mthca_main.c revision 68a3c21203b73696769dfdceb68d2d7f5d3c20a0
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
49MODULE_AUTHOR("Roland Dreier");
50MODULE_DESCRIPTION("Mellanox InfiniBand HCA low-level driver");
51MODULE_LICENSE("Dual BSD/GPL");
52MODULE_VERSION(DRV_VERSION);
53
54#ifdef CONFIG_PCI_MSI
55
56static int msi_x = 0;
57module_param(msi_x, int, 0444);
58MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
59
60static int msi = 0;
61module_param(msi, int, 0444);
62MODULE_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
71static const char mthca_version[] __devinitdata =
72	"ib_mthca: Mellanox InfiniBand HCA driver v"
73	DRV_VERSION " (" DRV_RELDATE ")\n";
74
75static 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
87static 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->mthca_flags & MTHCA_FLAG_PCIE))
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->mthca_flags & MTHCA_FLAG_PCIE)
123		mthca_info(mdev, "No PCI Express capability, "
124			   "not setting Max Read Request Size.\n");
125
126	return 0;
127}
128
129static 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
207static 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
289err_close:
290	mthca_CLOSE_HCA(mdev, 0, &status);
291
292err_disable:
293	mthca_SYS_DIS(mdev, &status);
294
295	return err;
296}
297
298static 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
336err_unmap_fa:
337	mthca_UNMAP_FA(mdev, &status);
338
339err_free:
340	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
341	return err;
342}
343
344static 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							 MTHCA_MTT_SEG_SIZE,
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->qp_table.rdb_table = mthca_alloc_icm_table(mdev, init_hca->rdb_base,
433							 MTHCA_RDB_ENTRY_SIZE,
434							 mdev->limits.num_qps <<
435							 mdev->qp_table.rdb_shift,
436							 0, 0);
437	if (!mdev->qp_table.rdb_table) {
438		mthca_err(mdev, "Failed to map RDB context memory, aborting\n");
439		err = -ENOMEM;
440		goto err_unmap_rdb;
441	}
442
443       mdev->cq_table.table = mthca_alloc_icm_table(mdev, init_hca->cqc_base,
444						     dev_lim->cqc_entry_sz,
445						     mdev->limits.num_cqs,
446						     mdev->limits.reserved_cqs, 0);
447	if (!mdev->cq_table.table) {
448		mthca_err(mdev, "Failed to map CQ context memory, aborting.\n");
449		err = -ENOMEM;
450		goto err_unmap_rdb;
451	}
452
453	/*
454	 * It's not strictly required, but for simplicity just map the
455	 * whole multicast group table now.  The table isn't very big
456	 * and it's a lot easier than trying to track ref counts.
457	 */
458	mdev->mcg_table.table = mthca_alloc_icm_table(mdev, init_hca->mc_base,
459						      MTHCA_MGM_ENTRY_SIZE,
460						      mdev->limits.num_mgms +
461						      mdev->limits.num_amgms,
462						      mdev->limits.num_mgms +
463						      mdev->limits.num_amgms,
464						      0);
465	if (!mdev->mcg_table.table) {
466		mthca_err(mdev, "Failed to map MCG context memory, aborting.\n");
467		err = -ENOMEM;
468		goto err_unmap_cq;
469	}
470
471	return 0;
472
473err_unmap_cq:
474	mthca_free_icm_table(mdev, mdev->cq_table.table);
475
476err_unmap_rdb:
477	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
478
479err_unmap_eqp:
480	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
481
482err_unmap_qp:
483	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
484
485err_unmap_mpt:
486	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
487
488err_unmap_mtt:
489	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
490
491err_unmap_eq:
492	mthca_unmap_eq_icm(mdev);
493
494err_unmap_aux:
495	mthca_UNMAP_ICM_AUX(mdev, &status);
496
497err_free_aux:
498	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm);
499
500	return err;
501}
502
503static int __devinit mthca_init_arbel(struct mthca_dev *mdev)
504{
505	struct mthca_dev_lim        dev_lim;
506	struct mthca_profile        profile;
507	struct mthca_init_hca_param init_hca;
508	struct mthca_adapter        adapter;
509	u64 icm_size;
510	u8 status;
511	int err;
512
513	err = mthca_QUERY_FW(mdev, &status);
514	if (err) {
515		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
516		return err;
517	}
518	if (status) {
519		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
520			  "aborting.\n", status);
521		return -EINVAL;
522	}
523
524	err = mthca_ENABLE_LAM(mdev, &status);
525	if (err) {
526		mthca_err(mdev, "ENABLE_LAM command failed, aborting.\n");
527		return err;
528	}
529	if (status == MTHCA_CMD_STAT_LAM_NOT_PRE) {
530		mthca_dbg(mdev, "No HCA-attached memory (running in MemFree mode)\n");
531		mdev->mthca_flags |= MTHCA_FLAG_NO_LAM;
532	} else if (status) {
533		mthca_err(mdev, "ENABLE_LAM returned status 0x%02x, "
534			  "aborting.\n", status);
535		return -EINVAL;
536	}
537
538	err = mthca_load_fw(mdev);
539	if (err) {
540		mthca_err(mdev, "Failed to start FW, aborting.\n");
541		goto err_disable;
542	}
543
544	err = mthca_dev_lim(mdev, &dev_lim);
545	if (err) {
546		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
547		goto err_stop_fw;
548	}
549
550	profile = default_profile;
551	profile.num_uar  = dev_lim.uar_size / PAGE_SIZE;
552	profile.num_udav = 0;
553
554	icm_size = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
555	if ((int) icm_size < 0) {
556		err = icm_size;
557		goto err_stop_fw;
558	}
559
560	err = mthca_init_icm(mdev, &dev_lim, &init_hca, icm_size);
561	if (err)
562		goto err_stop_fw;
563
564	err = mthca_INIT_HCA(mdev, &init_hca, &status);
565	if (err) {
566		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
567		goto err_free_icm;
568	}
569	if (status) {
570		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
571			  "aborting.\n", status);
572		err = -EINVAL;
573		goto err_free_icm;
574	}
575
576	err = mthca_QUERY_ADAPTER(mdev, &adapter, &status);
577	if (err) {
578		mthca_err(mdev, "QUERY_ADAPTER command failed, aborting.\n");
579		goto err_free_icm;
580	}
581	if (status) {
582		mthca_err(mdev, "QUERY_ADAPTER returned status 0x%02x, "
583			  "aborting.\n", status);
584		err = -EINVAL;
585		goto err_free_icm;
586	}
587
588	mdev->eq_table.inta_pin = adapter.inta_pin;
589	mdev->rev_id            = adapter.revision_id;
590
591	return 0;
592
593err_free_icm:
594	mthca_free_icm_table(mdev, mdev->cq_table.table);
595	mthca_free_icm_table(mdev, mdev->qp_table.rdb_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
605err_stop_fw:
606	mthca_UNMAP_FA(mdev, &status);
607	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
608
609err_disable:
610	if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
611		mthca_DISABLE_LAM(mdev, &status);
612
613	return err;
614}
615
616static 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
624static 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
734err_av_table_free:
735	mthca_cleanup_av_table(dev);
736
737err_qp_table_free:
738	mthca_cleanup_qp_table(dev);
739
740err_cq_table_free:
741	mthca_cleanup_cq_table(dev);
742
743err_cmd_poll:
744	mthca_cmd_use_polling(dev);
745
746err_eq_table_free:
747	mthca_cleanup_eq_table(dev);
748
749err_pd_free:
750	mthca_pd_free(dev, &dev->driver_pd);
751
752err_mr_table_free:
753	mthca_cleanup_mr_table(dev);
754
755err_pd_table_free:
756	mthca_cleanup_pd_table(dev);
757
758err_kar_unmap:
759	iounmap(dev->kar);
760
761err_uar_free:
762	mthca_uar_free(dev, &dev->driver_uar);
763
764err_uar_table_free:
765	mthca_cleanup_uar_table(dev);
766	return err;
767}
768
769static 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
800err_bar4_failed:
801	pci_release_region(pdev, 2);
802
803err_bar2_failed:
804	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
805			   MTHCA_HCR_SIZE);
806
807	return err;
808}
809
810static 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
822static 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
846static 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.rdb_table);
855		mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
856		mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
857		mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
858		mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
859		mthca_unmap_eq_icm(mdev);
860
861		mthca_UNMAP_ICM_AUX(mdev, &status);
862		mthca_free_icm(mdev, mdev->fw.arbel.aux_icm);
863
864		mthca_UNMAP_FA(mdev, &status);
865		mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
866
867		if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
868			mthca_DISABLE_LAM(mdev, &status);
869	} else
870		mthca_SYS_DIS(mdev, &status);
871}
872
873/* Types of supported HCA */
874enum {
875	TAVOR,			/* MT23108                        */
876	ARBEL_COMPAT,		/* MT25208 in Tavor compat mode   */
877	ARBEL_NATIVE,		/* MT25208 with extended features */
878	SINAI			/* MT25204 */
879};
880
881#define MTHCA_FW_VER(major, minor, subminor) \
882	(((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
883
884static struct {
885	u64 latest_fw;
886	int is_memfree;
887	int is_pcie;
888} mthca_hca_table[] = {
889	[TAVOR]        = { .latest_fw = MTHCA_FW_VER(3, 3, 2), .is_memfree = 0, .is_pcie = 0 },
890	[ARBEL_COMPAT] = { .latest_fw = MTHCA_FW_VER(4, 6, 2), .is_memfree = 0, .is_pcie = 1 },
891	[ARBEL_NATIVE] = { .latest_fw = MTHCA_FW_VER(5, 0, 1), .is_memfree = 1, .is_pcie = 1 },
892	[SINAI]        = { .latest_fw = MTHCA_FW_VER(1, 0, 1), .is_memfree = 1, .is_pcie = 1 }
893};
894
895static int __devinit mthca_init_one(struct pci_dev *pdev,
896				    const struct pci_device_id *id)
897{
898	static int mthca_version_printed = 0;
899	int ddr_hidden = 0;
900	int err;
901	struct mthca_dev *mdev;
902
903	if (!mthca_version_printed) {
904		printk(KERN_INFO "%s", mthca_version);
905		++mthca_version_printed;
906	}
907
908	printk(KERN_INFO PFX "Initializing %s (%s)\n",
909	       pci_pretty_name(pdev), pci_name(pdev));
910
911	if (id->driver_data >= ARRAY_SIZE(mthca_hca_table)) {
912		printk(KERN_ERR PFX "%s (%s) has invalid driver data %lx\n",
913		       pci_pretty_name(pdev), pci_name(pdev), id->driver_data);
914		return -ENODEV;
915	}
916
917	err = pci_enable_device(pdev);
918	if (err) {
919		dev_err(&pdev->dev, "Cannot enable PCI device, "
920			"aborting.\n");
921		return err;
922	}
923
924	/*
925	 * Check for BARs.  We expect 0: 1MB, 2: 8MB, 4: DDR (may not
926	 * be present)
927	 */
928	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
929	    pci_resource_len(pdev, 0) != 1 << 20) {
930		dev_err(&pdev->dev, "Missing DCS, aborting.");
931		err = -ENODEV;
932		goto err_disable_pdev;
933	}
934	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM) ||
935	    pci_resource_len(pdev, 2) != 1 << 23) {
936		dev_err(&pdev->dev, "Missing UAR, aborting.");
937		err = -ENODEV;
938		goto err_disable_pdev;
939	}
940	if (!(pci_resource_flags(pdev, 4) & IORESOURCE_MEM))
941		ddr_hidden = 1;
942
943	err = mthca_request_regions(pdev, ddr_hidden);
944	if (err) {
945		dev_err(&pdev->dev, "Cannot obtain PCI resources, "
946			"aborting.\n");
947		goto err_disable_pdev;
948	}
949
950	pci_set_master(pdev);
951
952	err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
953	if (err) {
954		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
955		err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
956		if (err) {
957			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
958			goto err_free_res;
959		}
960	}
961	err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
962	if (err) {
963		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
964			 "consistent PCI DMA mask.\n");
965		err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
966		if (err) {
967			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
968				"aborting.\n");
969			goto err_free_res;
970		}
971	}
972
973	mdev = (struct mthca_dev *) ib_alloc_device(sizeof *mdev);
974	if (!mdev) {
975		dev_err(&pdev->dev, "Device struct alloc failed, "
976			"aborting.\n");
977		err = -ENOMEM;
978		goto err_free_res;
979	}
980
981	mdev->pdev = pdev;
982
983	if (ddr_hidden)
984		mdev->mthca_flags |= MTHCA_FLAG_DDR_HIDDEN;
985	if (mthca_hca_table[id->driver_data].is_memfree)
986		mdev->mthca_flags |= MTHCA_FLAG_MEMFREE;
987	if (mthca_hca_table[id->driver_data].is_pcie)
988		mdev->mthca_flags |= MTHCA_FLAG_PCIE;
989
990	/*
991	 * Now reset the HCA before we touch the PCI capabilities or
992	 * attempt a firmware command, since a boot ROM may have left
993	 * the HCA in an undefined state.
994	 */
995	err = mthca_reset(mdev);
996	if (err) {
997		mthca_err(mdev, "Failed to reset HCA, aborting.\n");
998		goto err_free_dev;
999	}
1000
1001	if (msi_x && !mthca_enable_msi_x(mdev))
1002		mdev->mthca_flags |= MTHCA_FLAG_MSI_X;
1003	if (msi && !(mdev->mthca_flags & MTHCA_FLAG_MSI_X) &&
1004	    !pci_enable_msi(pdev))
1005		mdev->mthca_flags |= MTHCA_FLAG_MSI;
1006
1007	sema_init(&mdev->cmd.hcr_sem, 1);
1008	sema_init(&mdev->cmd.poll_sem, 1);
1009	mdev->cmd.use_events = 0;
1010
1011	mdev->hcr = ioremap(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE, MTHCA_HCR_SIZE);
1012	if (!mdev->hcr) {
1013		mthca_err(mdev, "Couldn't map command register, "
1014			  "aborting.\n");
1015		err = -ENOMEM;
1016		goto err_free_dev;
1017	}
1018
1019	err = mthca_tune_pci(mdev);
1020	if (err)
1021		goto err_iounmap;
1022
1023	err = mthca_init_hca(mdev);
1024	if (err)
1025		goto err_iounmap;
1026
1027	if (mdev->fw_ver < mthca_hca_table[id->driver_data].latest_fw) {
1028		mthca_warn(mdev, "HCA FW version %x.%x.%x is old (%x.%x.%x is current).\n",
1029			   (int) (mdev->fw_ver >> 32), (int) (mdev->fw_ver >> 16) & 0xffff,
1030			   (int) (mdev->fw_ver & 0xffff),
1031			   (int) (mthca_hca_table[id->driver_data].latest_fw >> 32),
1032			   (int) (mthca_hca_table[id->driver_data].latest_fw >> 16) & 0xffff,
1033			   (int) (mthca_hca_table[id->driver_data].latest_fw & 0xffff));
1034		mthca_warn(mdev, "If you have problems, try updating your HCA FW.\n");
1035	}
1036
1037	err = mthca_setup_hca(mdev);
1038	if (err)
1039		goto err_close;
1040
1041	err = mthca_register_device(mdev);
1042	if (err)
1043		goto err_cleanup;
1044
1045	err = mthca_create_agents(mdev);
1046	if (err)
1047		goto err_unregister;
1048
1049	pci_set_drvdata(pdev, mdev);
1050
1051	return 0;
1052
1053err_unregister:
1054	mthca_unregister_device(mdev);
1055
1056err_cleanup:
1057	mthca_cleanup_mcg_table(mdev);
1058	mthca_cleanup_av_table(mdev);
1059	mthca_cleanup_qp_table(mdev);
1060	mthca_cleanup_cq_table(mdev);
1061	mthca_cmd_use_polling(mdev);
1062	mthca_cleanup_eq_table(mdev);
1063
1064	mthca_pd_free(mdev, &mdev->driver_pd);
1065
1066	mthca_cleanup_mr_table(mdev);
1067	mthca_cleanup_pd_table(mdev);
1068	mthca_cleanup_uar_table(mdev);
1069
1070err_close:
1071	mthca_close_hca(mdev);
1072
1073err_iounmap:
1074	iounmap(mdev->hcr);
1075
1076err_free_dev:
1077	if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1078		pci_disable_msix(pdev);
1079	if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1080		pci_disable_msi(pdev);
1081
1082	ib_dealloc_device(&mdev->ib_dev);
1083
1084err_free_res:
1085	mthca_release_regions(pdev, ddr_hidden);
1086
1087err_disable_pdev:
1088	pci_disable_device(pdev);
1089	pci_set_drvdata(pdev, NULL);
1090	return err;
1091}
1092
1093static void __devexit mthca_remove_one(struct pci_dev *pdev)
1094{
1095	struct mthca_dev *mdev = pci_get_drvdata(pdev);
1096	u8 status;
1097	int p;
1098
1099	if (mdev) {
1100		mthca_free_agents(mdev);
1101		mthca_unregister_device(mdev);
1102
1103		for (p = 1; p <= mdev->limits.num_ports; ++p)
1104			mthca_CLOSE_IB(mdev, p, &status);
1105
1106		mthca_cleanup_mcg_table(mdev);
1107		mthca_cleanup_av_table(mdev);
1108		mthca_cleanup_qp_table(mdev);
1109		mthca_cleanup_cq_table(mdev);
1110		mthca_cmd_use_polling(mdev);
1111		mthca_cleanup_eq_table(mdev);
1112
1113		mthca_pd_free(mdev, &mdev->driver_pd);
1114
1115		mthca_cleanup_mr_table(mdev);
1116		mthca_cleanup_pd_table(mdev);
1117
1118		iounmap(mdev->kar);
1119		mthca_uar_free(mdev, &mdev->driver_uar);
1120		mthca_cleanup_uar_table(mdev);
1121
1122		mthca_close_hca(mdev);
1123
1124		iounmap(mdev->hcr);
1125
1126		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1127			pci_disable_msix(pdev);
1128		if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1129			pci_disable_msi(pdev);
1130
1131		ib_dealloc_device(&mdev->ib_dev);
1132		mthca_release_regions(pdev, mdev->mthca_flags &
1133				      MTHCA_FLAG_DDR_HIDDEN);
1134		pci_disable_device(pdev);
1135		pci_set_drvdata(pdev, NULL);
1136	}
1137}
1138
1139static struct pci_device_id mthca_pci_table[] = {
1140	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_TAVOR),
1141	  .driver_data = TAVOR },
1142	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_TAVOR),
1143	  .driver_data = TAVOR },
1144	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1145	  .driver_data = ARBEL_COMPAT },
1146	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1147	  .driver_data = ARBEL_COMPAT },
1148	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL),
1149	  .driver_data = ARBEL_NATIVE },
1150	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL),
1151	  .driver_data = ARBEL_NATIVE },
1152	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI),
1153	  .driver_data = SINAI },
1154	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI),
1155	  .driver_data = SINAI },
1156	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1157	  .driver_data = SINAI },
1158	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1159	  .driver_data = SINAI },
1160	{ 0, }
1161};
1162
1163MODULE_DEVICE_TABLE(pci, mthca_pci_table);
1164
1165static struct pci_driver mthca_driver = {
1166	.name		= "ib_mthca",
1167	.id_table	= mthca_pci_table,
1168	.probe		= mthca_init_one,
1169	.remove		= __devexit_p(mthca_remove_one)
1170};
1171
1172static int __init mthca_init(void)
1173{
1174	int ret;
1175
1176	ret = pci_register_driver(&mthca_driver);
1177	return ret < 0 ? ret : 0;
1178}
1179
1180static void __exit mthca_cleanup(void)
1181{
1182	pci_unregister_driver(&mthca_driver);
1183}
1184
1185module_init(mthca_init);
1186module_exit(mthca_cleanup);
1187