1/*
2 * Copyright (c) 2013-2015, Mellanox Technologies. 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
33#include <asm-generic/kmap_types.h>
34#include <linux/module.h>
35#include <linux/init.h>
36#include <linux/errno.h>
37#include <linux/pci.h>
38#include <linux/dma-mapping.h>
39#include <linux/slab.h>
40#include <linux/io-mapping.h>
41#include <linux/mlx5/driver.h>
42#include <linux/mlx5/cq.h>
43#include <linux/mlx5/qp.h>
44#include <linux/mlx5/srq.h>
45#include <linux/debugfs.h>
46#include <linux/kmod.h>
47#include <linux/mlx5/mlx5_ifc.h>
48#include "mlx5_core.h"
49
50#define DRIVER_NAME "mlx5_core"
51#define DRIVER_VERSION "3.0"
52#define DRIVER_RELDATE  "January 2015"
53
54MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
55MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
56MODULE_LICENSE("Dual BSD/GPL");
57MODULE_VERSION(DRIVER_VERSION);
58
59int mlx5_core_debug_mask;
60module_param_named(debug_mask, mlx5_core_debug_mask, int, 0644);
61MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
62
63#define MLX5_DEFAULT_PROF	2
64static int prof_sel = MLX5_DEFAULT_PROF;
65module_param_named(prof_sel, prof_sel, int, 0444);
66MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
67
68struct workqueue_struct *mlx5_core_wq;
69static LIST_HEAD(intf_list);
70static LIST_HEAD(dev_list);
71static DEFINE_MUTEX(intf_mutex);
72
73struct mlx5_device_context {
74	struct list_head	list;
75	struct mlx5_interface  *intf;
76	void		       *context;
77};
78
79static struct mlx5_profile profile[] = {
80	[0] = {
81		.mask           = 0,
82	},
83	[1] = {
84		.mask		= MLX5_PROF_MASK_QP_SIZE,
85		.log_max_qp	= 12,
86	},
87	[2] = {
88		.mask		= MLX5_PROF_MASK_QP_SIZE |
89				  MLX5_PROF_MASK_MR_CACHE,
90		.log_max_qp	= 17,
91		.mr_cache[0]	= {
92			.size	= 500,
93			.limit	= 250
94		},
95		.mr_cache[1]	= {
96			.size	= 500,
97			.limit	= 250
98		},
99		.mr_cache[2]	= {
100			.size	= 500,
101			.limit	= 250
102		},
103		.mr_cache[3]	= {
104			.size	= 500,
105			.limit	= 250
106		},
107		.mr_cache[4]	= {
108			.size	= 500,
109			.limit	= 250
110		},
111		.mr_cache[5]	= {
112			.size	= 500,
113			.limit	= 250
114		},
115		.mr_cache[6]	= {
116			.size	= 500,
117			.limit	= 250
118		},
119		.mr_cache[7]	= {
120			.size	= 500,
121			.limit	= 250
122		},
123		.mr_cache[8]	= {
124			.size	= 500,
125			.limit	= 250
126		},
127		.mr_cache[9]	= {
128			.size	= 500,
129			.limit	= 250
130		},
131		.mr_cache[10]	= {
132			.size	= 500,
133			.limit	= 250
134		},
135		.mr_cache[11]	= {
136			.size	= 500,
137			.limit	= 250
138		},
139		.mr_cache[12]	= {
140			.size	= 64,
141			.limit	= 32
142		},
143		.mr_cache[13]	= {
144			.size	= 32,
145			.limit	= 16
146		},
147		.mr_cache[14]	= {
148			.size	= 16,
149			.limit	= 8
150		},
151		.mr_cache[15]	= {
152			.size	= 8,
153			.limit	= 4
154		},
155	},
156};
157
158static int set_dma_caps(struct pci_dev *pdev)
159{
160	int err;
161
162	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
163	if (err) {
164		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
165		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
166		if (err) {
167			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
168			return err;
169		}
170	}
171
172	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
173	if (err) {
174		dev_warn(&pdev->dev,
175			 "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
176		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
177		if (err) {
178			dev_err(&pdev->dev,
179				"Can't set consistent PCI DMA mask, aborting\n");
180			return err;
181		}
182	}
183
184	dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
185	return err;
186}
187
188static int request_bar(struct pci_dev *pdev)
189{
190	int err = 0;
191
192	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
193		dev_err(&pdev->dev, "Missing registers BAR, aborting\n");
194		return -ENODEV;
195	}
196
197	err = pci_request_regions(pdev, DRIVER_NAME);
198	if (err)
199		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
200
201	return err;
202}
203
204static void release_bar(struct pci_dev *pdev)
205{
206	pci_release_regions(pdev);
207}
208
209static int mlx5_enable_msix(struct mlx5_core_dev *dev)
210{
211	struct mlx5_eq_table *table = &dev->priv.eq_table;
212	int num_eqs = 1 << dev->caps.gen.log_max_eq;
213	int nvec;
214	int i;
215
216	nvec = dev->caps.gen.num_ports * num_online_cpus() + MLX5_EQ_VEC_COMP_BASE;
217	nvec = min_t(int, nvec, num_eqs);
218	if (nvec <= MLX5_EQ_VEC_COMP_BASE)
219		return -ENOMEM;
220
221	table->msix_arr = kzalloc(nvec * sizeof(*table->msix_arr), GFP_KERNEL);
222	if (!table->msix_arr)
223		return -ENOMEM;
224
225	for (i = 0; i < nvec; i++)
226		table->msix_arr[i].entry = i;
227
228	nvec = pci_enable_msix_range(dev->pdev, table->msix_arr,
229				     MLX5_EQ_VEC_COMP_BASE + 1, nvec);
230	if (nvec < 0)
231		return nvec;
232
233	table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
234
235	return 0;
236}
237
238static void mlx5_disable_msix(struct mlx5_core_dev *dev)
239{
240	struct mlx5_eq_table *table = &dev->priv.eq_table;
241
242	pci_disable_msix(dev->pdev);
243	kfree(table->msix_arr);
244}
245
246struct mlx5_reg_host_endianess {
247	u8	he;
248	u8      rsvd[15];
249};
250
251
252#define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos))
253
254enum {
255	MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) |
256				MLX5_DEV_CAP_FLAG_DCT,
257};
258
259static u16 to_fw_pkey_sz(u32 size)
260{
261	switch (size) {
262	case 128:
263		return 0;
264	case 256:
265		return 1;
266	case 512:
267		return 2;
268	case 1024:
269		return 3;
270	case 2048:
271		return 4;
272	case 4096:
273		return 5;
274	default:
275		pr_warn("invalid pkey table size %d\n", size);
276		return 0;
277	}
278}
279
280/* selectively copy writable fields clearing any reserved area
281 */
282static void copy_rw_fields(void *to, struct mlx5_caps *from)
283{
284	__be64 *flags_off = (__be64 *)MLX5_ADDR_OF(cmd_hca_cap, to, reserved_22);
285	u64 v64;
286
287	MLX5_SET(cmd_hca_cap, to, log_max_qp, from->gen.log_max_qp);
288	MLX5_SET(cmd_hca_cap, to, log_max_ra_req_qp, from->gen.log_max_ra_req_qp);
289	MLX5_SET(cmd_hca_cap, to, log_max_ra_res_qp, from->gen.log_max_ra_res_qp);
290	MLX5_SET(cmd_hca_cap, to, pkey_table_size, from->gen.pkey_table_size);
291	MLX5_SET(cmd_hca_cap, to, pkey_table_size, to_fw_pkey_sz(from->gen.pkey_table_size));
292	MLX5_SET(cmd_hca_cap, to, log_uar_page_sz, PAGE_SHIFT - 12);
293	v64 = from->gen.flags & MLX5_CAP_BITS_RW_MASK;
294	*flags_off = cpu_to_be64(v64);
295}
296
297static u16 get_pkey_table_size(int pkey)
298{
299	if (pkey > MLX5_MAX_LOG_PKEY_TABLE)
300		return 0;
301
302	return MLX5_MIN_PKEY_TABLE_SIZE << pkey;
303}
304
305static void fw2drv_caps(struct mlx5_caps *caps, void *out)
306{
307	struct mlx5_general_caps *gen = &caps->gen;
308
309	gen->max_srq_wqes = 1 << MLX5_GET_PR(cmd_hca_cap, out, log_max_srq_sz);
310	gen->max_wqes = 1 << MLX5_GET_PR(cmd_hca_cap, out, log_max_qp_sz);
311	gen->log_max_qp = MLX5_GET_PR(cmd_hca_cap, out, log_max_qp);
312	gen->log_max_strq = MLX5_GET_PR(cmd_hca_cap, out, log_max_strq_sz);
313	gen->log_max_srq = MLX5_GET_PR(cmd_hca_cap, out, log_max_srqs);
314	gen->max_cqes = 1 << MLX5_GET_PR(cmd_hca_cap, out, log_max_cq_sz);
315	gen->log_max_cq = MLX5_GET_PR(cmd_hca_cap, out, log_max_cq);
316	gen->max_eqes = 1 << MLX5_GET_PR(cmd_hca_cap, out, log_max_eq_sz);
317	gen->log_max_mkey = MLX5_GET_PR(cmd_hca_cap, out, log_max_mkey);
318	gen->log_max_eq = MLX5_GET_PR(cmd_hca_cap, out, log_max_eq);
319	gen->max_indirection = MLX5_GET_PR(cmd_hca_cap, out, max_indirection);
320	gen->log_max_mrw_sz = MLX5_GET_PR(cmd_hca_cap, out, log_max_mrw_sz);
321	gen->log_max_bsf_list_size = MLX5_GET_PR(cmd_hca_cap, out, log_max_bsf_list_size);
322	gen->log_max_klm_list_size = MLX5_GET_PR(cmd_hca_cap, out, log_max_klm_list_size);
323	gen->log_max_ra_req_dc = MLX5_GET_PR(cmd_hca_cap, out, log_max_ra_req_dc);
324	gen->log_max_ra_res_dc = MLX5_GET_PR(cmd_hca_cap, out, log_max_ra_res_dc);
325	gen->log_max_ra_req_qp = MLX5_GET_PR(cmd_hca_cap, out, log_max_ra_req_qp);
326	gen->log_max_ra_res_qp = MLX5_GET_PR(cmd_hca_cap, out, log_max_ra_res_qp);
327	gen->max_qp_counters = MLX5_GET_PR(cmd_hca_cap, out, max_qp_cnt);
328	gen->pkey_table_size = get_pkey_table_size(MLX5_GET_PR(cmd_hca_cap, out, pkey_table_size));
329	gen->local_ca_ack_delay = MLX5_GET_PR(cmd_hca_cap, out, local_ca_ack_delay);
330	gen->num_ports = MLX5_GET_PR(cmd_hca_cap, out, num_ports);
331	gen->log_max_msg = MLX5_GET_PR(cmd_hca_cap, out, log_max_msg);
332	gen->stat_rate_support = MLX5_GET_PR(cmd_hca_cap, out, stat_rate_support);
333	gen->flags = be64_to_cpu(*(__be64 *)MLX5_ADDR_OF(cmd_hca_cap, out, reserved_22));
334	pr_debug("flags = 0x%llx\n", gen->flags);
335	gen->uar_sz = MLX5_GET_PR(cmd_hca_cap, out, uar_sz);
336	gen->min_log_pg_sz = MLX5_GET_PR(cmd_hca_cap, out, log_pg_sz);
337	gen->bf_reg_size = MLX5_GET_PR(cmd_hca_cap, out, bf);
338	gen->bf_reg_size = 1 << MLX5_GET_PR(cmd_hca_cap, out, log_bf_reg_size);
339	gen->max_sq_desc_sz = MLX5_GET_PR(cmd_hca_cap, out, max_wqe_sz_sq);
340	gen->max_rq_desc_sz = MLX5_GET_PR(cmd_hca_cap, out, max_wqe_sz_rq);
341	gen->max_dc_sq_desc_sz = MLX5_GET_PR(cmd_hca_cap, out, max_wqe_sz_sq_dc);
342	gen->max_qp_mcg = MLX5_GET_PR(cmd_hca_cap, out, max_qp_mcg);
343	gen->log_max_pd = MLX5_GET_PR(cmd_hca_cap, out, log_max_pd);
344	gen->log_max_xrcd = MLX5_GET_PR(cmd_hca_cap, out, log_max_xrcd);
345	gen->log_uar_page_sz = MLX5_GET_PR(cmd_hca_cap, out, log_uar_page_sz);
346}
347
348static const char *caps_opmod_str(u16 opmod)
349{
350	switch (opmod) {
351	case HCA_CAP_OPMOD_GET_MAX:
352		return "GET_MAX";
353	case HCA_CAP_OPMOD_GET_CUR:
354		return "GET_CUR";
355	default:
356		return "Invalid";
357	}
358}
359
360int mlx5_core_get_caps(struct mlx5_core_dev *dev, struct mlx5_caps *caps,
361		       u16 opmod)
362{
363	u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
364	int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
365	void *out;
366	int err;
367
368	memset(in, 0, sizeof(in));
369	out = kzalloc(out_sz, GFP_KERNEL);
370	if (!out)
371		return -ENOMEM;
372	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
373	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
374	err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz);
375	if (err)
376		goto query_ex;
377
378	err = mlx5_cmd_status_to_err_v2(out);
379	if (err) {
380		mlx5_core_warn(dev, "query max hca cap failed, %d\n", err);
381		goto query_ex;
382	}
383	mlx5_core_dbg(dev, "%s\n", caps_opmod_str(opmod));
384	fw2drv_caps(caps, MLX5_ADDR_OF(query_hca_cap_out, out, capability_struct));
385
386query_ex:
387	kfree(out);
388	return err;
389}
390
391static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz)
392{
393	u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)];
394	int err;
395
396	memset(out, 0, sizeof(out));
397
398	MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
399	err = mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out));
400	if (err)
401		return err;
402
403	err = mlx5_cmd_status_to_err_v2(out);
404
405	return err;
406}
407
408static int handle_hca_cap(struct mlx5_core_dev *dev)
409{
410	void *set_ctx = NULL;
411	struct mlx5_profile *prof = dev->profile;
412	struct mlx5_caps *cur_caps = NULL;
413	struct mlx5_caps *max_caps = NULL;
414	int err = -ENOMEM;
415	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
416
417	set_ctx = kzalloc(set_sz, GFP_KERNEL);
418	if (!set_ctx)
419		goto query_ex;
420
421	max_caps = kzalloc(sizeof(*max_caps), GFP_KERNEL);
422	if (!max_caps)
423		goto query_ex;
424
425	cur_caps = kzalloc(sizeof(*cur_caps), GFP_KERNEL);
426	if (!cur_caps)
427		goto query_ex;
428
429	err = mlx5_core_get_caps(dev, max_caps, HCA_CAP_OPMOD_GET_MAX);
430	if (err)
431		goto query_ex;
432
433	err = mlx5_core_get_caps(dev, cur_caps, HCA_CAP_OPMOD_GET_CUR);
434	if (err)
435		goto query_ex;
436
437	/* we limit the size of the pkey table to 128 entries for now */
438	cur_caps->gen.pkey_table_size = 128;
439
440	if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
441		cur_caps->gen.log_max_qp = prof->log_max_qp;
442
443	/* disable checksum */
444	cur_caps->gen.flags &= ~MLX5_DEV_CAP_FLAG_CMDIF_CSUM;
445
446	copy_rw_fields(MLX5_ADDR_OF(set_hca_cap_in, set_ctx, hca_capability_struct),
447		       cur_caps);
448	err = set_caps(dev, set_ctx, set_sz);
449
450query_ex:
451	kfree(cur_caps);
452	kfree(max_caps);
453	kfree(set_ctx);
454
455	return err;
456}
457
458static int set_hca_ctrl(struct mlx5_core_dev *dev)
459{
460	struct mlx5_reg_host_endianess he_in;
461	struct mlx5_reg_host_endianess he_out;
462	int err;
463
464	memset(&he_in, 0, sizeof(he_in));
465	he_in.he = MLX5_SET_HOST_ENDIANNESS;
466	err = mlx5_core_access_reg(dev, &he_in,  sizeof(he_in),
467					&he_out, sizeof(he_out),
468					MLX5_REG_HOST_ENDIANNESS, 0, 1);
469	return err;
470}
471
472static int mlx5_core_enable_hca(struct mlx5_core_dev *dev)
473{
474	int err;
475	struct mlx5_enable_hca_mbox_in in;
476	struct mlx5_enable_hca_mbox_out out;
477
478	memset(&in, 0, sizeof(in));
479	memset(&out, 0, sizeof(out));
480	in.hdr.opcode = cpu_to_be16(MLX5_CMD_OP_ENABLE_HCA);
481	err = mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
482	if (err)
483		return err;
484
485	if (out.hdr.status)
486		return mlx5_cmd_status_to_err(&out.hdr);
487
488	return 0;
489}
490
491static int mlx5_core_disable_hca(struct mlx5_core_dev *dev)
492{
493	int err;
494	struct mlx5_disable_hca_mbox_in in;
495	struct mlx5_disable_hca_mbox_out out;
496
497	memset(&in, 0, sizeof(in));
498	memset(&out, 0, sizeof(out));
499	in.hdr.opcode = cpu_to_be16(MLX5_CMD_OP_DISABLE_HCA);
500	err = mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
501	if (err)
502		return err;
503
504	if (out.hdr.status)
505		return mlx5_cmd_status_to_err(&out.hdr);
506
507	return 0;
508}
509
510int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn, int *irqn)
511{
512	struct mlx5_eq_table *table = &dev->priv.eq_table;
513	struct mlx5_eq *eq, *n;
514	int err = -ENOENT;
515
516	spin_lock(&table->lock);
517	list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
518		if (eq->index == vector) {
519			*eqn = eq->eqn;
520			*irqn = eq->irqn;
521			err = 0;
522			break;
523		}
524	}
525	spin_unlock(&table->lock);
526
527	return err;
528}
529EXPORT_SYMBOL(mlx5_vector2eqn);
530
531static void free_comp_eqs(struct mlx5_core_dev *dev)
532{
533	struct mlx5_eq_table *table = &dev->priv.eq_table;
534	struct mlx5_eq *eq, *n;
535
536	spin_lock(&table->lock);
537	list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
538		list_del(&eq->list);
539		spin_unlock(&table->lock);
540		if (mlx5_destroy_unmap_eq(dev, eq))
541			mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n",
542				       eq->eqn);
543		kfree(eq);
544		spin_lock(&table->lock);
545	}
546	spin_unlock(&table->lock);
547}
548
549static int alloc_comp_eqs(struct mlx5_core_dev *dev)
550{
551	struct mlx5_eq_table *table = &dev->priv.eq_table;
552	char name[MLX5_MAX_EQ_NAME];
553	struct mlx5_eq *eq;
554	int ncomp_vec;
555	int nent;
556	int err;
557	int i;
558
559	INIT_LIST_HEAD(&table->comp_eqs_list);
560	ncomp_vec = table->num_comp_vectors;
561	nent = MLX5_COMP_EQ_SIZE;
562	for (i = 0; i < ncomp_vec; i++) {
563		eq = kzalloc(sizeof(*eq), GFP_KERNEL);
564		if (!eq) {
565			err = -ENOMEM;
566			goto clean;
567		}
568
569		snprintf(name, MLX5_MAX_EQ_NAME, "mlx5_comp%d", i);
570		err = mlx5_create_map_eq(dev, eq,
571					 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
572					 name, &dev->priv.uuari.uars[0]);
573		if (err) {
574			kfree(eq);
575			goto clean;
576		}
577		mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
578		eq->index = i;
579		spin_lock(&table->lock);
580		list_add_tail(&eq->list, &table->comp_eqs_list);
581		spin_unlock(&table->lock);
582	}
583
584	return 0;
585
586clean:
587	free_comp_eqs(dev);
588	return err;
589}
590
591static int mlx5_dev_init(struct mlx5_core_dev *dev, struct pci_dev *pdev)
592{
593	struct mlx5_priv *priv = &dev->priv;
594	int err;
595
596	dev->pdev = pdev;
597	pci_set_drvdata(dev->pdev, dev);
598	strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
599	priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
600
601	mutex_init(&priv->pgdir_mutex);
602	INIT_LIST_HEAD(&priv->pgdir_list);
603	spin_lock_init(&priv->mkey_lock);
604
605	priv->dbg_root = debugfs_create_dir(dev_name(&pdev->dev), mlx5_debugfs_root);
606	if (!priv->dbg_root)
607		return -ENOMEM;
608
609	err = pci_enable_device(pdev);
610	if (err) {
611		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
612		goto err_dbg;
613	}
614
615	err = request_bar(pdev);
616	if (err) {
617		dev_err(&pdev->dev, "error requesting BARs, aborting\n");
618		goto err_disable;
619	}
620
621	pci_set_master(pdev);
622
623	err = set_dma_caps(pdev);
624	if (err) {
625		dev_err(&pdev->dev, "Failed setting DMA capabilities mask, aborting\n");
626		goto err_clr_master;
627	}
628
629	dev->iseg_base = pci_resource_start(dev->pdev, 0);
630	dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
631	if (!dev->iseg) {
632		err = -ENOMEM;
633		dev_err(&pdev->dev, "Failed mapping initialization segment, aborting\n");
634		goto err_clr_master;
635	}
636	dev_info(&pdev->dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
637		 fw_rev_min(dev), fw_rev_sub(dev));
638
639	err = mlx5_cmd_init(dev);
640	if (err) {
641		dev_err(&pdev->dev, "Failed initializing command interface, aborting\n");
642		goto err_unmap;
643	}
644
645	mlx5_pagealloc_init(dev);
646
647	err = mlx5_core_enable_hca(dev);
648	if (err) {
649		dev_err(&pdev->dev, "enable hca failed\n");
650		goto err_pagealloc_cleanup;
651	}
652
653	err = mlx5_satisfy_startup_pages(dev, 1);
654	if (err) {
655		dev_err(&pdev->dev, "failed to allocate boot pages\n");
656		goto err_disable_hca;
657	}
658
659	err = set_hca_ctrl(dev);
660	if (err) {
661		dev_err(&pdev->dev, "set_hca_ctrl failed\n");
662		goto reclaim_boot_pages;
663	}
664
665	err = handle_hca_cap(dev);
666	if (err) {
667		dev_err(&pdev->dev, "handle_hca_cap failed\n");
668		goto reclaim_boot_pages;
669	}
670
671	err = mlx5_satisfy_startup_pages(dev, 0);
672	if (err) {
673		dev_err(&pdev->dev, "failed to allocate init pages\n");
674		goto reclaim_boot_pages;
675	}
676
677	err = mlx5_pagealloc_start(dev);
678	if (err) {
679		dev_err(&pdev->dev, "mlx5_pagealloc_start failed\n");
680		goto reclaim_boot_pages;
681	}
682
683	err = mlx5_cmd_init_hca(dev);
684	if (err) {
685		dev_err(&pdev->dev, "init hca failed\n");
686		goto err_pagealloc_stop;
687	}
688
689	mlx5_start_health_poll(dev);
690
691	err = mlx5_cmd_query_hca_cap(dev, &dev->caps);
692	if (err) {
693		dev_err(&pdev->dev, "query hca failed\n");
694		goto err_stop_poll;
695	}
696
697	err = mlx5_cmd_query_adapter(dev);
698	if (err) {
699		dev_err(&pdev->dev, "query adapter failed\n");
700		goto err_stop_poll;
701	}
702
703	err = mlx5_enable_msix(dev);
704	if (err) {
705		dev_err(&pdev->dev, "enable msix failed\n");
706		goto err_stop_poll;
707	}
708
709	err = mlx5_eq_init(dev);
710	if (err) {
711		dev_err(&pdev->dev, "failed to initialize eq\n");
712		goto disable_msix;
713	}
714
715	err = mlx5_alloc_uuars(dev, &priv->uuari);
716	if (err) {
717		dev_err(&pdev->dev, "Failed allocating uar, aborting\n");
718		goto err_eq_cleanup;
719	}
720
721	err = mlx5_start_eqs(dev);
722	if (err) {
723		dev_err(&pdev->dev, "Failed to start pages and async EQs\n");
724		goto err_free_uar;
725	}
726
727	err = alloc_comp_eqs(dev);
728	if (err) {
729		dev_err(&pdev->dev, "Failed to alloc completion EQs\n");
730		goto err_stop_eqs;
731	}
732
733	MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock);
734
735	mlx5_init_cq_table(dev);
736	mlx5_init_qp_table(dev);
737	mlx5_init_srq_table(dev);
738	mlx5_init_mr_table(dev);
739
740	return 0;
741
742err_stop_eqs:
743	mlx5_stop_eqs(dev);
744
745err_free_uar:
746	mlx5_free_uuars(dev, &priv->uuari);
747
748err_eq_cleanup:
749	mlx5_eq_cleanup(dev);
750
751disable_msix:
752	mlx5_disable_msix(dev);
753
754err_stop_poll:
755	mlx5_stop_health_poll(dev);
756	if (mlx5_cmd_teardown_hca(dev)) {
757		dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
758		return err;
759	}
760
761err_pagealloc_stop:
762	mlx5_pagealloc_stop(dev);
763
764reclaim_boot_pages:
765	mlx5_reclaim_startup_pages(dev);
766
767err_disable_hca:
768	mlx5_core_disable_hca(dev);
769
770err_pagealloc_cleanup:
771	mlx5_pagealloc_cleanup(dev);
772	mlx5_cmd_cleanup(dev);
773
774err_unmap:
775	iounmap(dev->iseg);
776
777err_clr_master:
778	pci_clear_master(dev->pdev);
779	release_bar(dev->pdev);
780
781err_disable:
782	pci_disable_device(dev->pdev);
783
784err_dbg:
785	debugfs_remove(priv->dbg_root);
786	return err;
787}
788
789static void mlx5_dev_cleanup(struct mlx5_core_dev *dev)
790{
791	struct mlx5_priv *priv = &dev->priv;
792
793	mlx5_cleanup_srq_table(dev);
794	mlx5_cleanup_qp_table(dev);
795	mlx5_cleanup_cq_table(dev);
796	free_comp_eqs(dev);
797	mlx5_stop_eqs(dev);
798	mlx5_free_uuars(dev, &priv->uuari);
799	mlx5_eq_cleanup(dev);
800	mlx5_disable_msix(dev);
801	mlx5_stop_health_poll(dev);
802	if (mlx5_cmd_teardown_hca(dev)) {
803		dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
804		return;
805	}
806	mlx5_pagealloc_stop(dev);
807	mlx5_reclaim_startup_pages(dev);
808	mlx5_core_disable_hca(dev);
809	mlx5_pagealloc_cleanup(dev);
810	mlx5_cmd_cleanup(dev);
811	iounmap(dev->iseg);
812	pci_clear_master(dev->pdev);
813	release_bar(dev->pdev);
814	pci_disable_device(dev->pdev);
815	debugfs_remove(priv->dbg_root);
816}
817
818static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
819{
820	struct mlx5_device_context *dev_ctx;
821	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
822
823	dev_ctx = kmalloc(sizeof(*dev_ctx), GFP_KERNEL);
824	if (!dev_ctx) {
825		pr_warn("mlx5_add_device: alloc context failed\n");
826		return;
827	}
828
829	dev_ctx->intf    = intf;
830	dev_ctx->context = intf->add(dev);
831
832	if (dev_ctx->context) {
833		spin_lock_irq(&priv->ctx_lock);
834		list_add_tail(&dev_ctx->list, &priv->ctx_list);
835		spin_unlock_irq(&priv->ctx_lock);
836	} else {
837		kfree(dev_ctx);
838	}
839}
840
841static void mlx5_remove_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
842{
843	struct mlx5_device_context *dev_ctx;
844	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
845
846	list_for_each_entry(dev_ctx, &priv->ctx_list, list)
847		if (dev_ctx->intf == intf) {
848			spin_lock_irq(&priv->ctx_lock);
849			list_del(&dev_ctx->list);
850			spin_unlock_irq(&priv->ctx_lock);
851
852			intf->remove(dev, dev_ctx->context);
853			kfree(dev_ctx);
854			return;
855		}
856}
857static int mlx5_register_device(struct mlx5_core_dev *dev)
858{
859	struct mlx5_priv *priv = &dev->priv;
860	struct mlx5_interface *intf;
861
862	mutex_lock(&intf_mutex);
863	list_add_tail(&priv->dev_list, &dev_list);
864	list_for_each_entry(intf, &intf_list, list)
865		mlx5_add_device(intf, priv);
866	mutex_unlock(&intf_mutex);
867
868	return 0;
869}
870static void mlx5_unregister_device(struct mlx5_core_dev *dev)
871{
872	struct mlx5_priv *priv = &dev->priv;
873	struct mlx5_interface *intf;
874
875	mutex_lock(&intf_mutex);
876	list_for_each_entry(intf, &intf_list, list)
877		mlx5_remove_device(intf, priv);
878	list_del(&priv->dev_list);
879	mutex_unlock(&intf_mutex);
880}
881
882int mlx5_register_interface(struct mlx5_interface *intf)
883{
884	struct mlx5_priv *priv;
885
886	if (!intf->add || !intf->remove)
887		return -EINVAL;
888
889	mutex_lock(&intf_mutex);
890	list_add_tail(&intf->list, &intf_list);
891	list_for_each_entry(priv, &dev_list, dev_list)
892		mlx5_add_device(intf, priv);
893	mutex_unlock(&intf_mutex);
894
895	return 0;
896}
897EXPORT_SYMBOL(mlx5_register_interface);
898
899void mlx5_unregister_interface(struct mlx5_interface *intf)
900{
901	struct mlx5_priv *priv;
902
903	mutex_lock(&intf_mutex);
904	list_for_each_entry(priv, &dev_list, dev_list)
905	       mlx5_remove_device(intf, priv);
906	list_del(&intf->list);
907	mutex_unlock(&intf_mutex);
908}
909EXPORT_SYMBOL(mlx5_unregister_interface);
910
911void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol)
912{
913	struct mlx5_priv *priv = &mdev->priv;
914	struct mlx5_device_context *dev_ctx;
915	unsigned long flags;
916	void *result = NULL;
917
918	spin_lock_irqsave(&priv->ctx_lock, flags);
919
920	list_for_each_entry(dev_ctx, &mdev->priv.ctx_list, list)
921		if ((dev_ctx->intf->protocol == protocol) &&
922		    dev_ctx->intf->get_dev) {
923			result = dev_ctx->intf->get_dev(dev_ctx->context);
924			break;
925		}
926
927	spin_unlock_irqrestore(&priv->ctx_lock, flags);
928
929	return result;
930}
931EXPORT_SYMBOL(mlx5_get_protocol_dev);
932
933static void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event,
934			    unsigned long param)
935{
936	struct mlx5_priv *priv = &dev->priv;
937	struct mlx5_device_context *dev_ctx;
938	unsigned long flags;
939
940	spin_lock_irqsave(&priv->ctx_lock, flags);
941
942	list_for_each_entry(dev_ctx, &priv->ctx_list, list)
943		if (dev_ctx->intf->event)
944			dev_ctx->intf->event(dev, dev_ctx->context, event, param);
945
946	spin_unlock_irqrestore(&priv->ctx_lock, flags);
947}
948
949struct mlx5_core_event_handler {
950	void (*event)(struct mlx5_core_dev *dev,
951		      enum mlx5_dev_event event,
952		      void *data);
953};
954
955#define MLX5_IB_MOD "mlx5_ib"
956
957static int init_one(struct pci_dev *pdev,
958		    const struct pci_device_id *id)
959{
960	struct mlx5_core_dev *dev;
961	struct mlx5_priv *priv;
962	int err;
963
964	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
965	if (!dev) {
966		dev_err(&pdev->dev, "kzalloc failed\n");
967		return -ENOMEM;
968	}
969	priv = &dev->priv;
970
971	pci_set_drvdata(pdev, dev);
972
973	if (prof_sel < 0 || prof_sel >= ARRAY_SIZE(profile)) {
974		pr_warn("selected profile out of range, selecting default (%d)\n",
975			MLX5_DEFAULT_PROF);
976		prof_sel = MLX5_DEFAULT_PROF;
977	}
978	dev->profile = &profile[prof_sel];
979	dev->event = mlx5_core_event;
980
981	INIT_LIST_HEAD(&priv->ctx_list);
982	spin_lock_init(&priv->ctx_lock);
983	err = mlx5_dev_init(dev, pdev);
984	if (err) {
985		dev_err(&pdev->dev, "mlx5_dev_init failed %d\n", err);
986		goto out;
987	}
988
989	err = mlx5_register_device(dev);
990	if (err) {
991		dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err);
992		goto out_init;
993	}
994
995	err = request_module_nowait(MLX5_IB_MOD);
996	if (err)
997		pr_info("failed request module on %s\n", MLX5_IB_MOD);
998
999	return 0;
1000
1001out_init:
1002	mlx5_dev_cleanup(dev);
1003out:
1004	kfree(dev);
1005	return err;
1006}
1007static void remove_one(struct pci_dev *pdev)
1008{
1009	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1010
1011	mlx5_unregister_device(dev);
1012	mlx5_dev_cleanup(dev);
1013	kfree(dev);
1014}
1015
1016static const struct pci_device_id mlx5_core_pci_table[] = {
1017	{ PCI_VDEVICE(MELLANOX, 0x1011) }, /* Connect-IB */
1018	{ PCI_VDEVICE(MELLANOX, 0x1012) }, /* Connect-IB VF */
1019	{ PCI_VDEVICE(MELLANOX, 0x1013) }, /* ConnectX-4 */
1020	{ PCI_VDEVICE(MELLANOX, 0x1014) }, /* ConnectX-4 VF */
1021	{ PCI_VDEVICE(MELLANOX, 0x1015) }, /* ConnectX-4LX */
1022	{ PCI_VDEVICE(MELLANOX, 0x1016) }, /* ConnectX-4LX VF */
1023	{ 0, }
1024};
1025
1026MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1027
1028static struct pci_driver mlx5_core_driver = {
1029	.name           = DRIVER_NAME,
1030	.id_table       = mlx5_core_pci_table,
1031	.probe          = init_one,
1032	.remove         = remove_one
1033};
1034
1035static int __init init(void)
1036{
1037	int err;
1038
1039	mlx5_register_debugfs();
1040	mlx5_core_wq = create_singlethread_workqueue("mlx5_core_wq");
1041	if (!mlx5_core_wq) {
1042		err = -ENOMEM;
1043		goto err_debug;
1044	}
1045	mlx5_health_init();
1046
1047	err = pci_register_driver(&mlx5_core_driver);
1048	if (err)
1049		goto err_health;
1050
1051	return 0;
1052
1053err_health:
1054	mlx5_health_cleanup();
1055	destroy_workqueue(mlx5_core_wq);
1056err_debug:
1057	mlx5_unregister_debugfs();
1058	return err;
1059}
1060
1061static void __exit cleanup(void)
1062{
1063	pci_unregister_driver(&mlx5_core_driver);
1064	mlx5_health_cleanup();
1065	destroy_workqueue(mlx5_core_wq);
1066	mlx5_unregister_debugfs();
1067}
1068
1069module_init(init);
1070module_exit(cleanup);
1071