1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  * $Id: mthca_main.c 1396 2004-12-28 04:10:27Z roland $
33  */
34 
35 #include <linux/config.h>
36 #include <linux/version.h>
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_PCI_MSI
55 
56 static int msi_x = 0;
57 module_param(msi_x, int, 0444);
58 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
59 
60 static int msi = 0;
61 module_param(msi, int, 0444);
62 MODULE_PARM_DESC(msi, "attempt to use MSI if nonzero");
63 
64 #else /* CONFIG_PCI_MSI */
65 
66 #define msi_x (0)
67 #define msi   (0)
68 
69 #endif /* CONFIG_PCI_MSI */
70 
71 static const char mthca_version[] __devinitdata =
72 	"ib_mthca: Mellanox InfiniBand HCA driver v"
73 	DRV_VERSION " (" DRV_RELDATE ")\n";
74 
75 static struct mthca_profile default_profile = {
76 	.num_qp		   = 1 << 16,
77 	.rdb_per_qp	   = 4,
78 	.num_cq		   = 1 << 16,
79 	.num_mcg	   = 1 << 13,
80 	.num_mpt	   = 1 << 17,
81 	.num_mtt	   = 1 << 20,
82 	.num_udav	   = 1 << 15,	/* Tavor only */
83 	.fmr_reserved_mtts = 1 << 18,	/* Tavor only */
84 	.uarc_size	   = 1 << 18,	/* Arbel only */
85 };
86 
87 static int __devinit mthca_tune_pci(struct mthca_dev *mdev)
88 {
89 	int cap;
90 	u16 val;
91 
92 	/* First try to max out Read Byte Count */
93 	cap = pci_find_capability(mdev->pdev, PCI_CAP_ID_PCIX);
94 	if (cap) {
95 		if (pci_read_config_word(mdev->pdev, cap + PCI_X_CMD, &val)) {
96 			mthca_err(mdev, "Couldn't read PCI-X command register, "
97 				  "aborting.\n");
98 			return -ENODEV;
99 		}
100 		val = (val & ~PCI_X_CMD_MAX_READ) | (3 << 2);
101 		if (pci_write_config_word(mdev->pdev, cap + PCI_X_CMD, val)) {
102 			mthca_err(mdev, "Couldn't write PCI-X command register, "
103 				  "aborting.\n");
104 			return -ENODEV;
105 		}
106 	} else if (mdev->hca_type == TAVOR)
107 		mthca_info(mdev, "No PCI-X capability, not setting RBC.\n");
108 
109 	cap = pci_find_capability(mdev->pdev, PCI_CAP_ID_EXP);
110 	if (cap) {
111 		if (pci_read_config_word(mdev->pdev, cap + PCI_EXP_DEVCTL, &val)) {
112 			mthca_err(mdev, "Couldn't read PCI Express device control "
113 				  "register, aborting.\n");
114 			return -ENODEV;
115 		}
116 		val = (val & ~PCI_EXP_DEVCTL_READRQ) | (5 << 12);
117 		if (pci_write_config_word(mdev->pdev, cap + PCI_EXP_DEVCTL, val)) {
118 			mthca_err(mdev, "Couldn't write PCI Express device control "
119 				  "register, aborting.\n");
120 			return -ENODEV;
121 		}
122 	} else if (mdev->hca_type == ARBEL_NATIVE ||
123 		   mdev->hca_type == ARBEL_COMPAT)
124 		mthca_info(mdev, "No PCI Express capability, "
125 			   "not setting Max Read Request Size.\n");
126 
127 	return 0;
128 }
129 
130 static int __devinit mthca_dev_lim(struct mthca_dev *mdev, struct mthca_dev_lim *dev_lim)
131 {
132 	int err;
133 	u8 status;
134 
135 	err = mthca_QUERY_DEV_LIM(mdev, dev_lim, &status);
136 	if (err) {
137 		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
138 		return err;
139 	}
140 	if (status) {
141 		mthca_err(mdev, "QUERY_DEV_LIM returned status 0x%02x, "
142 			  "aborting.\n", status);
143 		return -EINVAL;
144 	}
145 	if (dev_lim->min_page_sz > PAGE_SIZE) {
146 		mthca_err(mdev, "HCA minimum page size of %d bigger than "
147 			  "kernel PAGE_SIZE of %ld, aborting.\n",
148 			  dev_lim->min_page_sz, PAGE_SIZE);
149 		return -ENODEV;
150 	}
151 	if (dev_lim->num_ports > MTHCA_MAX_PORTS) {
152 		mthca_err(mdev, "HCA has %d ports, but we only support %d, "
153 			  "aborting.\n",
154 			  dev_lim->num_ports, MTHCA_MAX_PORTS);
155 		return -ENODEV;
156 	}
157 
158 	mdev->limits.num_ports      	= dev_lim->num_ports;
159 	mdev->limits.vl_cap             = dev_lim->max_vl;
160 	mdev->limits.mtu_cap            = dev_lim->max_mtu;
161 	mdev->limits.gid_table_len  	= dev_lim->max_gids;
162 	mdev->limits.pkey_table_len 	= dev_lim->max_pkeys;
163 	mdev->limits.local_ca_ack_delay = dev_lim->local_ca_ack_delay;
164 	mdev->limits.max_sg             = dev_lim->max_sg;
165 	mdev->limits.reserved_qps       = dev_lim->reserved_qps;
166 	mdev->limits.reserved_srqs      = dev_lim->reserved_srqs;
167 	mdev->limits.reserved_eecs      = dev_lim->reserved_eecs;
168 	mdev->limits.reserved_cqs       = dev_lim->reserved_cqs;
169 	mdev->limits.reserved_eqs       = dev_lim->reserved_eqs;
170 	mdev->limits.reserved_mtts      = dev_lim->reserved_mtts;
171 	mdev->limits.reserved_mrws      = dev_lim->reserved_mrws;
172 	mdev->limits.reserved_uars      = dev_lim->reserved_uars;
173 	mdev->limits.reserved_pds       = dev_lim->reserved_pds;
174 
175 	/* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
176 	   May be doable since hardware supports it for SRQ.
177 
178 	   IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
179 
180 	   IB_DEVICE_SRQ_RESIZE is supported by hardware but SRQ is not
181 	   supported by driver. */
182 	mdev->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
183 		IB_DEVICE_PORT_ACTIVE_EVENT |
184 		IB_DEVICE_SYS_IMAGE_GUID |
185 		IB_DEVICE_RC_RNR_NAK_GEN;
186 
187 	if (dev_lim->flags & DEV_LIM_FLAG_BAD_PKEY_CNTR)
188 		mdev->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
189 
190 	if (dev_lim->flags & DEV_LIM_FLAG_BAD_QKEY_CNTR)
191 		mdev->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
192 
193 	if (dev_lim->flags & DEV_LIM_FLAG_RAW_MULTI)
194 		mdev->device_cap_flags |= IB_DEVICE_RAW_MULTI;
195 
196 	if (dev_lim->flags & DEV_LIM_FLAG_AUTO_PATH_MIG)
197 		mdev->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
198 
199 	if (dev_lim->flags & DEV_LIM_FLAG_UD_AV_PORT_ENFORCE)
200 		mdev->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
201 
202 	if (dev_lim->flags & DEV_LIM_FLAG_SRQ)
203 		mdev->mthca_flags |= MTHCA_FLAG_SRQ;
204 
205 	return 0;
206 }
207 
208 static int __devinit mthca_init_tavor(struct mthca_dev *mdev)
209 {
210 	u8 status;
211 	int err;
212 	struct mthca_dev_lim        dev_lim;
213 	struct mthca_profile        profile;
214 	struct mthca_init_hca_param init_hca;
215 	struct mthca_adapter        adapter;
216 
217 	err = mthca_SYS_EN(mdev, &status);
218 	if (err) {
219 		mthca_err(mdev, "SYS_EN command failed, aborting.\n");
220 		return err;
221 	}
222 	if (status) {
223 		mthca_err(mdev, "SYS_EN returned status 0x%02x, "
224 			  "aborting.\n", status);
225 		return -EINVAL;
226 	}
227 
228 	err = mthca_QUERY_FW(mdev, &status);
229 	if (err) {
230 		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
231 		goto err_disable;
232 	}
233 	if (status) {
234 		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
235 			  "aborting.\n", status);
236 		err = -EINVAL;
237 		goto err_disable;
238 	}
239 	err = mthca_QUERY_DDR(mdev, &status);
240 	if (err) {
241 		mthca_err(mdev, "QUERY_DDR command failed, aborting.\n");
242 		goto err_disable;
243 	}
244 	if (status) {
245 		mthca_err(mdev, "QUERY_DDR returned status 0x%02x, "
246 			  "aborting.\n", status);
247 		err = -EINVAL;
248 		goto err_disable;
249 	}
250 
251 	err = mthca_dev_lim(mdev, &dev_lim);
252 
253 	profile = default_profile;
254 	profile.num_uar   = dev_lim.uar_size / PAGE_SIZE;
255 	profile.uarc_size = 0;
256 
257 	err = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
258 	if (err < 0)
259 		goto err_disable;
260 
261 	err = mthca_INIT_HCA(mdev, &init_hca, &status);
262 	if (err) {
263 		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
264 		goto err_disable;
265 	}
266 	if (status) {
267 		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
268 			  "aborting.\n", status);
269 		err = -EINVAL;
270 		goto err_disable;
271 	}
272 
273 	err = mthca_QUERY_ADAPTER(mdev, &adapter, &status);
274 	if (err) {
275 		mthca_err(mdev, "QUERY_ADAPTER command failed, aborting.\n");
276 		goto err_close;
277 	}
278 	if (status) {
279 		mthca_err(mdev, "QUERY_ADAPTER returned status 0x%02x, "
280 			  "aborting.\n", status);
281 		err = -EINVAL;
282 		goto err_close;
283 	}
284 
285 	mdev->eq_table.inta_pin = adapter.inta_pin;
286 	mdev->rev_id            = adapter.revision_id;
287 
288 	return 0;
289 
290 err_close:
291 	mthca_CLOSE_HCA(mdev, 0, &status);
292 
293 err_disable:
294 	mthca_SYS_DIS(mdev, &status);
295 
296 	return err;
297 }
298 
299 static int __devinit mthca_load_fw(struct mthca_dev *mdev)
300 {
301 	u8 status;
302 	int err;
303 
304 	/* FIXME: use HCA-attached memory for FW if present */
305 
306 	mdev->fw.arbel.fw_icm =
307 		mthca_alloc_icm(mdev, mdev->fw.arbel.fw_pages,
308 				GFP_HIGHUSER | __GFP_NOWARN);
309 	if (!mdev->fw.arbel.fw_icm) {
310 		mthca_err(mdev, "Couldn't allocate FW area, aborting.\n");
311 		return -ENOMEM;
312 	}
313 
314 	err = mthca_MAP_FA(mdev, mdev->fw.arbel.fw_icm, &status);
315 	if (err) {
316 		mthca_err(mdev, "MAP_FA command failed, aborting.\n");
317 		goto err_free;
318 	}
319 	if (status) {
320 		mthca_err(mdev, "MAP_FA returned status 0x%02x, aborting.\n", status);
321 		err = -EINVAL;
322 		goto err_free;
323 	}
324 	err = mthca_RUN_FW(mdev, &status);
325 	if (err) {
326 		mthca_err(mdev, "RUN_FW command failed, aborting.\n");
327 		goto err_unmap_fa;
328 	}
329 	if (status) {
330 		mthca_err(mdev, "RUN_FW returned status 0x%02x, aborting.\n", status);
331 		err = -EINVAL;
332 		goto err_unmap_fa;
333 	}
334 
335 	return 0;
336 
337 err_unmap_fa:
338 	mthca_UNMAP_FA(mdev, &status);
339 
340 err_free:
341 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
342 	return err;
343 }
344 
345 static int __devinit mthca_init_icm(struct mthca_dev *mdev,
346 				    struct mthca_dev_lim *dev_lim,
347 				    struct mthca_init_hca_param *init_hca,
348 				    u64 icm_size)
349 {
350 	u64 aux_pages;
351 	u8 status;
352 	int err;
353 
354 	err = mthca_SET_ICM_SIZE(mdev, icm_size, &aux_pages, &status);
355 	if (err) {
356 		mthca_err(mdev, "SET_ICM_SIZE command failed, aborting.\n");
357 		return err;
358 	}
359 	if (status) {
360 		mthca_err(mdev, "SET_ICM_SIZE returned status 0x%02x, "
361 			  "aborting.\n", status);
362 		return -EINVAL;
363 	}
364 
365 	mthca_dbg(mdev, "%lld KB of HCA context requires %lld KB aux memory.\n",
366 		  (unsigned long long) icm_size >> 10,
367 		  (unsigned long long) aux_pages << 2);
368 
369 	mdev->fw.arbel.aux_icm = mthca_alloc_icm(mdev, aux_pages,
370 						 GFP_HIGHUSER | __GFP_NOWARN);
371 	if (!mdev->fw.arbel.aux_icm) {
372 		mthca_err(mdev, "Couldn't allocate aux memory, aborting.\n");
373 		return -ENOMEM;
374 	}
375 
376 	err = mthca_MAP_ICM_AUX(mdev, mdev->fw.arbel.aux_icm, &status);
377 	if (err) {
378 		mthca_err(mdev, "MAP_ICM_AUX command failed, aborting.\n");
379 		goto err_free_aux;
380 	}
381 	if (status) {
382 		mthca_err(mdev, "MAP_ICM_AUX returned status 0x%02x, aborting.\n", status);
383 		err = -EINVAL;
384 		goto err_free_aux;
385 	}
386 
387 	err = mthca_map_eq_icm(mdev, init_hca->eqc_base);
388 	if (err) {
389 		mthca_err(mdev, "Failed to map EQ context memory, aborting.\n");
390 		goto err_unmap_aux;
391 	}
392 
393 	mdev->mr_table.mtt_table = mthca_alloc_icm_table(mdev, init_hca->mtt_base,
394 							 MTHCA_MTT_SEG_SIZE,
395 							 mdev->limits.num_mtt_segs,
396 							 mdev->limits.reserved_mtts, 1);
397 	if (!mdev->mr_table.mtt_table) {
398 		mthca_err(mdev, "Failed to map MTT context memory, aborting.\n");
399 		err = -ENOMEM;
400 		goto err_unmap_eq;
401 	}
402 
403 	mdev->mr_table.mpt_table = mthca_alloc_icm_table(mdev, init_hca->mpt_base,
404 							 dev_lim->mpt_entry_sz,
405 							 mdev->limits.num_mpts,
406 							 mdev->limits.reserved_mrws, 1);
407 	if (!mdev->mr_table.mpt_table) {
408 		mthca_err(mdev, "Failed to map MPT context memory, aborting.\n");
409 		err = -ENOMEM;
410 		goto err_unmap_mtt;
411 	}
412 
413 	mdev->qp_table.qp_table = mthca_alloc_icm_table(mdev, init_hca->qpc_base,
414 							dev_lim->qpc_entry_sz,
415 							mdev->limits.num_qps,
416 							mdev->limits.reserved_qps, 0);
417 	if (!mdev->qp_table.qp_table) {
418 		mthca_err(mdev, "Failed to map QP context memory, aborting.\n");
419 		err = -ENOMEM;
420 		goto err_unmap_mpt;
421 	}
422 
423 	mdev->qp_table.eqp_table = mthca_alloc_icm_table(mdev, init_hca->eqpc_base,
424 							 dev_lim->eqpc_entry_sz,
425 							 mdev->limits.num_qps,
426 							 mdev->limits.reserved_qps, 0);
427 	if (!mdev->qp_table.eqp_table) {
428 		mthca_err(mdev, "Failed to map EQP context memory, aborting.\n");
429 		err = -ENOMEM;
430 		goto err_unmap_qp;
431 	}
432 
433 	mdev->qp_table.rdb_table = mthca_alloc_icm_table(mdev, init_hca->rdb_base,
434 							 MTHCA_RDB_ENTRY_SIZE,
435 							 mdev->limits.num_qps <<
436 							 mdev->qp_table.rdb_shift,
437 							 0, 0);
438 	if (!mdev->qp_table.rdb_table) {
439 		mthca_err(mdev, "Failed to map RDB context memory, aborting\n");
440 		err = -ENOMEM;
441 		goto err_unmap_eqp;
442 	}
443 
444        mdev->cq_table.table = mthca_alloc_icm_table(mdev, init_hca->cqc_base,
445 						     dev_lim->cqc_entry_sz,
446 						     mdev->limits.num_cqs,
447 						     mdev->limits.reserved_cqs, 0);
448 	if (!mdev->cq_table.table) {
449 		mthca_err(mdev, "Failed to map CQ context memory, aborting.\n");
450 		err = -ENOMEM;
451 		goto err_unmap_rdb;
452 	}
453 
454 	/*
455 	 * It's not strictly required, but for simplicity just map the
456 	 * whole multicast group table now.  The table isn't very big
457 	 * and it's a lot easier than trying to track ref counts.
458 	 */
459 	mdev->mcg_table.table = mthca_alloc_icm_table(mdev, init_hca->mc_base,
460 						      MTHCA_MGM_ENTRY_SIZE,
461 						      mdev->limits.num_mgms +
462 						      mdev->limits.num_amgms,
463 						      mdev->limits.num_mgms +
464 						      mdev->limits.num_amgms,
465 						      0);
466 	if (!mdev->mcg_table.table) {
467 		mthca_err(mdev, "Failed to map MCG context memory, aborting.\n");
468 		err = -ENOMEM;
469 		goto err_unmap_cq;
470 	}
471 
472 	return 0;
473 
474 err_unmap_cq:
475 	mthca_free_icm_table(mdev, mdev->cq_table.table);
476 
477 err_unmap_rdb:
478 	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
479 
480 err_unmap_eqp:
481 	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
482 
483 err_unmap_qp:
484 	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
485 
486 err_unmap_mpt:
487 	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
488 
489 err_unmap_mtt:
490 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
491 
492 err_unmap_eq:
493 	mthca_unmap_eq_icm(mdev);
494 
495 err_unmap_aux:
496 	mthca_UNMAP_ICM_AUX(mdev, &status);
497 
498 err_free_aux:
499 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm);
500 
501 	return err;
502 }
503 
504 static int __devinit mthca_init_arbel(struct mthca_dev *mdev)
505 {
506 	struct mthca_dev_lim        dev_lim;
507 	struct mthca_profile        profile;
508 	struct mthca_init_hca_param init_hca;
509 	struct mthca_adapter        adapter;
510 	u64 icm_size;
511 	u8 status;
512 	int err;
513 
514 	err = mthca_QUERY_FW(mdev, &status);
515 	if (err) {
516 		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
517 		return err;
518 	}
519 	if (status) {
520 		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
521 			  "aborting.\n", status);
522 		return -EINVAL;
523 	}
524 
525 	err = mthca_ENABLE_LAM(mdev, &status);
526 	if (err) {
527 		mthca_err(mdev, "ENABLE_LAM command failed, aborting.\n");
528 		return err;
529 	}
530 	if (status == MTHCA_CMD_STAT_LAM_NOT_PRE) {
531 		mthca_dbg(mdev, "No HCA-attached memory (running in MemFree mode)\n");
532 		mdev->mthca_flags |= MTHCA_FLAG_NO_LAM;
533 	} else if (status) {
534 		mthca_err(mdev, "ENABLE_LAM returned status 0x%02x, "
535 			  "aborting.\n", status);
536 		return -EINVAL;
537 	}
538 
539 	err = mthca_load_fw(mdev);
540 	if (err) {
541 		mthca_err(mdev, "Failed to start FW, aborting.\n");
542 		goto err_disable;
543 	}
544 
545 	err = mthca_dev_lim(mdev, &dev_lim);
546 	if (err) {
547 		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
548 		goto err_stop_fw;
549 	}
550 
551 	profile = default_profile;
552 	profile.num_uar  = dev_lim.uar_size / PAGE_SIZE;
553 	profile.num_udav = 0;
554 
555 	icm_size = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
556 	if ((int) icm_size < 0) {
557 		err = icm_size;
558 		goto err_stop_fw;
559 	}
560 
561 	err = mthca_init_icm(mdev, &dev_lim, &init_hca, icm_size);
562 	if (err)
563 		goto err_stop_fw;
564 
565 	err = mthca_INIT_HCA(mdev, &init_hca, &status);
566 	if (err) {
567 		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
568 		goto err_free_icm;
569 	}
570 	if (status) {
571 		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
572 			  "aborting.\n", status);
573 		err = -EINVAL;
574 		goto err_free_icm;
575 	}
576 
577 	err = mthca_QUERY_ADAPTER(mdev, &adapter, &status);
578 	if (err) {
579 		mthca_err(mdev, "QUERY_ADAPTER command failed, aborting.\n");
580 		goto err_free_icm;
581 	}
582 	if (status) {
583 		mthca_err(mdev, "QUERY_ADAPTER returned status 0x%02x, "
584 			  "aborting.\n", status);
585 		err = -EINVAL;
586 		goto err_free_icm;
587 	}
588 
589 	mdev->eq_table.inta_pin = adapter.inta_pin;
590 	mdev->rev_id            = adapter.revision_id;
591 
592 	return 0;
593 
594 err_free_icm:
595 	mthca_free_icm_table(mdev, mdev->cq_table.table);
596 	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
597 	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
598 	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
599 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
600 	mthca_unmap_eq_icm(mdev);
601 
602 	mthca_UNMAP_ICM_AUX(mdev, &status);
603 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm);
604 
605 err_stop_fw:
606 	mthca_UNMAP_FA(mdev, &status);
607 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
608 
609 err_disable:
610 	if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
611 		mthca_DISABLE_LAM(mdev, &status);
612 
613 	return err;
614 }
615 
616 static int __devinit mthca_init_hca(struct mthca_dev *mdev)
617 {
618 	if (mthca_is_memfree(mdev))
619 		return mthca_init_arbel(mdev);
620 	else
621 		return mthca_init_tavor(mdev);
622 }
623 
624 static int __devinit mthca_setup_hca(struct mthca_dev *dev)
625 {
626 	int err;
627 	u8 status;
628 
629 	MTHCA_INIT_DOORBELL_LOCK(&dev->doorbell_lock);
630 
631 	err = mthca_init_uar_table(dev);
632 	if (err) {
633 		mthca_err(dev, "Failed to initialize "
634 			  "user access region table, aborting.\n");
635 		return err;
636 	}
637 
638 	err = mthca_uar_alloc(dev, &dev->driver_uar);
639 	if (err) {
640 		mthca_err(dev, "Failed to allocate driver access region, "
641 			  "aborting.\n");
642 		goto err_uar_table_free;
643 	}
644 
645 	dev->kar = ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
646 	if (!dev->kar) {
647 		mthca_err(dev, "Couldn't map kernel access region, "
648 			  "aborting.\n");
649 		err = -ENOMEM;
650 		goto err_uar_free;
651 	}
652 
653 	err = mthca_init_pd_table(dev);
654 	if (err) {
655 		mthca_err(dev, "Failed to initialize "
656 			  "protection domain table, aborting.\n");
657 		goto err_kar_unmap;
658 	}
659 
660 	err = mthca_init_mr_table(dev);
661 	if (err) {
662 		mthca_err(dev, "Failed to initialize "
663 			  "memory region table, aborting.\n");
664 		goto err_pd_table_free;
665 	}
666 
667 	err = mthca_pd_alloc(dev, &dev->driver_pd);
668 	if (err) {
669 		mthca_err(dev, "Failed to create driver PD, "
670 			  "aborting.\n");
671 		goto err_mr_table_free;
672 	}
673 
674 	err = mthca_init_eq_table(dev);
675 	if (err) {
676 		mthca_err(dev, "Failed to initialize "
677 			  "event queue table, aborting.\n");
678 		goto err_pd_free;
679 	}
680 
681 	err = mthca_cmd_use_events(dev);
682 	if (err) {
683 		mthca_err(dev, "Failed to switch to event-driven "
684 			  "firmware commands, aborting.\n");
685 		goto err_eq_table_free;
686 	}
687 
688 	err = mthca_NOP(dev, &status);
689 	if (err || status) {
690 		mthca_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting.\n",
691 			  dev->mthca_flags & MTHCA_FLAG_MSI_X ?
692 			  dev->eq_table.eq[MTHCA_EQ_CMD].msi_x_vector :
693 			  dev->pdev->irq);
694 		if (dev->mthca_flags & (MTHCA_FLAG_MSI | MTHCA_FLAG_MSI_X))
695 			mthca_err(dev, "Try again with MSI/MSI-X disabled.\n");
696 		else
697 			mthca_err(dev, "BIOS or ACPI interrupt routing problem?\n");
698 
699 		goto err_cmd_poll;
700 	}
701 
702 	mthca_dbg(dev, "NOP command IRQ test passed\n");
703 
704 	err = mthca_init_cq_table(dev);
705 	if (err) {
706 		mthca_err(dev, "Failed to initialize "
707 			  "completion queue table, aborting.\n");
708 		goto err_cmd_poll;
709 	}
710 
711 	err = mthca_init_qp_table(dev);
712 	if (err) {
713 		mthca_err(dev, "Failed to initialize "
714 			  "queue pair table, aborting.\n");
715 		goto err_cq_table_free;
716 	}
717 
718 	err = mthca_init_av_table(dev);
719 	if (err) {
720 		mthca_err(dev, "Failed to initialize "
721 			  "address vector table, aborting.\n");
722 		goto err_qp_table_free;
723 	}
724 
725 	err = mthca_init_mcg_table(dev);
726 	if (err) {
727 		mthca_err(dev, "Failed to initialize "
728 			  "multicast group table, aborting.\n");
729 		goto err_av_table_free;
730 	}
731 
732 	return 0;
733 
734 err_av_table_free:
735 	mthca_cleanup_av_table(dev);
736 
737 err_qp_table_free:
738 	mthca_cleanup_qp_table(dev);
739 
740 err_cq_table_free:
741 	mthca_cleanup_cq_table(dev);
742 
743 err_cmd_poll:
744 	mthca_cmd_use_polling(dev);
745 
746 err_eq_table_free:
747 	mthca_cleanup_eq_table(dev);
748 
749 err_pd_free:
750 	mthca_pd_free(dev, &dev->driver_pd);
751 
752 err_mr_table_free:
753 	mthca_cleanup_mr_table(dev);
754 
755 err_pd_table_free:
756 	mthca_cleanup_pd_table(dev);
757 
758 err_kar_unmap:
759 	iounmap(dev->kar);
760 
761 err_uar_free:
762 	mthca_uar_free(dev, &dev->driver_uar);
763 
764 err_uar_table_free:
765 	mthca_cleanup_uar_table(dev);
766 	return err;
767 }
768 
769 static int __devinit mthca_request_regions(struct pci_dev *pdev,
770 					   int ddr_hidden)
771 {
772 	int err;
773 
774 	/*
775 	 * We can't just use pci_request_regions() because the MSI-X
776 	 * table is right in the middle of the first BAR.  If we did
777 	 * pci_request_region and grab all of the first BAR, then
778 	 * setting up MSI-X would fail, since the PCI core wants to do
779 	 * request_mem_region on the MSI-X vector table.
780 	 *
781 	 * So just request what we need right now, and request any
782 	 * other regions we need when setting up EQs.
783 	 */
784 	if (!request_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
785 				MTHCA_HCR_SIZE, DRV_NAME))
786 		return -EBUSY;
787 
788 	err = pci_request_region(pdev, 2, DRV_NAME);
789 	if (err)
790 		goto err_bar2_failed;
791 
792 	if (!ddr_hidden) {
793 		err = pci_request_region(pdev, 4, DRV_NAME);
794 		if (err)
795 			goto err_bar4_failed;
796 	}
797 
798 	return 0;
799 
800 err_bar4_failed:
801 	pci_release_region(pdev, 2);
802 
803 err_bar2_failed:
804 	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
805 			   MTHCA_HCR_SIZE);
806 
807 	return err;
808 }
809 
810 static void mthca_release_regions(struct pci_dev *pdev,
811 				  int ddr_hidden)
812 {
813 	if (!ddr_hidden)
814 		pci_release_region(pdev, 4);
815 
816 	pci_release_region(pdev, 2);
817 
818 	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
819 			   MTHCA_HCR_SIZE);
820 }
821 
822 static int __devinit mthca_enable_msi_x(struct mthca_dev *mdev)
823 {
824 	struct msix_entry entries[3];
825 	int err;
826 
827 	entries[0].entry = 0;
828 	entries[1].entry = 1;
829 	entries[2].entry = 2;
830 
831 	err = pci_enable_msix(mdev->pdev, entries, ARRAY_SIZE(entries));
832 	if (err) {
833 		if (err > 0)
834 			mthca_info(mdev, "Only %d MSI-X vectors available, "
835 				   "not using MSI-X\n", err);
836 		return err;
837 	}
838 
839 	mdev->eq_table.eq[MTHCA_EQ_COMP ].msi_x_vector = entries[0].vector;
840 	mdev->eq_table.eq[MTHCA_EQ_ASYNC].msi_x_vector = entries[1].vector;
841 	mdev->eq_table.eq[MTHCA_EQ_CMD  ].msi_x_vector = entries[2].vector;
842 
843 	return 0;
844 }
845 
846 static void mthca_close_hca(struct mthca_dev *mdev)
847 {
848 	u8 status;
849 
850 	mthca_CLOSE_HCA(mdev, 0, &status);
851 
852 	if (mthca_is_memfree(mdev)) {
853 		mthca_free_icm_table(mdev, mdev->cq_table.table);
854 		mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
855 		mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
856 		mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
857 		mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
858 		mthca_unmap_eq_icm(mdev);
859 
860 		mthca_UNMAP_ICM_AUX(mdev, &status);
861 		mthca_free_icm(mdev, mdev->fw.arbel.aux_icm);
862 
863 		mthca_UNMAP_FA(mdev, &status);
864 		mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
865 
866 		if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
867 			mthca_DISABLE_LAM(mdev, &status);
868 	} else
869 		mthca_SYS_DIS(mdev, &status);
870 }
871 
872 static int __devinit mthca_init_one(struct pci_dev *pdev,
873 				    const struct pci_device_id *id)
874 {
875 	static int mthca_version_printed = 0;
876 	static int mthca_memfree_warned = 0;
877 	int ddr_hidden = 0;
878 	int err;
879 	struct mthca_dev *mdev;
880 
881 	if (!mthca_version_printed) {
882 		printk(KERN_INFO "%s", mthca_version);
883 		++mthca_version_printed;
884 	}
885 
886 	printk(KERN_INFO PFX "Initializing %s (%s)\n",
887 	       pci_pretty_name(pdev), pci_name(pdev));
888 
889 	err = pci_enable_device(pdev);
890 	if (err) {
891 		dev_err(&pdev->dev, "Cannot enable PCI device, "
892 			"aborting.\n");
893 		return err;
894 	}
895 
896 	/*
897 	 * Check for BARs.  We expect 0: 1MB, 2: 8MB, 4: DDR (may not
898 	 * be present)
899 	 */
900 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
901 	    pci_resource_len(pdev, 0) != 1 << 20) {
902 		dev_err(&pdev->dev, "Missing DCS, aborting.");
903 		err = -ENODEV;
904 		goto err_disable_pdev;
905 	}
906 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM) ||
907 	    pci_resource_len(pdev, 2) != 1 << 23) {
908 		dev_err(&pdev->dev, "Missing UAR, aborting.");
909 		err = -ENODEV;
910 		goto err_disable_pdev;
911 	}
912 	if (!(pci_resource_flags(pdev, 4) & IORESOURCE_MEM))
913 		ddr_hidden = 1;
914 
915 	err = mthca_request_regions(pdev, ddr_hidden);
916 	if (err) {
917 		dev_err(&pdev->dev, "Cannot obtain PCI resources, "
918 			"aborting.\n");
919 		goto err_disable_pdev;
920 	}
921 
922 	pci_set_master(pdev);
923 
924 	err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
925 	if (err) {
926 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
927 		err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
928 		if (err) {
929 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
930 			goto err_free_res;
931 		}
932 	}
933 	err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
934 	if (err) {
935 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
936 			 "consistent PCI DMA mask.\n");
937 		err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
938 		if (err) {
939 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
940 				"aborting.\n");
941 			goto err_free_res;
942 		}
943 	}
944 
945 	mdev = (struct mthca_dev *) ib_alloc_device(sizeof *mdev);
946 	if (!mdev) {
947 		dev_err(&pdev->dev, "Device struct alloc failed, "
948 			"aborting.\n");
949 		err = -ENOMEM;
950 		goto err_free_res;
951 	}
952 
953 	mdev->pdev     = pdev;
954 	mdev->hca_type = id->driver_data;
955 
956 	if (mthca_is_memfree(mdev) && !mthca_memfree_warned++)
957 		mthca_warn(mdev, "Warning: native MT25208 mode support is incomplete.  "
958 			   "Your HCA may not work properly.\n");
959 
960 	if (ddr_hidden)
961 		mdev->mthca_flags |= MTHCA_FLAG_DDR_HIDDEN;
962 
963 	/*
964 	 * Now reset the HCA before we touch the PCI capabilities or
965 	 * attempt a firmware command, since a boot ROM may have left
966 	 * the HCA in an undefined state.
967 	 */
968 	err = mthca_reset(mdev);
969 	if (err) {
970 		mthca_err(mdev, "Failed to reset HCA, aborting.\n");
971 		goto err_free_dev;
972 	}
973 
974 	if (msi_x && !mthca_enable_msi_x(mdev))
975 		mdev->mthca_flags |= MTHCA_FLAG_MSI_X;
976 	if (msi && !(mdev->mthca_flags & MTHCA_FLAG_MSI_X) &&
977 	    !pci_enable_msi(pdev))
978 		mdev->mthca_flags |= MTHCA_FLAG_MSI;
979 
980 	sema_init(&mdev->cmd.hcr_sem, 1);
981 	sema_init(&mdev->cmd.poll_sem, 1);
982 	mdev->cmd.use_events = 0;
983 
984 	mdev->hcr = ioremap(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE, MTHCA_HCR_SIZE);
985 	if (!mdev->hcr) {
986 		mthca_err(mdev, "Couldn't map command register, "
987 			  "aborting.\n");
988 		err = -ENOMEM;
989 		goto err_free_dev;
990 	}
991 
992 	err = mthca_tune_pci(mdev);
993 	if (err)
994 		goto err_iounmap;
995 
996 	err = mthca_init_hca(mdev);
997 	if (err)
998 		goto err_iounmap;
999 
1000 	err = mthca_setup_hca(mdev);
1001 	if (err)
1002 		goto err_close;
1003 
1004 	err = mthca_register_device(mdev);
1005 	if (err)
1006 		goto err_cleanup;
1007 
1008 	err = mthca_create_agents(mdev);
1009 	if (err)
1010 		goto err_unregister;
1011 
1012 	pci_set_drvdata(pdev, mdev);
1013 
1014 	return 0;
1015 
1016 err_unregister:
1017 	mthca_unregister_device(mdev);
1018 
1019 err_cleanup:
1020 	mthca_cleanup_mcg_table(mdev);
1021 	mthca_cleanup_av_table(mdev);
1022 	mthca_cleanup_qp_table(mdev);
1023 	mthca_cleanup_cq_table(mdev);
1024 	mthca_cmd_use_polling(mdev);
1025 	mthca_cleanup_eq_table(mdev);
1026 
1027 	mthca_pd_free(mdev, &mdev->driver_pd);
1028 
1029 	mthca_cleanup_mr_table(mdev);
1030 	mthca_cleanup_pd_table(mdev);
1031 	mthca_cleanup_uar_table(mdev);
1032 
1033 err_close:
1034 	mthca_close_hca(mdev);
1035 
1036 err_iounmap:
1037 	iounmap(mdev->hcr);
1038 
1039 err_free_dev:
1040 	if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1041 		pci_disable_msix(pdev);
1042 	if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1043 		pci_disable_msi(pdev);
1044 
1045 	ib_dealloc_device(&mdev->ib_dev);
1046 
1047 err_free_res:
1048 	mthca_release_regions(pdev, ddr_hidden);
1049 
1050 err_disable_pdev:
1051 	pci_disable_device(pdev);
1052 	pci_set_drvdata(pdev, NULL);
1053 	return err;
1054 }
1055 
1056 static void __devexit mthca_remove_one(struct pci_dev *pdev)
1057 {
1058 	struct mthca_dev *mdev = pci_get_drvdata(pdev);
1059 	u8 status;
1060 	int p;
1061 
1062 	if (mdev) {
1063 		mthca_free_agents(mdev);
1064 		mthca_unregister_device(mdev);
1065 
1066 		for (p = 1; p <= mdev->limits.num_ports; ++p)
1067 			mthca_CLOSE_IB(mdev, p, &status);
1068 
1069 		mthca_cleanup_mcg_table(mdev);
1070 		mthca_cleanup_av_table(mdev);
1071 		mthca_cleanup_qp_table(mdev);
1072 		mthca_cleanup_cq_table(mdev);
1073 		mthca_cmd_use_polling(mdev);
1074 		mthca_cleanup_eq_table(mdev);
1075 
1076 		mthca_pd_free(mdev, &mdev->driver_pd);
1077 
1078 		mthca_cleanup_mr_table(mdev);
1079 		mthca_cleanup_pd_table(mdev);
1080 
1081 		iounmap(mdev->kar);
1082 		mthca_uar_free(mdev, &mdev->driver_uar);
1083 		mthca_cleanup_uar_table(mdev);
1084 
1085 		mthca_close_hca(mdev);
1086 
1087 		iounmap(mdev->hcr);
1088 
1089 		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1090 			pci_disable_msix(pdev);
1091 		if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1092 			pci_disable_msi(pdev);
1093 
1094 		ib_dealloc_device(&mdev->ib_dev);
1095 		mthca_release_regions(pdev, mdev->mthca_flags &
1096 				      MTHCA_FLAG_DDR_HIDDEN);
1097 		pci_disable_device(pdev);
1098 		pci_set_drvdata(pdev, NULL);
1099 	}
1100 }
1101 
1102 static struct pci_device_id mthca_pci_table[] = {
1103 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_TAVOR),
1104 	  .driver_data = TAVOR },
1105 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_TAVOR),
1106 	  .driver_data = TAVOR },
1107 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1108 	  .driver_data = ARBEL_COMPAT },
1109 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1110 	  .driver_data = ARBEL_COMPAT },
1111 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL),
1112 	  .driver_data = ARBEL_NATIVE },
1113 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL),
1114 	  .driver_data = ARBEL_NATIVE },
1115 	{ 0, }
1116 };
1117 
1118 MODULE_DEVICE_TABLE(pci, mthca_pci_table);
1119 
1120 static struct pci_driver mthca_driver = {
1121 	.name		= "ib_mthca",
1122 	.id_table	= mthca_pci_table,
1123 	.probe		= mthca_init_one,
1124 	.remove		= __devexit_p(mthca_remove_one)
1125 };
1126 
1127 static int __init mthca_init(void)
1128 {
1129 	int ret;
1130 
1131 	ret = pci_register_driver(&mthca_driver);
1132 	return ret < 0 ? ret : 0;
1133 }
1134 
1135 static void __exit mthca_cleanup(void)
1136 {
1137 	pci_unregister_driver(&mthca_driver);
1138 }
1139 
1140 module_init(mthca_init);
1141 module_exit(mthca_cleanup);
1142