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>
41 #include "mthca_dev.h"
42 #include "mthca_config_reg.h"
43 #include "mthca_cmd.h"
44 #include "mthca_profile.h"
45 #include "mthca_memfree.h"
46 #include "mthca_wqe.h"
48 MODULE_AUTHOR("Roland Dreier");
49 MODULE_DESCRIPTION("Mellanox InfiniBand HCA low-level driver");
50 MODULE_LICENSE("Dual BSD/GPL");
51 MODULE_VERSION(DRV_VERSION
);
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
[] __devinitdata
=
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_find_capability(mdev
->pdev
, PCI_CAP_ID_EXP
)) {
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
)
169 mdev
->limits
.mtt_seg_size
= (1 << log_mtts_per_seg
) * 8;
170 err
= mthca_QUERY_DEV_LIM(mdev
, dev_lim
, &status
);
172 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
176 mthca_err(mdev
, "QUERY_DEV_LIM returned status 0x%02x, "
177 "aborting.\n", status
);
180 if (dev_lim
->min_page_sz
> PAGE_SIZE
) {
181 mthca_err(mdev
, "HCA minimum page size of %d bigger than "
182 "kernel PAGE_SIZE of %ld, aborting.\n",
183 dev_lim
->min_page_sz
, PAGE_SIZE
);
186 if (dev_lim
->num_ports
> MTHCA_MAX_PORTS
) {
187 mthca_err(mdev
, "HCA has %d ports, but we only support %d, "
189 dev_lim
->num_ports
, MTHCA_MAX_PORTS
);
193 if (dev_lim
->uar_size
> pci_resource_len(mdev
->pdev
, 2)) {
194 mthca_err(mdev
, "HCA reported UAR size of 0x%x bigger than "
195 "PCI resource 2 size of 0x%llx, aborting.\n",
197 (unsigned long long)pci_resource_len(mdev
->pdev
, 2));
201 mdev
->limits
.num_ports
= dev_lim
->num_ports
;
202 mdev
->limits
.vl_cap
= dev_lim
->max_vl
;
203 mdev
->limits
.mtu_cap
= dev_lim
->max_mtu
;
204 mdev
->limits
.gid_table_len
= dev_lim
->max_gids
;
205 mdev
->limits
.pkey_table_len
= dev_lim
->max_pkeys
;
206 mdev
->limits
.local_ca_ack_delay
= dev_lim
->local_ca_ack_delay
;
208 * Need to allow for worst case send WQE overhead and check
209 * whether max_desc_sz imposes a lower limit than max_sg; UD
210 * send has the biggest overhead.
212 mdev
->limits
.max_sg
= min_t(int, dev_lim
->max_sg
,
213 (dev_lim
->max_desc_sz
-
214 sizeof (struct mthca_next_seg
) -
215 (mthca_is_memfree(mdev
) ?
216 sizeof (struct mthca_arbel_ud_seg
) :
217 sizeof (struct mthca_tavor_ud_seg
))) /
218 sizeof (struct mthca_data_seg
));
219 mdev
->limits
.max_wqes
= dev_lim
->max_qp_sz
;
220 mdev
->limits
.max_qp_init_rdma
= dev_lim
->max_requester_per_qp
;
221 mdev
->limits
.reserved_qps
= dev_lim
->reserved_qps
;
222 mdev
->limits
.max_srq_wqes
= dev_lim
->max_srq_sz
;
223 mdev
->limits
.reserved_srqs
= dev_lim
->reserved_srqs
;
224 mdev
->limits
.reserved_eecs
= dev_lim
->reserved_eecs
;
225 mdev
->limits
.max_desc_sz
= dev_lim
->max_desc_sz
;
226 mdev
->limits
.max_srq_sge
= mthca_max_srq_sge(mdev
);
228 * Subtract 1 from the limit because we need to allocate a
229 * spare CQE so the HCA HW can tell the difference between an
230 * empty CQ and a full CQ.
232 mdev
->limits
.max_cqes
= dev_lim
->max_cq_sz
- 1;
233 mdev
->limits
.reserved_cqs
= dev_lim
->reserved_cqs
;
234 mdev
->limits
.reserved_eqs
= dev_lim
->reserved_eqs
;
235 mdev
->limits
.reserved_mtts
= dev_lim
->reserved_mtts
;
236 mdev
->limits
.reserved_mrws
= dev_lim
->reserved_mrws
;
237 mdev
->limits
.reserved_uars
= dev_lim
->reserved_uars
;
238 mdev
->limits
.reserved_pds
= dev_lim
->reserved_pds
;
239 mdev
->limits
.port_width_cap
= dev_lim
->max_port_width
;
240 mdev
->limits
.page_size_cap
= ~(u32
) (dev_lim
->min_page_sz
- 1);
241 mdev
->limits
.flags
= dev_lim
->flags
;
243 * For old FW that doesn't return static rate support, use a
244 * value of 0x3 (only static rate values of 0 or 1 are handled),
245 * except on Sinai, where even old FW can handle static rate
248 if (dev_lim
->stat_rate_support
)
249 mdev
->limits
.stat_rate_support
= dev_lim
->stat_rate_support
;
250 else if (mdev
->mthca_flags
& MTHCA_FLAG_SINAI_OPT
)
251 mdev
->limits
.stat_rate_support
= 0xf;
253 mdev
->limits
.stat_rate_support
= 0x3;
255 /* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
256 May be doable since hardware supports it for SRQ.
258 IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
260 IB_DEVICE_SRQ_RESIZE is supported by hardware but SRQ is not
261 supported by driver. */
262 mdev
->device_cap_flags
= IB_DEVICE_CHANGE_PHY_PORT
|
263 IB_DEVICE_PORT_ACTIVE_EVENT
|
264 IB_DEVICE_SYS_IMAGE_GUID
|
265 IB_DEVICE_RC_RNR_NAK_GEN
;
267 if (dev_lim
->flags
& DEV_LIM_FLAG_BAD_PKEY_CNTR
)
268 mdev
->device_cap_flags
|= IB_DEVICE_BAD_PKEY_CNTR
;
270 if (dev_lim
->flags
& DEV_LIM_FLAG_BAD_QKEY_CNTR
)
271 mdev
->device_cap_flags
|= IB_DEVICE_BAD_QKEY_CNTR
;
273 if (dev_lim
->flags
& DEV_LIM_FLAG_RAW_MULTI
)
274 mdev
->device_cap_flags
|= IB_DEVICE_RAW_MULTI
;
276 if (dev_lim
->flags
& DEV_LIM_FLAG_AUTO_PATH_MIG
)
277 mdev
->device_cap_flags
|= IB_DEVICE_AUTO_PATH_MIG
;
279 if (dev_lim
->flags
& DEV_LIM_FLAG_UD_AV_PORT_ENFORCE
)
280 mdev
->device_cap_flags
|= IB_DEVICE_UD_AV_PORT_ENFORCE
;
282 if (dev_lim
->flags
& DEV_LIM_FLAG_SRQ
)
283 mdev
->mthca_flags
|= MTHCA_FLAG_SRQ
;
285 if (mthca_is_memfree(mdev
))
286 if (dev_lim
->flags
& DEV_LIM_FLAG_IPOIB_CSUM
)
287 mdev
->device_cap_flags
|= IB_DEVICE_UD_IP_CSUM
;
292 static int mthca_init_tavor(struct mthca_dev
*mdev
)
297 struct mthca_dev_lim dev_lim
;
298 struct mthca_profile profile
;
299 struct mthca_init_hca_param init_hca
;
301 err
= mthca_SYS_EN(mdev
, &status
);
303 mthca_err(mdev
, "SYS_EN command failed, aborting.\n");
307 mthca_err(mdev
, "SYS_EN returned status 0x%02x, "
308 "aborting.\n", status
);
312 err
= mthca_QUERY_FW(mdev
, &status
);
314 mthca_err(mdev
, "QUERY_FW command failed, aborting.\n");
318 mthca_err(mdev
, "QUERY_FW returned status 0x%02x, "
319 "aborting.\n", status
);
323 err
= mthca_QUERY_DDR(mdev
, &status
);
325 mthca_err(mdev
, "QUERY_DDR command failed, aborting.\n");
329 mthca_err(mdev
, "QUERY_DDR returned status 0x%02x, "
330 "aborting.\n", status
);
335 err
= mthca_dev_lim(mdev
, &dev_lim
);
337 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
341 profile
= hca_profile
;
342 profile
.num_uar
= dev_lim
.uar_size
/ PAGE_SIZE
;
343 profile
.uarc_size
= 0;
344 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
345 profile
.num_srq
= dev_lim
.max_srqs
;
347 size
= mthca_make_profile(mdev
, &profile
, &dev_lim
, &init_hca
);
353 err
= mthca_INIT_HCA(mdev
, &init_hca
, &status
);
355 mthca_err(mdev
, "INIT_HCA command failed, aborting.\n");
359 mthca_err(mdev
, "INIT_HCA returned status 0x%02x, "
360 "aborting.\n", status
);
368 mthca_SYS_DIS(mdev
, &status
);
373 static int mthca_load_fw(struct mthca_dev
*mdev
)
378 /* FIXME: use HCA-attached memory for FW if present */
380 mdev
->fw
.arbel
.fw_icm
=
381 mthca_alloc_icm(mdev
, mdev
->fw
.arbel
.fw_pages
,
382 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
383 if (!mdev
->fw
.arbel
.fw_icm
) {
384 mthca_err(mdev
, "Couldn't allocate FW area, aborting.\n");
388 err
= mthca_MAP_FA(mdev
, mdev
->fw
.arbel
.fw_icm
, &status
);
390 mthca_err(mdev
, "MAP_FA command failed, aborting.\n");
394 mthca_err(mdev
, "MAP_FA returned status 0x%02x, aborting.\n", status
);
398 err
= mthca_RUN_FW(mdev
, &status
);
400 mthca_err(mdev
, "RUN_FW command failed, aborting.\n");
404 mthca_err(mdev
, "RUN_FW returned status 0x%02x, aborting.\n", status
);
412 mthca_UNMAP_FA(mdev
, &status
);
415 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
419 static int mthca_init_icm(struct mthca_dev
*mdev
,
420 struct mthca_dev_lim
*dev_lim
,
421 struct mthca_init_hca_param
*init_hca
,
428 err
= mthca_SET_ICM_SIZE(mdev
, icm_size
, &aux_pages
, &status
);
430 mthca_err(mdev
, "SET_ICM_SIZE command failed, aborting.\n");
434 mthca_err(mdev
, "SET_ICM_SIZE returned status 0x%02x, "
435 "aborting.\n", status
);
439 mthca_dbg(mdev
, "%lld KB of HCA context requires %lld KB aux memory.\n",
440 (unsigned long long) icm_size
>> 10,
441 (unsigned long long) aux_pages
<< 2);
443 mdev
->fw
.arbel
.aux_icm
= mthca_alloc_icm(mdev
, aux_pages
,
444 GFP_HIGHUSER
| __GFP_NOWARN
, 0);
445 if (!mdev
->fw
.arbel
.aux_icm
) {
446 mthca_err(mdev
, "Couldn't allocate aux memory, aborting.\n");
450 err
= mthca_MAP_ICM_AUX(mdev
, mdev
->fw
.arbel
.aux_icm
, &status
);
452 mthca_err(mdev
, "MAP_ICM_AUX command failed, aborting.\n");
456 mthca_err(mdev
, "MAP_ICM_AUX returned status 0x%02x, aborting.\n", status
);
461 err
= mthca_map_eq_icm(mdev
, init_hca
->eqc_base
);
463 mthca_err(mdev
, "Failed to map EQ context memory, aborting.\n");
467 /* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
468 mdev
->limits
.reserved_mtts
= ALIGN(mdev
->limits
.reserved_mtts
* mdev
->limits
.mtt_seg_size
,
469 dma_get_cache_alignment()) / mdev
->limits
.mtt_seg_size
;
471 mdev
->mr_table
.mtt_table
= mthca_alloc_icm_table(mdev
, init_hca
->mtt_base
,
472 mdev
->limits
.mtt_seg_size
,
473 mdev
->limits
.num_mtt_segs
,
474 mdev
->limits
.reserved_mtts
,
476 if (!mdev
->mr_table
.mtt_table
) {
477 mthca_err(mdev
, "Failed to map MTT context memory, aborting.\n");
482 mdev
->mr_table
.mpt_table
= mthca_alloc_icm_table(mdev
, init_hca
->mpt_base
,
483 dev_lim
->mpt_entry_sz
,
484 mdev
->limits
.num_mpts
,
485 mdev
->limits
.reserved_mrws
,
487 if (!mdev
->mr_table
.mpt_table
) {
488 mthca_err(mdev
, "Failed to map MPT context memory, aborting.\n");
493 mdev
->qp_table
.qp_table
= mthca_alloc_icm_table(mdev
, init_hca
->qpc_base
,
494 dev_lim
->qpc_entry_sz
,
495 mdev
->limits
.num_qps
,
496 mdev
->limits
.reserved_qps
,
498 if (!mdev
->qp_table
.qp_table
) {
499 mthca_err(mdev
, "Failed to map QP context memory, aborting.\n");
504 mdev
->qp_table
.eqp_table
= mthca_alloc_icm_table(mdev
, init_hca
->eqpc_base
,
505 dev_lim
->eqpc_entry_sz
,
506 mdev
->limits
.num_qps
,
507 mdev
->limits
.reserved_qps
,
509 if (!mdev
->qp_table
.eqp_table
) {
510 mthca_err(mdev
, "Failed to map EQP context memory, aborting.\n");
515 mdev
->qp_table
.rdb_table
= mthca_alloc_icm_table(mdev
, init_hca
->rdb_base
,
516 MTHCA_RDB_ENTRY_SIZE
,
517 mdev
->limits
.num_qps
<<
518 mdev
->qp_table
.rdb_shift
, 0,
520 if (!mdev
->qp_table
.rdb_table
) {
521 mthca_err(mdev
, "Failed to map RDB context memory, aborting\n");
526 mdev
->cq_table
.table
= mthca_alloc_icm_table(mdev
, init_hca
->cqc_base
,
527 dev_lim
->cqc_entry_sz
,
528 mdev
->limits
.num_cqs
,
529 mdev
->limits
.reserved_cqs
,
531 if (!mdev
->cq_table
.table
) {
532 mthca_err(mdev
, "Failed to map CQ context memory, aborting.\n");
537 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
538 mdev
->srq_table
.table
=
539 mthca_alloc_icm_table(mdev
, init_hca
->srqc_base
,
540 dev_lim
->srq_entry_sz
,
541 mdev
->limits
.num_srqs
,
542 mdev
->limits
.reserved_srqs
,
544 if (!mdev
->srq_table
.table
) {
545 mthca_err(mdev
, "Failed to map SRQ context memory, "
553 * It's not strictly required, but for simplicity just map the
554 * whole multicast group table now. The table isn't very big
555 * and it's a lot easier than trying to track ref counts.
557 mdev
->mcg_table
.table
= mthca_alloc_icm_table(mdev
, init_hca
->mc_base
,
558 MTHCA_MGM_ENTRY_SIZE
,
559 mdev
->limits
.num_mgms
+
560 mdev
->limits
.num_amgms
,
561 mdev
->limits
.num_mgms
+
562 mdev
->limits
.num_amgms
,
564 if (!mdev
->mcg_table
.table
) {
565 mthca_err(mdev
, "Failed to map MCG context memory, aborting.\n");
573 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
574 mthca_free_icm_table(mdev
, mdev
->srq_table
.table
);
577 mthca_free_icm_table(mdev
, mdev
->cq_table
.table
);
580 mthca_free_icm_table(mdev
, mdev
->qp_table
.rdb_table
);
583 mthca_free_icm_table(mdev
, mdev
->qp_table
.eqp_table
);
586 mthca_free_icm_table(mdev
, mdev
->qp_table
.qp_table
);
589 mthca_free_icm_table(mdev
, mdev
->mr_table
.mpt_table
);
592 mthca_free_icm_table(mdev
, mdev
->mr_table
.mtt_table
);
595 mthca_unmap_eq_icm(mdev
);
598 mthca_UNMAP_ICM_AUX(mdev
, &status
);
601 mthca_free_icm(mdev
, mdev
->fw
.arbel
.aux_icm
, 0);
606 static void mthca_free_icms(struct mthca_dev
*mdev
)
610 mthca_free_icm_table(mdev
, mdev
->mcg_table
.table
);
611 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
612 mthca_free_icm_table(mdev
, mdev
->srq_table
.table
);
613 mthca_free_icm_table(mdev
, mdev
->cq_table
.table
);
614 mthca_free_icm_table(mdev
, mdev
->qp_table
.rdb_table
);
615 mthca_free_icm_table(mdev
, mdev
->qp_table
.eqp_table
);
616 mthca_free_icm_table(mdev
, mdev
->qp_table
.qp_table
);
617 mthca_free_icm_table(mdev
, mdev
->mr_table
.mpt_table
);
618 mthca_free_icm_table(mdev
, mdev
->mr_table
.mtt_table
);
619 mthca_unmap_eq_icm(mdev
);
621 mthca_UNMAP_ICM_AUX(mdev
, &status
);
622 mthca_free_icm(mdev
, mdev
->fw
.arbel
.aux_icm
, 0);
625 static int mthca_init_arbel(struct mthca_dev
*mdev
)
627 struct mthca_dev_lim dev_lim
;
628 struct mthca_profile profile
;
629 struct mthca_init_hca_param init_hca
;
634 err
= mthca_QUERY_FW(mdev
, &status
);
636 mthca_err(mdev
, "QUERY_FW command failed, aborting.\n");
640 mthca_err(mdev
, "QUERY_FW returned status 0x%02x, "
641 "aborting.\n", status
);
645 err
= mthca_ENABLE_LAM(mdev
, &status
);
647 mthca_err(mdev
, "ENABLE_LAM command failed, aborting.\n");
650 if (status
== MTHCA_CMD_STAT_LAM_NOT_PRE
) {
651 mthca_dbg(mdev
, "No HCA-attached memory (running in MemFree mode)\n");
652 mdev
->mthca_flags
|= MTHCA_FLAG_NO_LAM
;
654 mthca_err(mdev
, "ENABLE_LAM returned status 0x%02x, "
655 "aborting.\n", status
);
659 err
= mthca_load_fw(mdev
);
661 mthca_err(mdev
, "Failed to start FW, aborting.\n");
665 err
= mthca_dev_lim(mdev
, &dev_lim
);
667 mthca_err(mdev
, "QUERY_DEV_LIM command failed, aborting.\n");
671 profile
= hca_profile
;
672 profile
.num_uar
= dev_lim
.uar_size
/ PAGE_SIZE
;
673 profile
.num_udav
= 0;
674 if (mdev
->mthca_flags
& MTHCA_FLAG_SRQ
)
675 profile
.num_srq
= dev_lim
.max_srqs
;
677 icm_size
= mthca_make_profile(mdev
, &profile
, &dev_lim
, &init_hca
);
683 err
= mthca_init_icm(mdev
, &dev_lim
, &init_hca
, icm_size
);
687 err
= mthca_INIT_HCA(mdev
, &init_hca
, &status
);
689 mthca_err(mdev
, "INIT_HCA command failed, aborting.\n");
693 mthca_err(mdev
, "INIT_HCA returned status 0x%02x, "
694 "aborting.\n", status
);
702 mthca_free_icms(mdev
);
705 mthca_UNMAP_FA(mdev
, &status
);
706 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
709 if (!(mdev
->mthca_flags
& MTHCA_FLAG_NO_LAM
))
710 mthca_DISABLE_LAM(mdev
, &status
);
715 static void mthca_close_hca(struct mthca_dev
*mdev
)
719 mthca_CLOSE_HCA(mdev
, 0, &status
);
721 if (mthca_is_memfree(mdev
)) {
722 mthca_free_icms(mdev
);
724 mthca_UNMAP_FA(mdev
, &status
);
725 mthca_free_icm(mdev
, mdev
->fw
.arbel
.fw_icm
, 0);
727 if (!(mdev
->mthca_flags
& MTHCA_FLAG_NO_LAM
))
728 mthca_DISABLE_LAM(mdev
, &status
);
730 mthca_SYS_DIS(mdev
, &status
);
733 static int mthca_init_hca(struct mthca_dev
*mdev
)
737 struct mthca_adapter adapter
;
739 if (mthca_is_memfree(mdev
))
740 err
= mthca_init_arbel(mdev
);
742 err
= mthca_init_tavor(mdev
);
747 err
= mthca_QUERY_ADAPTER(mdev
, &adapter
, &status
);
749 mthca_err(mdev
, "QUERY_ADAPTER command failed, aborting.\n");
753 mthca_err(mdev
, "QUERY_ADAPTER returned status 0x%02x, "
754 "aborting.\n", status
);
759 mdev
->eq_table
.inta_pin
= adapter
.inta_pin
;
760 if (!mthca_is_memfree(mdev
))
761 mdev
->rev_id
= adapter
.revision_id
;
762 memcpy(mdev
->board_id
, adapter
.board_id
, sizeof mdev
->board_id
);
767 mthca_close_hca(mdev
);
771 static int mthca_setup_hca(struct mthca_dev
*dev
)
776 MTHCA_INIT_DOORBELL_LOCK(&dev
->doorbell_lock
);
778 err
= mthca_init_uar_table(dev
);
780 mthca_err(dev
, "Failed to initialize "
781 "user access region table, aborting.\n");
785 err
= mthca_uar_alloc(dev
, &dev
->driver_uar
);
787 mthca_err(dev
, "Failed to allocate driver access region, "
789 goto err_uar_table_free
;
792 dev
->kar
= ioremap(dev
->driver_uar
.pfn
<< PAGE_SHIFT
, PAGE_SIZE
);
794 mthca_err(dev
, "Couldn't map kernel access region, "
800 err
= mthca_init_pd_table(dev
);
802 mthca_err(dev
, "Failed to initialize "
803 "protection domain table, aborting.\n");
807 err
= mthca_init_mr_table(dev
);
809 mthca_err(dev
, "Failed to initialize "
810 "memory region table, aborting.\n");
811 goto err_pd_table_free
;
814 err
= mthca_pd_alloc(dev
, 1, &dev
->driver_pd
);
816 mthca_err(dev
, "Failed to create driver PD, "
818 goto err_mr_table_free
;
821 err
= mthca_init_eq_table(dev
);
823 mthca_err(dev
, "Failed to initialize "
824 "event queue table, aborting.\n");
828 err
= mthca_cmd_use_events(dev
);
830 mthca_err(dev
, "Failed to switch to event-driven "
831 "firmware commands, aborting.\n");
832 goto err_eq_table_free
;
835 err
= mthca_NOP(dev
, &status
);
837 if (dev
->mthca_flags
& MTHCA_FLAG_MSI_X
) {
838 mthca_warn(dev
, "NOP command failed to generate interrupt "
840 dev
->eq_table
.eq
[MTHCA_EQ_CMD
].msi_x_vector
);
841 mthca_warn(dev
, "Trying again with MSI-X disabled.\n");
843 mthca_err(dev
, "NOP command failed to generate interrupt "
844 "(IRQ %d), aborting.\n",
846 mthca_err(dev
, "BIOS or ACPI interrupt routing problem?\n");
852 mthca_dbg(dev
, "NOP command IRQ test passed\n");
854 err
= mthca_init_cq_table(dev
);
856 mthca_err(dev
, "Failed to initialize "
857 "completion queue table, aborting.\n");
861 err
= mthca_init_srq_table(dev
);
863 mthca_err(dev
, "Failed to initialize "
864 "shared receive queue table, aborting.\n");
865 goto err_cq_table_free
;
868 err
= mthca_init_qp_table(dev
);
870 mthca_err(dev
, "Failed to initialize "
871 "queue pair table, aborting.\n");
872 goto err_srq_table_free
;
875 err
= mthca_init_av_table(dev
);
877 mthca_err(dev
, "Failed to initialize "
878 "address vector table, aborting.\n");
879 goto err_qp_table_free
;
882 err
= mthca_init_mcg_table(dev
);
884 mthca_err(dev
, "Failed to initialize "
885 "multicast group table, aborting.\n");
886 goto err_av_table_free
;
892 mthca_cleanup_av_table(dev
);
895 mthca_cleanup_qp_table(dev
);
898 mthca_cleanup_srq_table(dev
);
901 mthca_cleanup_cq_table(dev
);
904 mthca_cmd_use_polling(dev
);
907 mthca_cleanup_eq_table(dev
);
910 mthca_pd_free(dev
, &dev
->driver_pd
);
913 mthca_cleanup_mr_table(dev
);
916 mthca_cleanup_pd_table(dev
);
922 mthca_uar_free(dev
, &dev
->driver_uar
);
925 mthca_cleanup_uar_table(dev
);
929 static int mthca_enable_msi_x(struct mthca_dev
*mdev
)
931 struct msix_entry entries
[3];
934 entries
[0].entry
= 0;
935 entries
[1].entry
= 1;
936 entries
[2].entry
= 2;
938 err
= pci_enable_msix(mdev
->pdev
, entries
, ARRAY_SIZE(entries
));
941 mthca_info(mdev
, "Only %d MSI-X vectors available, "
942 "not using MSI-X\n", err
);
946 mdev
->eq_table
.eq
[MTHCA_EQ_COMP
].msi_x_vector
= entries
[0].vector
;
947 mdev
->eq_table
.eq
[MTHCA_EQ_ASYNC
].msi_x_vector
= entries
[1].vector
;
948 mdev
->eq_table
.eq
[MTHCA_EQ_CMD
].msi_x_vector
= entries
[2].vector
;
953 /* Types of supported HCA */
956 ARBEL_COMPAT
, /* MT25208 in Tavor compat mode */
957 ARBEL_NATIVE
, /* MT25208 with extended features */
961 #define MTHCA_FW_VER(major, minor, subminor) \
962 (((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
967 } mthca_hca_table
[] = {
968 [TAVOR
] = { .latest_fw
= MTHCA_FW_VER(3, 5, 0),
970 [ARBEL_COMPAT
] = { .latest_fw
= MTHCA_FW_VER(4, 8, 200),
971 .flags
= MTHCA_FLAG_PCIE
},
972 [ARBEL_NATIVE
] = { .latest_fw
= MTHCA_FW_VER(5, 3, 0),
973 .flags
= MTHCA_FLAG_MEMFREE
|
975 [SINAI
] = { .latest_fw
= MTHCA_FW_VER(1, 2, 0),
976 .flags
= MTHCA_FLAG_MEMFREE
|
978 MTHCA_FLAG_SINAI_OPT
}
981 static int __mthca_init_one(struct pci_dev
*pdev
, int hca_type
)
985 struct mthca_dev
*mdev
;
987 printk(KERN_INFO PFX
"Initializing %s\n",
990 err
= pci_enable_device(pdev
);
992 dev_err(&pdev
->dev
, "Cannot enable PCI device, "
998 * Check for BARs. We expect 0: 1MB, 2: 8MB, 4: DDR (may not
1001 if (!(pci_resource_flags(pdev
, 0) & IORESOURCE_MEM
) ||
1002 pci_resource_len(pdev
, 0) != 1 << 20) {
1003 dev_err(&pdev
->dev
, "Missing DCS, aborting.\n");
1005 goto err_disable_pdev
;
1007 if (!(pci_resource_flags(pdev
, 2) & IORESOURCE_MEM
)) {
1008 dev_err(&pdev
->dev
, "Missing UAR, aborting.\n");
1010 goto err_disable_pdev
;
1012 if (!(pci_resource_flags(pdev
, 4) & IORESOURCE_MEM
))
1015 err
= pci_request_regions(pdev
, DRV_NAME
);
1017 dev_err(&pdev
->dev
, "Cannot obtain PCI resources, "
1019 goto err_disable_pdev
;
1022 pci_set_master(pdev
);
1024 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
1026 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1027 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
1029 dev_err(&pdev
->dev
, "Can't set PCI DMA mask, aborting.\n");
1033 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64));
1035 dev_warn(&pdev
->dev
, "Warning: couldn't set 64-bit "
1036 "consistent PCI DMA mask.\n");
1037 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(32));
1039 dev_err(&pdev
->dev
, "Can't set consistent PCI DMA mask, "
1045 mdev
= (struct mthca_dev
*) ib_alloc_device(sizeof *mdev
);
1047 dev_err(&pdev
->dev
, "Device struct alloc failed, "
1055 mdev
->mthca_flags
= mthca_hca_table
[hca_type
].flags
;
1057 mdev
->mthca_flags
|= MTHCA_FLAG_DDR_HIDDEN
;
1060 * Now reset the HCA before we touch the PCI capabilities or
1061 * attempt a firmware command, since a boot ROM may have left
1062 * the HCA in an undefined state.
1064 err
= mthca_reset(mdev
);
1066 mthca_err(mdev
, "Failed to reset HCA, aborting.\n");
1070 if (mthca_cmd_init(mdev
)) {
1071 mthca_err(mdev
, "Failed to init command interface, aborting.\n");
1075 err
= mthca_tune_pci(mdev
);
1079 err
= mthca_init_hca(mdev
);
1083 if (mdev
->fw_ver
< mthca_hca_table
[hca_type
].latest_fw
) {
1084 mthca_warn(mdev
, "HCA FW version %d.%d.%03d is old (%d.%d.%03d is current).\n",
1085 (int) (mdev
->fw_ver
>> 32), (int) (mdev
->fw_ver
>> 16) & 0xffff,
1086 (int) (mdev
->fw_ver
& 0xffff),
1087 (int) (mthca_hca_table
[hca_type
].latest_fw
>> 32),
1088 (int) (mthca_hca_table
[hca_type
].latest_fw
>> 16) & 0xffff,
1089 (int) (mthca_hca_table
[hca_type
].latest_fw
& 0xffff));
1090 mthca_warn(mdev
, "If you have problems, try updating your HCA FW.\n");
1093 if (msi_x
&& !mthca_enable_msi_x(mdev
))
1094 mdev
->mthca_flags
|= MTHCA_FLAG_MSI_X
;
1096 err
= mthca_setup_hca(mdev
);
1097 if (err
== -EBUSY
&& (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)) {
1098 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1099 pci_disable_msix(pdev
);
1100 mdev
->mthca_flags
&= ~MTHCA_FLAG_MSI_X
;
1102 err
= mthca_setup_hca(mdev
);
1108 err
= mthca_register_device(mdev
);
1112 err
= mthca_create_agents(mdev
);
1114 goto err_unregister
;
1116 pci_set_drvdata(pdev
, mdev
);
1117 mdev
->hca_type
= hca_type
;
1122 mthca_unregister_device(mdev
);
1125 mthca_cleanup_mcg_table(mdev
);
1126 mthca_cleanup_av_table(mdev
);
1127 mthca_cleanup_qp_table(mdev
);
1128 mthca_cleanup_srq_table(mdev
);
1129 mthca_cleanup_cq_table(mdev
);
1130 mthca_cmd_use_polling(mdev
);
1131 mthca_cleanup_eq_table(mdev
);
1133 mthca_pd_free(mdev
, &mdev
->driver_pd
);
1135 mthca_cleanup_mr_table(mdev
);
1136 mthca_cleanup_pd_table(mdev
);
1137 mthca_cleanup_uar_table(mdev
);
1140 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1141 pci_disable_msix(pdev
);
1143 mthca_close_hca(mdev
);
1146 mthca_cmd_cleanup(mdev
);
1149 ib_dealloc_device(&mdev
->ib_dev
);
1152 pci_release_regions(pdev
);
1155 pci_disable_device(pdev
);
1156 pci_set_drvdata(pdev
, NULL
);
1160 static void __mthca_remove_one(struct pci_dev
*pdev
)
1162 struct mthca_dev
*mdev
= pci_get_drvdata(pdev
);
1167 mthca_free_agents(mdev
);
1168 mthca_unregister_device(mdev
);
1170 for (p
= 1; p
<= mdev
->limits
.num_ports
; ++p
)
1171 mthca_CLOSE_IB(mdev
, p
, &status
);
1173 mthca_cleanup_mcg_table(mdev
);
1174 mthca_cleanup_av_table(mdev
);
1175 mthca_cleanup_qp_table(mdev
);
1176 mthca_cleanup_srq_table(mdev
);
1177 mthca_cleanup_cq_table(mdev
);
1178 mthca_cmd_use_polling(mdev
);
1179 mthca_cleanup_eq_table(mdev
);
1181 mthca_pd_free(mdev
, &mdev
->driver_pd
);
1183 mthca_cleanup_mr_table(mdev
);
1184 mthca_cleanup_pd_table(mdev
);
1187 mthca_uar_free(mdev
, &mdev
->driver_uar
);
1188 mthca_cleanup_uar_table(mdev
);
1189 mthca_close_hca(mdev
);
1190 mthca_cmd_cleanup(mdev
);
1192 if (mdev
->mthca_flags
& MTHCA_FLAG_MSI_X
)
1193 pci_disable_msix(pdev
);
1195 ib_dealloc_device(&mdev
->ib_dev
);
1196 pci_release_regions(pdev
);
1197 pci_disable_device(pdev
);
1198 pci_set_drvdata(pdev
, NULL
);
1202 int __mthca_restart_one(struct pci_dev
*pdev
)
1204 struct mthca_dev
*mdev
;
1207 mdev
= pci_get_drvdata(pdev
);
1210 hca_type
= mdev
->hca_type
;
1211 __mthca_remove_one(pdev
);
1212 return __mthca_init_one(pdev
, hca_type
);
1215 static int __devinit
mthca_init_one(struct pci_dev
*pdev
,
1216 const struct pci_device_id
*id
)
1218 static int mthca_version_printed
= 0;
1221 mutex_lock(&mthca_device_mutex
);
1223 if (!mthca_version_printed
) {
1224 printk(KERN_INFO
"%s", mthca_version
);
1225 ++mthca_version_printed
;
1228 if (id
->driver_data
>= ARRAY_SIZE(mthca_hca_table
)) {
1229 printk(KERN_ERR PFX
"%s has invalid driver data %lx\n",
1230 pci_name(pdev
), id
->driver_data
);
1231 mutex_unlock(&mthca_device_mutex
);
1235 ret
= __mthca_init_one(pdev
, id
->driver_data
);
1237 mutex_unlock(&mthca_device_mutex
);
1242 static void __devexit
mthca_remove_one(struct pci_dev
*pdev
)
1244 mutex_lock(&mthca_device_mutex
);
1245 __mthca_remove_one(pdev
);
1246 mutex_unlock(&mthca_device_mutex
);
1249 static struct pci_device_id mthca_pci_table
[] = {
1250 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_TAVOR
),
1251 .driver_data
= TAVOR
},
1252 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_TAVOR
),
1253 .driver_data
= TAVOR
},
1254 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
),
1255 .driver_data
= ARBEL_COMPAT
},
1256 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
),
1257 .driver_data
= ARBEL_COMPAT
},
1258 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_ARBEL
),
1259 .driver_data
= ARBEL_NATIVE
},
1260 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_ARBEL
),
1261 .driver_data
= ARBEL_NATIVE
},
1262 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_SINAI
),
1263 .driver_data
= SINAI
},
1264 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_SINAI
),
1265 .driver_data
= SINAI
},
1266 { PCI_DEVICE(PCI_VENDOR_ID_MELLANOX
, PCI_DEVICE_ID_MELLANOX_SINAI_OLD
),
1267 .driver_data
= SINAI
},
1268 { PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN
, PCI_DEVICE_ID_MELLANOX_SINAI_OLD
),
1269 .driver_data
= SINAI
},
1273 MODULE_DEVICE_TABLE(pci
, mthca_pci_table
);
1275 static struct pci_driver mthca_driver
= {
1277 .id_table
= mthca_pci_table
,
1278 .probe
= mthca_init_one
,
1279 .remove
= __devexit_p(mthca_remove_one
)
1282 static void __init
__mthca_check_profile_val(const char *name
, int *pval
,
1285 /* value must be positive and power of 2 */
1286 int old_pval
= *pval
;
1289 *pval
= pval_default
;
1291 *pval
= roundup_pow_of_two(old_pval
);
1293 if (old_pval
!= *pval
) {
1294 printk(KERN_WARNING PFX
"Invalid value %d for %s in module parameter.\n",
1296 printk(KERN_WARNING PFX
"Corrected %s to %d.\n", name
, *pval
);
1300 #define mthca_check_profile_val(name, default) \
1301 __mthca_check_profile_val(#name, &hca_profile.name, default)
1303 static void __init
mthca_validate_profile(void)
1305 mthca_check_profile_val(num_qp
, MTHCA_DEFAULT_NUM_QP
);
1306 mthca_check_profile_val(rdb_per_qp
, MTHCA_DEFAULT_RDB_PER_QP
);
1307 mthca_check_profile_val(num_cq
, MTHCA_DEFAULT_NUM_CQ
);
1308 mthca_check_profile_val(num_mcg
, MTHCA_DEFAULT_NUM_MCG
);
1309 mthca_check_profile_val(num_mpt
, MTHCA_DEFAULT_NUM_MPT
);
1310 mthca_check_profile_val(num_mtt
, MTHCA_DEFAULT_NUM_MTT
);
1311 mthca_check_profile_val(num_udav
, MTHCA_DEFAULT_NUM_UDAV
);
1312 mthca_check_profile_val(fmr_reserved_mtts
, MTHCA_DEFAULT_NUM_RESERVED_MTTS
);
1314 if (hca_profile
.fmr_reserved_mtts
>= hca_profile
.num_mtt
) {
1315 printk(KERN_WARNING PFX
"Invalid fmr_reserved_mtts module parameter %d.\n",
1316 hca_profile
.fmr_reserved_mtts
);
1317 printk(KERN_WARNING PFX
"(Must be smaller than num_mtt %d)\n",
1318 hca_profile
.num_mtt
);
1319 hca_profile
.fmr_reserved_mtts
= hca_profile
.num_mtt
/ 2;
1320 printk(KERN_WARNING PFX
"Corrected fmr_reserved_mtts to %d.\n",
1321 hca_profile
.fmr_reserved_mtts
);
1324 if ((log_mtts_per_seg
< 1) || (log_mtts_per_seg
> 5)) {
1325 printk(KERN_WARNING PFX
"bad log_mtts_per_seg (%d). Using default - %d\n",
1326 log_mtts_per_seg
, ilog2(MTHCA_MTT_SEG_SIZE
/ 8));
1327 log_mtts_per_seg
= ilog2(MTHCA_MTT_SEG_SIZE
/ 8);
1331 static int __init
mthca_init(void)
1335 mthca_validate_profile();
1337 ret
= mthca_catas_init();
1341 ret
= pci_register_driver(&mthca_driver
);
1343 mthca_catas_cleanup();
1350 static void __exit
mthca_cleanup(void)
1352 pci_unregister_driver(&mthca_driver
);
1353 mthca_catas_cleanup();
1356 module_init(mthca_init
);
1357 module_exit(mthca_cleanup
);