= flags
;
114 wqe
->send
.wrid
= wrid
;
118 wqe
->send
.len16
= len16
;
121 /* CQE/AE status codes */
122 #define T4_ERR_SUCCESS 0x0
123 #define T4_ERR_STAG 0x1 /* STAG invalid: either the */
124 /* STAG is offlimt, being 0, */
125 /* or STAG_key mismatch */
126 #define T4_ERR_PDID 0x2 /* PDID mismatch */
127 #define T4_ERR_QPID 0x3 /* QPID mismatch */
128 #define T4_ERR_ACCESS 0x4 /* Invalid access right */
129 #define T4_ERR_WRAP 0x5 /* Wrap error */
130 #define T4_ERR_BOUND 0x6 /* base and bounds voilation */
131 #define T4_ERR_INVALIDATE_SHARED_MR 0x7 /* attempt to invalidate a */
132 /* shared memory region */
133 #define T4_ERR_INVALIDATE_MR_WITH_MW_BOUND 0x8 /* attempt to invalidate a */
134 /* shared memory region */
135 #define T4_ERR_ECC 0x9 /* ECC error detected */
136 #define T4_ERR_ECC_PSTAG 0xA /* ECC error detected when */
137 /* reading PSTAG for a MW */
139 #define T4_ERR_PBL_ADDR_BOUND 0xB /* pbl addr out of bounds: */
141 #define T4_ERR_SWFLUSH 0xC /* SW FLUSHED */
142 #define T4_ERR_CRC 0x10 /* CRC error */
143 #define T4_ERR_MARKER 0x11 /* Marker error */
144 #define T4_ERR_PDU_LEN_ERR 0x12 /* invalid PDU length */
145 #define T4_ERR_OUT_OF_RQE 0x13 /* out of RQE */
146 #define T4_ERR_DDP_VERSION 0x14 /* wrong DDP version */
147 #define T4_ERR_RDMA_VERSION 0x15 /* wrong RDMA version */
148 #define T4_ERR_OPCODE 0x16 /* invalid rdma opcode */
149 #define T4_ERR_DDP_QUEUE_NUM 0x17 /* invalid ddp queue number */
150 #define T4_ERR_MSN 0x18 /* MSN error */
151 #define T4_ERR_TBIT 0x19 /* tag bit not set correctly */
152 #define T4_ERR_MO 0x1A /* MO not 0 for TERMINATE */
154 #define T4_ERR_MSN_GAP 0x1B
155 #define T4_ERR_MSN_RANGE 0x1C
156 #define T4_ERR_IRD_OVERFLOW 0x1D
157 #define T4_ERR_RQE_ADDR_BOUND 0x1E /* RQE addr out of bounds: */
159 #define T4_ERR_INTERNAL_ERR 0x1F /* internal error (opcode */
186 /* macros for flit 0 of the cqe */
188 #define CQE_QPID_S 12
189 #define CQE_QPID_M 0xFFFFF
190 #define CQE_QPID_G(x) ((((x) >> CQE_QPID_S)) & CQE_QPID_M)
191 #define CQE_QPID_V(x) ((x)<<CQE_QPID_S)
193 #define CQE_SWCQE_S 11
194 #define CQE_SWCQE_M 0x1
195 #define CQE_SWCQE_G(x) ((((x) >> CQE_SWCQE_S)) & CQE_SWCQE_M)
196 #define CQE_SWCQE_V(x) ((x)<<CQE_SWCQE_S)
198 #define CQE_STATUS_S 5
199 #define CQE_STATUS_M 0x1F
200 #define CQE_STATUS_G(x) ((((x) >> CQE_STATUS_S)) & CQE_STATUS_M)
201 #define CQE_STATUS_V(x) ((x)<<CQE_STATUS_S)
204 #define CQE_TYPE_M 0x1
205 #define CQE_TYPE_G(x) ((((x) >> CQE_TYPE_S)) & CQE_TYPE_M)
206 #define CQE_TYPE_V(x) ((x)<<CQE_TYPE_S)
208 #define CQE_OPCODE_S 0
209 #define CQE_OPCODE_M 0xF
210 #define CQE_OPCODE_G(x) ((((x) >> CQE_OPCODE_S)) & CQE_OPCODE_M)
211 #define CQE_OPCODE_V(x) ((x)<<CQE_OPCODE_S)
213 #define SW_CQE(x) (CQE_SWCQE_G(be32_to_cpu((x)->header)))
214 #define CQE_QPID(x) (CQE_QPID_G(be32_to_cpu((x)->header)))
215 #define CQE_TYPE(x) (CQE_TYPE_G(be32_to_cpu((x)->header)))
216 #define SQ_TYPE(x) (CQE_TYPE((x)))
217 #define RQ_TYPE(x) (!CQE_TYPE((x)))
218 #define CQE_STATUS(x) (CQE_STATUS_G(be32_to_cpu((x)->header)))
219 #define CQE_OPCODE(x) (CQE_OPCODE_G(be32_to_cpu((x)->header)))
221 #define CQE_SEND_OPCODE(x)( \
222 (CQE_OPCODE_G(be32_to_cpu((x)->header)) == FW_RI_SEND) || \
223 (CQE_OPCODE_G(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_SE) || \
224 (CQE_OPCODE_G(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_INV) || \
225 (CQE_OPCODE_G(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_SE_INV))
227 #define CQE_LEN(x) (be32_to_cpu((x)->len))
229 /* used for RQ completion processing */
230 #define CQE_WRID_STAG(x) (be32_to_cpu((x)->u.rcqe.stag))
231 #define CQE_WRID_MSN(x) (be32_to_cpu((x)->u.rcqe.msn))
233 /* used for SQ completion processing */
234 #define CQE_WRID_SQ_IDX(x) ((x)->u.scqe.cidx)
236 /* generic accessor macros */
237 #define CQE_WRID_HI(x) (be32_to_cpu((x)->u.gen.wrid_hi))
238 #define CQE_WRID_LOW(x) (be32_to_cpu((x)->u.gen.wrid_low))
240 /* macros for flit 3 of the cqe */
241 #define CQE_GENBIT_S 63
242 #define CQE_GENBIT_M 0x1
243 #define CQE_GENBIT_G(x) (((x) >> CQE_GENBIT_S) & CQE_GENBIT_M)
244 #define CQE_GENBIT_V(x) ((x)<<CQE_GENBIT_S)
246 #define CQE_OVFBIT_S 62
247 #define CQE_OVFBIT_M 0x1
248 #define CQE_OVFBIT_G(x) ((((x) >> CQE_OVFBIT_S)) & CQE_OVFBIT_M)
250 #define CQE_IQTYPE_S 60
251 #define CQE_IQTYPE_M 0x3
252 #define CQE_IQTYPE_G(x) ((((x) >> CQE_IQTYPE_S)) & CQE_IQTYPE_M)
254 #define CQE_TS_M 0x0fffffffffffffffULL
255 #define CQE_TS_G(x) ((x) & CQE_TS_M)
257 #define CQE_OVFBIT(x) ((unsigned)CQE_OVFBIT_G(be64_to_cpu((x)->bits_type_ts)))
258 #define CQE_GENBIT(x) ((unsigned)CQE_GENBIT_G(be64_to_cpu((x)->bits_type_ts)))
259 #define CQE_TS(x) (CQE_TS_G(be64_to_cpu((x)->bits_type_ts)))
270 struct timespec host_ts
;
274 static inline pgprot_t
t4_pgprot_wc(pgprot_t prot
)
276 #if defined(__i386__) || defined(__x86_64__) || defined(CONFIG_PPC64)
277 return pgprot_writecombine(prot
);
279 return pgprot_noncached(prot
);
284 T4_SQ_ONCHIP
= (1<<0),
290 DEFINE_DMA_UNMAP_ADDR(mapping
);
291 unsigned long phys_addr
;
292 struct t4_swsqe
*sw_sq
;
293 struct t4_swsqe
*oldest_read
;
294 void __iomem
*bar2_va
;
311 struct timespec host_ts
;
316 union t4_recv_wr
*queue
;
318 DEFINE_DMA_UNMAP_ADDR(mapping
);
319 struct t4_swrqe
*sw_rq
;
320 void __iomem
*bar2_va
;
340 struct c4iw_rdev
*rdev
;
344 static inline int t4_rqes_posted(struct t4_wq
*wq
)
346 return wq
->rq
.in_use
;
349 static inline int t4_rq_empty(struct t4_wq
*wq
)
351 return wq
->rq
.in_use
== 0;
354 static inline int t4_rq_full(struct t4_wq
*wq
)
356 return wq
->rq
.in_use
== (wq
->rq
.size
- 1);
359 static inline u32
t4_rq_avail(struct t4_wq
*wq
)
361 return wq
->rq
.size
- 1 - wq
->rq
.in_use
;
364 static inline void t4_rq_produce(struct t4_wq
*wq
, u8 len16
)
367 if (++wq
->rq
.pidx
== wq
->rq
.size
)
369 wq
->rq
.wq_pidx
+= DIV_ROUND_UP(len16
*16, T4_EQ_ENTRY_SIZE
);
370 if (wq
->rq
.wq_pidx
>= wq
->rq
.size
* T4_RQ_NUM_SLOTS
)
371 wq
->rq
.wq_pidx
%= wq
->rq
.size
* T4_RQ_NUM_SLOTS
;
374 static inline void t4_rq_consume(struct t4_wq
*wq
)
378 if (++wq
->rq
.cidx
== wq
->rq
.size
)
382 static inline u16
t4_rq_host_wq_pidx(struct t4_wq
*wq
)
384 return wq
->rq
.queue
[wq
->rq
.size
].status
.host_wq_pidx
;
387 static inline u16
t4_rq_wq_size(struct t4_wq
*wq
)
389 return wq
->rq
.size
* T4_RQ_NUM_SLOTS
;
392 static inline int t4_sq_onchip(struct t4_sq
*sq
)
394 return sq
->flags
& T4_SQ_ONCHIP
;
397 static inline int t4_sq_empty(struct t4_wq
*wq
)
399 return wq
->sq
.in_use
== 0;
402 static inline int t4_sq_full(struct t4_wq
*wq
)
404 return wq
->sq
.in_use
== (wq
->sq
.size
- 1);
407 static inline u32
t4_sq_avail(struct t4_wq
*wq
)
409 return wq
->sq
.size
- 1 - wq
->sq
.in_use
;
412 static inline void t4_sq_produce(struct t4_wq
*wq
, u8 len16
)
415 if (++wq
->sq
.pidx
== wq
->sq
.size
)
417 wq
->sq
.wq_pidx
+= DIV_ROUND_UP(len16
*16, T4_EQ_ENTRY_SIZE
);
418 if (wq
->sq
.wq_pidx
>= wq
->sq
.size
* T4_SQ_NUM_SLOTS
)
419 wq
->sq
.wq_pidx
%= wq
->sq
.size
* T4_SQ_NUM_SLOTS
;
422 static inline void t4_sq_consume(struct t4_wq
*wq
)
424 BUG_ON(wq
->sq
.in_use
< 1);
425 if (wq
->sq
.cidx
== wq
->sq
.flush_cidx
)
426 wq
->sq
.flush_cidx
= -1;
428 if (++wq
->sq
.cidx
== wq
->sq
.size
)
432 static inline u16
t4_sq_host_wq_pidx(struct t4_wq
*wq
)
434 return wq
->sq
.queue
[wq
->sq
.size
].status
.host_wq_pidx
;
437 static inline u16
t4_sq_wq_size(struct t4_wq
*wq
)
439 return wq
->sq
.size
* T4_SQ_NUM_SLOTS
;
442 /* This function copies 64 byte coalesced work request to memory
443 * mapped BAR2 space. For coalesced WRs, the SGE fetches data
444 * from the FIFO instead of from Host.
446 static inline void pio_copy(u64 __iomem
*dst
, u64
*src
)
458 static inline void t4_ring_sq_db(struct t4_wq
*wq
, u16 inc
, union t4_wr
*wqe
)
461 /* Flush host queue memory writes. */
463 if (wq
->sq
.bar2_va
) {
464 if (inc
== 1 && wq
->sq
.bar2_qid
== 0 && wqe
) {
465 PDBG("%s: WC wq->sq.pidx = %d\n",
466 __func__
, wq
->sq
.pidx
);
467 pio_copy((u64 __iomem
*)
468 (wq
->sq
.bar2_va
+ SGE_UDB_WCDOORBELL
),
471 PDBG("%s: DB wq->sq.pidx = %d\n",
472 __func__
, wq
->sq
.pidx
);
473 writel(PIDX_T5_V(inc
) | QID_V(wq
->sq
.bar2_qid
),
474 wq
->sq
.bar2_va
+ SGE_UDB_KDOORBELL
);
477 /* Flush user doorbell area writes. */
481 writel(QID_V(wq
->sq
.qid
) | PIDX_V(inc
), wq
->db
);
484 static inline void t4_ring_rq_db(struct t4_wq
*wq
, u16 inc
,
485 union t4_recv_wr
*wqe
)
488 /* Flush host queue memory writes. */
490 if (wq
->rq
.bar2_va
) {
491 if (inc
== 1 && wq
->rq
.bar2_qid
== 0 && wqe
) {
492 PDBG("%s: WC wq->rq.pidx = %d\n",
493 __func__
, wq
->rq
.pidx
);
494 pio_copy((u64 __iomem
*)
495 (wq
->rq
.bar2_va
+ SGE_UDB_WCDOORBELL
),
498 PDBG("%s: DB wq->rq.pidx = %d\n",
499 __func__
, wq
->rq
.pidx
);
500 writel(PIDX_T5_V(inc
) | QID_V(wq
->rq
.bar2_qid
),
501 wq
->rq
.bar2_va
+ SGE_UDB_KDOORBELL
);
504 /* Flush user doorbell area writes. */
508 writel(QID_V(wq
->rq
.qid
) | PIDX_V(inc
), wq
->db
);
511 static inline int t4_wq_in_error(struct t4_wq
*wq
)
513 return wq
->rq
.queue
[wq
->rq
.size
].status
.qp_err
;
516 static inline void t4_set_wq_in_error(struct t4_wq
*wq
)
518 wq
->rq
.queue
[wq
->rq
.size
].status
.qp_err
= 1;
521 static inline void t4_disable_wq_db(struct t4_wq
*wq
)
523 wq
->rq
.queue
[wq
->rq
.size
].status
.db_off
= 1;
526 static inline void t4_enable_wq_db(struct t4_wq
*wq
)
528 wq
->rq
.queue
[wq
->rq
.size
].status
.db_off
= 0;
531 static inline int t4_wq_db_enabled(struct t4_wq
*wq
)
533 return !wq
->rq
.queue
[wq
->rq
.size
].status
.db_off
;
541 struct t4_cqe
*queue
;
543 DEFINE_DMA_UNMAP_ADDR(mapping
);
544 struct t4_cqe
*sw_queue
;
546 void __iomem
*bar2_va
;
549 struct c4iw_rdev
*rdev
;
555 u16 size
; /* including status page */
566 static inline void write_gts(struct t4_cq
*cq
, u32 val
)
569 writel(val
| INGRESSQID_V(cq
->bar2_qid
),
570 cq
->bar2_va
+ SGE_UDB_GTS
);
572 writel(val
| INGRESSQID_V(cq
->cqid
), cq
->gts
);
575 static inline int t4_clear_cq_armed(struct t4_cq
*cq
)
577 return test_and_clear_bit(CQ_ARMED
, &cq
->flags
);
580 static inline int t4_arm_cq(struct t4_cq
*cq
, int se
)
584 set_bit(CQ_ARMED
, &cq
->flags
);
585 while (cq
->cidx_inc
> CIDXINC_M
) {
586 val
= SEINTARM_V(0) | CIDXINC_V(CIDXINC_M
) | TIMERREG_V(7);
588 cq
->cidx_inc
-= CIDXINC_M
;
590 val
= SEINTARM_V(se
) | CIDXINC_V(cq
->cidx_inc
) | TIMERREG_V(6);
596 static inline void t4_swcq_produce(struct t4_cq
*cq
)
599 if (cq
->sw_in_use
== cq
->size
) {
600 PDBG("%s cxgb4 sw cq overflow cqid %u\n", __func__
, cq
->cqid
);
604 if (++cq
->sw_pidx
== cq
->size
)
608 static inline void t4_swcq_consume(struct t4_cq
*cq
)
610 BUG_ON(cq
->sw_in_use
< 1);
612 if (++cq
->sw_cidx
== cq
->size
)
616 static inline void t4_hwcq_consume(struct t4_cq
*cq
)
618 cq
->bits_type_ts
= cq
->queue
[cq
->cidx
].bits_type_ts
;
619 if (++cq
->cidx_inc
== (cq
->size
>> 4) || cq
->cidx_inc
== CIDXINC_M
) {
622 val
= SEINTARM_V(0) | CIDXINC_V(cq
->cidx_inc
) | TIMERREG_V(7);
626 if (++cq
->cidx
== cq
->size
) {
632 static inline int t4_valid_cqe(struct t4_cq
*cq
, struct t4_cqe
*cqe
)
634 return (CQE_GENBIT(cqe
) == cq
->gen
);
637 static inline int t4_next_hw_cqe(struct t4_cq
*cq
, struct t4_cqe
**cqe
)
643 prev_cidx
= cq
->size
- 1;
645 prev_cidx
= cq
->cidx
- 1;
647 if (cq
->queue
[prev_cidx
].bits_type_ts
!= cq
->bits_type_ts
) {
650 printk(KERN_ERR MOD
"cq overflow cqid %u\n", cq
->cqid
);
652 } else if (t4_valid_cqe(cq
, &cq
->queue
[cq
->cidx
])) {
654 /* Ensure CQE is flushed to memory */
656 *cqe
= &cq
->queue
[cq
->cidx
];
663 static inline struct t4_cqe
*t4_next_sw_cqe(struct t4_cq
*cq
)
665 if (cq
->sw_in_use
== cq
->size
) {
666 PDBG("%s cxgb4 sw cq overflow cqid %u\n", __func__
, cq
->cqid
);
672 return &cq
->sw_queue
[cq
->sw_cidx
];
676 static inline int t4_next_cqe(struct t4_cq
*cq
, struct t4_cqe
**cqe
)
682 else if (cq
->sw_in_use
)
683 *cqe
= &cq
->sw_queue
[cq
->sw_cidx
];
685 ret
= t4_next_hw_cqe(cq
, cqe
);
689 static inline int t4_cq_in_error(struct t4_cq
*cq
)
691 return ((struct t4_status_page
*)&cq
->queue
[cq
->size
])->qp_err
;
694 static inline void t4_set_cq_in_error(struct t4_cq
*cq
)
696 ((struct t4_status_page
*)&cq
->queue
[cq
->size
])->qp_err
= 1;
700 struct t4_dev_status_page
{