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
34 * $Id: mthca_main.c 1396 2004-12-28 04:10:27Z roland $
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/errno.h>
40 #include <linux/pci.h>
41 #include <linux/interrupt.h>
43 #include "mthca_dev.h"
44 #include "mthca_config_reg.h"
45 #include "mthca_cmd.h"
46 #include "mthca_profile.h"
47 #include "mthca_memfree.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 char mthca_version
[] __devinitdata
=
130 DRV_NAME
": Mellanox InfiniBand HCA driver v"
131 DRV_VERSION
" (" DRV_RELDATE
")\n";
133 static int mthca_tune_pci(struct mthca_dev
*mdev
)
138 /* First try to max out Read Byte Count */
139 if (pci_find_capability(mdev
->pdev
, PCI_CAP_ID_PCIX
)) {
140 if (pcix_set_mmrbc(mdev
->pdev
, pcix_get_max_mmrbc(mdev
->pdev
))) {
141 mthca_err(mdev
, "Couldn't set PCI-X max read count, "
145 } else if (!(mdev
->mthca_flags
& MTHCA_FLAG_PCIE
))
146 mthca_info(mdev
, "No PCI-X capability, not setting RBC.\n");
148 if (pci_find_capability(mdev
->pdev
, PCI_CAP_ID_EXP
)) {
149 if (pcie_set_readrq(mdev
->pdev
, 4096)) {
150 mthca_err(mdev
, "Couldn't write PCI Express read request, "
154 } else if (mdev
->mthca_flags
& MTHCA_FLAG_PCIE
)
155 mthca_info(mdev
, "No PCI Express capability, "
156 "not setting Max Read Request Size.\n");
161 static int mthca_dev_lim(struct mthca_dev
*mdev
, struct mthca_dev_lim
*dev_lim
)
166 err
= mthca_QUERY_DEV_LIM(mdev
, dev_lim
, &status
);
168 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
172 mthca_err(mdev
, "QUERY_DEV_LIM returned status 0x%02x, "
173 "aborting.\n", status
);
176 if (dev_lim
->min_page_sz
> PAGE_SIZE
) {
177 mthca_err(mdev
, "HCA minimum page size of %d bigger than "
178 "kernel PAGE_SIZE of %ld, aborting.\n",
179 dev_lim
->min_page_sz
, PAGE_SIZE
);
182 if (dev_lim
->num_ports
> MTHCA_MAX_PORTS
) {
183 mthca_err(mdev
, "HCA has %d ports, but we only support %d, "
185 dev_lim
->num_ports
, MTHCA_MAX_PORTS
);
189 if (dev_lim
->uar_size
> pci_resource_len(mdev
->pdev
, 2)) {
190 mthca_err(mdev
, "HCA reported UAR size of 0x%x bigger than "
191 "PCI resource 2 size of 0x%llx, aborting.\n",
193 (unsigned long long)pci_resource_len(mdev
->pdev
, 2));
197 mdev
->limits
.num_ports
= dev_lim
->num_ports
;
198 mdev
->limits
.vl_cap
= dev_lim
->max_vl
;
199 mdev
->limits
.mtu_cap
= dev_lim
->max_mtu
;
200 mdev
->limits
.gid_table_len
= dev_lim
->max_gids
;
201 mdev
->limits
.pkey_table_len
= dev_lim
->max_pkeys
;
202 mdev
->limits
.local_ca_ack_delay
= dev_lim
->local_ca_ack_delay
;
203 mdev
->limits
.max_sg
= dev_lim
->max_sg
;
204 mdev
->limits
.max_wqes
= dev_lim
->max_qp_sz
;
205 mdev
->limits
.max_qp_init_rdma
= dev_lim
->max_requester_per_qp
;
206 mdev
->limits
.reserved_qps
= dev_lim
->reserved_qps
;
207 mdev
->limits
.max_srq_wqes
= dev_lim
->max_srq_sz
;
208 mdev
->limits
.reserved_srqs
= dev_lim
->reserved_srqs
;
209 mdev
->limits
.reserved_eecs
= dev_lim
->reserved_eecs
;
210 mdev
->limits
.max_desc_sz
= dev_lim
->max_desc_sz
;
211 mdev
->limits
.max_srq_sge
= mthca_max_srq_sge(mdev
);
213 * Subtract 1 from the limit because we need to allocate a
214 * spare CQE so the HCA HW can tell the difference between an
215 * empty CQ and a full CQ.
217 mdev
->limits
.max_cqes
= dev_lim
->max_cq_sz
- 1;
218 mdev
->limits
.reserved_cqs
= dev_lim
->reserved_cqs
;
219 mdev
->limits
.reserved_eqs
= dev_lim
->reserved_eqs
;
220 mdev
->limits
.reserved_mtts
= dev_lim
->reserved_mtts
;
221 mdev
->limits
.reserved_mrws
= dev_lim
->reserved_mrws
;
222 mdev
->limits
.reserved_uars
= dev_lim
->reserved_uars
;
223 mdev
->limits
.reserved_pds
= dev_lim
->reserved_pds
;
224 mdev
->limits
.port_width_cap
= dev_lim
->max_port_width
;
225 mdev
->limits
.page_size_cap
= ~(u32
) (dev_lim
->min_page_sz
- 1);
226 mdev
->limits
.flags
= dev_lim
->flags
;
228 * For old FW that doesn't return static rate support, use a
229 * value of 0x3 (only static rate values of 0 or 1 are handled),
230 * except on Sinai, where even old FW can handle static rate
233 if (dev_lim
->stat_rate_support
)
234 mdev
->limits
.stat_rate_support
= dev_lim
->stat_rate_support
;
235 else if (mdev
->mthca_flags
& MTHCA_FLAG_SINAI_OPT
)
236 mdev
->limits
.stat_rate_support
= 0xf;
238 mdev
->limits
.stat_rate_support
= 0x3;
240 /* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
241 May be doable since hardware supports it for SRQ.
243 IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
245 IB_DEVICE_SRQ_RESIZE is supported by hardware but SRQ is not
246 supported by driver. */
247 mdev
->device_cap_flags
= IB_DEVICE_CHANGE_PHY_PORT
|
248 IB_DEVICE_PORT_ACTIVE_EVENT
|
249 IB_DEVICE_SYS_IMAGE_GUID
|
250 IB_DEVICE_RC_RNR_NAK_GEN
;
252 if (dev_lim
->flags
& DEV_LIM_FLAG_BAD_PKEY_CNTR
)
253 mdev
->device_cap_flags
|= IB_DEVICE_BAD_PKEY_CNTR
;
255 if (dev_lim
->flags
& DEV_LIM_FLAG_BAD_QKEY_CNTR
)
256 mdev
->device_cap_flags
|= IB_DEVICE_BAD_QKEY_CNTR
;
258 if (dev_lim
->flags
& DEV_LIM_FLAG_RAW_MULTI
)
259 mdev
->device_cap_flags
|= IB_DEVICE_RAW_MULTI
;
261 if (dev_lim
->flags
& DEV_LIM_FLAG_AUTO_PATH_MIG
)
262 mdev
->device_cap_flags
|= IB_DEVICE_AUTO_PATH_MIG
;
264 if (dev_lim
->flags
& DEV_LIM_FLAG_UD_AV_PORT_ENFORCE
)
265 mdev
->device_cap_flags
|= IB_DEVICE_UD_AV_PORT_ENFORCE
;
267 if (dev_lim
->flags
& DEV_LIM_FLAG_SRQ
)
268 mdev
->mthca_flags
|= MTHCA_FLAG_SRQ
;
270 if (mthca_is_memfree(mdev
))
271 if (dev_lim
->flags
& DEV_LIM_FLAG_IPOIB_CSUM
)
272 mdev
->device_cap_flags
|= IB_DEVICE_UD_IP_CSUM
;
277 static int mthca_init_tavor(struct mthca_dev
*mdev
)
282 struct mthca_dev_lim dev_lim
;
283 struct mthca_profile profile
;
284 struct mthca_init_hca_param init_hca
;
286 err
= mthca_SYS_EN(mdev
, &status
);
288 mthca_err(mdev
, "SYS_EN command failed, aborting.\n");
292 mthca_err(mdev
, "SYS_EN returned status 0x%02x, "
293 "aborting.\n", status
);
297 err
= mthca_QUERY_FW(mdev
, &status
);
299 mthca_err(mdev
, "QUERY_FW command failed, aborting.\n");
303 mthca_err(mdev
, "QUERY_FW returned status 0x%02x, "
304 "aborting.\n", status
);
308 err
= mthca_QUERY_DDR(mdev
, &status
);
310 mthca_err(mdev
, "QUERY_DDR command failed, aborting.\n");
314 mthca_err(mdev
, "QUERY_DDR returned status 0x%02x, "
315 "aborting.\n", status
);
320 err
= mthca_dev_lim(mdev
, &dev_lim
);
322 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
326 profile
= hca_profile
;
327 profile
.num_uar
= dev_lim
.uar_size
/ PAGE_SIZE
;
328 profile
.uarc_size
= 0;
329 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
330 profile
.num_srq
= dev_lim
.max_srqs
;
332 size
= mthca_make_profile(mdev
, &profile
, &dev_lim
, &init_hca
);
338 err
= mthca_INIT_HCA(mdev
, &init_hca
, &status
);
340 mthca_err(mdev
, "INIT_HCA command failed, aborting.\n");
344 mthca_err(mdev
, "INIT_HCA returned status 0x%02x, "
345 "aborting.\n", status
);
353 mthca_SYS_DIS(mdev
, &status
);
358 static int mthca_load_fw(struct mthca_dev
*mdev
)
363 /* FIXME: use HCA-attached memory for FW if present */
365 mdev
->fw
.arbel
.fw_icm
=
366 mthca_alloc_icm(mdev
, mdev
->fw
.arbel
.fw_pages
,
367 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
368 if (!mdev
->fw
.arbel
.fw_icm
) {
369 mthca_err(mdev
, "Couldn't allocate FW area, aborting.\n");
373 err
= mthca_MAP_FA(mdev
, mdev
->fw
.arbel
.fw_icm
, &status
);
375 mthca_err(mdev
, "MAP_FA command failed, aborting.\n");
379 mthca_err(mdev
, "MAP_FA returned status 0x%02x, aborting.\n", status
);
383 err
= mthca_RUN_FW(mdev
, &status
);
385 mthca_err(mdev
, "RUN_FW command failed, aborting.\n");
389 mthca_err(mdev
, "RUN_FW returned status 0x%02x, aborting.\n", status
);
397 mthca_UNMAP_FA(mdev
, &status
);
400 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
404 static int mthca_init_icm(struct mthca_dev
*mdev
,
405 struct mthca_dev_lim
*dev_lim
,
406 struct mthca_init_hca_param
*init_hca
,
413 err
= mthca_SET_ICM_SIZE(mdev
, icm_size
, &aux_pages
, &status
);
415 mthca_err(mdev
, "SET_ICM_SIZE command failed, aborting.\n");
419 mthca_err(mdev
, "SET_ICM_SIZE returned status 0x%02x, "
420 "aborting.\n", status
);
424 mthca_dbg(mdev
, "%lld KB of HCA context requires %lld KB aux memory.\n",
425 (unsigned long long) icm_size
>> 10,
426 (unsigned long long) aux_pages
<< 2);
428 mdev
->fw
.arbel
.aux_icm
= mthca_alloc_icm(mdev
, aux_pages
,
429 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
430 if (!mdev
->fw
.arbel
.aux_icm
) {
431 mthca_err(mdev
, "Couldn't allocate aux memory, aborting.\n");
435 err
= mthca_MAP_ICM_AUX(mdev
, mdev
->fw
.arbel
.aux_icm
, &status
);
437 mthca_err(mdev
, "MAP_ICM_AUX command failed, aborting.\n");
441 mthca_err(mdev
, "MAP_ICM_AUX returned status 0x%02x, aborting.\n", status
);
446 err
= mthca_map_eq_icm(mdev
, init_hca
->eqc_base
);
448 mthca_err(mdev
, "Failed to map EQ context memory, aborting.\n");
452 /* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
453 mdev
->limits
.reserved_mtts
= ALIGN(mdev
->limits
.reserved_mtts
* MTHCA_MTT_SEG_SIZE
,
454 dma_get_cache_alignment()) / MTHCA_MTT_SEG_SIZE
;
456 mdev
->mr_table
.mtt_table
= mthca_alloc_icm_table(mdev
, init_hca
->mtt_base
,
458 mdev
->limits
.num_mtt_segs
,
459 mdev
->limits
.reserved_mtts
,
461 if (!mdev
->mr_table
.mtt_table
) {
462 mthca_err(mdev
, "Failed to map MTT context memory, aborting.\n");
467 mdev
->mr_table
.mpt_table
= mthca_alloc_icm_table(mdev
, init_hca
->mpt_base
,
468 dev_lim
->mpt_entry_sz
,
469 mdev
->limits
.num_mpts
,
470 mdev
->limits
.reserved_mrws
,
472 if (!mdev
->mr_table
.mpt_table
) {
473 mthca_err(mdev
, "Failed to map MPT context memory, aborting.\n");
478 mdev
->qp_table
.qp_table
= mthca_alloc_icm_table(mdev
, init_hca
->qpc_base
,
479 dev_lim
->qpc_entry_sz
,
480 mdev
->limits
.num_qps
,
481 mdev
->limits
.reserved_qps
,
483 if (!mdev
->qp_table
.qp_table
) {
484 mthca_err(mdev
, "Failed to map QP context memory, aborting.\n");
489 mdev
->qp_table
.eqp_table
= mthca_alloc_icm_table(mdev
, init_hca
->eqpc_base
,
490 dev_lim
->eqpc_entry_sz
,
491 mdev
->limits
.num_qps
,
492 mdev
->limits
.reserved_qps
,
494 if (!mdev
->qp_table
.eqp_table
) {
495 mthca_err(mdev
, "Failed to map EQP context memory, aborting.\n");
500 mdev
->qp_table
.rdb_table
= mthca_alloc_icm_table(mdev
, init_hca
->rdb_base
,
501 MTHCA_RDB_ENTRY_SIZE
,
502 mdev
->limits
.num_qps
<<
503 mdev
->qp_table
.rdb_shift
, 0,
505 if (!mdev
->qp_table
.rdb_table
) {
506 mthca_err(mdev
, "Failed to map RDB context memory, aborting\n");
511 mdev
->cq_table
.table
= mthca_alloc_icm_table(mdev
, init_hca
->cqc_base
,
512 dev_lim
->cqc_entry_sz
,
513 mdev
->limits
.num_cqs
,
514 mdev
->limits
.reserved_cqs
,
516 if (!mdev
->cq_table
.table
) {
517 mthca_err(mdev
, "Failed to map CQ context memory, aborting.\n");
522 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
523 mdev
->srq_table
.table
=
524 mthca_alloc_icm_table(mdev
, init_hca
->srqc_base
,
525 dev_lim
->srq_entry_sz
,
526 mdev
->limits
.num_srqs
,
527 mdev
->limits
.reserved_srqs
,
529 if (!mdev
->srq_table
.table
) {
530 mthca_err(mdev
, "Failed to map SRQ context memory, "
538 * It's not strictly required, but for simplicity just map the
539 * whole multicast group table now. The table isn't very big
540 * and it's a lot easier than trying to track ref counts.
542 mdev
->mcg_table
.table
= mthca_alloc_icm_table(mdev
, init_hca
->mc_base
,
543 MTHCA_MGM_ENTRY_SIZE
,
544 mdev
->limits
.num_mgms
+
545 mdev
->limits
.num_amgms
,
546 mdev
->limits
.num_mgms
+
547 mdev
->limits
.num_amgms
,
549 if (!mdev
->mcg_table
.table
) {
550 mthca_err(mdev
, "Failed to map MCG context memory, aborting.\n");
558 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
559 mthca_free_icm_table(mdev
, mdev
->srq_table
.table
);
562 mthca_free_icm_table(mdev
, mdev
->cq_table
.table
);
565 mthca_free_icm_table(mdev
, mdev
->qp_table
.rdb_table
);
568 mthca_free_icm_table(mdev
, mdev
->qp_table
.eqp_table
);
571 mthca_free_icm_table(mdev
, mdev
->qp_table
.qp_table
);
574 mthca_free_icm_table(mdev
, mdev
->mr_table
.mpt_table
);
577 mthca_free_icm_table(mdev
, mdev
->mr_table
.mtt_table
);
580 mthca_unmap_eq_icm(mdev
);
583 mthca_UNMAP_ICM_AUX(mdev
, &status
);
586 mthca_free_icm(mdev
, mdev
->fw
.arbel
.aux_icm
, 0);
591 static void mthca_free_icms(struct mthca_dev
*mdev
)
595 mthca_free_icm_table(mdev
, mdev
->mcg_table
.table
);
596 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
597 mthca_free_icm_table(mdev
, mdev
->srq_table
.table
);
598 mthca_free_icm_table(mdev
, mdev
->cq_table
.table
);
599 mthca_free_icm_table(mdev
, mdev
->qp_table
.rdb_table
);
600 mthca_free_icm_table(mdev
, mdev
->qp_table
.eqp_table
);
601 mthca_free_icm_table(mdev
, mdev
->qp_table
.qp_table
);
602 mthca_free_icm_table(mdev
, mdev
->mr_table
.mpt_table
);
603 mthca_free_icm_table(mdev
, mdev
->mr_table
.mtt_table
);
604 mthca_unmap_eq_icm(mdev
);
606 mthca_UNMAP_ICM_AUX(mdev
, &status
);
607 mthca_free_icm(mdev
, mdev
->fw
.arbel
.aux_icm
, 0);
610 static int mthca_init_arbel(struct mthca_dev
*mdev
)
612 struct mthca_dev_lim dev_lim
;
613 struct mthca_profile profile
;
614 struct mthca_init_hca_param init_hca
;
619 err
= mthca_QUERY_FW(mdev
, &status
);
621 mthca_err(mdev
, "QUERY_FW command failed, aborting.\n");
625 mthca_err(mdev
, "QUERY_FW returned status 0x%02x, "
626 "aborting.\n", status
);
630 err
= mthca_ENABLE_LAM(mdev
, &status
);
632 mthca_err(mdev
, "ENABLE_LAM command failed, aborting.\n");
635 if (status
== MTHCA_CMD_STAT_LAM_NOT_PRE
) {
636 mthca_dbg(mdev
, "No HCA-attached memory (running in MemFree mode)\n");
637 mdev
->mthca_flags
|= MTHCA_FLAG_NO_LAM
;
639 mthca_err(mdev
, "ENABLE_LAM returned status 0x%02x, "
640 "aborting.\n", status
);
644 err
= mthca_load_fw(mdev
);
646 mthca_err(mdev
, "Failed to start FW, aborting.\n");
650 err
= mthca_dev_lim(mdev
, &dev_lim
);
652 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
656 profile
= hca_profile
;
657 profile
.num_uar
= dev_lim
.uar_size
/ PAGE_SIZE
;
658 profile
.num_udav
= 0;
659 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
660 profile
.num_srq
= dev_lim
.max_srqs
;
662 icm_size
= mthca_make_profile(mdev
, &profile
, &dev_lim
, &init_hca
);
668 err
= mthca_init_icm(mdev
, &dev_lim
, &init_hca
, icm_size
);
672 err
= mthca_INIT_HCA(mdev
, &init_hca
, &status
);
674 mthca_err(mdev
, "INIT_HCA command failed, aborting.\n");
678 mthca_err(mdev
, "INIT_HCA returned status 0x%02x, "
679 "aborting.\n", status
);
687 mthca_free_icms(mdev
);
690 mthca_UNMAP_FA(mdev
, &status
);
691 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
694 if (!(mdev
->mthca_flags
& MTHCA_FLAG_NO_LAM
))
695 mthca_DISABLE_LAM(mdev
, &status
);
700 static void mthca_close_hca(struct mthca_dev
*mdev
)
704 mthca_CLOSE_HCA(mdev
, 0, &status
);
706 if (mthca_is_memfree(mdev
)) {
707 mthca_free_icms(mdev
);
709 mthca_UNMAP_FA(mdev
, &status
);
710 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
712 if (!(mdev
->mthca_flags
& MTHCA_FLAG_NO_LAM
))
713 mthca_DISABLE_LAM(mdev
, &status
);
715 mthca_SYS_DIS(mdev
, &status
);
718 static int mthca_init_hca(struct mthca_dev
*mdev
)
722 struct mthca_adapter adapter
;
724 if (mthca_is_memfree(mdev
))
725 err
= mthca_init_arbel(mdev
);
727 err
= mthca_init_tavor(mdev
);
732 err
= mthca_QUERY_ADAPTER(mdev
, &adapter
, &status
);
734 mthca_err(mdev
, "QUERY_ADAPTER command failed, aborting.\n");
738 mthca_err(mdev
, "QUERY_ADAPTER returned status 0x%02x, "
739 "aborting.\n", status
);
744 mdev
->eq_table
.inta_pin
= adapter
.inta_pin
;
745 if (!mthca_is_memfree(mdev
))
746 mdev
->rev_id
= adapter
.revision_id
;
747 memcpy(mdev
->board_id
, adapter
.board_id
, sizeof mdev
->board_id
);
752 mthca_close_hca(mdev
);
756 static int mthca_setup_hca(struct mthca_dev
*dev
)
761 MTHCA_INIT_DOORBELL_LOCK(&dev
->doorbell_lock
);
763 err
= mthca_init_uar_table(dev
);
765 mthca_err(dev
, "Failed to initialize "
766 "user access region table, aborting.\n");
770 err
= mthca_uar_alloc(dev
, &dev
->driver_uar
);
772 mthca_err(dev
, "Failed to allocate driver access region, "
774 goto err_uar_table_free
;
777 dev
->kar
= ioremap(dev
->driver_uar
.pfn
<< PAGE_SHIFT
, PAGE_SIZE
);
779 mthca_err(dev
, "Couldn't map kernel access region, "
785 err
= mthca_init_pd_table(dev
);
787 mthca_err(dev
, "Failed to initialize "
788 "protection domain table, aborting.\n");
792 err
= mthca_init_mr_table(dev
);
794 mthca_err(dev
, "Failed to initialize "
795 "memory region table, aborting.\n");
796 goto err_pd_table_free
;
799 err
= mthca_pd_alloc(dev
, 1, &dev
->driver_pd
);
801 mthca_err(dev
, "Failed to create driver PD, "
803 goto err_mr_table_free
;
806 err
= mthca_init_eq_table(dev
);
808 mthca_err(dev
, "Failed to initialize "
809 "event queue table, aborting.\n");
813 err
= mthca_cmd_use_events(dev
);
815 mthca_err(dev
, "Failed to switch to event-driven "
816 "firmware commands, aborting.\n");
817 goto err_eq_table_free
;
820 err
= mthca_NOP(dev
, &status
);
822 if (dev
->mthca_flags
& MTHCA_FLAG_MSI_X
) {
823 mthca_warn(dev
, "NOP command failed to generate interrupt "
825 dev
->eq_table
.eq
[MTHCA_EQ_CMD
].msi_x_vector
);
826 mthca_warn(dev
, "Trying again with MSI-X disabled.\n");
828 mthca_err(dev
, "NOP command failed to generate interrupt "
829 "(IRQ %d), aborting.\n",
831 mthca_err(dev
, "BIOS or ACPI interrupt routing problem?\n");
837 mthca_dbg(dev
, "NOP command IRQ test passed\n");
839 err
= mthca_init_cq_table(dev
);
841 mthca_err(dev
, "Failed to initialize "
842 "completion queue table, aborting.\n");
846 err
= mthca_init_srq_table(dev
);
848 mthca_err(dev
, "Failed to initialize "
849 "shared receive queue table, aborting.\n");
850 goto err_cq_table_free
;
853 err
= mthca_init_qp_table(dev
);
855 mthca_err(dev
, "Failed to initialize "
856 "queue pair table, aborting.\n");
857 goto err_srq_table_free
;
860 err
= mthca_init_av_table(dev
);
862 mthca_err(dev
, "Failed to initialize "
863 "address vector table, aborting.\n");
864 goto err_qp_table_free
;
867 err
= mthca_init_mcg_table(dev
);
869 mthca_err(dev
, "Failed to initialize "
870 "multicast group table, aborting.\n");
871 goto err_av_table_free
;
877 mthca_cleanup_av_table(dev
);
880 mthca_cleanup_qp_table(dev
);
883 mthca_cleanup_srq_table(dev
);
886 mthca_cleanup_cq_table(dev
);
889 mthca_cmd_use_polling(dev
);
892 mthca_cleanup_eq_table(dev
);
895 mthca_pd_free(dev
, &dev
->driver_pd
);
898 mthca_cleanup_mr_table(dev
);
901 mthca_cleanup_pd_table(dev
);
907 mthca_uar_free(dev
, &dev
->driver_uar
);
910 mthca_cleanup_uar_table(dev
);
914 static int mthca_request_regions(struct pci_dev
*pdev
, int ddr_hidden
)
919 * We can't just use pci_request_regions() because the MSI-X
920 * table is right in the middle of the first BAR. If we did
921 * pci_request_region and grab all of the first BAR, then
922 * setting up MSI-X would fail, since the PCI core wants to do
923 * request_mem_region on the MSI-X vector table.
925 * So just request what we need right now, and request any
926 * other regions we need when setting up EQs.
928 if (!request_mem_region(pci_resource_start(pdev
, 0) + MTHCA_HCR_BASE
,
929 MTHCA_HCR_SIZE
, DRV_NAME
))
932 err
= pci_request_region(pdev
, 2, DRV_NAME
);
934 goto err_bar2_failed
;
937 err
= pci_request_region(pdev
, 4, DRV_NAME
);
939 goto err_bar4_failed
;
945 pci_release_region(pdev
, 2);
948 release_mem_region(pci_resource_start(pdev
, 0) + MTHCA_HCR_BASE
,
954 static void mthca_release_regions(struct pci_dev
*pdev
,
958 pci_release_region(pdev
, 4);
960 pci_release_region(pdev
, 2);
962 release_mem_region(pci_resource_start(pdev
, 0) + MTHCA_HCR_BASE
,
966 static int mthca_enable_msi_x(struct mthca_dev
*mdev
)
968 struct msix_entry entries
[3];
971 entries
[0].entry
= 0;
972 entries
[1].entry
= 1;
973 entries
[2].entry
= 2;
975 err
= pci_enable_msix(mdev
->pdev
, entries
, ARRAY_SIZE(entries
));
978 mthca_info(mdev
, "Only %d MSI-X vectors available, "
979 "not using MSI-X\n", err
);
983 mdev
->eq_table
.eq
[MTHCA_EQ_COMP
].msi_x_vector
= entries
[0].vector
;
984 mdev
->eq_table
.eq
[MTHCA_EQ_ASYNC
].msi_x_vector
= entries
[1].vector
;
985 mdev
->eq_table
.eq
[MTHCA_EQ_CMD
].msi_x_vector
= entries
[2].vector
;
990 /* Types of supported HCA */
993 ARBEL_COMPAT
, /* MT25208 in Tavor compat mode */
994 ARBEL_NATIVE
, /* MT25208 with extended features */
998 #define MTHCA_FW_VER(major, minor, subminor) \
999 (((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
1004 } mthca_hca_table
[] = {
1005 [TAVOR
] = { .latest_fw
= MTHCA_FW_VER(3, 5, 0),
1007 [ARBEL_COMPAT
] = { .latest_fw
= MTHCA_FW_VER(4, 8, 200),
1008 .flags
= MTHCA_FLAG_PCIE
},
1009 [ARBEL_NATIVE
] = { .latest_fw
= MTHCA_FW_VER(5, 3, 0),
1010 .flags
= MTHCA_FLAG_MEMFREE
|
1012 [SINAI
] = { .latest_fw
= MTHCA_FW_VER(1, 2, 0),
1013 .flags
= MTHCA_FLAG_MEMFREE
|
1015 MTHCA_FLAG_SINAI_OPT
}
1018 static int __mthca_init_one(struct pci_dev
*pdev
, int hca_type
)
1022 struct mthca_dev
*mdev
;
1024 printk(KERN_INFO PFX
"Initializing %s\n",
1027 err
= pci_enable_device(pdev
);
1029 dev_err(&pdev
->dev
, "Cannot enable PCI device, "
1035 * Check for BARs. We expect 0: 1MB, 2: 8MB, 4: DDR (may not
1038 if (!(pci_resource_flags(pdev
, 0) & IORESOURCE_MEM
) ||
1039 pci_resource_len(pdev
, 0) != 1 << 20) {
1040 dev_err(&pdev
->dev
, "Missing DCS, aborting.\n");
1042 goto err_disable_pdev
;
1044 if (!(pci_resource_flags(pdev
, 2) & IORESOURCE_MEM
)) {
1045 dev_err(&pdev
->dev
, "Missing UAR, aborting.\n");
1047 goto err_disable_pdev
;
1049 if (!(pci_resource_flags(pdev
, 4) & IORESOURCE_MEM
))
1052 err
= mthca_request_regions(pdev
, ddr_hidden
);
1054 dev_err(&pdev
->dev
, "Cannot obtain PCI resources, "
1056 goto err_disable_pdev
;
1059 pci_set_master(pdev
);
1061 err
= pci_set_dma_mask(pdev
, DMA_64BIT_MASK
);
1063 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1064 err
= pci_set_dma_mask(pdev
, DMA_32BIT_MASK
);
1066 dev_err(&pdev
->dev
, "Can't set PCI DMA mask, aborting.\n");
1070 err
= pci_set_consistent_dma_mask(pdev
, DMA_64BIT_MASK
);
1072 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit "
1073 "consistent PCI DMA mask.\n");
1074 err
= pci_set_consistent_dma_mask(pdev
, DMA_32BIT_MASK
);
1076 dev_err(&pdev
->dev
, "Can't set consistent PCI DMA mask, "
1082 mdev
= (struct mthca_dev
*) ib_alloc_device(sizeof *mdev
);
1084 dev_err(&pdev
->dev
, "Device struct alloc failed, "
1092 mdev
->mthca_flags
= mthca_hca_table
[hca_type
].flags
;
1094 mdev
->mthca_flags
|= MTHCA_FLAG_DDR_HIDDEN
;
1097 * Now reset the HCA before we touch the PCI capabilities or
1098 * attempt a firmware command, since a boot ROM may have left
1099 * the HCA in an undefined state.
1101 err
= mthca_reset(mdev
);
1103 mthca_err(mdev
, "Failed to reset HCA, aborting.\n");
1107 if (mthca_cmd_init(mdev
)) {
1108 mthca_err(mdev
, "Failed to init command interface, aborting.\n");
1112 err
= mthca_tune_pci(mdev
);
1116 err
= mthca_init_hca(mdev
);
1120 if (mdev
->fw_ver
< mthca_hca_table
[hca_type
].latest_fw
) {
1121 mthca_warn(mdev
, "HCA FW version %d.%d.%03d is old (%d.%d.%03d is current).\n",
1122 (int) (mdev
->fw_ver
>> 32), (int) (mdev
->fw_ver
>> 16) & 0xffff,
1123 (int) (mdev
->fw_ver
& 0xffff),
1124 (int) (mthca_hca_table
[hca_type
].latest_fw
>> 32),
1125 (int) (mthca_hca_table
[hca_type
].latest_fw
>> 16) & 0xffff,
1126 (int) (mthca_hca_table
[hca_type
].latest_fw
& 0xffff));
1127 mthca_warn(mdev
, "If you have problems, try updating your HCA FW.\n");
1130 if (msi_x
&& !mthca_enable_msi_x(mdev
))
1131 mdev
->mthca_flags
|= MTHCA_FLAG_MSI_X
;
1133 err
= mthca_setup_hca(mdev
);
1134 if (err
== -EBUSY
&& (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)) {
1135 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1136 pci_disable_msix(pdev
);
1137 mdev
->mthca_flags
&= ~MTHCA_FLAG_MSI_X
;
1139 err
= mthca_setup_hca(mdev
);
1145 err
= mthca_register_device(mdev
);
1149 err
= mthca_create_agents(mdev
);
1151 goto err_unregister
;
1153 pci_set_drvdata(pdev
, mdev
);
1154 mdev
->hca_type
= hca_type
;
1159 mthca_unregister_device(mdev
);
1162 mthca_cleanup_mcg_table(mdev
);
1163 mthca_cleanup_av_table(mdev
);
1164 mthca_cleanup_qp_table(mdev
);
1165 mthca_cleanup_srq_table(mdev
);
1166 mthca_cleanup_cq_table(mdev
);
1167 mthca_cmd_use_polling(mdev
);
1168 mthca_cleanup_eq_table(mdev
);
1170 mthca_pd_free(mdev
, &mdev
->driver_pd
);
1172 mthca_cleanup_mr_table(mdev
);
1173 mthca_cleanup_pd_table(mdev
);
1174 mthca_cleanup_uar_table(mdev
);
1177 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1178 pci_disable_msix(pdev
);
1180 mthca_close_hca(mdev
);
1183 mthca_cmd_cleanup(mdev
);
1186 ib_dealloc_device(&mdev
->ib_dev
);
1189 mthca_release_regions(pdev
, ddr_hidden
);
1192 pci_disable_device(pdev
);
1193 pci_set_drvdata(pdev
, NULL
);
1197 static void __mthca_remove_one(struct pci_dev
*pdev
)
1199 struct mthca_dev
*mdev
= pci_get_drvdata(pdev
);
1204 mthca_free_agents(mdev
);
1205 mthca_unregister_device(mdev
);
1207 for (p
= 1; p
<= mdev
->limits
.num_ports
; ++p
)
1208 mthca_CLOSE_IB(mdev
, p
, &status
);
1210 mthca_cleanup_mcg_table(mdev
);
1211 mthca_cleanup_av_table(mdev
);
1212 mthca_cleanup_qp_table(mdev
);
1213 mthca_cleanup_srq_table(mdev
);
1214 mthca_cleanup_cq_table(mdev
);
1215 mthca_cmd_use_polling(mdev
);
1216 mthca_cleanup_eq_table(mdev
);
1218 mthca_pd_free(mdev
, &mdev
->driver_pd
);
1220 mthca_cleanup_mr_table(mdev
);
1221 mthca_cleanup_pd_table(mdev
);
1224 mthca_uar_free(mdev
, &mdev
->driver_uar
);
1225 mthca_cleanup_uar_table(mdev
);
1226 mthca_close_hca(mdev
);
1227 mthca_cmd_cleanup(mdev
);
1229 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1230 pci_disable_msix(pdev
);
1232 ib_dealloc_device(&mdev
->ib_dev
);
1233 mthca_release_regions(pdev
, mdev
->mthca_flags
&
1234 MTHCA_FLAG_DDR_HIDDEN
);
1235 pci_disable_device(pdev
);
1236 pci_set_drvdata(pdev
, NULL
);
1240 int __mthca_restart_one(struct pci_dev
*pdev
)
1242 struct mthca_dev
*mdev
;
1245 mdev
= pci_get_drvdata(pdev
);
1248 hca_type
= mdev
->hca_type
;
1249 __mthca_remove_one(pdev
);
1250 return __mthca_init_one(pdev
, hca_type
);
1253 static int __devinit
mthca_init_one(struct pci_dev
*pdev
,
1254 const struct pci_device_id
*id
)
1256 static int mthca_version_printed
= 0;
1259 mutex_lock(&mthca_device_mutex
);
1261 if (!mthca_version_printed
) {
1262 printk(KERN_INFO
"%s", mthca_version
);
1263 ++mthca_version_printed
;
1266 if (id
->driver_data
>= ARRAY_SIZE(mthca_hca_table
)) {
1267 printk(KERN_ERR PFX
"%s has invalid driver data %lx\n",
1268 pci_name(pdev
), id
->driver_data
);
1269 mutex_unlock(&mthca_device_mutex
);
1273 ret
= __mthca_init_one(pdev
, id
->driver_data
);
1275 mutex_unlock(&mthca_device_mutex
);
1280 static void __devexit
mthca_remove_one(struct pci_dev
*pdev
)
1282 mutex_lock(&mthca_device_mutex
);
1283 __mthca_remove_one(pdev
);
1284 mutex_unlock(&mthca_device_mutex
);
1287 static struct pci_device_id mthca_pci_table
[] = {
1288 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_TAVOR
),
1289 .driver_data
= TAVOR
},
1290 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_TAVOR
),
1291 .driver_data
= TAVOR
},
1292 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
),
1293 .driver_data
= ARBEL_COMPAT
},
1294 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
),
1295 .driver_data
= ARBEL_COMPAT
},
1296 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_ARBEL
),
1297 .driver_data
= ARBEL_NATIVE
},
1298 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_ARBEL
),
1299 .driver_data
= ARBEL_NATIVE
},
1300 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_SINAI
),
1301 .driver_data
= SINAI
},
1302 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_SINAI
),
1303 .driver_data
= SINAI
},
1304 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_SINAI_OLD
),
1305 .driver_data
= SINAI
},
1306 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_SINAI_OLD
),
1307 .driver_data
= SINAI
},
1311 MODULE_DEVICE_TABLE(pci
, mthca_pci_table
);
1313 static struct pci_driver mthca_driver
= {
1315 .id_table
= mthca_pci_table
,
1316 .probe
= mthca_init_one
,
1317 .remove
= __devexit_p(mthca_remove_one
)
1320 static void __init
__mthca_check_profile_val(const char *name
, int *pval
,
1323 /* value must be positive and power of 2 */
1324 int old_pval
= *pval
;
1327 *pval
= pval_default
;
1329 *pval
= roundup_pow_of_two(old_pval
);
1331 if (old_pval
!= *pval
) {
1332 printk(KERN_WARNING PFX
"Invalid value %d for %s in module parameter.\n",
1334 printk(KERN_WARNING PFX
"Corrected %s to %d.\n", name
, *pval
);
1338 #define mthca_check_profile_val(name, default) \
1339 __mthca_check_profile_val(#name, &hca_profile.name, default)
1341 static void __init
mthca_validate_profile(void)
1343 mthca_check_profile_val(num_qp
, MTHCA_DEFAULT_NUM_QP
);
1344 mthca_check_profile_val(rdb_per_qp
, MTHCA_DEFAULT_RDB_PER_QP
);
1345 mthca_check_profile_val(num_cq
, MTHCA_DEFAULT_NUM_CQ
);
1346 mthca_check_profile_val(num_mcg
, MTHCA_DEFAULT_NUM_MCG
);
1347 mthca_check_profile_val(num_mpt
, MTHCA_DEFAULT_NUM_MPT
);
1348 mthca_check_profile_val(num_mtt
, MTHCA_DEFAULT_NUM_MTT
);
1349 mthca_check_profile_val(num_udav
, MTHCA_DEFAULT_NUM_UDAV
);
1350 mthca_check_profile_val(fmr_reserved_mtts
, MTHCA_DEFAULT_NUM_RESERVED_MTTS
);
1352 if (hca_profile
.fmr_reserved_mtts
>= hca_profile
.num_mtt
) {
1353 printk(KERN_WARNING PFX
"Invalid fmr_reserved_mtts module parameter %d.\n",
1354 hca_profile
.fmr_reserved_mtts
);
1355 printk(KERN_WARNING PFX
"(Must be smaller than num_mtt %d)\n",
1356 hca_profile
.num_mtt
);
1357 hca_profile
.fmr_reserved_mtts
= hca_profile
.num_mtt
/ 2;
1358 printk(KERN_WARNING PFX
"Corrected fmr_reserved_mtts to %d.\n",
1359 hca_profile
.fmr_reserved_mtts
);
1363 static int __init
mthca_init(void)
1367 mthca_validate_profile();
1369 ret
= mthca_catas_init();
1373 ret
= pci_register_driver(&mthca_driver
);
1375 mthca_catas_cleanup();
1382 static void __exit
mthca_cleanup(void)
1384 pci_unregister_driver(&mthca_driver
);
1385 mthca_catas_cleanup();
1388 module_init(mthca_init
);
1389 module_exit(mthca_cleanup
);