2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
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:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/errno.h>
38 #include <linux/pci.h>
39 #include <linux/interrupt.h>
40 #include <linux/gfp.h>
42 #include "mthca_dev.h"
43 #include "mthca_config_reg.h"
44 #include "mthca_cmd.h"
45 #include "mthca_profile.h"
46 #include "mthca_memfree.h"
47 #include "mthca_wqe.h"
49 MODULE_AUTHOR("Roland Dreier");
50 MODULE_DESCRIPTION("Mellanox InfiniBand HCA low-level driver");
51 MODULE_LICENSE("Dual BSD/GPL");
52 MODULE_VERSION(DRV_VERSION
);
54 #ifdef CONFIG_INFINIBAND_MTHCA_DEBUG
56 int mthca_debug_level
= 0;
57 module_param_named(debug_level
, mthca_debug_level
, int, 0644);
58 MODULE_PARM_DESC(debug_level
, "Enable debug tracing if > 0");
60 #endif /* CONFIG_INFINIBAND_MTHCA_DEBUG */
65 module_param(msi_x
, int, 0444);
66 MODULE_PARM_DESC(msi_x
, "attempt to use MSI-X if nonzero");
68 #else /* CONFIG_PCI_MSI */
72 #endif /* CONFIG_PCI_MSI */
74 static int tune_pci
= 0;
75 module_param(tune_pci
, int, 0444);
76 MODULE_PARM_DESC(tune_pci
, "increase PCI burst from the default set by BIOS if nonzero");
78 DEFINE_MUTEX(mthca_device_mutex
);
80 #define MTHCA_DEFAULT_NUM_QP (1 << 16)
81 #define MTHCA_DEFAULT_RDB_PER_QP (1 << 2)
82 #define MTHCA_DEFAULT_NUM_CQ (1 << 16)
83 #define MTHCA_DEFAULT_NUM_MCG (1 << 13)
84 #define MTHCA_DEFAULT_NUM_MPT (1 << 17)
85 #define MTHCA_DEFAULT_NUM_MTT (1 << 20)
86 #define MTHCA_DEFAULT_NUM_UDAV (1 << 15)
87 #define MTHCA_DEFAULT_NUM_RESERVED_MTTS (1 << 18)
88 #define MTHCA_DEFAULT_NUM_UARC_SIZE (1 << 18)
90 static struct mthca_profile hca_profile
= {
91 .num_qp
= MTHCA_DEFAULT_NUM_QP
,
92 .rdb_per_qp
= MTHCA_DEFAULT_RDB_PER_QP
,
93 .num_cq
= MTHCA_DEFAULT_NUM_CQ
,
94 .num_mcg
= MTHCA_DEFAULT_NUM_MCG
,
95 .num_mpt
= MTHCA_DEFAULT_NUM_MPT
,
96 .num_mtt
= MTHCA_DEFAULT_NUM_MTT
,
97 .num_udav
= MTHCA_DEFAULT_NUM_UDAV
, /* Tavor only */
98 .fmr_reserved_mtts
= MTHCA_DEFAULT_NUM_RESERVED_MTTS
, /* Tavor only */
99 .uarc_size
= MTHCA_DEFAULT_NUM_UARC_SIZE
, /* Arbel only */
102 module_param_named(num_qp
, hca_profile
.num_qp
, int, 0444);
103 MODULE_PARM_DESC(num_qp
, "maximum number of QPs per HCA");
105 module_param_named(rdb_per_qp
, hca_profile
.rdb_per_qp
, int, 0444);
106 MODULE_PARM_DESC(rdb_per_qp
, "number of RDB buffers per QP");
108 module_param_named(num_cq
, hca_profile
.num_cq
, int, 0444);
109 MODULE_PARM_DESC(num_cq
, "maximum number of CQs per HCA");
111 module_param_named(num_mcg
, hca_profile
.num_mcg
, int, 0444);
112 MODULE_PARM_DESC(num_mcg
, "maximum number of multicast groups per HCA");
114 module_param_named(num_mpt
, hca_profile
.num_mpt
, int, 0444);
115 MODULE_PARM_DESC(num_mpt
,
116 "maximum number of memory protection table entries per HCA");
118 module_param_named(num_mtt
, hca_profile
.num_mtt
, int, 0444);
119 MODULE_PARM_DESC(num_mtt
,
120 "maximum number of memory translation table segments per HCA");
122 module_param_named(num_udav
, hca_profile
.num_udav
, int, 0444);
123 MODULE_PARM_DESC(num_udav
, "maximum number of UD address vectors per HCA");
125 module_param_named(fmr_reserved_mtts
, hca_profile
.fmr_reserved_mtts
, int, 0444);
126 MODULE_PARM_DESC(fmr_reserved_mtts
,
127 "number of memory translation table segments reserved for FMR");
129 static int log_mtts_per_seg
= ilog2(MTHCA_MTT_SEG_SIZE
/ 8);
130 module_param_named(log_mtts_per_seg
, log_mtts_per_seg
, int, 0444);
131 MODULE_PARM_DESC(log_mtts_per_seg
, "Log2 number of MTT entries per segment (1-5)");
133 static char mthca_version
[] __devinitdata
=
134 DRV_NAME
": Mellanox InfiniBand HCA driver v"
135 DRV_VERSION
" (" DRV_RELDATE
")\n";
137 static int mthca_tune_pci(struct mthca_dev
*mdev
)
142 /* First try to max out Read Byte Count */
143 if (pci_find_capability(mdev
->pdev
, PCI_CAP_ID_PCIX
)) {
144 if (pcix_set_mmrbc(mdev
->pdev
, pcix_get_max_mmrbc(mdev
->pdev
))) {
145 mthca_err(mdev
, "Couldn't set PCI-X max read count, "
149 } else if (!(mdev
->mthca_flags
& MTHCA_FLAG_PCIE
))
150 mthca_info(mdev
, "No PCI-X capability, not setting RBC.\n");
152 if (pci_find_capability(mdev
->pdev
, PCI_CAP_ID_EXP
)) {
153 if (pcie_set_readrq(mdev
->pdev
, 4096)) {
154 mthca_err(mdev
, "Couldn't write PCI Express read request, "
158 } else if (mdev
->mthca_flags
& MTHCA_FLAG_PCIE
)
159 mthca_info(mdev
, "No PCI Express capability, "
160 "not setting Max Read Request Size.\n");
165 static int mthca_dev_lim(struct mthca_dev
*mdev
, struct mthca_dev_lim
*dev_lim
)
170 mdev
->limits
.mtt_seg_size
= (1 << log_mtts_per_seg
) * 8;
171 err
= mthca_QUERY_DEV_LIM(mdev
, dev_lim
, &status
);
173 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
177 mthca_err(mdev
, "QUERY_DEV_LIM returned status 0x%02x, "
178 "aborting.\n", status
);
181 if (dev_lim
->min_page_sz
> PAGE_SIZE
) {
182 mthca_err(mdev
, "HCA minimum page size of %d bigger than "
183 "kernel PAGE_SIZE of %ld, aborting.\n",
184 dev_lim
->min_page_sz
, PAGE_SIZE
);
187 if (dev_lim
->num_ports
> MTHCA_MAX_PORTS
) {
188 mthca_err(mdev
, "HCA has %d ports, but we only support %d, "
190 dev_lim
->num_ports
, MTHCA_MAX_PORTS
);
194 if (dev_lim
->uar_size
> pci_resource_len(mdev
->pdev
, 2)) {
195 mthca_err(mdev
, "HCA reported UAR size of 0x%x bigger than "
196 "PCI resource 2 size of 0x%llx, aborting.\n",
198 (unsigned long long)pci_resource_len(mdev
->pdev
, 2));
202 mdev
->limits
.num_ports
= dev_lim
->num_ports
;
203 mdev
->limits
.vl_cap
= dev_lim
->max_vl
;
204 mdev
->limits
.mtu_cap
= dev_lim
->max_mtu
;
205 mdev
->limits
.gid_table_len
= dev_lim
->max_gids
;
206 mdev
->limits
.pkey_table_len
= dev_lim
->max_pkeys
;
207 mdev
->limits
.local_ca_ack_delay
= dev_lim
->local_ca_ack_delay
;
209 * Need to allow for worst case send WQE overhead and check
210 * whether max_desc_sz imposes a lower limit than max_sg; UD
211 * send has the biggest overhead.
213 mdev
->limits
.max_sg
= min_t(int, dev_lim
->max_sg
,
214 (dev_lim
->max_desc_sz
-
215 sizeof (struct mthca_next_seg
) -
216 (mthca_is_memfree(mdev
) ?
217 sizeof (struct mthca_arbel_ud_seg
) :
218 sizeof (struct mthca_tavor_ud_seg
))) /
219 sizeof (struct mthca_data_seg
));
220 mdev
->limits
.max_wqes
= dev_lim
->max_qp_sz
;
221 mdev
->limits
.max_qp_init_rdma
= dev_lim
->max_requester_per_qp
;
222 mdev
->limits
.reserved_qps
= dev_lim
->reserved_qps
;
223 mdev
->limits
.max_srq_wqes
= dev_lim
->max_srq_sz
;
224 mdev
->limits
.reserved_srqs
= dev_lim
->reserved_srqs
;
225 mdev
->limits
.reserved_eecs
= dev_lim
->reserved_eecs
;
226 mdev
->limits
.max_desc_sz
= dev_lim
->max_desc_sz
;
227 mdev
->limits
.max_srq_sge
= mthca_max_srq_sge(mdev
);
229 * Subtract 1 from the limit because we need to allocate a
230 * spare CQE so the HCA HW can tell the difference between an
231 * empty CQ and a full CQ.
233 mdev
->limits
.max_cqes
= dev_lim
->max_cq_sz
- 1;
234 mdev
->limits
.reserved_cqs
= dev_lim
->reserved_cqs
;
235 mdev
->limits
.reserved_eqs
= dev_lim
->reserved_eqs
;
236 mdev
->limits
.reserved_mtts
= dev_lim
->reserved_mtts
;
237 mdev
->limits
.reserved_mrws
= dev_lim
->reserved_mrws
;
238 mdev
->limits
.reserved_uars
= dev_lim
->reserved_uars
;
239 mdev
->limits
.reserved_pds
= dev_lim
->reserved_pds
;
240 mdev
->limits
.port_width_cap
= dev_lim
->max_port_width
;
241 mdev
->limits
.page_size_cap
= ~(u32
) (dev_lim
->min_page_sz
- 1);
242 mdev
->limits
.flags
= dev_lim
->flags
;
244 * For old FW that doesn't return static rate support, use a
245 * value of 0x3 (only static rate values of 0 or 1 are handled),
246 * except on Sinai, where even old FW can handle static rate
249 if (dev_lim
->stat_rate_support
)
250 mdev
->limits
.stat_rate_support
= dev_lim
->stat_rate_support
;
251 else if (mdev
->mthca_flags
& MTHCA_FLAG_SINAI_OPT
)
252 mdev
->limits
.stat_rate_support
= 0xf;
254 mdev
->limits
.stat_rate_support
= 0x3;
256 /* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
257 May be doable since hardware supports it for SRQ.
259 IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
261 IB_DEVICE_SRQ_RESIZE is supported by hardware but SRQ is not
262 supported by driver. */
263 mdev
->device_cap_flags
= IB_DEVICE_CHANGE_PHY_PORT
|
264 IB_DEVICE_PORT_ACTIVE_EVENT
|
265 IB_DEVICE_SYS_IMAGE_GUID
|
266 IB_DEVICE_RC_RNR_NAK_GEN
;
268 if (dev_lim
->flags
& DEV_LIM_FLAG_BAD_PKEY_CNTR
)
269 mdev
->device_cap_flags
|= IB_DEVICE_BAD_PKEY_CNTR
;
271 if (dev_lim
->flags
& DEV_LIM_FLAG_BAD_QKEY_CNTR
)
272 mdev
->device_cap_flags
|= IB_DEVICE_BAD_QKEY_CNTR
;
274 if (dev_lim
->flags
& DEV_LIM_FLAG_RAW_MULTI
)
275 mdev
->device_cap_flags
|= IB_DEVICE_RAW_MULTI
;
277 if (dev_lim
->flags
& DEV_LIM_FLAG_AUTO_PATH_MIG
)
278 mdev
->device_cap_flags
|= IB_DEVICE_AUTO_PATH_MIG
;
280 if (dev_lim
->flags
& DEV_LIM_FLAG_UD_AV_PORT_ENFORCE
)
281 mdev
->device_cap_flags
|= IB_DEVICE_UD_AV_PORT_ENFORCE
;
283 if (dev_lim
->flags
& DEV_LIM_FLAG_SRQ
)
284 mdev
->mthca_flags
|= MTHCA_FLAG_SRQ
;
286 if (mthca_is_memfree(mdev
))
287 if (dev_lim
->flags
& DEV_LIM_FLAG_IPOIB_CSUM
)
288 mdev
->device_cap_flags
|= IB_DEVICE_UD_IP_CSUM
;
293 static int mthca_init_tavor(struct mthca_dev
*mdev
)
298 struct mthca_dev_lim dev_lim
;
299 struct mthca_profile profile
;
300 struct mthca_init_hca_param init_hca
;
302 err
= mthca_SYS_EN(mdev
, &status
);
304 mthca_err(mdev
, "SYS_EN command failed, aborting.\n");
308 mthca_err(mdev
, "SYS_EN returned status 0x%02x, "
309 "aborting.\n", status
);
313 err
= mthca_QUERY_FW(mdev
, &status
);
315 mthca_err(mdev
, "QUERY_FW command failed, aborting.\n");
319 mthca_err(mdev
, "QUERY_FW returned status 0x%02x, "
320 "aborting.\n", status
);
324 err
= mthca_QUERY_DDR(mdev
, &status
);
326 mthca_err(mdev
, "QUERY_DDR command failed, aborting.\n");
330 mthca_err(mdev
, "QUERY_DDR returned status 0x%02x, "
331 "aborting.\n", status
);
336 err
= mthca_dev_lim(mdev
, &dev_lim
);
338 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
342 profile
= hca_profile
;
343 profile
.num_uar
= dev_lim
.uar_size
/ PAGE_SIZE
;
344 profile
.uarc_size
= 0;
345 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
346 profile
.num_srq
= dev_lim
.max_srqs
;
348 size
= mthca_make_profile(mdev
, &profile
, &dev_lim
, &init_hca
);
354 err
= mthca_INIT_HCA(mdev
, &init_hca
, &status
);
356 mthca_err(mdev
, "INIT_HCA command failed, aborting.\n");
360 mthca_err(mdev
, "INIT_HCA returned status 0x%02x, "
361 "aborting.\n", status
);
369 mthca_SYS_DIS(mdev
, &status
);
374 static int mthca_load_fw(struct mthca_dev
*mdev
)
379 /* FIXME: use HCA-attached memory for FW if present */
381 mdev
->fw
.arbel
.fw_icm
=
382 mthca_alloc_icm(mdev
, mdev
->fw
.arbel
.fw_pages
,
383 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
384 if (!mdev
->fw
.arbel
.fw_icm
) {
385 mthca_err(mdev
, "Couldn't allocate FW area, aborting.\n");
389 err
= mthca_MAP_FA(mdev
, mdev
->fw
.arbel
.fw_icm
, &status
);
391 mthca_err(mdev
, "MAP_FA command failed, aborting.\n");
395 mthca_err(mdev
, "MAP_FA returned status 0x%02x, aborting.\n", status
);
399 err
= mthca_RUN_FW(mdev
, &status
);
401 mthca_err(mdev
, "RUN_FW command failed, aborting.\n");
405 mthca_err(mdev
, "RUN_FW returned status 0x%02x, aborting.\n", status
);
413 mthca_UNMAP_FA(mdev
, &status
);
416 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
420 static int mthca_init_icm(struct mthca_dev
*mdev
,
421 struct mthca_dev_lim
*dev_lim
,
422 struct mthca_init_hca_param
*init_hca
,
429 err
= mthca_SET_ICM_SIZE(mdev
, icm_size
, &aux_pages
, &status
);
431 mthca_err(mdev
, "SET_ICM_SIZE command failed, aborting.\n");
435 mthca_err(mdev
, "SET_ICM_SIZE returned status 0x%02x, "
436 "aborting.\n", status
);
440 mthca_dbg(mdev
, "%lld KB of HCA context requires %lld KB aux memory.\n",
441 (unsigned long long) icm_size
>> 10,
442 (unsigned long long) aux_pages
<< 2);
444 mdev
->fw
.arbel
.aux_icm
= mthca_alloc_icm(mdev
, aux_pages
,
445 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
446 if (!mdev
->fw
.arbel
.aux_icm
) {
447 mthca_err(mdev
, "Couldn't allocate aux memory, aborting.\n");
451 err
= mthca_MAP_ICM_AUX(mdev
, mdev
->fw
.arbel
.aux_icm
, &status
);
453 mthca_err(mdev
, "MAP_ICM_AUX command failed, aborting.\n");
457 mthca_err(mdev
, "MAP_ICM_AUX returned status 0x%02x, aborting.\n", status
);
462 err
= mthca_map_eq_icm(mdev
, init_hca
->eqc_base
);
464 mthca_err(mdev
, "Failed to map EQ context memory, aborting.\n");
468 /* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
469 mdev
->limits
.reserved_mtts
= ALIGN(mdev
->limits
.reserved_mtts
* mdev
->limits
.mtt_seg_size
,
470 dma_get_cache_alignment()) / mdev
->limits
.mtt_seg_size
;
472 mdev
->mr_table
.mtt_table
= mthca_alloc_icm_table(mdev
, init_hca
->mtt_base
,
473 mdev
->limits
.mtt_seg_size
,
474 mdev
->limits
.num_mtt_segs
,
475 mdev
->limits
.reserved_mtts
,
477 if (!mdev
->mr_table
.mtt_table
) {
478 mthca_err(mdev
, "Failed to map MTT context memory, aborting.\n");
483 mdev
->mr_table
.mpt_table
= mthca_alloc_icm_table(mdev
, init_hca
->mpt_base
,
484 dev_lim
->mpt_entry_sz
,
485 mdev
->limits
.num_mpts
,
486 mdev
->limits
.reserved_mrws
,
488 if (!mdev
->mr_table
.mpt_table
) {
489 mthca_err(mdev
, "Failed to map MPT context memory, aborting.\n");
494 mdev
->qp_table
.qp_table
= mthca_alloc_icm_table(mdev
, init_hca
->qpc_base
,
495 dev_lim
->qpc_entry_sz
,
496 mdev
->limits
.num_qps
,
497 mdev
->limits
.reserved_qps
,
499 if (!mdev
->qp_table
.qp_table
) {
500 mthca_err(mdev
, "Failed to map QP context memory, aborting.\n");
505 mdev
->qp_table
.eqp_table
= mthca_alloc_icm_table(mdev
, init_hca
->eqpc_base
,
506 dev_lim
->eqpc_entry_sz
,
507 mdev
->limits
.num_qps
,
508 mdev
->limits
.reserved_qps
,
510 if (!mdev
->qp_table
.eqp_table
) {
511 mthca_err(mdev
, "Failed to map EQP context memory, aborting.\n");
516 mdev
->qp_table
.rdb_table
= mthca_alloc_icm_table(mdev
, init_hca
->rdb_base
,
517 MTHCA_RDB_ENTRY_SIZE
,
518 mdev
->limits
.num_qps
<<
519 mdev
->qp_table
.rdb_shift
, 0,
521 if (!mdev
->qp_table
.rdb_table
) {
522 mthca_err(mdev
, "Failed to map RDB context memory, aborting\n");
527 mdev
->cq_table
.table
= mthca_alloc_icm_table(mdev
, init_hca
->cqc_base
,
528 dev_lim
->cqc_entry_sz
,
529 mdev
->limits
.num_cqs
,
530 mdev
->limits
.reserved_cqs
,
532 if (!mdev
->cq_table
.table
) {
533 mthca_err(mdev
, "Failed to map CQ context memory, aborting.\n");
538 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
539 mdev
->srq_table
.table
=
540 mthca_alloc_icm_table(mdev
, init_hca
->srqc_base
,
541 dev_lim
->srq_entry_sz
,
542 mdev
->limits
.num_srqs
,
543 mdev
->limits
.reserved_srqs
,
545 if (!mdev
->srq_table
.table
) {
546 mthca_err(mdev
, "Failed to map SRQ context memory, "
554 * It's not strictly required, but for simplicity just map the
555 * whole multicast group table now. The table isn't very big
556 * and it's a lot easier than trying to track ref counts.
558 mdev
->mcg_table
.table
= mthca_alloc_icm_table(mdev
, init_hca
->mc_base
,
559 MTHCA_MGM_ENTRY_SIZE
,
560 mdev
->limits
.num_mgms
+
561 mdev
->limits
.num_amgms
,
562 mdev
->limits
.num_mgms
+
563 mdev
->limits
.num_amgms
,
565 if (!mdev
->mcg_table
.table
) {
566 mthca_err(mdev
, "Failed to map MCG context memory, aborting.\n");
574 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
575 mthca_free_icm_table(mdev
, mdev
->srq_table
.table
);
578 mthca_free_icm_table(mdev
, mdev
->cq_table
.table
);
581 mthca_free_icm_table(mdev
, mdev
->qp_table
.rdb_table
);
584 mthca_free_icm_table(mdev
, mdev
->qp_table
.eqp_table
);
587 mthca_free_icm_table(mdev
, mdev
->qp_table
.qp_table
);
590 mthca_free_icm_table(mdev
, mdev
->mr_table
.mpt_table
);
593 mthca_free_icm_table(mdev
, mdev
->mr_table
.mtt_table
);
596 mthca_unmap_eq_icm(mdev
);
599 mthca_UNMAP_ICM_AUX(mdev
, &status
);
602 mthca_free_icm(mdev
, mdev
->fw
.arbel
.aux_icm
, 0);
607 static void mthca_free_icms(struct mthca_dev
*mdev
)
611 mthca_free_icm_table(mdev
, mdev
->mcg_table
.table
);
612 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
613 mthca_free_icm_table(mdev
, mdev
->srq_table
.table
);
614 mthca_free_icm_table(mdev
, mdev
->cq_table
.table
);
615 mthca_free_icm_table(mdev
, mdev
->qp_table
.rdb_table
);
616 mthca_free_icm_table(mdev
, mdev
->qp_table
.eqp_table
);
617 mthca_free_icm_table(mdev
, mdev
->qp_table
.qp_table
);
618 mthca_free_icm_table(mdev
, mdev
->mr_table
.mpt_table
);
619 mthca_free_icm_table(mdev
, mdev
->mr_table
.mtt_table
);
620 mthca_unmap_eq_icm(mdev
);
622 mthca_UNMAP_ICM_AUX(mdev
, &status
);
623 mthca_free_icm(mdev
, mdev
->fw
.arbel
.aux_icm
, 0);
626 static int mthca_init_arbel(struct mthca_dev
*mdev
)
628 struct mthca_dev_lim dev_lim
;
629 struct mthca_profile profile
;
630 struct mthca_init_hca_param init_hca
;
635 err
= mthca_QUERY_FW(mdev
, &status
);
637 mthca_err(mdev
, "QUERY_FW command failed, aborting.\n");
641 mthca_err(mdev
, "QUERY_FW returned status 0x%02x, "
642 "aborting.\n", status
);
646 err
= mthca_ENABLE_LAM(mdev
, &status
);
648 mthca_err(mdev
, "ENABLE_LAM command failed, aborting.\n");
651 if (status
== MTHCA_CMD_STAT_LAM_NOT_PRE
) {
652 mthca_dbg(mdev
, "No HCA-attached memory (running in MemFree mode)\n");
653 mdev
->mthca_flags
|= MTHCA_FLAG_NO_LAM
;
655 mthca_err(mdev
, "ENABLE_LAM returned status 0x%02x, "
656 "aborting.\n", status
);
660 err
= mthca_load_fw(mdev
);
662 mthca_err(mdev
, "Failed to start FW, aborting.\n");
666 err
= mthca_dev_lim(mdev
, &dev_lim
);
668 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
672 profile
= hca_profile
;
673 profile
.num_uar
= dev_lim
.uar_size
/ PAGE_SIZE
;
674 profile
.num_udav
= 0;
675 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
676 profile
.num_srq
= dev_lim
.max_srqs
;
678 icm_size
= mthca_make_profile(mdev
, &profile
, &dev_lim
, &init_hca
);
684 err
= mthca_init_icm(mdev
, &dev_lim
, &init_hca
, icm_size
);
688 err
= mthca_INIT_HCA(mdev
, &init_hca
, &status
);
690 mthca_err(mdev
, "INIT_HCA command failed, aborting.\n");
694 mthca_err(mdev
, "INIT_HCA returned status 0x%02x, "
695 "aborting.\n", status
);
703 mthca_free_icms(mdev
);
706 mthca_UNMAP_FA(mdev
, &status
);
707 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
710 if (!(mdev
->mthca_flags
& MTHCA_FLAG_NO_LAM
))
711 mthca_DISABLE_LAM(mdev
, &status
);
716 static void mthca_close_hca(struct mthca_dev
*mdev
)
720 mthca_CLOSE_HCA(mdev
, 0, &status
);
722 if (mthca_is_memfree(mdev
)) {
723 mthca_free_icms(mdev
);
725 mthca_UNMAP_FA(mdev
, &status
);
726 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
728 if (!(mdev
->mthca_flags
& MTHCA_FLAG_NO_LAM
))
729 mthca_DISABLE_LAM(mdev
, &status
);
731 mthca_SYS_DIS(mdev
, &status
);
734 static int mthca_init_hca(struct mthca_dev
*mdev
)
738 struct mthca_adapter adapter
;
740 if (mthca_is_memfree(mdev
))
741 err
= mthca_init_arbel(mdev
);
743 err
= mthca_init_tavor(mdev
);
748 err
= mthca_QUERY_ADAPTER(mdev
, &adapter
, &status
);
750 mthca_err(mdev
, "QUERY_ADAPTER command failed, aborting.\n");
754 mthca_err(mdev
, "QUERY_ADAPTER returned status 0x%02x, "
755 "aborting.\n", status
);
760 mdev
->eq_table
.inta_pin
= adapter
.inta_pin
;
761 if (!mthca_is_memfree(mdev
))
762 mdev
->rev_id
= adapter
.revision_id
;
763 memcpy(mdev
->board_id
, adapter
.board_id
, sizeof mdev
->board_id
);
768 mthca_close_hca(mdev
);
772 static int mthca_setup_hca(struct mthca_dev
*dev
)
777 MTHCA_INIT_DOORBELL_LOCK(&dev
->doorbell_lock
);
779 err
= mthca_init_uar_table(dev
);
781 mthca_err(dev
, "Failed to initialize "
782 "user access region table, aborting.\n");
786 err
= mthca_uar_alloc(dev
, &dev
->driver_uar
);
788 mthca_err(dev
, "Failed to allocate driver access region, "
790 goto err_uar_table_free
;
793 dev
->kar
= ioremap((phys_addr_t
) dev
->driver_uar
.pfn
<< PAGE_SHIFT
, PAGE_SIZE
);
795 mthca_err(dev
, "Couldn't map kernel access region, "
801 err
= mthca_init_pd_table(dev
);
803 mthca_err(dev
, "Failed to initialize "
804 "protection domain table, aborting.\n");
808 err
= mthca_init_mr_table(dev
);
810 mthca_err(dev
, "Failed to initialize "
811 "memory region table, aborting.\n");
812 goto err_pd_table_free
;
815 err
= mthca_pd_alloc(dev
, 1, &dev
->driver_pd
);
817 mthca_err(dev
, "Failed to create driver PD, "
819 goto err_mr_table_free
;
822 err
= mthca_init_eq_table(dev
);
824 mthca_err(dev
, "Failed to initialize "
825 "event queue table, aborting.\n");
829 err
= mthca_cmd_use_events(dev
);
831 mthca_err(dev
, "Failed to switch to event-driven "
832 "firmware commands, aborting.\n");
833 goto err_eq_table_free
;
836 err
= mthca_NOP(dev
, &status
);
838 if (dev
->mthca_flags
& MTHCA_FLAG_MSI_X
) {
839 mthca_warn(dev
, "NOP command failed to generate interrupt "
841 dev
->eq_table
.eq
[MTHCA_EQ_CMD
].msi_x_vector
);
842 mthca_warn(dev
, "Trying again with MSI-X disabled.\n");
844 mthca_err(dev
, "NOP command failed to generate interrupt "
845 "(IRQ %d), aborting.\n",
847 mthca_err(dev
, "BIOS or ACPI interrupt routing problem?\n");
853 mthca_dbg(dev
, "NOP command IRQ test passed\n");
855 err
= mthca_init_cq_table(dev
);
857 mthca_err(dev
, "Failed to initialize "
858 "completion queue table, aborting.\n");
862 err
= mthca_init_srq_table(dev
);
864 mthca_err(dev
, "Failed to initialize "
865 "shared receive queue table, aborting.\n");
866 goto err_cq_table_free
;
869 err
= mthca_init_qp_table(dev
);
871 mthca_err(dev
, "Failed to initialize "
872 "queue pair table, aborting.\n");
873 goto err_srq_table_free
;
876 err
= mthca_init_av_table(dev
);
878 mthca_err(dev
, "Failed to initialize "
879 "address vector table, aborting.\n");
880 goto err_qp_table_free
;
883 err
= mthca_init_mcg_table(dev
);
885 mthca_err(dev
, "Failed to initialize "
886 "multicast group table, aborting.\n");
887 goto err_av_table_free
;
893 mthca_cleanup_av_table(dev
);
896 mthca_cleanup_qp_table(dev
);
899 mthca_cleanup_srq_table(dev
);
902 mthca_cleanup_cq_table(dev
);
905 mthca_cmd_use_polling(dev
);
908 mthca_cleanup_eq_table(dev
);
911 mthca_pd_free(dev
, &dev
->driver_pd
);
914 mthca_cleanup_mr_table(dev
);
917 mthca_cleanup_pd_table(dev
);
923 mthca_uar_free(dev
, &dev
->driver_uar
);
926 mthca_cleanup_uar_table(dev
);
930 static int mthca_enable_msi_x(struct mthca_dev
*mdev
)
932 struct msix_entry entries
[3];
935 entries
[0].entry
= 0;
936 entries
[1].entry
= 1;
937 entries
[2].entry
= 2;
939 err
= pci_enable_msix(mdev
->pdev
, entries
, ARRAY_SIZE(entries
));
942 mthca_info(mdev
, "Only %d MSI-X vectors available, "
943 "not using MSI-X\n", err
);
947 mdev
->eq_table
.eq
[MTHCA_EQ_COMP
].msi_x_vector
= entries
[0].vector
;
948 mdev
->eq_table
.eq
[MTHCA_EQ_ASYNC
].msi_x_vector
= entries
[1].vector
;
949 mdev
->eq_table
.eq
[MTHCA_EQ_CMD
].msi_x_vector
= entries
[2].vector
;
954 /* Types of supported HCA */
957 ARBEL_COMPAT
, /* MT25208 in Tavor compat mode */
958 ARBEL_NATIVE
, /* MT25208 with extended features */
962 #define MTHCA_FW_VER(major, minor, subminor) \
963 (((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
968 } mthca_hca_table
[] = {
969 [TAVOR
] = { .latest_fw
= MTHCA_FW_VER(3, 5, 0),
971 [ARBEL_COMPAT
] = { .latest_fw
= MTHCA_FW_VER(4, 8, 200),
972 .flags
= MTHCA_FLAG_PCIE
},
973 [ARBEL_NATIVE
] = { .latest_fw
= MTHCA_FW_VER(5, 3, 0),
974 .flags
= MTHCA_FLAG_MEMFREE
|
976 [SINAI
] = { .latest_fw
= MTHCA_FW_VER(1, 2, 0),
977 .flags
= MTHCA_FLAG_MEMFREE
|
979 MTHCA_FLAG_SINAI_OPT
}
982 static int __mthca_init_one(struct pci_dev
*pdev
, int hca_type
)
986 struct mthca_dev
*mdev
;
988 printk(KERN_INFO PFX
"Initializing %s\n",
991 err
= pci_enable_device(pdev
);
993 dev_err(&pdev
->dev
, "Cannot enable PCI device, "
999 * Check for BARs. We expect 0: 1MB, 2: 8MB, 4: DDR (may not
1002 if (!(pci_resource_flags(pdev
, 0) & IORESOURCE_MEM
) ||
1003 pci_resource_len(pdev
, 0) != 1 << 20) {
1004 dev_err(&pdev
->dev
, "Missing DCS, aborting.\n");
1006 goto err_disable_pdev
;
1008 if (!(pci_resource_flags(pdev
, 2) & IORESOURCE_MEM
)) {
1009 dev_err(&pdev
->dev
, "Missing UAR, aborting.\n");
1011 goto err_disable_pdev
;
1013 if (!(pci_resource_flags(pdev
, 4) & IORESOURCE_MEM
))
1016 err
= pci_request_regions(pdev
, DRV_NAME
);
1018 dev_err(&pdev
->dev
, "Cannot obtain PCI resources, "
1020 goto err_disable_pdev
;
1023 pci_set_master(pdev
);
1025 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
1027 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1028 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
1030 dev_err(&pdev
->dev
, "Can't set PCI DMA mask, aborting.\n");
1034 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
1036 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit "
1037 "consistent PCI DMA mask.\n");
1038 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(32));
1040 dev_err(&pdev
->dev
, "Can't set consistent PCI DMA mask, "
1046 /* We can handle large RDMA requests, so allow larger segments. */
1047 dma_set_max_seg_size(&pdev
->dev
, 1024 * 1024 * 1024);
1049 mdev
= (struct mthca_dev
*) ib_alloc_device(sizeof *mdev
);
1051 dev_err(&pdev
->dev
, "Device struct alloc failed, "
1059 mdev
->mthca_flags
= mthca_hca_table
[hca_type
].flags
;
1061 mdev
->mthca_flags
|= MTHCA_FLAG_DDR_HIDDEN
;
1064 * Now reset the HCA before we touch the PCI capabilities or
1065 * attempt a firmware command, since a boot ROM may have left
1066 * the HCA in an undefined state.
1068 err
= mthca_reset(mdev
);
1070 mthca_err(mdev
, "Failed to reset HCA, aborting.\n");
1074 if (mthca_cmd_init(mdev
)) {
1075 mthca_err(mdev
, "Failed to init command interface, aborting.\n");
1079 err
= mthca_tune_pci(mdev
);
1083 err
= mthca_init_hca(mdev
);
1087 if (mdev
->fw_ver
< mthca_hca_table
[hca_type
].latest_fw
) {
1088 mthca_warn(mdev
, "HCA FW version %d.%d.%03d is old (%d.%d.%03d is current).\n",
1089 (int) (mdev
->fw_ver
>> 32), (int) (mdev
->fw_ver
>> 16) & 0xffff,
1090 (int) (mdev
->fw_ver
& 0xffff),
1091 (int) (mthca_hca_table
[hca_type
].latest_fw
>> 32),
1092 (int) (mthca_hca_table
[hca_type
].latest_fw
>> 16) & 0xffff,
1093 (int) (mthca_hca_table
[hca_type
].latest_fw
& 0xffff));
1094 mthca_warn(mdev
, "If you have problems, try updating your HCA FW.\n");
1097 if (msi_x
&& !mthca_enable_msi_x(mdev
))
1098 mdev
->mthca_flags
|= MTHCA_FLAG_MSI_X
;
1100 err
= mthca_setup_hca(mdev
);
1101 if (err
== -EBUSY
&& (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)) {
1102 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1103 pci_disable_msix(pdev
);
1104 mdev
->mthca_flags
&= ~MTHCA_FLAG_MSI_X
;
1106 err
= mthca_setup_hca(mdev
);
1112 err
= mthca_register_device(mdev
);
1116 err
= mthca_create_agents(mdev
);
1118 goto err_unregister
;
1120 pci_set_drvdata(pdev
, mdev
);
1121 mdev
->hca_type
= hca_type
;
1123 mdev
->active
= true;
1128 mthca_unregister_device(mdev
);
1131 mthca_cleanup_mcg_table(mdev
);
1132 mthca_cleanup_av_table(mdev
);
1133 mthca_cleanup_qp_table(mdev
);
1134 mthca_cleanup_srq_table(mdev
);
1135 mthca_cleanup_cq_table(mdev
);
1136 mthca_cmd_use_polling(mdev
);
1137 mthca_cleanup_eq_table(mdev
);
1139 mthca_pd_free(mdev
, &mdev
->driver_pd
);
1141 mthca_cleanup_mr_table(mdev
);
1142 mthca_cleanup_pd_table(mdev
);
1143 mthca_cleanup_uar_table(mdev
);
1146 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1147 pci_disable_msix(pdev
);
1149 mthca_close_hca(mdev
);
1152 mthca_cmd_cleanup(mdev
);
1155 ib_dealloc_device(&mdev
->ib_dev
);
1158 pci_release_regions(pdev
);
1161 pci_disable_device(pdev
);
1162 pci_set_drvdata(pdev
, NULL
);
1166 static void __mthca_remove_one(struct pci_dev
*pdev
)
1168 struct mthca_dev
*mdev
= pci_get_drvdata(pdev
);
1173 mthca_free_agents(mdev
);
1174 mthca_unregister_device(mdev
);
1176 for (p
= 1; p
<= mdev
->limits
.num_ports
; ++p
)
1177 mthca_CLOSE_IB(mdev
, p
, &status
);
1179 mthca_cleanup_mcg_table(mdev
);
1180 mthca_cleanup_av_table(mdev
);
1181 mthca_cleanup_qp_table(mdev
);
1182 mthca_cleanup_srq_table(mdev
);
1183 mthca_cleanup_cq_table(mdev
);
1184 mthca_cmd_use_polling(mdev
);
1185 mthca_cleanup_eq_table(mdev
);
1187 mthca_pd_free(mdev
, &mdev
->driver_pd
);
1189 mthca_cleanup_mr_table(mdev
);
1190 mthca_cleanup_pd_table(mdev
);
1193 mthca_uar_free(mdev
, &mdev
->driver_uar
);
1194 mthca_cleanup_uar_table(mdev
);
1195 mthca_close_hca(mdev
);
1196 mthca_cmd_cleanup(mdev
);
1198 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1199 pci_disable_msix(pdev
);
1201 ib_dealloc_device(&mdev
->ib_dev
);
1202 pci_release_regions(pdev
);
1203 pci_disable_device(pdev
);
1204 pci_set_drvdata(pdev
, NULL
);
1208 int __mthca_restart_one(struct pci_dev
*pdev
)
1210 struct mthca_dev
*mdev
;
1213 mdev
= pci_get_drvdata(pdev
);
1216 hca_type
= mdev
->hca_type
;
1217 __mthca_remove_one(pdev
);
1218 return __mthca_init_one(pdev
, hca_type
);
1221 static int __devinit
mthca_init_one(struct pci_dev
*pdev
,
1222 const struct pci_device_id
*id
)
1226 mutex_lock(&mthca_device_mutex
);
1228 printk_once(KERN_INFO
"%s", mthca_version
);
1230 if (id
->driver_data
>= ARRAY_SIZE(mthca_hca_table
)) {
1231 printk(KERN_ERR PFX
"%s has invalid driver data %lx\n",
1232 pci_name(pdev
), id
->driver_data
);
1233 mutex_unlock(&mthca_device_mutex
);
1237 ret
= __mthca_init_one(pdev
, id
->driver_data
);
1239 mutex_unlock(&mthca_device_mutex
);
1244 static void __devexit
mthca_remove_one(struct pci_dev
*pdev
)
1246 mutex_lock(&mthca_device_mutex
);
1247 __mthca_remove_one(pdev
);
1248 mutex_unlock(&mthca_device_mutex
);
1251 static struct pci_device_id mthca_pci_table
[] = {
1252 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_TAVOR
),
1253 .driver_data
= TAVOR
},
1254 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_TAVOR
),
1255 .driver_data
= TAVOR
},
1256 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
),
1257 .driver_data
= ARBEL_COMPAT
},
1258 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
),
1259 .driver_data
= ARBEL_COMPAT
},
1260 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_ARBEL
),
1261 .driver_data
= ARBEL_NATIVE
},
1262 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_ARBEL
),
1263 .driver_data
= ARBEL_NATIVE
},
1264 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_SINAI
),
1265 .driver_data
= SINAI
},
1266 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_SINAI
),
1267 .driver_data
= SINAI
},
1268 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_SINAI_OLD
),
1269 .driver_data
= SINAI
},
1270 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_SINAI_OLD
),
1271 .driver_data
= SINAI
},
1275 MODULE_DEVICE_TABLE(pci
, mthca_pci_table
);
1277 static struct pci_driver mthca_driver
= {
1279 .id_table
= mthca_pci_table
,
1280 .probe
= mthca_init_one
,
1281 .remove
= __devexit_p(mthca_remove_one
)
1284 static void __init
__mthca_check_profile_val(const char *name
, int *pval
,
1287 /* value must be positive and power of 2 */
1288 int old_pval
= *pval
;
1291 *pval
= pval_default
;
1293 *pval
= roundup_pow_of_two(old_pval
);
1295 if (old_pval
!= *pval
) {
1296 printk(KERN_WARNING PFX
"Invalid value %d for %s in module parameter.\n",
1298 printk(KERN_WARNING PFX
"Corrected %s to %d.\n", name
, *pval
);
1302 #define mthca_check_profile_val(name, default) \
1303 __mthca_check_profile_val(#name, &hca_profile.name, default)
1305 static void __init
mthca_validate_profile(void)
1307 mthca_check_profile_val(num_qp
, MTHCA_DEFAULT_NUM_QP
);
1308 mthca_check_profile_val(rdb_per_qp
, MTHCA_DEFAULT_RDB_PER_QP
);
1309 mthca_check_profile_val(num_cq
, MTHCA_DEFAULT_NUM_CQ
);
1310 mthca_check_profile_val(num_mcg
, MTHCA_DEFAULT_NUM_MCG
);
1311 mthca_check_profile_val(num_mpt
, MTHCA_DEFAULT_NUM_MPT
);
1312 mthca_check_profile_val(num_mtt
, MTHCA_DEFAULT_NUM_MTT
);
1313 mthca_check_profile_val(num_udav
, MTHCA_DEFAULT_NUM_UDAV
);
1314 mthca_check_profile_val(fmr_reserved_mtts
, MTHCA_DEFAULT_NUM_RESERVED_MTTS
);
1316 if (hca_profile
.fmr_reserved_mtts
>= hca_profile
.num_mtt
) {
1317 printk(KERN_WARNING PFX
"Invalid fmr_reserved_mtts module parameter %d.\n",
1318 hca_profile
.fmr_reserved_mtts
);
1319 printk(KERN_WARNING PFX
"(Must be smaller than num_mtt %d)\n",
1320 hca_profile
.num_mtt
);
1321 hca_profile
.fmr_reserved_mtts
= hca_profile
.num_mtt
/ 2;
1322 printk(KERN_WARNING PFX
"Corrected fmr_reserved_mtts to %d.\n",
1323 hca_profile
.fmr_reserved_mtts
);
1326 if ((log_mtts_per_seg
< 1) || (log_mtts_per_seg
> 5)) {
1327 printk(KERN_WARNING PFX
"bad log_mtts_per_seg (%d). Using default - %d\n",
1328 log_mtts_per_seg
, ilog2(MTHCA_MTT_SEG_SIZE
/ 8));
1329 log_mtts_per_seg
= ilog2(MTHCA_MTT_SEG_SIZE
/ 8);
1333 static int __init
mthca_init(void)
1337 mthca_validate_profile();
1339 ret
= mthca_catas_init();
1343 ret
= pci_register_driver(&mthca_driver
);
1345 mthca_catas_cleanup();
1352 static void __exit
mthca_cleanup(void)
1354 pci_unregister_driver(&mthca_driver
);
1355 mthca_catas_cleanup();
1358 module_init(mthca_init
);
1359 module_exit(mthca_cleanup
);