xref: /openbmc/linux/drivers/infiniband/hw/mthca/mthca_main.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
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 static int msi = 0;
69 module_param(msi, int, 0444);
70 MODULE_PARM_DESC(msi, "attempt to use MSI if nonzero (deprecated, use MSI-X instead)");
71 
72 #else /* CONFIG_PCI_MSI */
73 
74 #define msi_x (0)
75 #define msi   (0)
76 
77 #endif /* CONFIG_PCI_MSI */
78 
79 static int tune_pci = 0;
80 module_param(tune_pci, int, 0444);
81 MODULE_PARM_DESC(tune_pci, "increase PCI burst from the default set by BIOS if nonzero");
82 
83 DEFINE_MUTEX(mthca_device_mutex);
84 
85 #define MTHCA_DEFAULT_NUM_QP            (1 << 16)
86 #define MTHCA_DEFAULT_RDB_PER_QP        (1 << 2)
87 #define MTHCA_DEFAULT_NUM_CQ            (1 << 16)
88 #define MTHCA_DEFAULT_NUM_MCG           (1 << 13)
89 #define MTHCA_DEFAULT_NUM_MPT           (1 << 17)
90 #define MTHCA_DEFAULT_NUM_MTT           (1 << 20)
91 #define MTHCA_DEFAULT_NUM_UDAV          (1 << 15)
92 #define MTHCA_DEFAULT_NUM_RESERVED_MTTS (1 << 18)
93 #define MTHCA_DEFAULT_NUM_UARC_SIZE     (1 << 18)
94 
95 static struct mthca_profile hca_profile = {
96 	.num_qp             = MTHCA_DEFAULT_NUM_QP,
97 	.rdb_per_qp         = MTHCA_DEFAULT_RDB_PER_QP,
98 	.num_cq             = MTHCA_DEFAULT_NUM_CQ,
99 	.num_mcg            = MTHCA_DEFAULT_NUM_MCG,
100 	.num_mpt            = MTHCA_DEFAULT_NUM_MPT,
101 	.num_mtt            = MTHCA_DEFAULT_NUM_MTT,
102 	.num_udav           = MTHCA_DEFAULT_NUM_UDAV,          /* Tavor only */
103 	.fmr_reserved_mtts  = MTHCA_DEFAULT_NUM_RESERVED_MTTS, /* Tavor only */
104 	.uarc_size          = MTHCA_DEFAULT_NUM_UARC_SIZE,     /* Arbel only */
105 };
106 
107 module_param_named(num_qp, hca_profile.num_qp, int, 0444);
108 MODULE_PARM_DESC(num_qp, "maximum number of QPs per HCA");
109 
110 module_param_named(rdb_per_qp, hca_profile.rdb_per_qp, int, 0444);
111 MODULE_PARM_DESC(rdb_per_qp, "number of RDB buffers per QP");
112 
113 module_param_named(num_cq, hca_profile.num_cq, int, 0444);
114 MODULE_PARM_DESC(num_cq, "maximum number of CQs per HCA");
115 
116 module_param_named(num_mcg, hca_profile.num_mcg, int, 0444);
117 MODULE_PARM_DESC(num_mcg, "maximum number of multicast groups per HCA");
118 
119 module_param_named(num_mpt, hca_profile.num_mpt, int, 0444);
120 MODULE_PARM_DESC(num_mpt,
121 		"maximum number of memory protection table entries per HCA");
122 
123 module_param_named(num_mtt, hca_profile.num_mtt, int, 0444);
124 MODULE_PARM_DESC(num_mtt,
125 		 "maximum number of memory translation table segments per HCA");
126 
127 module_param_named(num_udav, hca_profile.num_udav, int, 0444);
128 MODULE_PARM_DESC(num_udav, "maximum number of UD address vectors per HCA");
129 
130 module_param_named(fmr_reserved_mtts, hca_profile.fmr_reserved_mtts, int, 0444);
131 MODULE_PARM_DESC(fmr_reserved_mtts,
132 		 "number of memory translation table segments reserved for FMR");
133 
134 static const char mthca_version[] __devinitdata =
135 	DRV_NAME ": Mellanox InfiniBand HCA driver v"
136 	DRV_VERSION " (" DRV_RELDATE ")\n";
137 
138 static int mthca_tune_pci(struct mthca_dev *mdev)
139 {
140 	if (!tune_pci)
141 		return 0;
142 
143 	/* First try to max out Read Byte Count */
144 	if (pci_find_capability(mdev->pdev, PCI_CAP_ID_PCIX)) {
145 		if (pcix_set_mmrbc(mdev->pdev, pcix_get_max_mmrbc(mdev->pdev))) {
146 			mthca_err(mdev, "Couldn't set PCI-X max read count, "
147 				"aborting.\n");
148 			return -ENODEV;
149 		}
150 	} else if (!(mdev->mthca_flags & MTHCA_FLAG_PCIE))
151 		mthca_info(mdev, "No PCI-X capability, not setting RBC.\n");
152 
153 	if (pci_find_capability(mdev->pdev, PCI_CAP_ID_EXP)) {
154 		if (pcie_set_readrq(mdev->pdev, 4096)) {
155 			mthca_err(mdev, "Couldn't write PCI Express read request, "
156 				"aborting.\n");
157 			return -ENODEV;
158 		}
159 	} else if (mdev->mthca_flags & MTHCA_FLAG_PCIE)
160 		mthca_info(mdev, "No PCI Express capability, "
161 			   "not setting Max Read Request Size.\n");
162 
163 	return 0;
164 }
165 
166 static int mthca_dev_lim(struct mthca_dev *mdev, struct mthca_dev_lim *dev_lim)
167 {
168 	int err;
169 	u8 status;
170 
171 	err = mthca_QUERY_DEV_LIM(mdev, dev_lim, &status);
172 	if (err) {
173 		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
174 		return err;
175 	}
176 	if (status) {
177 		mthca_err(mdev, "QUERY_DEV_LIM returned status 0x%02x, "
178 			  "aborting.\n", status);
179 		return -EINVAL;
180 	}
181 	if (dev_lim->min_page_sz > PAGE_SIZE) {
182 		mthca_err(mdev, "HCA minimum page size of %d bigger than "
183 			  "kernel PAGE_SIZE of %ld, aborting.\n",
184 			  dev_lim->min_page_sz, PAGE_SIZE);
185 		return -ENODEV;
186 	}
187 	if (dev_lim->num_ports > MTHCA_MAX_PORTS) {
188 		mthca_err(mdev, "HCA has %d ports, but we only support %d, "
189 			  "aborting.\n",
190 			  dev_lim->num_ports, MTHCA_MAX_PORTS);
191 		return -ENODEV;
192 	}
193 
194 	if (dev_lim->uar_size > pci_resource_len(mdev->pdev, 2)) {
195 		mthca_err(mdev, "HCA reported UAR size of 0x%x bigger than "
196 			  "PCI resource 2 size of 0x%llx, aborting.\n",
197 			  dev_lim->uar_size,
198 			  (unsigned long long)pci_resource_len(mdev->pdev, 2));
199 		return -ENODEV;
200 	}
201 
202 	mdev->limits.num_ports      	= dev_lim->num_ports;
203 	mdev->limits.vl_cap             = dev_lim->max_vl;
204 	mdev->limits.mtu_cap            = dev_lim->max_mtu;
205 	mdev->limits.gid_table_len  	= dev_lim->max_gids;
206 	mdev->limits.pkey_table_len 	= dev_lim->max_pkeys;
207 	mdev->limits.local_ca_ack_delay = dev_lim->local_ca_ack_delay;
208 	mdev->limits.max_sg             = dev_lim->max_sg;
209 	mdev->limits.max_wqes           = dev_lim->max_qp_sz;
210 	mdev->limits.max_qp_init_rdma   = dev_lim->max_requester_per_qp;
211 	mdev->limits.reserved_qps       = dev_lim->reserved_qps;
212 	mdev->limits.max_srq_wqes       = dev_lim->max_srq_sz;
213 	mdev->limits.reserved_srqs      = dev_lim->reserved_srqs;
214 	mdev->limits.reserved_eecs      = dev_lim->reserved_eecs;
215 	mdev->limits.max_desc_sz        = dev_lim->max_desc_sz;
216 	mdev->limits.max_srq_sge	= mthca_max_srq_sge(mdev);
217 	/*
218 	 * Subtract 1 from the limit because we need to allocate a
219 	 * spare CQE so the HCA HW can tell the difference between an
220 	 * empty CQ and a full CQ.
221 	 */
222 	mdev->limits.max_cqes           = dev_lim->max_cq_sz - 1;
223 	mdev->limits.reserved_cqs       = dev_lim->reserved_cqs;
224 	mdev->limits.reserved_eqs       = dev_lim->reserved_eqs;
225 	mdev->limits.reserved_mtts      = dev_lim->reserved_mtts;
226 	mdev->limits.reserved_mrws      = dev_lim->reserved_mrws;
227 	mdev->limits.reserved_uars      = dev_lim->reserved_uars;
228 	mdev->limits.reserved_pds       = dev_lim->reserved_pds;
229 	mdev->limits.port_width_cap     = dev_lim->max_port_width;
230 	mdev->limits.page_size_cap      = ~(u32) (dev_lim->min_page_sz - 1);
231 	mdev->limits.flags              = dev_lim->flags;
232 	/*
233 	 * For old FW that doesn't return static rate support, use a
234 	 * value of 0x3 (only static rate values of 0 or 1 are handled),
235 	 * except on Sinai, where even old FW can handle static rate
236 	 * values of 2 and 3.
237 	 */
238 	if (dev_lim->stat_rate_support)
239 		mdev->limits.stat_rate_support = dev_lim->stat_rate_support;
240 	else if (mdev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
241 		mdev->limits.stat_rate_support = 0xf;
242 	else
243 		mdev->limits.stat_rate_support = 0x3;
244 
245 	/* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
246 	   May be doable since hardware supports it for SRQ.
247 
248 	   IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
249 
250 	   IB_DEVICE_SRQ_RESIZE is supported by hardware but SRQ is not
251 	   supported by driver. */
252 	mdev->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
253 		IB_DEVICE_PORT_ACTIVE_EVENT |
254 		IB_DEVICE_SYS_IMAGE_GUID |
255 		IB_DEVICE_RC_RNR_NAK_GEN;
256 
257 	if (dev_lim->flags & DEV_LIM_FLAG_BAD_PKEY_CNTR)
258 		mdev->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
259 
260 	if (dev_lim->flags & DEV_LIM_FLAG_BAD_QKEY_CNTR)
261 		mdev->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
262 
263 	if (dev_lim->flags & DEV_LIM_FLAG_RAW_MULTI)
264 		mdev->device_cap_flags |= IB_DEVICE_RAW_MULTI;
265 
266 	if (dev_lim->flags & DEV_LIM_FLAG_AUTO_PATH_MIG)
267 		mdev->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
268 
269 	if (dev_lim->flags & DEV_LIM_FLAG_UD_AV_PORT_ENFORCE)
270 		mdev->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
271 
272 	if (dev_lim->flags & DEV_LIM_FLAG_SRQ)
273 		mdev->mthca_flags |= MTHCA_FLAG_SRQ;
274 
275 	return 0;
276 }
277 
278 static int mthca_init_tavor(struct mthca_dev *mdev)
279 {
280 	u8 status;
281 	int err;
282 	struct mthca_dev_lim        dev_lim;
283 	struct mthca_profile        profile;
284 	struct mthca_init_hca_param init_hca;
285 
286 	err = mthca_SYS_EN(mdev, &status);
287 	if (err) {
288 		mthca_err(mdev, "SYS_EN command failed, aborting.\n");
289 		return err;
290 	}
291 	if (status) {
292 		mthca_err(mdev, "SYS_EN returned status 0x%02x, "
293 			  "aborting.\n", status);
294 		return -EINVAL;
295 	}
296 
297 	err = mthca_QUERY_FW(mdev, &status);
298 	if (err) {
299 		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
300 		goto err_disable;
301 	}
302 	if (status) {
303 		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
304 			  "aborting.\n", status);
305 		err = -EINVAL;
306 		goto err_disable;
307 	}
308 	err = mthca_QUERY_DDR(mdev, &status);
309 	if (err) {
310 		mthca_err(mdev, "QUERY_DDR command failed, aborting.\n");
311 		goto err_disable;
312 	}
313 	if (status) {
314 		mthca_err(mdev, "QUERY_DDR returned status 0x%02x, "
315 			  "aborting.\n", status);
316 		err = -EINVAL;
317 		goto err_disable;
318 	}
319 
320 	err = mthca_dev_lim(mdev, &dev_lim);
321 	if (err) {
322 		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
323 		goto err_disable;
324 	}
325 
326 	profile = hca_profile;
327 	profile.num_uar   = dev_lim.uar_size / PAGE_SIZE;
328 	profile.uarc_size = 0;
329 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
330 		profile.num_srq = dev_lim.max_srqs;
331 
332 	err = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
333 	if (err < 0)
334 		goto err_disable;
335 
336 	err = mthca_INIT_HCA(mdev, &init_hca, &status);
337 	if (err) {
338 		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
339 		goto err_disable;
340 	}
341 	if (status) {
342 		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
343 			  "aborting.\n", status);
344 		err = -EINVAL;
345 		goto err_disable;
346 	}
347 
348 	return 0;
349 
350 err_disable:
351 	mthca_SYS_DIS(mdev, &status);
352 
353 	return err;
354 }
355 
356 static int mthca_load_fw(struct mthca_dev *mdev)
357 {
358 	u8 status;
359 	int err;
360 
361 	/* FIXME: use HCA-attached memory for FW if present */
362 
363 	mdev->fw.arbel.fw_icm =
364 		mthca_alloc_icm(mdev, mdev->fw.arbel.fw_pages,
365 				GFP_HIGHUSER | __GFP_NOWARN, 0);
366 	if (!mdev->fw.arbel.fw_icm) {
367 		mthca_err(mdev, "Couldn't allocate FW area, aborting.\n");
368 		return -ENOMEM;
369 	}
370 
371 	err = mthca_MAP_FA(mdev, mdev->fw.arbel.fw_icm, &status);
372 	if (err) {
373 		mthca_err(mdev, "MAP_FA command failed, aborting.\n");
374 		goto err_free;
375 	}
376 	if (status) {
377 		mthca_err(mdev, "MAP_FA returned status 0x%02x, aborting.\n", status);
378 		err = -EINVAL;
379 		goto err_free;
380 	}
381 	err = mthca_RUN_FW(mdev, &status);
382 	if (err) {
383 		mthca_err(mdev, "RUN_FW command failed, aborting.\n");
384 		goto err_unmap_fa;
385 	}
386 	if (status) {
387 		mthca_err(mdev, "RUN_FW returned status 0x%02x, aborting.\n", status);
388 		err = -EINVAL;
389 		goto err_unmap_fa;
390 	}
391 
392 	return 0;
393 
394 err_unmap_fa:
395 	mthca_UNMAP_FA(mdev, &status);
396 
397 err_free:
398 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
399 	return err;
400 }
401 
402 static int mthca_init_icm(struct mthca_dev *mdev,
403 			  struct mthca_dev_lim *dev_lim,
404 			  struct mthca_init_hca_param *init_hca,
405 			  u64 icm_size)
406 {
407 	u64 aux_pages;
408 	u8 status;
409 	int err;
410 
411 	err = mthca_SET_ICM_SIZE(mdev, icm_size, &aux_pages, &status);
412 	if (err) {
413 		mthca_err(mdev, "SET_ICM_SIZE command failed, aborting.\n");
414 		return err;
415 	}
416 	if (status) {
417 		mthca_err(mdev, "SET_ICM_SIZE returned status 0x%02x, "
418 			  "aborting.\n", status);
419 		return -EINVAL;
420 	}
421 
422 	mthca_dbg(mdev, "%lld KB of HCA context requires %lld KB aux memory.\n",
423 		  (unsigned long long) icm_size >> 10,
424 		  (unsigned long long) aux_pages << 2);
425 
426 	mdev->fw.arbel.aux_icm = mthca_alloc_icm(mdev, aux_pages,
427 						 GFP_HIGHUSER | __GFP_NOWARN, 0);
428 	if (!mdev->fw.arbel.aux_icm) {
429 		mthca_err(mdev, "Couldn't allocate aux memory, aborting.\n");
430 		return -ENOMEM;
431 	}
432 
433 	err = mthca_MAP_ICM_AUX(mdev, mdev->fw.arbel.aux_icm, &status);
434 	if (err) {
435 		mthca_err(mdev, "MAP_ICM_AUX command failed, aborting.\n");
436 		goto err_free_aux;
437 	}
438 	if (status) {
439 		mthca_err(mdev, "MAP_ICM_AUX returned status 0x%02x, aborting.\n", status);
440 		err = -EINVAL;
441 		goto err_free_aux;
442 	}
443 
444 	err = mthca_map_eq_icm(mdev, init_hca->eqc_base);
445 	if (err) {
446 		mthca_err(mdev, "Failed to map EQ context memory, aborting.\n");
447 		goto err_unmap_aux;
448 	}
449 
450 	/* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
451 	mdev->limits.reserved_mtts = ALIGN(mdev->limits.reserved_mtts * MTHCA_MTT_SEG_SIZE,
452 					   dma_get_cache_alignment()) / MTHCA_MTT_SEG_SIZE;
453 
454 	mdev->mr_table.mtt_table = mthca_alloc_icm_table(mdev, init_hca->mtt_base,
455 							 MTHCA_MTT_SEG_SIZE,
456 							 mdev->limits.num_mtt_segs,
457 							 mdev->limits.reserved_mtts,
458 							 1, 0);
459 	if (!mdev->mr_table.mtt_table) {
460 		mthca_err(mdev, "Failed to map MTT context memory, aborting.\n");
461 		err = -ENOMEM;
462 		goto err_unmap_eq;
463 	}
464 
465 	mdev->mr_table.mpt_table = mthca_alloc_icm_table(mdev, init_hca->mpt_base,
466 							 dev_lim->mpt_entry_sz,
467 							 mdev->limits.num_mpts,
468 							 mdev->limits.reserved_mrws,
469 							 1, 1);
470 	if (!mdev->mr_table.mpt_table) {
471 		mthca_err(mdev, "Failed to map MPT context memory, aborting.\n");
472 		err = -ENOMEM;
473 		goto err_unmap_mtt;
474 	}
475 
476 	mdev->qp_table.qp_table = mthca_alloc_icm_table(mdev, init_hca->qpc_base,
477 							dev_lim->qpc_entry_sz,
478 							mdev->limits.num_qps,
479 							mdev->limits.reserved_qps,
480 							0, 0);
481 	if (!mdev->qp_table.qp_table) {
482 		mthca_err(mdev, "Failed to map QP context memory, aborting.\n");
483 		err = -ENOMEM;
484 		goto err_unmap_mpt;
485 	}
486 
487 	mdev->qp_table.eqp_table = mthca_alloc_icm_table(mdev, init_hca->eqpc_base,
488 							 dev_lim->eqpc_entry_sz,
489 							 mdev->limits.num_qps,
490 							 mdev->limits.reserved_qps,
491 							 0, 0);
492 	if (!mdev->qp_table.eqp_table) {
493 		mthca_err(mdev, "Failed to map EQP context memory, aborting.\n");
494 		err = -ENOMEM;
495 		goto err_unmap_qp;
496 	}
497 
498 	mdev->qp_table.rdb_table = mthca_alloc_icm_table(mdev, init_hca->rdb_base,
499 							 MTHCA_RDB_ENTRY_SIZE,
500 							 mdev->limits.num_qps <<
501 							 mdev->qp_table.rdb_shift, 0,
502 							 0, 0);
503 	if (!mdev->qp_table.rdb_table) {
504 		mthca_err(mdev, "Failed to map RDB context memory, aborting\n");
505 		err = -ENOMEM;
506 		goto err_unmap_eqp;
507 	}
508 
509        mdev->cq_table.table = mthca_alloc_icm_table(mdev, init_hca->cqc_base,
510 						    dev_lim->cqc_entry_sz,
511 						    mdev->limits.num_cqs,
512 						    mdev->limits.reserved_cqs,
513 						    0, 0);
514 	if (!mdev->cq_table.table) {
515 		mthca_err(mdev, "Failed to map CQ context memory, aborting.\n");
516 		err = -ENOMEM;
517 		goto err_unmap_rdb;
518 	}
519 
520 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ) {
521 		mdev->srq_table.table =
522 			mthca_alloc_icm_table(mdev, init_hca->srqc_base,
523 					      dev_lim->srq_entry_sz,
524 					      mdev->limits.num_srqs,
525 					      mdev->limits.reserved_srqs,
526 					      0, 0);
527 		if (!mdev->srq_table.table) {
528 			mthca_err(mdev, "Failed to map SRQ context memory, "
529 				  "aborting.\n");
530 			err = -ENOMEM;
531 			goto err_unmap_cq;
532 		}
533 	}
534 
535 	/*
536 	 * It's not strictly required, but for simplicity just map the
537 	 * whole multicast group table now.  The table isn't very big
538 	 * and it's a lot easier than trying to track ref counts.
539 	 */
540 	mdev->mcg_table.table = mthca_alloc_icm_table(mdev, init_hca->mc_base,
541 						      MTHCA_MGM_ENTRY_SIZE,
542 						      mdev->limits.num_mgms +
543 						      mdev->limits.num_amgms,
544 						      mdev->limits.num_mgms +
545 						      mdev->limits.num_amgms,
546 						      0, 0);
547 	if (!mdev->mcg_table.table) {
548 		mthca_err(mdev, "Failed to map MCG context memory, aborting.\n");
549 		err = -ENOMEM;
550 		goto err_unmap_srq;
551 	}
552 
553 	return 0;
554 
555 err_unmap_srq:
556 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
557 		mthca_free_icm_table(mdev, mdev->srq_table.table);
558 
559 err_unmap_cq:
560 	mthca_free_icm_table(mdev, mdev->cq_table.table);
561 
562 err_unmap_rdb:
563 	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
564 
565 err_unmap_eqp:
566 	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
567 
568 err_unmap_qp:
569 	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
570 
571 err_unmap_mpt:
572 	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
573 
574 err_unmap_mtt:
575 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
576 
577 err_unmap_eq:
578 	mthca_unmap_eq_icm(mdev);
579 
580 err_unmap_aux:
581 	mthca_UNMAP_ICM_AUX(mdev, &status);
582 
583 err_free_aux:
584 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm, 0);
585 
586 	return err;
587 }
588 
589 static void mthca_free_icms(struct mthca_dev *mdev)
590 {
591 	u8 status;
592 
593 	mthca_free_icm_table(mdev, mdev->mcg_table.table);
594 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
595 		mthca_free_icm_table(mdev, mdev->srq_table.table);
596 	mthca_free_icm_table(mdev, mdev->cq_table.table);
597 	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
598 	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
599 	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
600 	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
601 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
602 	mthca_unmap_eq_icm(mdev);
603 
604 	mthca_UNMAP_ICM_AUX(mdev, &status);
605 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm, 0);
606 }
607 
608 static int mthca_init_arbel(struct mthca_dev *mdev)
609 {
610 	struct mthca_dev_lim        dev_lim;
611 	struct mthca_profile        profile;
612 	struct mthca_init_hca_param init_hca;
613 	u64 icm_size;
614 	u8 status;
615 	int err;
616 
617 	err = mthca_QUERY_FW(mdev, &status);
618 	if (err) {
619 		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
620 		return err;
621 	}
622 	if (status) {
623 		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
624 			  "aborting.\n", status);
625 		return -EINVAL;
626 	}
627 
628 	err = mthca_ENABLE_LAM(mdev, &status);
629 	if (err) {
630 		mthca_err(mdev, "ENABLE_LAM command failed, aborting.\n");
631 		return err;
632 	}
633 	if (status == MTHCA_CMD_STAT_LAM_NOT_PRE) {
634 		mthca_dbg(mdev, "No HCA-attached memory (running in MemFree mode)\n");
635 		mdev->mthca_flags |= MTHCA_FLAG_NO_LAM;
636 	} else if (status) {
637 		mthca_err(mdev, "ENABLE_LAM returned status 0x%02x, "
638 			  "aborting.\n", status);
639 		return -EINVAL;
640 	}
641 
642 	err = mthca_load_fw(mdev);
643 	if (err) {
644 		mthca_err(mdev, "Failed to start FW, aborting.\n");
645 		goto err_disable;
646 	}
647 
648 	err = mthca_dev_lim(mdev, &dev_lim);
649 	if (err) {
650 		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
651 		goto err_stop_fw;
652 	}
653 
654 	profile = hca_profile;
655 	profile.num_uar  = dev_lim.uar_size / PAGE_SIZE;
656 	profile.num_udav = 0;
657 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
658 		profile.num_srq = dev_lim.max_srqs;
659 
660 	icm_size = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
661 	if ((int) icm_size < 0) {
662 		err = icm_size;
663 		goto err_stop_fw;
664 	}
665 
666 	err = mthca_init_icm(mdev, &dev_lim, &init_hca, icm_size);
667 	if (err)
668 		goto err_stop_fw;
669 
670 	err = mthca_INIT_HCA(mdev, &init_hca, &status);
671 	if (err) {
672 		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
673 		goto err_free_icm;
674 	}
675 	if (status) {
676 		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
677 			  "aborting.\n", status);
678 		err = -EINVAL;
679 		goto err_free_icm;
680 	}
681 
682 	return 0;
683 
684 err_free_icm:
685 	mthca_free_icms(mdev);
686 
687 err_stop_fw:
688 	mthca_UNMAP_FA(mdev, &status);
689 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
690 
691 err_disable:
692 	if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
693 		mthca_DISABLE_LAM(mdev, &status);
694 
695 	return err;
696 }
697 
698 static void mthca_close_hca(struct mthca_dev *mdev)
699 {
700 	u8 status;
701 
702 	mthca_CLOSE_HCA(mdev, 0, &status);
703 
704 	if (mthca_is_memfree(mdev)) {
705 		mthca_free_icms(mdev);
706 
707 		mthca_UNMAP_FA(mdev, &status);
708 		mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
709 
710 		if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
711 			mthca_DISABLE_LAM(mdev, &status);
712 	} else
713 		mthca_SYS_DIS(mdev, &status);
714 }
715 
716 static int mthca_init_hca(struct mthca_dev *mdev)
717 {
718 	u8 status;
719 	int err;
720 	struct mthca_adapter adapter;
721 
722 	if (mthca_is_memfree(mdev))
723 		err = mthca_init_arbel(mdev);
724 	else
725 		err = mthca_init_tavor(mdev);
726 
727 	if (err)
728 		return err;
729 
730 	err = mthca_QUERY_ADAPTER(mdev, &adapter, &status);
731 	if (err) {
732 		mthca_err(mdev, "QUERY_ADAPTER command failed, aborting.\n");
733 		goto err_close;
734 	}
735 	if (status) {
736 		mthca_err(mdev, "QUERY_ADAPTER returned status 0x%02x, "
737 			  "aborting.\n", status);
738 		err = -EINVAL;
739 		goto err_close;
740 	}
741 
742 	mdev->eq_table.inta_pin = adapter.inta_pin;
743 	mdev->rev_id            = adapter.revision_id;
744 	memcpy(mdev->board_id, adapter.board_id, sizeof mdev->board_id);
745 
746 	return 0;
747 
748 err_close:
749 	mthca_close_hca(mdev);
750 	return err;
751 }
752 
753 static int mthca_setup_hca(struct mthca_dev *dev)
754 {
755 	int err;
756 	u8 status;
757 
758 	MTHCA_INIT_DOORBELL_LOCK(&dev->doorbell_lock);
759 
760 	err = mthca_init_uar_table(dev);
761 	if (err) {
762 		mthca_err(dev, "Failed to initialize "
763 			  "user access region table, aborting.\n");
764 		return err;
765 	}
766 
767 	err = mthca_uar_alloc(dev, &dev->driver_uar);
768 	if (err) {
769 		mthca_err(dev, "Failed to allocate driver access region, "
770 			  "aborting.\n");
771 		goto err_uar_table_free;
772 	}
773 
774 	dev->kar = ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
775 	if (!dev->kar) {
776 		mthca_err(dev, "Couldn't map kernel access region, "
777 			  "aborting.\n");
778 		err = -ENOMEM;
779 		goto err_uar_free;
780 	}
781 
782 	err = mthca_init_pd_table(dev);
783 	if (err) {
784 		mthca_err(dev, "Failed to initialize "
785 			  "protection domain table, aborting.\n");
786 		goto err_kar_unmap;
787 	}
788 
789 	err = mthca_init_mr_table(dev);
790 	if (err) {
791 		mthca_err(dev, "Failed to initialize "
792 			  "memory region table, aborting.\n");
793 		goto err_pd_table_free;
794 	}
795 
796 	err = mthca_pd_alloc(dev, 1, &dev->driver_pd);
797 	if (err) {
798 		mthca_err(dev, "Failed to create driver PD, "
799 			  "aborting.\n");
800 		goto err_mr_table_free;
801 	}
802 
803 	err = mthca_init_eq_table(dev);
804 	if (err) {
805 		mthca_err(dev, "Failed to initialize "
806 			  "event queue table, aborting.\n");
807 		goto err_pd_free;
808 	}
809 
810 	err = mthca_cmd_use_events(dev);
811 	if (err) {
812 		mthca_err(dev, "Failed to switch to event-driven "
813 			  "firmware commands, aborting.\n");
814 		goto err_eq_table_free;
815 	}
816 
817 	err = mthca_NOP(dev, &status);
818 	if (err || status) {
819 		if (dev->mthca_flags & (MTHCA_FLAG_MSI | MTHCA_FLAG_MSI_X)) {
820 			mthca_warn(dev, "NOP command failed to generate interrupt "
821 				   "(IRQ %d).\n",
822 				   dev->mthca_flags & MTHCA_FLAG_MSI_X ?
823 				   dev->eq_table.eq[MTHCA_EQ_CMD].msi_x_vector :
824 				   dev->pdev->irq);
825 			mthca_warn(dev, "Trying again with MSI/MSI-X disabled.\n");
826 		} else {
827 			mthca_err(dev, "NOP command failed to generate interrupt "
828 				  "(IRQ %d), aborting.\n",
829 				  dev->pdev->irq);
830 			mthca_err(dev, "BIOS or ACPI interrupt routing problem?\n");
831 		}
832 
833 		goto err_cmd_poll;
834 	}
835 
836 	mthca_dbg(dev, "NOP command IRQ test passed\n");
837 
838 	err = mthca_init_cq_table(dev);
839 	if (err) {
840 		mthca_err(dev, "Failed to initialize "
841 			  "completion queue table, aborting.\n");
842 		goto err_cmd_poll;
843 	}
844 
845 	err = mthca_init_srq_table(dev);
846 	if (err) {
847 		mthca_err(dev, "Failed to initialize "
848 			  "shared receive queue table, aborting.\n");
849 		goto err_cq_table_free;
850 	}
851 
852 	err = mthca_init_qp_table(dev);
853 	if (err) {
854 		mthca_err(dev, "Failed to initialize "
855 			  "queue pair table, aborting.\n");
856 		goto err_srq_table_free;
857 	}
858 
859 	err = mthca_init_av_table(dev);
860 	if (err) {
861 		mthca_err(dev, "Failed to initialize "
862 			  "address vector table, aborting.\n");
863 		goto err_qp_table_free;
864 	}
865 
866 	err = mthca_init_mcg_table(dev);
867 	if (err) {
868 		mthca_err(dev, "Failed to initialize "
869 			  "multicast group table, aborting.\n");
870 		goto err_av_table_free;
871 	}
872 
873 	return 0;
874 
875 err_av_table_free:
876 	mthca_cleanup_av_table(dev);
877 
878 err_qp_table_free:
879 	mthca_cleanup_qp_table(dev);
880 
881 err_srq_table_free:
882 	mthca_cleanup_srq_table(dev);
883 
884 err_cq_table_free:
885 	mthca_cleanup_cq_table(dev);
886 
887 err_cmd_poll:
888 	mthca_cmd_use_polling(dev);
889 
890 err_eq_table_free:
891 	mthca_cleanup_eq_table(dev);
892 
893 err_pd_free:
894 	mthca_pd_free(dev, &dev->driver_pd);
895 
896 err_mr_table_free:
897 	mthca_cleanup_mr_table(dev);
898 
899 err_pd_table_free:
900 	mthca_cleanup_pd_table(dev);
901 
902 err_kar_unmap:
903 	iounmap(dev->kar);
904 
905 err_uar_free:
906 	mthca_uar_free(dev, &dev->driver_uar);
907 
908 err_uar_table_free:
909 	mthca_cleanup_uar_table(dev);
910 	return err;
911 }
912 
913 static int mthca_request_regions(struct pci_dev *pdev, int ddr_hidden)
914 {
915 	int err;
916 
917 	/*
918 	 * We can't just use pci_request_regions() because the MSI-X
919 	 * table is right in the middle of the first BAR.  If we did
920 	 * pci_request_region and grab all of the first BAR, then
921 	 * setting up MSI-X would fail, since the PCI core wants to do
922 	 * request_mem_region on the MSI-X vector table.
923 	 *
924 	 * So just request what we need right now, and request any
925 	 * other regions we need when setting up EQs.
926 	 */
927 	if (!request_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
928 				MTHCA_HCR_SIZE, DRV_NAME))
929 		return -EBUSY;
930 
931 	err = pci_request_region(pdev, 2, DRV_NAME);
932 	if (err)
933 		goto err_bar2_failed;
934 
935 	if (!ddr_hidden) {
936 		err = pci_request_region(pdev, 4, DRV_NAME);
937 		if (err)
938 			goto err_bar4_failed;
939 	}
940 
941 	return 0;
942 
943 err_bar4_failed:
944 	pci_release_region(pdev, 2);
945 
946 err_bar2_failed:
947 	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
948 			   MTHCA_HCR_SIZE);
949 
950 	return err;
951 }
952 
953 static void mthca_release_regions(struct pci_dev *pdev,
954 				  int ddr_hidden)
955 {
956 	if (!ddr_hidden)
957 		pci_release_region(pdev, 4);
958 
959 	pci_release_region(pdev, 2);
960 
961 	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
962 			   MTHCA_HCR_SIZE);
963 }
964 
965 static int mthca_enable_msi_x(struct mthca_dev *mdev)
966 {
967 	struct msix_entry entries[3];
968 	int err;
969 
970 	entries[0].entry = 0;
971 	entries[1].entry = 1;
972 	entries[2].entry = 2;
973 
974 	err = pci_enable_msix(mdev->pdev, entries, ARRAY_SIZE(entries));
975 	if (err) {
976 		if (err > 0)
977 			mthca_info(mdev, "Only %d MSI-X vectors available, "
978 				   "not using MSI-X\n", err);
979 		return err;
980 	}
981 
982 	mdev->eq_table.eq[MTHCA_EQ_COMP ].msi_x_vector = entries[0].vector;
983 	mdev->eq_table.eq[MTHCA_EQ_ASYNC].msi_x_vector = entries[1].vector;
984 	mdev->eq_table.eq[MTHCA_EQ_CMD  ].msi_x_vector = entries[2].vector;
985 
986 	return 0;
987 }
988 
989 /* Types of supported HCA */
990 enum {
991 	TAVOR,			/* MT23108                        */
992 	ARBEL_COMPAT,		/* MT25208 in Tavor compat mode   */
993 	ARBEL_NATIVE,		/* MT25208 with extended features */
994 	SINAI			/* MT25204 */
995 };
996 
997 #define MTHCA_FW_VER(major, minor, subminor) \
998 	(((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
999 
1000 static struct {
1001 	u64 latest_fw;
1002 	u32 flags;
1003 } mthca_hca_table[] = {
1004 	[TAVOR]        = { .latest_fw = MTHCA_FW_VER(3, 5, 0),
1005 			   .flags     = 0 },
1006 	[ARBEL_COMPAT] = { .latest_fw = MTHCA_FW_VER(4, 8, 200),
1007 			   .flags     = MTHCA_FLAG_PCIE },
1008 	[ARBEL_NATIVE] = { .latest_fw = MTHCA_FW_VER(5, 2, 0),
1009 			   .flags     = MTHCA_FLAG_MEMFREE |
1010 					MTHCA_FLAG_PCIE },
1011 	[SINAI]        = { .latest_fw = MTHCA_FW_VER(1, 2, 0),
1012 			   .flags     = MTHCA_FLAG_MEMFREE |
1013 					MTHCA_FLAG_PCIE    |
1014 					MTHCA_FLAG_SINAI_OPT }
1015 };
1016 
1017 static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
1018 {
1019 	int ddr_hidden = 0;
1020 	int err;
1021 	struct mthca_dev *mdev;
1022 
1023 	printk(KERN_INFO PFX "Initializing %s\n",
1024 	       pci_name(pdev));
1025 
1026 	err = pci_enable_device(pdev);
1027 	if (err) {
1028 		dev_err(&pdev->dev, "Cannot enable PCI device, "
1029 			"aborting.\n");
1030 		return err;
1031 	}
1032 
1033 	/*
1034 	 * Check for BARs.  We expect 0: 1MB, 2: 8MB, 4: DDR (may not
1035 	 * be present)
1036 	 */
1037 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
1038 	    pci_resource_len(pdev, 0) != 1 << 20) {
1039 		dev_err(&pdev->dev, "Missing DCS, aborting.\n");
1040 		err = -ENODEV;
1041 		goto err_disable_pdev;
1042 	}
1043 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
1044 		dev_err(&pdev->dev, "Missing UAR, aborting.\n");
1045 		err = -ENODEV;
1046 		goto err_disable_pdev;
1047 	}
1048 	if (!(pci_resource_flags(pdev, 4) & IORESOURCE_MEM))
1049 		ddr_hidden = 1;
1050 
1051 	err = mthca_request_regions(pdev, ddr_hidden);
1052 	if (err) {
1053 		dev_err(&pdev->dev, "Cannot obtain PCI resources, "
1054 			"aborting.\n");
1055 		goto err_disable_pdev;
1056 	}
1057 
1058 	pci_set_master(pdev);
1059 
1060 	err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
1061 	if (err) {
1062 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1063 		err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1064 		if (err) {
1065 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
1066 			goto err_free_res;
1067 		}
1068 	}
1069 	err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
1070 	if (err) {
1071 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
1072 			 "consistent PCI DMA mask.\n");
1073 		err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1074 		if (err) {
1075 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
1076 				"aborting.\n");
1077 			goto err_free_res;
1078 		}
1079 	}
1080 
1081 	mdev = (struct mthca_dev *) ib_alloc_device(sizeof *mdev);
1082 	if (!mdev) {
1083 		dev_err(&pdev->dev, "Device struct alloc failed, "
1084 			"aborting.\n");
1085 		err = -ENOMEM;
1086 		goto err_free_res;
1087 	}
1088 
1089 	mdev->pdev = pdev;
1090 
1091 	mdev->mthca_flags = mthca_hca_table[hca_type].flags;
1092 	if (ddr_hidden)
1093 		mdev->mthca_flags |= MTHCA_FLAG_DDR_HIDDEN;
1094 
1095 	/*
1096 	 * Now reset the HCA before we touch the PCI capabilities or
1097 	 * attempt a firmware command, since a boot ROM may have left
1098 	 * the HCA in an undefined state.
1099 	 */
1100 	err = mthca_reset(mdev);
1101 	if (err) {
1102 		mthca_err(mdev, "Failed to reset HCA, aborting.\n");
1103 		goto err_free_dev;
1104 	}
1105 
1106 	if (mthca_cmd_init(mdev)) {
1107 		mthca_err(mdev, "Failed to init command interface, aborting.\n");
1108 		goto err_free_dev;
1109 	}
1110 
1111 	err = mthca_tune_pci(mdev);
1112 	if (err)
1113 		goto err_cmd;
1114 
1115 	err = mthca_init_hca(mdev);
1116 	if (err)
1117 		goto err_cmd;
1118 
1119 	if (mdev->fw_ver < mthca_hca_table[hca_type].latest_fw) {
1120 		mthca_warn(mdev, "HCA FW version %d.%d.%03d is old (%d.%d.%03d is current).\n",
1121 			   (int) (mdev->fw_ver >> 32), (int) (mdev->fw_ver >> 16) & 0xffff,
1122 			   (int) (mdev->fw_ver & 0xffff),
1123 			   (int) (mthca_hca_table[hca_type].latest_fw >> 32),
1124 			   (int) (mthca_hca_table[hca_type].latest_fw >> 16) & 0xffff,
1125 			   (int) (mthca_hca_table[hca_type].latest_fw & 0xffff));
1126 		mthca_warn(mdev, "If you have problems, try updating your HCA FW.\n");
1127 	}
1128 
1129 	if (msi_x && !mthca_enable_msi_x(mdev))
1130 		mdev->mthca_flags |= MTHCA_FLAG_MSI_X;
1131 	else if (msi) {
1132 		static int warned;
1133 
1134 		if (!warned) {
1135 			printk(KERN_WARNING PFX "WARNING: MSI support will be "
1136 			       "removed from the ib_mthca driver in January 2008.\n");
1137 			printk(KERN_WARNING "    If you are using MSI and cannot "
1138 			       "switch to MSI-X, please tell "
1139 			       "<general@lists.openfabrics.org>.\n");
1140 			++warned;
1141 		}
1142 
1143 		if (!pci_enable_msi(pdev))
1144 			mdev->mthca_flags |= MTHCA_FLAG_MSI;
1145 	}
1146 
1147 	err = mthca_setup_hca(mdev);
1148 	if (err == -EBUSY && (mdev->mthca_flags & (MTHCA_FLAG_MSI | MTHCA_FLAG_MSI_X))) {
1149 		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1150 			pci_disable_msix(pdev);
1151 		if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1152 			pci_disable_msi(pdev);
1153 		mdev->mthca_flags &= ~(MTHCA_FLAG_MSI_X | MTHCA_FLAG_MSI);
1154 
1155 		err = mthca_setup_hca(mdev);
1156 	}
1157 
1158 	if (err)
1159 		goto err_close;
1160 
1161 	err = mthca_register_device(mdev);
1162 	if (err)
1163 		goto err_cleanup;
1164 
1165 	err = mthca_create_agents(mdev);
1166 	if (err)
1167 		goto err_unregister;
1168 
1169 	pci_set_drvdata(pdev, mdev);
1170 	mdev->hca_type = hca_type;
1171 
1172 	return 0;
1173 
1174 err_unregister:
1175 	mthca_unregister_device(mdev);
1176 
1177 err_cleanup:
1178 	mthca_cleanup_mcg_table(mdev);
1179 	mthca_cleanup_av_table(mdev);
1180 	mthca_cleanup_qp_table(mdev);
1181 	mthca_cleanup_srq_table(mdev);
1182 	mthca_cleanup_cq_table(mdev);
1183 	mthca_cmd_use_polling(mdev);
1184 	mthca_cleanup_eq_table(mdev);
1185 
1186 	mthca_pd_free(mdev, &mdev->driver_pd);
1187 
1188 	mthca_cleanup_mr_table(mdev);
1189 	mthca_cleanup_pd_table(mdev);
1190 	mthca_cleanup_uar_table(mdev);
1191 
1192 err_close:
1193 	if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1194 		pci_disable_msix(pdev);
1195 	if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1196 		pci_disable_msi(pdev);
1197 
1198 	mthca_close_hca(mdev);
1199 
1200 err_cmd:
1201 	mthca_cmd_cleanup(mdev);
1202 
1203 err_free_dev:
1204 	ib_dealloc_device(&mdev->ib_dev);
1205 
1206 err_free_res:
1207 	mthca_release_regions(pdev, ddr_hidden);
1208 
1209 err_disable_pdev:
1210 	pci_disable_device(pdev);
1211 	pci_set_drvdata(pdev, NULL);
1212 	return err;
1213 }
1214 
1215 static void __mthca_remove_one(struct pci_dev *pdev)
1216 {
1217 	struct mthca_dev *mdev = pci_get_drvdata(pdev);
1218 	u8 status;
1219 	int p;
1220 
1221 	if (mdev) {
1222 		mthca_free_agents(mdev);
1223 		mthca_unregister_device(mdev);
1224 
1225 		for (p = 1; p <= mdev->limits.num_ports; ++p)
1226 			mthca_CLOSE_IB(mdev, p, &status);
1227 
1228 		mthca_cleanup_mcg_table(mdev);
1229 		mthca_cleanup_av_table(mdev);
1230 		mthca_cleanup_qp_table(mdev);
1231 		mthca_cleanup_srq_table(mdev);
1232 		mthca_cleanup_cq_table(mdev);
1233 		mthca_cmd_use_polling(mdev);
1234 		mthca_cleanup_eq_table(mdev);
1235 
1236 		mthca_pd_free(mdev, &mdev->driver_pd);
1237 
1238 		mthca_cleanup_mr_table(mdev);
1239 		mthca_cleanup_pd_table(mdev);
1240 
1241 		iounmap(mdev->kar);
1242 		mthca_uar_free(mdev, &mdev->driver_uar);
1243 		mthca_cleanup_uar_table(mdev);
1244 		mthca_close_hca(mdev);
1245 		mthca_cmd_cleanup(mdev);
1246 
1247 		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1248 			pci_disable_msix(pdev);
1249 		if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1250 			pci_disable_msi(pdev);
1251 
1252 		ib_dealloc_device(&mdev->ib_dev);
1253 		mthca_release_regions(pdev, mdev->mthca_flags &
1254 				      MTHCA_FLAG_DDR_HIDDEN);
1255 		pci_disable_device(pdev);
1256 		pci_set_drvdata(pdev, NULL);
1257 	}
1258 }
1259 
1260 int __mthca_restart_one(struct pci_dev *pdev)
1261 {
1262 	struct mthca_dev *mdev;
1263 	int hca_type;
1264 
1265 	mdev = pci_get_drvdata(pdev);
1266 	if (!mdev)
1267 		return -ENODEV;
1268 	hca_type = mdev->hca_type;
1269 	__mthca_remove_one(pdev);
1270 	return __mthca_init_one(pdev, hca_type);
1271 }
1272 
1273 static int __devinit mthca_init_one(struct pci_dev *pdev,
1274 				    const struct pci_device_id *id)
1275 {
1276 	static int mthca_version_printed = 0;
1277 	int ret;
1278 
1279 	mutex_lock(&mthca_device_mutex);
1280 
1281 	if (!mthca_version_printed) {
1282 		printk(KERN_INFO "%s", mthca_version);
1283 		++mthca_version_printed;
1284 	}
1285 
1286 	if (id->driver_data >= ARRAY_SIZE(mthca_hca_table)) {
1287 		printk(KERN_ERR PFX "%s has invalid driver data %lx\n",
1288 		       pci_name(pdev), id->driver_data);
1289 		mutex_unlock(&mthca_device_mutex);
1290 		return -ENODEV;
1291 	}
1292 
1293 	ret = __mthca_init_one(pdev, id->driver_data);
1294 
1295 	mutex_unlock(&mthca_device_mutex);
1296 
1297 	return ret;
1298 }
1299 
1300 static void __devexit mthca_remove_one(struct pci_dev *pdev)
1301 {
1302 	mutex_lock(&mthca_device_mutex);
1303 	__mthca_remove_one(pdev);
1304 	mutex_unlock(&mthca_device_mutex);
1305 }
1306 
1307 static struct pci_device_id mthca_pci_table[] = {
1308 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_TAVOR),
1309 	  .driver_data = TAVOR },
1310 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_TAVOR),
1311 	  .driver_data = TAVOR },
1312 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1313 	  .driver_data = ARBEL_COMPAT },
1314 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1315 	  .driver_data = ARBEL_COMPAT },
1316 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL),
1317 	  .driver_data = ARBEL_NATIVE },
1318 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL),
1319 	  .driver_data = ARBEL_NATIVE },
1320 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI),
1321 	  .driver_data = SINAI },
1322 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI),
1323 	  .driver_data = SINAI },
1324 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1325 	  .driver_data = SINAI },
1326 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1327 	  .driver_data = SINAI },
1328 	{ 0, }
1329 };
1330 
1331 MODULE_DEVICE_TABLE(pci, mthca_pci_table);
1332 
1333 static struct pci_driver mthca_driver = {
1334 	.name		= DRV_NAME,
1335 	.id_table	= mthca_pci_table,
1336 	.probe		= mthca_init_one,
1337 	.remove		= __devexit_p(mthca_remove_one)
1338 };
1339 
1340 static void __init __mthca_check_profile_val(const char *name, int *pval,
1341 					     int pval_default)
1342 {
1343 	/* value must be positive and power of 2 */
1344 	int old_pval = *pval;
1345 
1346 	if (old_pval <= 0)
1347 		*pval = pval_default;
1348 	else
1349 		*pval = roundup_pow_of_two(old_pval);
1350 
1351 	if (old_pval != *pval) {
1352 		printk(KERN_WARNING PFX "Invalid value %d for %s in module parameter.\n",
1353 		       old_pval, name);
1354 		printk(KERN_WARNING PFX "Corrected %s to %d.\n", name, *pval);
1355 	}
1356 }
1357 
1358 #define mthca_check_profile_val(name, default)				\
1359 	__mthca_check_profile_val(#name, &hca_profile.name, default)
1360 
1361 static void __init mthca_validate_profile(void)
1362 {
1363 	mthca_check_profile_val(num_qp,            MTHCA_DEFAULT_NUM_QP);
1364 	mthca_check_profile_val(rdb_per_qp,        MTHCA_DEFAULT_RDB_PER_QP);
1365 	mthca_check_profile_val(num_cq,            MTHCA_DEFAULT_NUM_CQ);
1366 	mthca_check_profile_val(num_mcg, 	   MTHCA_DEFAULT_NUM_MCG);
1367 	mthca_check_profile_val(num_mpt, 	   MTHCA_DEFAULT_NUM_MPT);
1368 	mthca_check_profile_val(num_mtt, 	   MTHCA_DEFAULT_NUM_MTT);
1369 	mthca_check_profile_val(num_udav,          MTHCA_DEFAULT_NUM_UDAV);
1370 	mthca_check_profile_val(fmr_reserved_mtts, MTHCA_DEFAULT_NUM_RESERVED_MTTS);
1371 
1372 	if (hca_profile.fmr_reserved_mtts >= hca_profile.num_mtt) {
1373 		printk(KERN_WARNING PFX "Invalid fmr_reserved_mtts module parameter %d.\n",
1374 		       hca_profile.fmr_reserved_mtts);
1375 		printk(KERN_WARNING PFX "(Must be smaller than num_mtt %d)\n",
1376 		       hca_profile.num_mtt);
1377 		hca_profile.fmr_reserved_mtts = hca_profile.num_mtt / 2;
1378 		printk(KERN_WARNING PFX "Corrected fmr_reserved_mtts to %d.\n",
1379 		       hca_profile.fmr_reserved_mtts);
1380 	}
1381 }
1382 
1383 static int __init mthca_init(void)
1384 {
1385 	int ret;
1386 
1387 	mthca_validate_profile();
1388 
1389 	ret = mthca_catas_init();
1390 	if (ret)
1391 		return ret;
1392 
1393 	ret = pci_register_driver(&mthca_driver);
1394 	if (ret < 0) {
1395 		mthca_catas_cleanup();
1396 		return ret;
1397 	}
1398 
1399 	return 0;
1400 }
1401 
1402 static void __exit mthca_cleanup(void)
1403 {
1404 	pci_unregister_driver(&mthca_driver);
1405 	mthca_catas_cleanup();
1406 }
1407 
1408 module_init(mthca_init);
1409 module_exit(mthca_cleanup);
1410