1 /*
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.
5  *
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:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
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.
24  *
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
32  * SOFTWARE.
33  *
34  * $Id: mthca_main.c 1396 2004-12-28 04:10:27Z roland $
35  */
36 
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>
42 
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"
48 
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);
53 
54 #ifdef CONFIG_INFINIBAND_MTHCA_DEBUG
55 
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");
59 
60 #endif /* CONFIG_INFINIBAND_MTHCA_DEBUG */
61 
62 #ifdef CONFIG_PCI_MSI
63 
64 static int msi_x = 1;
65 module_param(msi_x, int, 0444);
66 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
67 
68 #else /* CONFIG_PCI_MSI */
69 
70 #define msi_x (0)
71 
72 #endif /* CONFIG_PCI_MSI */
73 
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");
77 
78 DEFINE_MUTEX(mthca_device_mutex);
79 
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)
89 
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 */
100 };
101 
102 module_param_named(num_qp, hca_profile.num_qp, int, 0444);
103 MODULE_PARM_DESC(num_qp, "maximum number of QPs per HCA");
104 
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");
107 
108 module_param_named(num_cq, hca_profile.num_cq, int, 0444);
109 MODULE_PARM_DESC(num_cq, "maximum number of CQs per HCA");
110 
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");
113 
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");
117 
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");
121 
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");
124 
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");
128 
129 static const char mthca_version[] __devinitdata =
130 	DRV_NAME ": Mellanox InfiniBand HCA driver v"
131 	DRV_VERSION " (" DRV_RELDATE ")\n";
132 
133 static int mthca_tune_pci(struct mthca_dev *mdev)
134 {
135 	if (!tune_pci)
136 		return 0;
137 
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, "
142 				"aborting.\n");
143 			return -ENODEV;
144 		}
145 	} else if (!(mdev->mthca_flags & MTHCA_FLAG_PCIE))
146 		mthca_info(mdev, "No PCI-X capability, not setting RBC.\n");
147 
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, "
151 				"aborting.\n");
152 			return -ENODEV;
153 		}
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");
157 
158 	return 0;
159 }
160 
161 static int mthca_dev_lim(struct mthca_dev *mdev, struct mthca_dev_lim *dev_lim)
162 {
163 	int err;
164 	u8 status;
165 
166 	err = mthca_QUERY_DEV_LIM(mdev, dev_lim, &status);
167 	if (err) {
168 		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
169 		return err;
170 	}
171 	if (status) {
172 		mthca_err(mdev, "QUERY_DEV_LIM returned status 0x%02x, "
173 			  "aborting.\n", status);
174 		return -EINVAL;
175 	}
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);
180 		return -ENODEV;
181 	}
182 	if (dev_lim->num_ports > MTHCA_MAX_PORTS) {
183 		mthca_err(mdev, "HCA has %d ports, but we only support %d, "
184 			  "aborting.\n",
185 			  dev_lim->num_ports, MTHCA_MAX_PORTS);
186 		return -ENODEV;
187 	}
188 
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",
192 			  dev_lim->uar_size,
193 			  (unsigned long long)pci_resource_len(mdev->pdev, 2));
194 		return -ENODEV;
195 	}
196 
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);
212 	/*
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.
216 	 */
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;
227 	/*
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
231 	 * values of 2 and 3.
232 	 */
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;
237 	else
238 		mdev->limits.stat_rate_support = 0x3;
239 
240 	/* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
241 	   May be doable since hardware supports it for SRQ.
242 
243 	   IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
244 
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;
251 
252 	if (dev_lim->flags & DEV_LIM_FLAG_BAD_PKEY_CNTR)
253 		mdev->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
254 
255 	if (dev_lim->flags & DEV_LIM_FLAG_BAD_QKEY_CNTR)
256 		mdev->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
257 
258 	if (dev_lim->flags & DEV_LIM_FLAG_RAW_MULTI)
259 		mdev->device_cap_flags |= IB_DEVICE_RAW_MULTI;
260 
261 	if (dev_lim->flags & DEV_LIM_FLAG_AUTO_PATH_MIG)
262 		mdev->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
263 
264 	if (dev_lim->flags & DEV_LIM_FLAG_UD_AV_PORT_ENFORCE)
265 		mdev->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
266 
267 	if (dev_lim->flags & DEV_LIM_FLAG_SRQ)
268 		mdev->mthca_flags |= MTHCA_FLAG_SRQ;
269 
270 	return 0;
271 }
272 
273 static int mthca_init_tavor(struct mthca_dev *mdev)
274 {
275 	u8 status;
276 	int err;
277 	struct mthca_dev_lim        dev_lim;
278 	struct mthca_profile        profile;
279 	struct mthca_init_hca_param init_hca;
280 
281 	err = mthca_SYS_EN(mdev, &status);
282 	if (err) {
283 		mthca_err(mdev, "SYS_EN command failed, aborting.\n");
284 		return err;
285 	}
286 	if (status) {
287 		mthca_err(mdev, "SYS_EN returned status 0x%02x, "
288 			  "aborting.\n", status);
289 		return -EINVAL;
290 	}
291 
292 	err = mthca_QUERY_FW(mdev, &status);
293 	if (err) {
294 		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
295 		goto err_disable;
296 	}
297 	if (status) {
298 		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
299 			  "aborting.\n", status);
300 		err = -EINVAL;
301 		goto err_disable;
302 	}
303 	err = mthca_QUERY_DDR(mdev, &status);
304 	if (err) {
305 		mthca_err(mdev, "QUERY_DDR command failed, aborting.\n");
306 		goto err_disable;
307 	}
308 	if (status) {
309 		mthca_err(mdev, "QUERY_DDR returned status 0x%02x, "
310 			  "aborting.\n", status);
311 		err = -EINVAL;
312 		goto err_disable;
313 	}
314 
315 	err = mthca_dev_lim(mdev, &dev_lim);
316 	if (err) {
317 		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
318 		goto err_disable;
319 	}
320 
321 	profile = hca_profile;
322 	profile.num_uar   = dev_lim.uar_size / PAGE_SIZE;
323 	profile.uarc_size = 0;
324 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
325 		profile.num_srq = dev_lim.max_srqs;
326 
327 	err = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
328 	if (err < 0)
329 		goto err_disable;
330 
331 	err = mthca_INIT_HCA(mdev, &init_hca, &status);
332 	if (err) {
333 		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
334 		goto err_disable;
335 	}
336 	if (status) {
337 		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
338 			  "aborting.\n", status);
339 		err = -EINVAL;
340 		goto err_disable;
341 	}
342 
343 	return 0;
344 
345 err_disable:
346 	mthca_SYS_DIS(mdev, &status);
347 
348 	return err;
349 }
350 
351 static int mthca_load_fw(struct mthca_dev *mdev)
352 {
353 	u8 status;
354 	int err;
355 
356 	/* FIXME: use HCA-attached memory for FW if present */
357 
358 	mdev->fw.arbel.fw_icm =
359 		mthca_alloc_icm(mdev, mdev->fw.arbel.fw_pages,
360 				GFP_HIGHUSER | __GFP_NOWARN, 0);
361 	if (!mdev->fw.arbel.fw_icm) {
362 		mthca_err(mdev, "Couldn't allocate FW area, aborting.\n");
363 		return -ENOMEM;
364 	}
365 
366 	err = mthca_MAP_FA(mdev, mdev->fw.arbel.fw_icm, &status);
367 	if (err) {
368 		mthca_err(mdev, "MAP_FA command failed, aborting.\n");
369 		goto err_free;
370 	}
371 	if (status) {
372 		mthca_err(mdev, "MAP_FA returned status 0x%02x, aborting.\n", status);
373 		err = -EINVAL;
374 		goto err_free;
375 	}
376 	err = mthca_RUN_FW(mdev, &status);
377 	if (err) {
378 		mthca_err(mdev, "RUN_FW command failed, aborting.\n");
379 		goto err_unmap_fa;
380 	}
381 	if (status) {
382 		mthca_err(mdev, "RUN_FW returned status 0x%02x, aborting.\n", status);
383 		err = -EINVAL;
384 		goto err_unmap_fa;
385 	}
386 
387 	return 0;
388 
389 err_unmap_fa:
390 	mthca_UNMAP_FA(mdev, &status);
391 
392 err_free:
393 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
394 	return err;
395 }
396 
397 static int mthca_init_icm(struct mthca_dev *mdev,
398 			  struct mthca_dev_lim *dev_lim,
399 			  struct mthca_init_hca_param *init_hca,
400 			  u64 icm_size)
401 {
402 	u64 aux_pages;
403 	u8 status;
404 	int err;
405 
406 	err = mthca_SET_ICM_SIZE(mdev, icm_size, &aux_pages, &status);
407 	if (err) {
408 		mthca_err(mdev, "SET_ICM_SIZE command failed, aborting.\n");
409 		return err;
410 	}
411 	if (status) {
412 		mthca_err(mdev, "SET_ICM_SIZE returned status 0x%02x, "
413 			  "aborting.\n", status);
414 		return -EINVAL;
415 	}
416 
417 	mthca_dbg(mdev, "%lld KB of HCA context requires %lld KB aux memory.\n",
418 		  (unsigned long long) icm_size >> 10,
419 		  (unsigned long long) aux_pages << 2);
420 
421 	mdev->fw.arbel.aux_icm = mthca_alloc_icm(mdev, aux_pages,
422 						 GFP_HIGHUSER | __GFP_NOWARN, 0);
423 	if (!mdev->fw.arbel.aux_icm) {
424 		mthca_err(mdev, "Couldn't allocate aux memory, aborting.\n");
425 		return -ENOMEM;
426 	}
427 
428 	err = mthca_MAP_ICM_AUX(mdev, mdev->fw.arbel.aux_icm, &status);
429 	if (err) {
430 		mthca_err(mdev, "MAP_ICM_AUX command failed, aborting.\n");
431 		goto err_free_aux;
432 	}
433 	if (status) {
434 		mthca_err(mdev, "MAP_ICM_AUX returned status 0x%02x, aborting.\n", status);
435 		err = -EINVAL;
436 		goto err_free_aux;
437 	}
438 
439 	err = mthca_map_eq_icm(mdev, init_hca->eqc_base);
440 	if (err) {
441 		mthca_err(mdev, "Failed to map EQ context memory, aborting.\n");
442 		goto err_unmap_aux;
443 	}
444 
445 	/* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
446 	mdev->limits.reserved_mtts = ALIGN(mdev->limits.reserved_mtts * MTHCA_MTT_SEG_SIZE,
447 					   dma_get_cache_alignment()) / MTHCA_MTT_SEG_SIZE;
448 
449 	mdev->mr_table.mtt_table = mthca_alloc_icm_table(mdev, init_hca->mtt_base,
450 							 MTHCA_MTT_SEG_SIZE,
451 							 mdev->limits.num_mtt_segs,
452 							 mdev->limits.reserved_mtts,
453 							 1, 0);
454 	if (!mdev->mr_table.mtt_table) {
455 		mthca_err(mdev, "Failed to map MTT context memory, aborting.\n");
456 		err = -ENOMEM;
457 		goto err_unmap_eq;
458 	}
459 
460 	mdev->mr_table.mpt_table = mthca_alloc_icm_table(mdev, init_hca->mpt_base,
461 							 dev_lim->mpt_entry_sz,
462 							 mdev->limits.num_mpts,
463 							 mdev->limits.reserved_mrws,
464 							 1, 1);
465 	if (!mdev->mr_table.mpt_table) {
466 		mthca_err(mdev, "Failed to map MPT context memory, aborting.\n");
467 		err = -ENOMEM;
468 		goto err_unmap_mtt;
469 	}
470 
471 	mdev->qp_table.qp_table = mthca_alloc_icm_table(mdev, init_hca->qpc_base,
472 							dev_lim->qpc_entry_sz,
473 							mdev->limits.num_qps,
474 							mdev->limits.reserved_qps,
475 							0, 0);
476 	if (!mdev->qp_table.qp_table) {
477 		mthca_err(mdev, "Failed to map QP context memory, aborting.\n");
478 		err = -ENOMEM;
479 		goto err_unmap_mpt;
480 	}
481 
482 	mdev->qp_table.eqp_table = mthca_alloc_icm_table(mdev, init_hca->eqpc_base,
483 							 dev_lim->eqpc_entry_sz,
484 							 mdev->limits.num_qps,
485 							 mdev->limits.reserved_qps,
486 							 0, 0);
487 	if (!mdev->qp_table.eqp_table) {
488 		mthca_err(mdev, "Failed to map EQP context memory, aborting.\n");
489 		err = -ENOMEM;
490 		goto err_unmap_qp;
491 	}
492 
493 	mdev->qp_table.rdb_table = mthca_alloc_icm_table(mdev, init_hca->rdb_base,
494 							 MTHCA_RDB_ENTRY_SIZE,
495 							 mdev->limits.num_qps <<
496 							 mdev->qp_table.rdb_shift, 0,
497 							 0, 0);
498 	if (!mdev->qp_table.rdb_table) {
499 		mthca_err(mdev, "Failed to map RDB context memory, aborting\n");
500 		err = -ENOMEM;
501 		goto err_unmap_eqp;
502 	}
503 
504        mdev->cq_table.table = mthca_alloc_icm_table(mdev, init_hca->cqc_base,
505 						    dev_lim->cqc_entry_sz,
506 						    mdev->limits.num_cqs,
507 						    mdev->limits.reserved_cqs,
508 						    0, 0);
509 	if (!mdev->cq_table.table) {
510 		mthca_err(mdev, "Failed to map CQ context memory, aborting.\n");
511 		err = -ENOMEM;
512 		goto err_unmap_rdb;
513 	}
514 
515 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ) {
516 		mdev->srq_table.table =
517 			mthca_alloc_icm_table(mdev, init_hca->srqc_base,
518 					      dev_lim->srq_entry_sz,
519 					      mdev->limits.num_srqs,
520 					      mdev->limits.reserved_srqs,
521 					      0, 0);
522 		if (!mdev->srq_table.table) {
523 			mthca_err(mdev, "Failed to map SRQ context memory, "
524 				  "aborting.\n");
525 			err = -ENOMEM;
526 			goto err_unmap_cq;
527 		}
528 	}
529 
530 	/*
531 	 * It's not strictly required, but for simplicity just map the
532 	 * whole multicast group table now.  The table isn't very big
533 	 * and it's a lot easier than trying to track ref counts.
534 	 */
535 	mdev->mcg_table.table = mthca_alloc_icm_table(mdev, init_hca->mc_base,
536 						      MTHCA_MGM_ENTRY_SIZE,
537 						      mdev->limits.num_mgms +
538 						      mdev->limits.num_amgms,
539 						      mdev->limits.num_mgms +
540 						      mdev->limits.num_amgms,
541 						      0, 0);
542 	if (!mdev->mcg_table.table) {
543 		mthca_err(mdev, "Failed to map MCG context memory, aborting.\n");
544 		err = -ENOMEM;
545 		goto err_unmap_srq;
546 	}
547 
548 	return 0;
549 
550 err_unmap_srq:
551 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
552 		mthca_free_icm_table(mdev, mdev->srq_table.table);
553 
554 err_unmap_cq:
555 	mthca_free_icm_table(mdev, mdev->cq_table.table);
556 
557 err_unmap_rdb:
558 	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
559 
560 err_unmap_eqp:
561 	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
562 
563 err_unmap_qp:
564 	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
565 
566 err_unmap_mpt:
567 	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
568 
569 err_unmap_mtt:
570 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
571 
572 err_unmap_eq:
573 	mthca_unmap_eq_icm(mdev);
574 
575 err_unmap_aux:
576 	mthca_UNMAP_ICM_AUX(mdev, &status);
577 
578 err_free_aux:
579 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm, 0);
580 
581 	return err;
582 }
583 
584 static void mthca_free_icms(struct mthca_dev *mdev)
585 {
586 	u8 status;
587 
588 	mthca_free_icm_table(mdev, mdev->mcg_table.table);
589 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
590 		mthca_free_icm_table(mdev, mdev->srq_table.table);
591 	mthca_free_icm_table(mdev, mdev->cq_table.table);
592 	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
593 	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
594 	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
595 	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
596 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
597 	mthca_unmap_eq_icm(mdev);
598 
599 	mthca_UNMAP_ICM_AUX(mdev, &status);
600 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm, 0);
601 }
602 
603 static int mthca_init_arbel(struct mthca_dev *mdev)
604 {
605 	struct mthca_dev_lim        dev_lim;
606 	struct mthca_profile        profile;
607 	struct mthca_init_hca_param init_hca;
608 	u64 icm_size;
609 	u8 status;
610 	int err;
611 
612 	err = mthca_QUERY_FW(mdev, &status);
613 	if (err) {
614 		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
615 		return err;
616 	}
617 	if (status) {
618 		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
619 			  "aborting.\n", status);
620 		return -EINVAL;
621 	}
622 
623 	err = mthca_ENABLE_LAM(mdev, &status);
624 	if (err) {
625 		mthca_err(mdev, "ENABLE_LAM command failed, aborting.\n");
626 		return err;
627 	}
628 	if (status == MTHCA_CMD_STAT_LAM_NOT_PRE) {
629 		mthca_dbg(mdev, "No HCA-attached memory (running in MemFree mode)\n");
630 		mdev->mthca_flags |= MTHCA_FLAG_NO_LAM;
631 	} else if (status) {
632 		mthca_err(mdev, "ENABLE_LAM returned status 0x%02x, "
633 			  "aborting.\n", status);
634 		return -EINVAL;
635 	}
636 
637 	err = mthca_load_fw(mdev);
638 	if (err) {
639 		mthca_err(mdev, "Failed to start FW, aborting.\n");
640 		goto err_disable;
641 	}
642 
643 	err = mthca_dev_lim(mdev, &dev_lim);
644 	if (err) {
645 		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
646 		goto err_stop_fw;
647 	}
648 
649 	profile = hca_profile;
650 	profile.num_uar  = dev_lim.uar_size / PAGE_SIZE;
651 	profile.num_udav = 0;
652 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
653 		profile.num_srq = dev_lim.max_srqs;
654 
655 	icm_size = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
656 	if ((int) icm_size < 0) {
657 		err = icm_size;
658 		goto err_stop_fw;
659 	}
660 
661 	err = mthca_init_icm(mdev, &dev_lim, &init_hca, icm_size);
662 	if (err)
663 		goto err_stop_fw;
664 
665 	err = mthca_INIT_HCA(mdev, &init_hca, &status);
666 	if (err) {
667 		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
668 		goto err_free_icm;
669 	}
670 	if (status) {
671 		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
672 			  "aborting.\n", status);
673 		err = -EINVAL;
674 		goto err_free_icm;
675 	}
676 
677 	return 0;
678 
679 err_free_icm:
680 	mthca_free_icms(mdev);
681 
682 err_stop_fw:
683 	mthca_UNMAP_FA(mdev, &status);
684 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
685 
686 err_disable:
687 	if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
688 		mthca_DISABLE_LAM(mdev, &status);
689 
690 	return err;
691 }
692 
693 static void mthca_close_hca(struct mthca_dev *mdev)
694 {
695 	u8 status;
696 
697 	mthca_CLOSE_HCA(mdev, 0, &status);
698 
699 	if (mthca_is_memfree(mdev)) {
700 		mthca_free_icms(mdev);
701 
702 		mthca_UNMAP_FA(mdev, &status);
703 		mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
704 
705 		if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
706 			mthca_DISABLE_LAM(mdev, &status);
707 	} else
708 		mthca_SYS_DIS(mdev, &status);
709 }
710 
711 static int mthca_init_hca(struct mthca_dev *mdev)
712 {
713 	u8 status;
714 	int err;
715 	struct mthca_adapter adapter;
716 
717 	if (mthca_is_memfree(mdev))
718 		err = mthca_init_arbel(mdev);
719 	else
720 		err = mthca_init_tavor(mdev);
721 
722 	if (err)
723 		return err;
724 
725 	err = mthca_QUERY_ADAPTER(mdev, &adapter, &status);
726 	if (err) {
727 		mthca_err(mdev, "QUERY_ADAPTER command failed, aborting.\n");
728 		goto err_close;
729 	}
730 	if (status) {
731 		mthca_err(mdev, "QUERY_ADAPTER returned status 0x%02x, "
732 			  "aborting.\n", status);
733 		err = -EINVAL;
734 		goto err_close;
735 	}
736 
737 	mdev->eq_table.inta_pin = adapter.inta_pin;
738 	mdev->rev_id            = adapter.revision_id;
739 	memcpy(mdev->board_id, adapter.board_id, sizeof mdev->board_id);
740 
741 	return 0;
742 
743 err_close:
744 	mthca_close_hca(mdev);
745 	return err;
746 }
747 
748 static int mthca_setup_hca(struct mthca_dev *dev)
749 {
750 	int err;
751 	u8 status;
752 
753 	MTHCA_INIT_DOORBELL_LOCK(&dev->doorbell_lock);
754 
755 	err = mthca_init_uar_table(dev);
756 	if (err) {
757 		mthca_err(dev, "Failed to initialize "
758 			  "user access region table, aborting.\n");
759 		return err;
760 	}
761 
762 	err = mthca_uar_alloc(dev, &dev->driver_uar);
763 	if (err) {
764 		mthca_err(dev, "Failed to allocate driver access region, "
765 			  "aborting.\n");
766 		goto err_uar_table_free;
767 	}
768 
769 	dev->kar = ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
770 	if (!dev->kar) {
771 		mthca_err(dev, "Couldn't map kernel access region, "
772 			  "aborting.\n");
773 		err = -ENOMEM;
774 		goto err_uar_free;
775 	}
776 
777 	err = mthca_init_pd_table(dev);
778 	if (err) {
779 		mthca_err(dev, "Failed to initialize "
780 			  "protection domain table, aborting.\n");
781 		goto err_kar_unmap;
782 	}
783 
784 	err = mthca_init_mr_table(dev);
785 	if (err) {
786 		mthca_err(dev, "Failed to initialize "
787 			  "memory region table, aborting.\n");
788 		goto err_pd_table_free;
789 	}
790 
791 	err = mthca_pd_alloc(dev, 1, &dev->driver_pd);
792 	if (err) {
793 		mthca_err(dev, "Failed to create driver PD, "
794 			  "aborting.\n");
795 		goto err_mr_table_free;
796 	}
797 
798 	err = mthca_init_eq_table(dev);
799 	if (err) {
800 		mthca_err(dev, "Failed to initialize "
801 			  "event queue table, aborting.\n");
802 		goto err_pd_free;
803 	}
804 
805 	err = mthca_cmd_use_events(dev);
806 	if (err) {
807 		mthca_err(dev, "Failed to switch to event-driven "
808 			  "firmware commands, aborting.\n");
809 		goto err_eq_table_free;
810 	}
811 
812 	err = mthca_NOP(dev, &status);
813 	if (err || status) {
814 		if (dev->mthca_flags & MTHCA_FLAG_MSI_X) {
815 			mthca_warn(dev, "NOP command failed to generate interrupt "
816 				   "(IRQ %d).\n",
817 				   dev->eq_table.eq[MTHCA_EQ_CMD].msi_x_vector);
818 			mthca_warn(dev, "Trying again with MSI-X disabled.\n");
819 		} else {
820 			mthca_err(dev, "NOP command failed to generate interrupt "
821 				  "(IRQ %d), aborting.\n",
822 				  dev->pdev->irq);
823 			mthca_err(dev, "BIOS or ACPI interrupt routing problem?\n");
824 		}
825 
826 		goto err_cmd_poll;
827 	}
828 
829 	mthca_dbg(dev, "NOP command IRQ test passed\n");
830 
831 	err = mthca_init_cq_table(dev);
832 	if (err) {
833 		mthca_err(dev, "Failed to initialize "
834 			  "completion queue table, aborting.\n");
835 		goto err_cmd_poll;
836 	}
837 
838 	err = mthca_init_srq_table(dev);
839 	if (err) {
840 		mthca_err(dev, "Failed to initialize "
841 			  "shared receive queue table, aborting.\n");
842 		goto err_cq_table_free;
843 	}
844 
845 	err = mthca_init_qp_table(dev);
846 	if (err) {
847 		mthca_err(dev, "Failed to initialize "
848 			  "queue pair table, aborting.\n");
849 		goto err_srq_table_free;
850 	}
851 
852 	err = mthca_init_av_table(dev);
853 	if (err) {
854 		mthca_err(dev, "Failed to initialize "
855 			  "address vector table, aborting.\n");
856 		goto err_qp_table_free;
857 	}
858 
859 	err = mthca_init_mcg_table(dev);
860 	if (err) {
861 		mthca_err(dev, "Failed to initialize "
862 			  "multicast group table, aborting.\n");
863 		goto err_av_table_free;
864 	}
865 
866 	return 0;
867 
868 err_av_table_free:
869 	mthca_cleanup_av_table(dev);
870 
871 err_qp_table_free:
872 	mthca_cleanup_qp_table(dev);
873 
874 err_srq_table_free:
875 	mthca_cleanup_srq_table(dev);
876 
877 err_cq_table_free:
878 	mthca_cleanup_cq_table(dev);
879 
880 err_cmd_poll:
881 	mthca_cmd_use_polling(dev);
882 
883 err_eq_table_free:
884 	mthca_cleanup_eq_table(dev);
885 
886 err_pd_free:
887 	mthca_pd_free(dev, &dev->driver_pd);
888 
889 err_mr_table_free:
890 	mthca_cleanup_mr_table(dev);
891 
892 err_pd_table_free:
893 	mthca_cleanup_pd_table(dev);
894 
895 err_kar_unmap:
896 	iounmap(dev->kar);
897 
898 err_uar_free:
899 	mthca_uar_free(dev, &dev->driver_uar);
900 
901 err_uar_table_free:
902 	mthca_cleanup_uar_table(dev);
903 	return err;
904 }
905 
906 static int mthca_request_regions(struct pci_dev *pdev, int ddr_hidden)
907 {
908 	int err;
909 
910 	/*
911 	 * We can't just use pci_request_regions() because the MSI-X
912 	 * table is right in the middle of the first BAR.  If we did
913 	 * pci_request_region and grab all of the first BAR, then
914 	 * setting up MSI-X would fail, since the PCI core wants to do
915 	 * request_mem_region on the MSI-X vector table.
916 	 *
917 	 * So just request what we need right now, and request any
918 	 * other regions we need when setting up EQs.
919 	 */
920 	if (!request_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
921 				MTHCA_HCR_SIZE, DRV_NAME))
922 		return -EBUSY;
923 
924 	err = pci_request_region(pdev, 2, DRV_NAME);
925 	if (err)
926 		goto err_bar2_failed;
927 
928 	if (!ddr_hidden) {
929 		err = pci_request_region(pdev, 4, DRV_NAME);
930 		if (err)
931 			goto err_bar4_failed;
932 	}
933 
934 	return 0;
935 
936 err_bar4_failed:
937 	pci_release_region(pdev, 2);
938 
939 err_bar2_failed:
940 	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
941 			   MTHCA_HCR_SIZE);
942 
943 	return err;
944 }
945 
946 static void mthca_release_regions(struct pci_dev *pdev,
947 				  int ddr_hidden)
948 {
949 	if (!ddr_hidden)
950 		pci_release_region(pdev, 4);
951 
952 	pci_release_region(pdev, 2);
953 
954 	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
955 			   MTHCA_HCR_SIZE);
956 }
957 
958 static int mthca_enable_msi_x(struct mthca_dev *mdev)
959 {
960 	struct msix_entry entries[3];
961 	int err;
962 
963 	entries[0].entry = 0;
964 	entries[1].entry = 1;
965 	entries[2].entry = 2;
966 
967 	err = pci_enable_msix(mdev->pdev, entries, ARRAY_SIZE(entries));
968 	if (err) {
969 		if (err > 0)
970 			mthca_info(mdev, "Only %d MSI-X vectors available, "
971 				   "not using MSI-X\n", err);
972 		return err;
973 	}
974 
975 	mdev->eq_table.eq[MTHCA_EQ_COMP ].msi_x_vector = entries[0].vector;
976 	mdev->eq_table.eq[MTHCA_EQ_ASYNC].msi_x_vector = entries[1].vector;
977 	mdev->eq_table.eq[MTHCA_EQ_CMD  ].msi_x_vector = entries[2].vector;
978 
979 	return 0;
980 }
981 
982 /* Types of supported HCA */
983 enum {
984 	TAVOR,			/* MT23108                        */
985 	ARBEL_COMPAT,		/* MT25208 in Tavor compat mode   */
986 	ARBEL_NATIVE,		/* MT25208 with extended features */
987 	SINAI			/* MT25204 */
988 };
989 
990 #define MTHCA_FW_VER(major, minor, subminor) \
991 	(((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
992 
993 static struct {
994 	u64 latest_fw;
995 	u32 flags;
996 } mthca_hca_table[] = {
997 	[TAVOR]        = { .latest_fw = MTHCA_FW_VER(3, 5, 0),
998 			   .flags     = 0 },
999 	[ARBEL_COMPAT] = { .latest_fw = MTHCA_FW_VER(4, 8, 200),
1000 			   .flags     = MTHCA_FLAG_PCIE },
1001 	[ARBEL_NATIVE] = { .latest_fw = MTHCA_FW_VER(5, 3, 0),
1002 			   .flags     = MTHCA_FLAG_MEMFREE |
1003 					MTHCA_FLAG_PCIE },
1004 	[SINAI]        = { .latest_fw = MTHCA_FW_VER(1, 2, 0),
1005 			   .flags     = MTHCA_FLAG_MEMFREE |
1006 					MTHCA_FLAG_PCIE    |
1007 					MTHCA_FLAG_SINAI_OPT }
1008 };
1009 
1010 static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
1011 {
1012 	int ddr_hidden = 0;
1013 	int err;
1014 	struct mthca_dev *mdev;
1015 
1016 	printk(KERN_INFO PFX "Initializing %s\n",
1017 	       pci_name(pdev));
1018 
1019 	err = pci_enable_device(pdev);
1020 	if (err) {
1021 		dev_err(&pdev->dev, "Cannot enable PCI device, "
1022 			"aborting.\n");
1023 		return err;
1024 	}
1025 
1026 	/*
1027 	 * Check for BARs.  We expect 0: 1MB, 2: 8MB, 4: DDR (may not
1028 	 * be present)
1029 	 */
1030 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
1031 	    pci_resource_len(pdev, 0) != 1 << 20) {
1032 		dev_err(&pdev->dev, "Missing DCS, aborting.\n");
1033 		err = -ENODEV;
1034 		goto err_disable_pdev;
1035 	}
1036 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
1037 		dev_err(&pdev->dev, "Missing UAR, aborting.\n");
1038 		err = -ENODEV;
1039 		goto err_disable_pdev;
1040 	}
1041 	if (!(pci_resource_flags(pdev, 4) & IORESOURCE_MEM))
1042 		ddr_hidden = 1;
1043 
1044 	err = mthca_request_regions(pdev, ddr_hidden);
1045 	if (err) {
1046 		dev_err(&pdev->dev, "Cannot obtain PCI resources, "
1047 			"aborting.\n");
1048 		goto err_disable_pdev;
1049 	}
1050 
1051 	pci_set_master(pdev);
1052 
1053 	err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
1054 	if (err) {
1055 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1056 		err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1057 		if (err) {
1058 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
1059 			goto err_free_res;
1060 		}
1061 	}
1062 	err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
1063 	if (err) {
1064 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
1065 			 "consistent PCI DMA mask.\n");
1066 		err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1067 		if (err) {
1068 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
1069 				"aborting.\n");
1070 			goto err_free_res;
1071 		}
1072 	}
1073 
1074 	mdev = (struct mthca_dev *) ib_alloc_device(sizeof *mdev);
1075 	if (!mdev) {
1076 		dev_err(&pdev->dev, "Device struct alloc failed, "
1077 			"aborting.\n");
1078 		err = -ENOMEM;
1079 		goto err_free_res;
1080 	}
1081 
1082 	mdev->pdev = pdev;
1083 
1084 	mdev->mthca_flags = mthca_hca_table[hca_type].flags;
1085 	if (ddr_hidden)
1086 		mdev->mthca_flags |= MTHCA_FLAG_DDR_HIDDEN;
1087 
1088 	/*
1089 	 * Now reset the HCA before we touch the PCI capabilities or
1090 	 * attempt a firmware command, since a boot ROM may have left
1091 	 * the HCA in an undefined state.
1092 	 */
1093 	err = mthca_reset(mdev);
1094 	if (err) {
1095 		mthca_err(mdev, "Failed to reset HCA, aborting.\n");
1096 		goto err_free_dev;
1097 	}
1098 
1099 	if (mthca_cmd_init(mdev)) {
1100 		mthca_err(mdev, "Failed to init command interface, aborting.\n");
1101 		goto err_free_dev;
1102 	}
1103 
1104 	err = mthca_tune_pci(mdev);
1105 	if (err)
1106 		goto err_cmd;
1107 
1108 	err = mthca_init_hca(mdev);
1109 	if (err)
1110 		goto err_cmd;
1111 
1112 	if (mdev->fw_ver < mthca_hca_table[hca_type].latest_fw) {
1113 		mthca_warn(mdev, "HCA FW version %d.%d.%03d is old (%d.%d.%03d is current).\n",
1114 			   (int) (mdev->fw_ver >> 32), (int) (mdev->fw_ver >> 16) & 0xffff,
1115 			   (int) (mdev->fw_ver & 0xffff),
1116 			   (int) (mthca_hca_table[hca_type].latest_fw >> 32),
1117 			   (int) (mthca_hca_table[hca_type].latest_fw >> 16) & 0xffff,
1118 			   (int) (mthca_hca_table[hca_type].latest_fw & 0xffff));
1119 		mthca_warn(mdev, "If you have problems, try updating your HCA FW.\n");
1120 	}
1121 
1122 	if (msi_x && !mthca_enable_msi_x(mdev))
1123 		mdev->mthca_flags |= MTHCA_FLAG_MSI_X;
1124 
1125 	err = mthca_setup_hca(mdev);
1126 	if (err == -EBUSY && (mdev->mthca_flags & MTHCA_FLAG_MSI_X)) {
1127 		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1128 			pci_disable_msix(pdev);
1129 		mdev->mthca_flags &= ~MTHCA_FLAG_MSI_X;
1130 
1131 		err = mthca_setup_hca(mdev);
1132 	}
1133 
1134 	if (err)
1135 		goto err_close;
1136 
1137 	err = mthca_register_device(mdev);
1138 	if (err)
1139 		goto err_cleanup;
1140 
1141 	err = mthca_create_agents(mdev);
1142 	if (err)
1143 		goto err_unregister;
1144 
1145 	pci_set_drvdata(pdev, mdev);
1146 	mdev->hca_type = hca_type;
1147 
1148 	return 0;
1149 
1150 err_unregister:
1151 	mthca_unregister_device(mdev);
1152 
1153 err_cleanup:
1154 	mthca_cleanup_mcg_table(mdev);
1155 	mthca_cleanup_av_table(mdev);
1156 	mthca_cleanup_qp_table(mdev);
1157 	mthca_cleanup_srq_table(mdev);
1158 	mthca_cleanup_cq_table(mdev);
1159 	mthca_cmd_use_polling(mdev);
1160 	mthca_cleanup_eq_table(mdev);
1161 
1162 	mthca_pd_free(mdev, &mdev->driver_pd);
1163 
1164 	mthca_cleanup_mr_table(mdev);
1165 	mthca_cleanup_pd_table(mdev);
1166 	mthca_cleanup_uar_table(mdev);
1167 
1168 err_close:
1169 	if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1170 		pci_disable_msix(pdev);
1171 
1172 	mthca_close_hca(mdev);
1173 
1174 err_cmd:
1175 	mthca_cmd_cleanup(mdev);
1176 
1177 err_free_dev:
1178 	ib_dealloc_device(&mdev->ib_dev);
1179 
1180 err_free_res:
1181 	mthca_release_regions(pdev, ddr_hidden);
1182 
1183 err_disable_pdev:
1184 	pci_disable_device(pdev);
1185 	pci_set_drvdata(pdev, NULL);
1186 	return err;
1187 }
1188 
1189 static void __mthca_remove_one(struct pci_dev *pdev)
1190 {
1191 	struct mthca_dev *mdev = pci_get_drvdata(pdev);
1192 	u8 status;
1193 	int p;
1194 
1195 	if (mdev) {
1196 		mthca_free_agents(mdev);
1197 		mthca_unregister_device(mdev);
1198 
1199 		for (p = 1; p <= mdev->limits.num_ports; ++p)
1200 			mthca_CLOSE_IB(mdev, p, &status);
1201 
1202 		mthca_cleanup_mcg_table(mdev);
1203 		mthca_cleanup_av_table(mdev);
1204 		mthca_cleanup_qp_table(mdev);
1205 		mthca_cleanup_srq_table(mdev);
1206 		mthca_cleanup_cq_table(mdev);
1207 		mthca_cmd_use_polling(mdev);
1208 		mthca_cleanup_eq_table(mdev);
1209 
1210 		mthca_pd_free(mdev, &mdev->driver_pd);
1211 
1212 		mthca_cleanup_mr_table(mdev);
1213 		mthca_cleanup_pd_table(mdev);
1214 
1215 		iounmap(mdev->kar);
1216 		mthca_uar_free(mdev, &mdev->driver_uar);
1217 		mthca_cleanup_uar_table(mdev);
1218 		mthca_close_hca(mdev);
1219 		mthca_cmd_cleanup(mdev);
1220 
1221 		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1222 			pci_disable_msix(pdev);
1223 
1224 		ib_dealloc_device(&mdev->ib_dev);
1225 		mthca_release_regions(pdev, mdev->mthca_flags &
1226 				      MTHCA_FLAG_DDR_HIDDEN);
1227 		pci_disable_device(pdev);
1228 		pci_set_drvdata(pdev, NULL);
1229 	}
1230 }
1231 
1232 int __mthca_restart_one(struct pci_dev *pdev)
1233 {
1234 	struct mthca_dev *mdev;
1235 	int hca_type;
1236 
1237 	mdev = pci_get_drvdata(pdev);
1238 	if (!mdev)
1239 		return -ENODEV;
1240 	hca_type = mdev->hca_type;
1241 	__mthca_remove_one(pdev);
1242 	return __mthca_init_one(pdev, hca_type);
1243 }
1244 
1245 static int __devinit mthca_init_one(struct pci_dev *pdev,
1246 				    const struct pci_device_id *id)
1247 {
1248 	static int mthca_version_printed = 0;
1249 	int ret;
1250 
1251 	mutex_lock(&mthca_device_mutex);
1252 
1253 	if (!mthca_version_printed) {
1254 		printk(KERN_INFO "%s", mthca_version);
1255 		++mthca_version_printed;
1256 	}
1257 
1258 	if (id->driver_data >= ARRAY_SIZE(mthca_hca_table)) {
1259 		printk(KERN_ERR PFX "%s has invalid driver data %lx\n",
1260 		       pci_name(pdev), id->driver_data);
1261 		mutex_unlock(&mthca_device_mutex);
1262 		return -ENODEV;
1263 	}
1264 
1265 	ret = __mthca_init_one(pdev, id->driver_data);
1266 
1267 	mutex_unlock(&mthca_device_mutex);
1268 
1269 	return ret;
1270 }
1271 
1272 static void __devexit mthca_remove_one(struct pci_dev *pdev)
1273 {
1274 	mutex_lock(&mthca_device_mutex);
1275 	__mthca_remove_one(pdev);
1276 	mutex_unlock(&mthca_device_mutex);
1277 }
1278 
1279 static struct pci_device_id mthca_pci_table[] = {
1280 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_TAVOR),
1281 	  .driver_data = TAVOR },
1282 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_TAVOR),
1283 	  .driver_data = TAVOR },
1284 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1285 	  .driver_data = ARBEL_COMPAT },
1286 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1287 	  .driver_data = ARBEL_COMPAT },
1288 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL),
1289 	  .driver_data = ARBEL_NATIVE },
1290 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL),
1291 	  .driver_data = ARBEL_NATIVE },
1292 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI),
1293 	  .driver_data = SINAI },
1294 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI),
1295 	  .driver_data = SINAI },
1296 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1297 	  .driver_data = SINAI },
1298 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1299 	  .driver_data = SINAI },
1300 	{ 0, }
1301 };
1302 
1303 MODULE_DEVICE_TABLE(pci, mthca_pci_table);
1304 
1305 static struct pci_driver mthca_driver = {
1306 	.name		= DRV_NAME,
1307 	.id_table	= mthca_pci_table,
1308 	.probe		= mthca_init_one,
1309 	.remove		= __devexit_p(mthca_remove_one)
1310 };
1311 
1312 static void __init __mthca_check_profile_val(const char *name, int *pval,
1313 					     int pval_default)
1314 {
1315 	/* value must be positive and power of 2 */
1316 	int old_pval = *pval;
1317 
1318 	if (old_pval <= 0)
1319 		*pval = pval_default;
1320 	else
1321 		*pval = roundup_pow_of_two(old_pval);
1322 
1323 	if (old_pval != *pval) {
1324 		printk(KERN_WARNING PFX "Invalid value %d for %s in module parameter.\n",
1325 		       old_pval, name);
1326 		printk(KERN_WARNING PFX "Corrected %s to %d.\n", name, *pval);
1327 	}
1328 }
1329 
1330 #define mthca_check_profile_val(name, default)				\
1331 	__mthca_check_profile_val(#name, &hca_profile.name, default)
1332 
1333 static void __init mthca_validate_profile(void)
1334 {
1335 	mthca_check_profile_val(num_qp,            MTHCA_DEFAULT_NUM_QP);
1336 	mthca_check_profile_val(rdb_per_qp,        MTHCA_DEFAULT_RDB_PER_QP);
1337 	mthca_check_profile_val(num_cq,            MTHCA_DEFAULT_NUM_CQ);
1338 	mthca_check_profile_val(num_mcg, 	   MTHCA_DEFAULT_NUM_MCG);
1339 	mthca_check_profile_val(num_mpt, 	   MTHCA_DEFAULT_NUM_MPT);
1340 	mthca_check_profile_val(num_mtt, 	   MTHCA_DEFAULT_NUM_MTT);
1341 	mthca_check_profile_val(num_udav,          MTHCA_DEFAULT_NUM_UDAV);
1342 	mthca_check_profile_val(fmr_reserved_mtts, MTHCA_DEFAULT_NUM_RESERVED_MTTS);
1343 
1344 	if (hca_profile.fmr_reserved_mtts >= hca_profile.num_mtt) {
1345 		printk(KERN_WARNING PFX "Invalid fmr_reserved_mtts module parameter %d.\n",
1346 		       hca_profile.fmr_reserved_mtts);
1347 		printk(KERN_WARNING PFX "(Must be smaller than num_mtt %d)\n",
1348 		       hca_profile.num_mtt);
1349 		hca_profile.fmr_reserved_mtts = hca_profile.num_mtt / 2;
1350 		printk(KERN_WARNING PFX "Corrected fmr_reserved_mtts to %d.\n",
1351 		       hca_profile.fmr_reserved_mtts);
1352 	}
1353 }
1354 
1355 static int __init mthca_init(void)
1356 {
1357 	int ret;
1358 
1359 	mthca_validate_profile();
1360 
1361 	ret = mthca_catas_init();
1362 	if (ret)
1363 		return ret;
1364 
1365 	ret = pci_register_driver(&mthca_driver);
1366 	if (ret < 0) {
1367 		mthca_catas_cleanup();
1368 		return ret;
1369 	}
1370 
1371 	return 0;
1372 }
1373 
1374 static void __exit mthca_cleanup(void)
1375 {
1376 	pci_unregister_driver(&mthca_driver);
1377 	mthca_catas_cleanup();
1378 }
1379 
1380 module_init(mthca_init);
1381 module_exit(mthca_cleanup);
1382