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