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");
53 #ifdef CONFIG_INFINIBAND_MTHCA_DEBUG
55 int mthca_debug_level
= 0;
56 module_param_named(debug_level
, mthca_debug_level
, int, 0644);
57 MODULE_PARM_DESC(debug_level
, "Enable debug tracing if > 0");
59 #endif /* CONFIG_INFINIBAND_MTHCA_DEBUG */
64 module_param(msi_x
, int, 0444);
65 MODULE_PARM_DESC(msi_x
, "attempt to use MSI-X if nonzero");
67 #else /* CONFIG_PCI_MSI */
71 #endif /* CONFIG_PCI_MSI */
73 static int tune_pci
= 0;
74 module_param(tune_pci
, int, 0444);
75 MODULE_PARM_DESC(tune_pci
, "increase PCI burst from the default set by BIOS if nonzero");
77 DEFINE_MUTEX(mthca_device_mutex
);
79 #define MTHCA_DEFAULT_NUM_QP (1 << 16)
80 #define MTHCA_DEFAULT_RDB_PER_QP (1 << 2)
81 #define MTHCA_DEFAULT_NUM_CQ (1 << 16)
82 #define MTHCA_DEFAULT_NUM_MCG (1 << 13)
83 #define MTHCA_DEFAULT_NUM_MPT (1 << 17)
84 #define MTHCA_DEFAULT_NUM_MTT (1 << 20)
85 #define MTHCA_DEFAULT_NUM_UDAV (1 << 15)
86 #define MTHCA_DEFAULT_NUM_RESERVED_MTTS (1 << 18)
87 #define MTHCA_DEFAULT_NUM_UARC_SIZE (1 << 18)
89 static struct mthca_profile hca_profile
= {
90 .num_qp
= MTHCA_DEFAULT_NUM_QP
,
91 .rdb_per_qp
= MTHCA_DEFAULT_RDB_PER_QP
,
92 .num_cq
= MTHCA_DEFAULT_NUM_CQ
,
93 .num_mcg
= MTHCA_DEFAULT_NUM_MCG
,
94 .num_mpt
= MTHCA_DEFAULT_NUM_MPT
,
95 .num_mtt
= MTHCA_DEFAULT_NUM_MTT
,
96 .num_udav
= MTHCA_DEFAULT_NUM_UDAV
, /* Tavor only */
97 .fmr_reserved_mtts
= MTHCA_DEFAULT_NUM_RESERVED_MTTS
, /* Tavor only */
98 .uarc_size
= MTHCA_DEFAULT_NUM_UARC_SIZE
, /* Arbel only */
101 module_param_named(num_qp
, hca_profile
.num_qp
, int, 0444);
102 MODULE_PARM_DESC(num_qp
, "maximum number of QPs per HCA");
104 module_param_named(rdb_per_qp
, hca_profile
.rdb_per_qp
, int, 0444);
105 MODULE_PARM_DESC(rdb_per_qp
, "number of RDB buffers per QP");
107 module_param_named(num_cq
, hca_profile
.num_cq
, int, 0444);
108 MODULE_PARM_DESC(num_cq
, "maximum number of CQs per HCA");
110 module_param_named(num_mcg
, hca_profile
.num_mcg
, int, 0444);
111 MODULE_PARM_DESC(num_mcg
, "maximum number of multicast groups per HCA");
113 module_param_named(num_mpt
, hca_profile
.num_mpt
, int, 0444);
114 MODULE_PARM_DESC(num_mpt
,
115 "maximum number of memory protection table entries per HCA");
117 module_param_named(num_mtt
, hca_profile
.num_mtt
, int, 0444);
118 MODULE_PARM_DESC(num_mtt
,
119 "maximum number of memory translation table segments per HCA");
121 module_param_named(num_udav
, hca_profile
.num_udav
, int, 0444);
122 MODULE_PARM_DESC(num_udav
, "maximum number of UD address vectors per HCA");
124 module_param_named(fmr_reserved_mtts
, hca_profile
.fmr_reserved_mtts
, int, 0444);
125 MODULE_PARM_DESC(fmr_reserved_mtts
,
126 "number of memory translation table segments reserved for FMR");
128 static int log_mtts_per_seg
= ilog2(MTHCA_MTT_SEG_SIZE
/ 8);
129 module_param_named(log_mtts_per_seg
, log_mtts_per_seg
, int, 0444);
130 MODULE_PARM_DESC(log_mtts_per_seg
, "Log2 number of MTT entries per segment (1-5)");
132 static char mthca_version
[] =
133 DRV_NAME
": Mellanox InfiniBand HCA driver v"
134 DRV_VERSION
" (" DRV_RELDATE
")\n";
136 static int mthca_tune_pci(struct mthca_dev
*mdev
)
141 /* First try to max out Read Byte Count */
142 if (pci_find_capability(mdev
->pdev
, PCI_CAP_ID_PCIX
)) {
143 if (pcix_set_mmrbc(mdev
->pdev
, pcix_get_max_mmrbc(mdev
->pdev
))) {
144 mthca_err(mdev
, "Couldn't set PCI-X max read count, "
148 } else if (!(mdev
->mthca_flags
& MTHCA_FLAG_PCIE
))
149 mthca_info(mdev
, "No PCI-X capability, not setting RBC.\n");
151 if (pci_is_pcie(mdev
->pdev
)) {
152 if (pcie_set_readrq(mdev
->pdev
, 4096)) {
153 mthca_err(mdev
, "Couldn't write PCI Express read request, "
157 } else if (mdev
->mthca_flags
& MTHCA_FLAG_PCIE
)
158 mthca_info(mdev
, "No PCI Express capability, "
159 "not setting Max Read Request Size.\n");
164 static int mthca_dev_lim(struct mthca_dev
*mdev
, struct mthca_dev_lim
*dev_lim
)
168 mdev
->limits
.mtt_seg_size
= (1 << log_mtts_per_seg
) * 8;
169 err
= mthca_QUERY_DEV_LIM(mdev
, dev_lim
);
171 mthca_err(mdev
, "QUERY_DEV_LIM command returned %d"
172 ", aborting.\n", err
);
175 if (dev_lim
->min_page_sz
> PAGE_SIZE
) {
176 mthca_err(mdev
, "HCA minimum page size of %d bigger than "
177 "kernel PAGE_SIZE of %ld, aborting.\n",
178 dev_lim
->min_page_sz
, PAGE_SIZE
);
181 if (dev_lim
->num_ports
> MTHCA_MAX_PORTS
) {
182 mthca_err(mdev
, "HCA has %d ports, but we only support %d, "
184 dev_lim
->num_ports
, MTHCA_MAX_PORTS
);
188 if (dev_lim
->uar_size
> pci_resource_len(mdev
->pdev
, 2)) {
189 mthca_err(mdev
, "HCA reported UAR size of 0x%x bigger than "
190 "PCI resource 2 size of 0x%llx, aborting.\n",
192 (unsigned long long)pci_resource_len(mdev
->pdev
, 2));
196 mdev
->limits
.num_ports
= dev_lim
->num_ports
;
197 mdev
->limits
.vl_cap
= dev_lim
->max_vl
;
198 mdev
->limits
.mtu_cap
= dev_lim
->max_mtu
;
199 mdev
->limits
.gid_table_len
= dev_lim
->max_gids
;
200 mdev
->limits
.pkey_table_len
= dev_lim
->max_pkeys
;
201 mdev
->limits
.local_ca_ack_delay
= dev_lim
->local_ca_ack_delay
;
203 * Need to allow for worst case send WQE overhead and check
204 * whether max_desc_sz imposes a lower limit than max_sg; UD
205 * send has the biggest overhead.
207 mdev
->limits
.max_sg
= min_t(int, dev_lim
->max_sg
,
208 (dev_lim
->max_desc_sz
-
209 sizeof (struct mthca_next_seg
) -
210 (mthca_is_memfree(mdev
) ?
211 sizeof (struct mthca_arbel_ud_seg
) :
212 sizeof (struct mthca_tavor_ud_seg
))) /
213 sizeof (struct mthca_data_seg
));
214 mdev
->limits
.max_wqes
= dev_lim
->max_qp_sz
;
215 mdev
->limits
.max_qp_init_rdma
= dev_lim
->max_requester_per_qp
;
216 mdev
->limits
.reserved_qps
= dev_lim
->reserved_qps
;
217 mdev
->limits
.max_srq_wqes
= dev_lim
->max_srq_sz
;
218 mdev
->limits
.reserved_srqs
= dev_lim
->reserved_srqs
;
219 mdev
->limits
.reserved_eecs
= dev_lim
->reserved_eecs
;
220 mdev
->limits
.max_desc_sz
= dev_lim
->max_desc_sz
;
221 mdev
->limits
.max_srq_sge
= mthca_max_srq_sge(mdev
);
223 * Subtract 1 from the limit because we need to allocate a
224 * spare CQE so the HCA HW can tell the difference between an
225 * empty CQ and a full CQ.
227 mdev
->limits
.max_cqes
= dev_lim
->max_cq_sz
- 1;
228 mdev
->limits
.reserved_cqs
= dev_lim
->reserved_cqs
;
229 mdev
->limits
.reserved_eqs
= dev_lim
->reserved_eqs
;
230 mdev
->limits
.reserved_mtts
= dev_lim
->reserved_mtts
;
231 mdev
->limits
.reserved_mrws
= dev_lim
->reserved_mrws
;
232 mdev
->limits
.reserved_uars
= dev_lim
->reserved_uars
;
233 mdev
->limits
.reserved_pds
= dev_lim
->reserved_pds
;
234 mdev
->limits
.port_width_cap
= dev_lim
->max_port_width
;
235 mdev
->limits
.page_size_cap
= ~(u32
) (dev_lim
->min_page_sz
- 1);
236 mdev
->limits
.flags
= dev_lim
->flags
;
238 * For old FW that doesn't return static rate support, use a
239 * value of 0x3 (only static rate values of 0 or 1 are handled),
240 * except on Sinai, where even old FW can handle static rate
243 if (dev_lim
->stat_rate_support
)
244 mdev
->limits
.stat_rate_support
= dev_lim
->stat_rate_support
;
245 else if (mdev
->mthca_flags
& MTHCA_FLAG_SINAI_OPT
)
246 mdev
->limits
.stat_rate_support
= 0xf;
248 mdev
->limits
.stat_rate_support
= 0x3;
250 /* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
251 May be doable since hardware supports it for SRQ.
253 IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
255 IB_DEVICE_SRQ_RESIZE is supported by hardware but SRQ is not
256 supported by driver. */
257 mdev
->device_cap_flags
= IB_DEVICE_CHANGE_PHY_PORT
|
258 IB_DEVICE_PORT_ACTIVE_EVENT
|
259 IB_DEVICE_SYS_IMAGE_GUID
|
260 IB_DEVICE_RC_RNR_NAK_GEN
;
262 if (dev_lim
->flags
& DEV_LIM_FLAG_BAD_PKEY_CNTR
)
263 mdev
->device_cap_flags
|= IB_DEVICE_BAD_PKEY_CNTR
;
265 if (dev_lim
->flags
& DEV_LIM_FLAG_BAD_QKEY_CNTR
)
266 mdev
->device_cap_flags
|= IB_DEVICE_BAD_QKEY_CNTR
;
268 if (dev_lim
->flags
& DEV_LIM_FLAG_RAW_MULTI
)
269 mdev
->device_cap_flags
|= IB_DEVICE_RAW_MULTI
;
271 if (dev_lim
->flags
& DEV_LIM_FLAG_AUTO_PATH_MIG
)
272 mdev
->device_cap_flags
|= IB_DEVICE_AUTO_PATH_MIG
;
274 if (dev_lim
->flags
& DEV_LIM_FLAG_UD_AV_PORT_ENFORCE
)
275 mdev
->device_cap_flags
|= IB_DEVICE_UD_AV_PORT_ENFORCE
;
277 if (dev_lim
->flags
& DEV_LIM_FLAG_SRQ
)
278 mdev
->mthca_flags
|= MTHCA_FLAG_SRQ
;
280 if (mthca_is_memfree(mdev
))
281 if (dev_lim
->flags
& DEV_LIM_FLAG_IPOIB_CSUM
)
282 mdev
->device_cap_flags
|= IB_DEVICE_UD_IP_CSUM
;
287 static int mthca_init_tavor(struct mthca_dev
*mdev
)
291 struct mthca_dev_lim dev_lim
;
292 struct mthca_profile profile
;
293 struct mthca_init_hca_param init_hca
;
295 err
= mthca_SYS_EN(mdev
);
297 mthca_err(mdev
, "SYS_EN command returned %d, aborting.\n", err
);
301 err
= mthca_QUERY_FW(mdev
);
303 mthca_err(mdev
, "QUERY_FW command returned %d,"
304 " aborting.\n", err
);
307 err
= mthca_QUERY_DDR(mdev
);
309 mthca_err(mdev
, "QUERY_DDR command returned %d, aborting.\n", err
);
313 err
= mthca_dev_lim(mdev
, &dev_lim
);
315 mthca_err(mdev
, "QUERY_DEV_LIM command returned %d, aborting.\n", err
);
319 profile
= hca_profile
;
320 profile
.num_uar
= dev_lim
.uar_size
/ PAGE_SIZE
;
321 profile
.uarc_size
= 0;
322 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
323 profile
.num_srq
= dev_lim
.max_srqs
;
325 size
= mthca_make_profile(mdev
, &profile
, &dev_lim
, &init_hca
);
331 err
= mthca_INIT_HCA(mdev
, &init_hca
);
333 mthca_err(mdev
, "INIT_HCA command returned %d, aborting.\n", err
);
345 static int mthca_load_fw(struct mthca_dev
*mdev
)
349 /* FIXME: use HCA-attached memory for FW if present */
351 mdev
->fw
.arbel
.fw_icm
=
352 mthca_alloc_icm(mdev
, mdev
->fw
.arbel
.fw_pages
,
353 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
354 if (!mdev
->fw
.arbel
.fw_icm
) {
355 mthca_err(mdev
, "Couldn't allocate FW area, aborting.\n");
359 err
= mthca_MAP_FA(mdev
, mdev
->fw
.arbel
.fw_icm
);
361 mthca_err(mdev
, "MAP_FA command returned %d, aborting.\n", err
);
364 err
= mthca_RUN_FW(mdev
);
366 mthca_err(mdev
, "RUN_FW command returned %d, aborting.\n", err
);
373 mthca_UNMAP_FA(mdev
);
376 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
380 static int mthca_init_icm(struct mthca_dev
*mdev
,
381 struct mthca_dev_lim
*dev_lim
,
382 struct mthca_init_hca_param
*init_hca
,
388 err
= mthca_SET_ICM_SIZE(mdev
, icm_size
, &aux_pages
);
390 mthca_err(mdev
, "SET_ICM_SIZE command returned %d, aborting.\n", err
);
394 mthca_dbg(mdev
, "%lld KB of HCA context requires %lld KB aux memory.\n",
395 (unsigned long long) icm_size
>> 10,
396 (unsigned long long) aux_pages
<< 2);
398 mdev
->fw
.arbel
.aux_icm
= mthca_alloc_icm(mdev
, aux_pages
,
399 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
400 if (!mdev
->fw
.arbel
.aux_icm
) {
401 mthca_err(mdev
, "Couldn't allocate aux memory, aborting.\n");
405 err
= mthca_MAP_ICM_AUX(mdev
, mdev
->fw
.arbel
.aux_icm
);
407 mthca_err(mdev
, "MAP_ICM_AUX returned %d, aborting.\n", err
);
411 err
= mthca_map_eq_icm(mdev
, init_hca
->eqc_base
);
413 mthca_err(mdev
, "Failed to map EQ context memory, aborting.\n");
417 /* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
418 mdev
->limits
.reserved_mtts
= ALIGN(mdev
->limits
.reserved_mtts
* mdev
->limits
.mtt_seg_size
,
419 dma_get_cache_alignment()) / mdev
->limits
.mtt_seg_size
;
421 mdev
->mr_table
.mtt_table
= mthca_alloc_icm_table(mdev
, init_hca
->mtt_base
,
422 mdev
->limits
.mtt_seg_size
,
423 mdev
->limits
.num_mtt_segs
,
424 mdev
->limits
.reserved_mtts
,
426 if (!mdev
->mr_table
.mtt_table
) {
427 mthca_err(mdev
, "Failed to map MTT context memory, aborting.\n");
432 mdev
->mr_table
.mpt_table
= mthca_alloc_icm_table(mdev
, init_hca
->mpt_base
,
433 dev_lim
->mpt_entry_sz
,
434 mdev
->limits
.num_mpts
,
435 mdev
->limits
.reserved_mrws
,
437 if (!mdev
->mr_table
.mpt_table
) {
438 mthca_err(mdev
, "Failed to map MPT context memory, aborting.\n");
443 mdev
->qp_table
.qp_table
= mthca_alloc_icm_table(mdev
, init_hca
->qpc_base
,
444 dev_lim
->qpc_entry_sz
,
445 mdev
->limits
.num_qps
,
446 mdev
->limits
.reserved_qps
,
448 if (!mdev
->qp_table
.qp_table
) {
449 mthca_err(mdev
, "Failed to map QP context memory, aborting.\n");
454 mdev
->qp_table
.eqp_table
= mthca_alloc_icm_table(mdev
, init_hca
->eqpc_base
,
455 dev_lim
->eqpc_entry_sz
,
456 mdev
->limits
.num_qps
,
457 mdev
->limits
.reserved_qps
,
459 if (!mdev
->qp_table
.eqp_table
) {
460 mthca_err(mdev
, "Failed to map EQP context memory, aborting.\n");
465 mdev
->qp_table
.rdb_table
= mthca_alloc_icm_table(mdev
, init_hca
->rdb_base
,
466 MTHCA_RDB_ENTRY_SIZE
,
467 mdev
->limits
.num_qps
<<
468 mdev
->qp_table
.rdb_shift
, 0,
470 if (!mdev
->qp_table
.rdb_table
) {
471 mthca_err(mdev
, "Failed to map RDB context memory, aborting\n");
476 mdev
->cq_table
.table
= mthca_alloc_icm_table(mdev
, init_hca
->cqc_base
,
477 dev_lim
->cqc_entry_sz
,
478 mdev
->limits
.num_cqs
,
479 mdev
->limits
.reserved_cqs
,
481 if (!mdev
->cq_table
.table
) {
482 mthca_err(mdev
, "Failed to map CQ context memory, aborting.\n");
487 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
488 mdev
->srq_table
.table
=
489 mthca_alloc_icm_table(mdev
, init_hca
->srqc_base
,
490 dev_lim
->srq_entry_sz
,
491 mdev
->limits
.num_srqs
,
492 mdev
->limits
.reserved_srqs
,
494 if (!mdev
->srq_table
.table
) {
495 mthca_err(mdev
, "Failed to map SRQ context memory, "
503 * It's not strictly required, but for simplicity just map the
504 * whole multicast group table now. The table isn't very big
505 * and it's a lot easier than trying to track ref counts.
507 mdev
->mcg_table
.table
= mthca_alloc_icm_table(mdev
, init_hca
->mc_base
,
508 MTHCA_MGM_ENTRY_SIZE
,
509 mdev
->limits
.num_mgms
+
510 mdev
->limits
.num_amgms
,
511 mdev
->limits
.num_mgms
+
512 mdev
->limits
.num_amgms
,
514 if (!mdev
->mcg_table
.table
) {
515 mthca_err(mdev
, "Failed to map MCG context memory, aborting.\n");
523 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
524 mthca_free_icm_table(mdev
, mdev
->srq_table
.table
);
527 mthca_free_icm_table(mdev
, mdev
->cq_table
.table
);
530 mthca_free_icm_table(mdev
, mdev
->qp_table
.rdb_table
);
533 mthca_free_icm_table(mdev
, mdev
->qp_table
.eqp_table
);
536 mthca_free_icm_table(mdev
, mdev
->qp_table
.qp_table
);
539 mthca_free_icm_table(mdev
, mdev
->mr_table
.mpt_table
);
542 mthca_free_icm_table(mdev
, mdev
->mr_table
.mtt_table
);
545 mthca_unmap_eq_icm(mdev
);
548 mthca_UNMAP_ICM_AUX(mdev
);
551 mthca_free_icm(mdev
, mdev
->fw
.arbel
.aux_icm
, 0);
556 static void mthca_free_icms(struct mthca_dev
*mdev
)
559 mthca_free_icm_table(mdev
, mdev
->mcg_table
.table
);
560 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
561 mthca_free_icm_table(mdev
, mdev
->srq_table
.table
);
562 mthca_free_icm_table(mdev
, mdev
->cq_table
.table
);
563 mthca_free_icm_table(mdev
, mdev
->qp_table
.rdb_table
);
564 mthca_free_icm_table(mdev
, mdev
->qp_table
.eqp_table
);
565 mthca_free_icm_table(mdev
, mdev
->qp_table
.qp_table
);
566 mthca_free_icm_table(mdev
, mdev
->mr_table
.mpt_table
);
567 mthca_free_icm_table(mdev
, mdev
->mr_table
.mtt_table
);
568 mthca_unmap_eq_icm(mdev
);
570 mthca_UNMAP_ICM_AUX(mdev
);
571 mthca_free_icm(mdev
, mdev
->fw
.arbel
.aux_icm
, 0);
574 static int mthca_init_arbel(struct mthca_dev
*mdev
)
576 struct mthca_dev_lim dev_lim
;
577 struct mthca_profile profile
;
578 struct mthca_init_hca_param init_hca
;
582 err
= mthca_QUERY_FW(mdev
);
584 mthca_err(mdev
, "QUERY_FW command failed %d, aborting.\n", err
);
588 err
= mthca_ENABLE_LAM(mdev
);
589 if (err
== -EAGAIN
) {
590 mthca_dbg(mdev
, "No HCA-attached memory (running in MemFree mode)\n");
591 mdev
->mthca_flags
|= MTHCA_FLAG_NO_LAM
;
593 mthca_err(mdev
, "ENABLE_LAM returned %d, aborting.\n", err
);
597 err
= mthca_load_fw(mdev
);
599 mthca_err(mdev
, "Loading FW returned %d, aborting.\n", err
);
603 err
= mthca_dev_lim(mdev
, &dev_lim
);
605 mthca_err(mdev
, "QUERY_DEV_LIM returned %d, aborting.\n", err
);
609 profile
= hca_profile
;
610 profile
.num_uar
= dev_lim
.uar_size
/ PAGE_SIZE
;
611 profile
.num_udav
= 0;
612 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
613 profile
.num_srq
= dev_lim
.max_srqs
;
615 icm_size
= mthca_make_profile(mdev
, &profile
, &dev_lim
, &init_hca
);
621 err
= mthca_init_icm(mdev
, &dev_lim
, &init_hca
, icm_size
);
625 err
= mthca_INIT_HCA(mdev
, &init_hca
);
627 mthca_err(mdev
, "INIT_HCA command returned %d, aborting.\n", err
);
634 mthca_free_icms(mdev
);
637 mthca_UNMAP_FA(mdev
);
638 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
641 if (!(mdev
->mthca_flags
& MTHCA_FLAG_NO_LAM
))
642 mthca_DISABLE_LAM(mdev
);
647 static void mthca_close_hca(struct mthca_dev
*mdev
)
649 mthca_CLOSE_HCA(mdev
, 0);
651 if (mthca_is_memfree(mdev
)) {
652 mthca_free_icms(mdev
);
654 mthca_UNMAP_FA(mdev
);
655 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
657 if (!(mdev
->mthca_flags
& MTHCA_FLAG_NO_LAM
))
658 mthca_DISABLE_LAM(mdev
);
663 static int mthca_init_hca(struct mthca_dev
*mdev
)
666 struct mthca_adapter adapter
;
668 if (mthca_is_memfree(mdev
))
669 err
= mthca_init_arbel(mdev
);
671 err
= mthca_init_tavor(mdev
);
676 err
= mthca_QUERY_ADAPTER(mdev
, &adapter
);
678 mthca_err(mdev
, "QUERY_ADAPTER command returned %d, aborting.\n", err
);
682 mdev
->eq_table
.inta_pin
= adapter
.inta_pin
;
683 if (!mthca_is_memfree(mdev
))
684 mdev
->rev_id
= adapter
.revision_id
;
685 memcpy(mdev
->board_id
, adapter
.board_id
, sizeof mdev
->board_id
);
690 mthca_close_hca(mdev
);
694 static int mthca_setup_hca(struct mthca_dev
*dev
)
698 MTHCA_INIT_DOORBELL_LOCK(&dev
->doorbell_lock
);
700 err
= mthca_init_uar_table(dev
);
702 mthca_err(dev
, "Failed to initialize "
703 "user access region table, aborting.\n");
707 err
= mthca_uar_alloc(dev
, &dev
->driver_uar
);
709 mthca_err(dev
, "Failed to allocate driver access region, "
711 goto err_uar_table_free
;
714 dev
->kar
= ioremap((phys_addr_t
) dev
->driver_uar
.pfn
<< PAGE_SHIFT
, PAGE_SIZE
);
716 mthca_err(dev
, "Couldn't map kernel access region, "
722 err
= mthca_init_pd_table(dev
);
724 mthca_err(dev
, "Failed to initialize "
725 "protection domain table, aborting.\n");
729 err
= mthca_init_mr_table(dev
);
731 mthca_err(dev
, "Failed to initialize "
732 "memory region table, aborting.\n");
733 goto err_pd_table_free
;
736 err
= mthca_pd_alloc(dev
, 1, &dev
->driver_pd
);
738 mthca_err(dev
, "Failed to create driver PD, "
740 goto err_mr_table_free
;
743 err
= mthca_init_eq_table(dev
);
745 mthca_err(dev
, "Failed to initialize "
746 "event queue table, aborting.\n");
750 err
= mthca_cmd_use_events(dev
);
752 mthca_err(dev
, "Failed to switch to event-driven "
753 "firmware commands, aborting.\n");
754 goto err_eq_table_free
;
757 err
= mthca_NOP(dev
);
759 if (dev
->mthca_flags
& MTHCA_FLAG_MSI_X
) {
760 mthca_warn(dev
, "NOP command failed to generate interrupt "
762 dev
->eq_table
.eq
[MTHCA_EQ_CMD
].msi_x_vector
);
763 mthca_warn(dev
, "Trying again with MSI-X disabled.\n");
765 mthca_err(dev
, "NOP command failed to generate interrupt "
766 "(IRQ %d), aborting.\n",
768 mthca_err(dev
, "BIOS or ACPI interrupt routing problem?\n");
774 mthca_dbg(dev
, "NOP command IRQ test passed\n");
776 err
= mthca_init_cq_table(dev
);
778 mthca_err(dev
, "Failed to initialize "
779 "completion queue table, aborting.\n");
783 err
= mthca_init_srq_table(dev
);
785 mthca_err(dev
, "Failed to initialize "
786 "shared receive queue table, aborting.\n");
787 goto err_cq_table_free
;
790 err
= mthca_init_qp_table(dev
);
792 mthca_err(dev
, "Failed to initialize "
793 "queue pair table, aborting.\n");
794 goto err_srq_table_free
;
797 err
= mthca_init_av_table(dev
);
799 mthca_err(dev
, "Failed to initialize "
800 "address vector table, aborting.\n");
801 goto err_qp_table_free
;
804 err
= mthca_init_mcg_table(dev
);
806 mthca_err(dev
, "Failed to initialize "
807 "multicast group table, aborting.\n");
808 goto err_av_table_free
;
814 mthca_cleanup_av_table(dev
);
817 mthca_cleanup_qp_table(dev
);
820 mthca_cleanup_srq_table(dev
);
823 mthca_cleanup_cq_table(dev
);
826 mthca_cmd_use_polling(dev
);
829 mthca_cleanup_eq_table(dev
);
832 mthca_pd_free(dev
, &dev
->driver_pd
);
835 mthca_cleanup_mr_table(dev
);
838 mthca_cleanup_pd_table(dev
);
844 mthca_uar_free(dev
, &dev
->driver_uar
);
847 mthca_cleanup_uar_table(dev
);
851 static int mthca_enable_msi_x(struct mthca_dev
*mdev
)
855 err
= pci_alloc_irq_vectors(mdev
->pdev
, 3, 3, PCI_IRQ_MSIX
);
859 mdev
->eq_table
.eq
[MTHCA_EQ_COMP
].msi_x_vector
=
860 pci_irq_vector(mdev
->pdev
, 0);
861 mdev
->eq_table
.eq
[MTHCA_EQ_ASYNC
].msi_x_vector
=
862 pci_irq_vector(mdev
->pdev
, 1);
863 mdev
->eq_table
.eq
[MTHCA_EQ_CMD
].msi_x_vector
=
864 pci_irq_vector(mdev
->pdev
, 2);
869 /* Types of supported HCA */
872 ARBEL_COMPAT
, /* MT25208 in Tavor compat mode */
873 ARBEL_NATIVE
, /* MT25208 with extended features */
877 #define MTHCA_FW_VER(major, minor, subminor) \
878 (((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
883 } mthca_hca_table
[] = {
884 [TAVOR
] = { .latest_fw
= MTHCA_FW_VER(3, 5, 0),
886 [ARBEL_COMPAT
] = { .latest_fw
= MTHCA_FW_VER(4, 8, 200),
887 .flags
= MTHCA_FLAG_PCIE
},
888 [ARBEL_NATIVE
] = { .latest_fw
= MTHCA_FW_VER(5, 3, 0),
889 .flags
= MTHCA_FLAG_MEMFREE
|
891 [SINAI
] = { .latest_fw
= MTHCA_FW_VER(1, 2, 0),
892 .flags
= MTHCA_FLAG_MEMFREE
|
894 MTHCA_FLAG_SINAI_OPT
}
897 static int __mthca_init_one(struct pci_dev
*pdev
, int hca_type
)
901 struct mthca_dev
*mdev
;
903 printk(KERN_INFO PFX
"Initializing %s\n",
906 err
= pci_enable_device(pdev
);
908 dev_err(&pdev
->dev
, "Cannot enable PCI device, "
914 * Check for BARs. We expect 0: 1MB, 2: 8MB, 4: DDR (may not
917 if (!(pci_resource_flags(pdev
, 0) & IORESOURCE_MEM
) ||
918 pci_resource_len(pdev
, 0) != 1 << 20) {
919 dev_err(&pdev
->dev
, "Missing DCS, aborting.\n");
921 goto err_disable_pdev
;
923 if (!(pci_resource_flags(pdev
, 2) & IORESOURCE_MEM
)) {
924 dev_err(&pdev
->dev
, "Missing UAR, aborting.\n");
926 goto err_disable_pdev
;
928 if (!(pci_resource_flags(pdev
, 4) & IORESOURCE_MEM
))
931 err
= pci_request_regions(pdev
, DRV_NAME
);
933 dev_err(&pdev
->dev
, "Cannot obtain PCI resources, "
935 goto err_disable_pdev
;
938 pci_set_master(pdev
);
940 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
942 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit PCI DMA mask.\n");
943 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
945 dev_err(&pdev
->dev
, "Can't set PCI DMA mask, aborting.\n");
949 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
951 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit "
952 "consistent PCI DMA mask.\n");
953 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(32));
955 dev_err(&pdev
->dev
, "Can't set consistent PCI DMA mask, "
961 /* We can handle large RDMA requests, so allow larger segments. */
962 dma_set_max_seg_size(&pdev
->dev
, 1024 * 1024 * 1024);
964 mdev
= ib_alloc_device(mthca_dev
, ib_dev
);
966 dev_err(&pdev
->dev
, "Device struct alloc failed, "
974 mdev
->mthca_flags
= mthca_hca_table
[hca_type
].flags
;
976 mdev
->mthca_flags
|= MTHCA_FLAG_DDR_HIDDEN
;
979 * Now reset the HCA before we touch the PCI capabilities or
980 * attempt a firmware command, since a boot ROM may have left
981 * the HCA in an undefined state.
983 err
= mthca_reset(mdev
);
985 mthca_err(mdev
, "Failed to reset HCA, aborting.\n");
989 err
= mthca_cmd_init(mdev
);
991 mthca_err(mdev
, "Failed to init command interface, aborting.\n");
995 err
= mthca_tune_pci(mdev
);
999 err
= mthca_init_hca(mdev
);
1003 if (mdev
->fw_ver
< mthca_hca_table
[hca_type
].latest_fw
) {
1004 mthca_warn(mdev
, "HCA FW version %d.%d.%03d is old (%d.%d.%03d is current).\n",
1005 (int) (mdev
->fw_ver
>> 32), (int) (mdev
->fw_ver
>> 16) & 0xffff,
1006 (int) (mdev
->fw_ver
& 0xffff),
1007 (int) (mthca_hca_table
[hca_type
].latest_fw
>> 32),
1008 (int) (mthca_hca_table
[hca_type
].latest_fw
>> 16) & 0xffff,
1009 (int) (mthca_hca_table
[hca_type
].latest_fw
& 0xffff));
1010 mthca_warn(mdev
, "If you have problems, try updating your HCA FW.\n");
1013 if (msi_x
&& !mthca_enable_msi_x(mdev
))
1014 mdev
->mthca_flags
|= MTHCA_FLAG_MSI_X
;
1016 err
= mthca_setup_hca(mdev
);
1017 if (err
== -EBUSY
&& (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)) {
1018 pci_free_irq_vectors(pdev
);
1019 mdev
->mthca_flags
&= ~MTHCA_FLAG_MSI_X
;
1021 err
= mthca_setup_hca(mdev
);
1027 err
= mthca_register_device(mdev
);
1031 err
= mthca_create_agents(mdev
);
1033 goto err_unregister
;
1035 pci_set_drvdata(pdev
, mdev
);
1036 mdev
->hca_type
= hca_type
;
1038 mdev
->active
= true;
1043 mthca_unregister_device(mdev
);
1046 mthca_cleanup_mcg_table(mdev
);
1047 mthca_cleanup_av_table(mdev
);
1048 mthca_cleanup_qp_table(mdev
);
1049 mthca_cleanup_srq_table(mdev
);
1050 mthca_cleanup_cq_table(mdev
);
1051 mthca_cmd_use_polling(mdev
);
1052 mthca_cleanup_eq_table(mdev
);
1054 mthca_pd_free(mdev
, &mdev
->driver_pd
);
1056 mthca_cleanup_mr_table(mdev
);
1057 mthca_cleanup_pd_table(mdev
);
1058 mthca_cleanup_uar_table(mdev
);
1061 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1062 pci_free_irq_vectors(pdev
);
1064 mthca_close_hca(mdev
);
1067 mthca_cmd_cleanup(mdev
);
1070 ib_dealloc_device(&mdev
->ib_dev
);
1073 pci_release_regions(pdev
);
1076 pci_disable_device(pdev
);
1077 pci_set_drvdata(pdev
, NULL
);
1081 static void __mthca_remove_one(struct pci_dev
*pdev
)
1083 struct mthca_dev
*mdev
= pci_get_drvdata(pdev
);
1087 mthca_free_agents(mdev
);
1088 mthca_unregister_device(mdev
);
1090 for (p
= 1; p
<= mdev
->limits
.num_ports
; ++p
)
1091 mthca_CLOSE_IB(mdev
, p
);
1093 mthca_cleanup_mcg_table(mdev
);
1094 mthca_cleanup_av_table(mdev
);
1095 mthca_cleanup_qp_table(mdev
);
1096 mthca_cleanup_srq_table(mdev
);
1097 mthca_cleanup_cq_table(mdev
);
1098 mthca_cmd_use_polling(mdev
);
1099 mthca_cleanup_eq_table(mdev
);
1101 mthca_pd_free(mdev
, &mdev
->driver_pd
);
1103 mthca_cleanup_mr_table(mdev
);
1104 mthca_cleanup_pd_table(mdev
);
1107 mthca_uar_free(mdev
, &mdev
->driver_uar
);
1108 mthca_cleanup_uar_table(mdev
);
1109 mthca_close_hca(mdev
);
1110 mthca_cmd_cleanup(mdev
);
1112 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1113 pci_free_irq_vectors(pdev
);
1115 ib_dealloc_device(&mdev
->ib_dev
);
1116 pci_release_regions(pdev
);
1117 pci_disable_device(pdev
);
1118 pci_set_drvdata(pdev
, NULL
);
1122 int __mthca_restart_one(struct pci_dev
*pdev
)
1124 struct mthca_dev
*mdev
;
1127 mdev
= pci_get_drvdata(pdev
);
1130 hca_type
= mdev
->hca_type
;
1131 __mthca_remove_one(pdev
);
1132 return __mthca_init_one(pdev
, hca_type
);
1135 static int mthca_init_one(struct pci_dev
*pdev
, const struct pci_device_id
*id
)
1139 mutex_lock(&mthca_device_mutex
);
1141 printk_once(KERN_INFO
"%s", mthca_version
);
1143 if (id
->driver_data
>= ARRAY_SIZE(mthca_hca_table
)) {
1144 printk(KERN_ERR PFX
"%s has invalid driver data %lx\n",
1145 pci_name(pdev
), id
->driver_data
);
1146 mutex_unlock(&mthca_device_mutex
);
1150 ret
= __mthca_init_one(pdev
, id
->driver_data
);
1152 mutex_unlock(&mthca_device_mutex
);
1157 static void mthca_remove_one(struct pci_dev
*pdev
)
1159 mutex_lock(&mthca_device_mutex
);
1160 __mthca_remove_one(pdev
);
1161 mutex_unlock(&mthca_device_mutex
);
1164 static const struct pci_device_id mthca_pci_table
[] = {
1165 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_TAVOR
),
1166 .driver_data
= TAVOR
},
1167 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_TAVOR
),
1168 .driver_data
= TAVOR
},
1169 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
),
1170 .driver_data
= ARBEL_COMPAT
},
1171 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
),
1172 .driver_data
= ARBEL_COMPAT
},
1173 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_ARBEL
),
1174 .driver_data
= ARBEL_NATIVE
},
1175 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_ARBEL
),
1176 .driver_data
= ARBEL_NATIVE
},
1177 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_SINAI
),
1178 .driver_data
= SINAI
},
1179 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_SINAI
),
1180 .driver_data
= SINAI
},
1181 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_SINAI_OLD
),
1182 .driver_data
= SINAI
},
1183 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_SINAI_OLD
),
1184 .driver_data
= SINAI
},
1188 MODULE_DEVICE_TABLE(pci
, mthca_pci_table
);
1190 static struct pci_driver mthca_driver
= {
1192 .id_table
= mthca_pci_table
,
1193 .probe
= mthca_init_one
,
1194 .remove
= mthca_remove_one
,
1197 static void __init
__mthca_check_profile_val(const char *name
, int *pval
,
1200 /* value must be positive and power of 2 */
1201 int old_pval
= *pval
;
1204 *pval
= pval_default
;
1206 *pval
= roundup_pow_of_two(old_pval
);
1208 if (old_pval
!= *pval
) {
1209 printk(KERN_WARNING PFX
"Invalid value %d for %s in module parameter.\n",
1211 printk(KERN_WARNING PFX
"Corrected %s to %d.\n", name
, *pval
);
1215 #define mthca_check_profile_val(name, default) \
1216 __mthca_check_profile_val(#name, &hca_profile.name, default)
1218 static void __init
mthca_validate_profile(void)
1220 mthca_check_profile_val(num_qp
, MTHCA_DEFAULT_NUM_QP
);
1221 mthca_check_profile_val(rdb_per_qp
, MTHCA_DEFAULT_RDB_PER_QP
);
1222 mthca_check_profile_val(num_cq
, MTHCA_DEFAULT_NUM_CQ
);
1223 mthca_check_profile_val(num_mcg
, MTHCA_DEFAULT_NUM_MCG
);
1224 mthca_check_profile_val(num_mpt
, MTHCA_DEFAULT_NUM_MPT
);
1225 mthca_check_profile_val(num_mtt
, MTHCA_DEFAULT_NUM_MTT
);
1226 mthca_check_profile_val(num_udav
, MTHCA_DEFAULT_NUM_UDAV
);
1227 mthca_check_profile_val(fmr_reserved_mtts
, MTHCA_DEFAULT_NUM_RESERVED_MTTS
);
1229 if (hca_profile
.fmr_reserved_mtts
>= hca_profile
.num_mtt
) {
1230 printk(KERN_WARNING PFX
"Invalid fmr_reserved_mtts module parameter %d.\n",
1231 hca_profile
.fmr_reserved_mtts
);
1232 printk(KERN_WARNING PFX
"(Must be smaller than num_mtt %d)\n",
1233 hca_profile
.num_mtt
);
1234 hca_profile
.fmr_reserved_mtts
= hca_profile
.num_mtt
/ 2;
1235 printk(KERN_WARNING PFX
"Corrected fmr_reserved_mtts to %d.\n",
1236 hca_profile
.fmr_reserved_mtts
);
1239 if ((log_mtts_per_seg
< 1) || (log_mtts_per_seg
> 5)) {
1240 printk(KERN_WARNING PFX
"bad log_mtts_per_seg (%d). Using default - %d\n",
1241 log_mtts_per_seg
, ilog2(MTHCA_MTT_SEG_SIZE
/ 8));
1242 log_mtts_per_seg
= ilog2(MTHCA_MTT_SEG_SIZE
/ 8);
1246 static int __init
mthca_init(void)
1250 mthca_validate_profile();
1252 ret
= mthca_catas_init();
1256 ret
= pci_register_driver(&mthca_driver
);
1258 mthca_catas_cleanup();
1265 static void __exit
mthca_cleanup(void)
1267 pci_unregister_driver(&mthca_driver
);
1268 mthca_catas_cleanup();
1271 module_init(mthca_init
);
1272 module_exit(mthca_cleanup
);