1 // SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause
2 /*
3  * Copyright 2018-2025 Amazon.com, Inc. or its affiliates. All rights reserved.
4  */
5 
6 #include <linux/module.h>
7 #include <linux/pci.h>
8 #include <linux/utsname.h>
9 #include <linux/version.h>
10 
11 #include <rdma/ib_user_verbs.h>
12 #include <rdma/uverbs_ioctl.h>
13 
14 #include "efa.h"
15 
16 #define PCI_DEV_ID_EFA0_VF 0xefa0
17 #define PCI_DEV_ID_EFA1_VF 0xefa1
18 #define PCI_DEV_ID_EFA2_VF 0xefa2
19 #define PCI_DEV_ID_EFA3_VF 0xefa3
20 
21 static const struct pci_device_id efa_pci_tbl[] = {
22 	{ PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA0_VF) },
23 	{ PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA1_VF) },
24 	{ PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA2_VF) },
25 	{ PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA3_VF) },
26 	{ }
27 };
28 
29 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
30 MODULE_LICENSE("Dual BSD/GPL");
31 MODULE_DESCRIPTION(DEVICE_NAME);
32 MODULE_DEVICE_TABLE(pci, efa_pci_tbl);
33 
34 #define EFA_REG_BAR 0
35 #define EFA_MEM_BAR 2
36 #define EFA_BASE_BAR_MASK (BIT(EFA_REG_BAR) | BIT(EFA_MEM_BAR))
37 
38 #define EFA_AENQ_ENABLED_GROUPS \
39 	(BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \
40 	 BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE))
41 
42 extern const struct uapi_definition efa_uapi_defs[];
43 
44 /* This handler will called for unknown event group or unimplemented handlers */
unimplemented_aenq_handler(void * data,struct efa_admin_aenq_entry * aenq_e)45 static void unimplemented_aenq_handler(void *data,
46 				       struct efa_admin_aenq_entry *aenq_e)
47 {
48 	struct efa_dev *dev = (struct efa_dev *)data;
49 
50 	ibdev_err(&dev->ibdev,
51 		  "Unknown event was received or event with unimplemented handler\n");
52 }
53 
efa_keep_alive(void * data,struct efa_admin_aenq_entry * aenq_e)54 static void efa_keep_alive(void *data, struct efa_admin_aenq_entry *aenq_e)
55 {
56 	struct efa_dev *dev = (struct efa_dev *)data;
57 
58 	atomic64_inc(&dev->stats.keep_alive_rcvd);
59 }
60 
61 static struct efa_aenq_handlers aenq_handlers = {
62 	.handlers = {
63 		[EFA_ADMIN_KEEP_ALIVE] = efa_keep_alive,
64 	},
65 	.unimplemented_handler = unimplemented_aenq_handler
66 };
67 
efa_release_bars(struct efa_dev * dev,int bars_mask)68 static void efa_release_bars(struct efa_dev *dev, int bars_mask)
69 {
70 	struct pci_dev *pdev = dev->pdev;
71 	int release_bars;
72 
73 	release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & bars_mask;
74 	pci_release_selected_regions(pdev, release_bars);
75 }
76 
efa_process_comp_eqe(struct efa_dev * dev,struct efa_admin_eqe * eqe)77 static void efa_process_comp_eqe(struct efa_dev *dev, struct efa_admin_eqe *eqe)
78 {
79 	u16 cqn = eqe->u.comp_event.cqn;
80 	struct efa_cq *cq;
81 
82 	/* Safe to load as we're in irq and removal calls synchronize_irq() */
83 	cq = xa_load(&dev->cqs_xa, cqn);
84 	if (unlikely(!cq)) {
85 		ibdev_err_ratelimited(&dev->ibdev,
86 				      "Completion event on non-existent CQ[%u]",
87 				      cqn);
88 		return;
89 	}
90 
91 	cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
92 }
93 
efa_process_eqe(struct efa_com_eq * eeq,struct efa_admin_eqe * eqe)94 static void efa_process_eqe(struct efa_com_eq *eeq, struct efa_admin_eqe *eqe)
95 {
96 	struct efa_dev *dev = container_of(eeq->edev, struct efa_dev, edev);
97 
98 	if (likely(EFA_GET(&eqe->common, EFA_ADMIN_EQE_EVENT_TYPE) ==
99 			   EFA_ADMIN_EQE_EVENT_TYPE_COMPLETION))
100 		efa_process_comp_eqe(dev, eqe);
101 	else
102 		ibdev_err_ratelimited(&dev->ibdev,
103 				      "Unknown event type received %lu",
104 				      EFA_GET(&eqe->common,
105 					      EFA_ADMIN_EQE_EVENT_TYPE));
106 }
107 
efa_intr_msix_comp(int irq,void * data)108 static irqreturn_t efa_intr_msix_comp(int irq, void *data)
109 {
110 	struct efa_eq *eq = data;
111 	struct efa_com_dev *edev = eq->eeq.edev;
112 
113 	efa_com_eq_comp_intr_handler(edev, &eq->eeq);
114 
115 	return IRQ_HANDLED;
116 }
117 
efa_intr_msix_mgmnt(int irq,void * data)118 static irqreturn_t efa_intr_msix_mgmnt(int irq, void *data)
119 {
120 	struct efa_dev *dev = data;
121 
122 	efa_com_admin_q_comp_intr_handler(&dev->edev);
123 	efa_com_aenq_intr_handler(&dev->edev, data);
124 
125 	return IRQ_HANDLED;
126 }
127 
efa_request_irq(struct efa_dev * dev,struct efa_irq * irq)128 static int efa_request_irq(struct efa_dev *dev, struct efa_irq *irq)
129 {
130 	int err;
131 
132 	err = request_irq(irq->irqn, irq->handler, 0, irq->name, irq->data);
133 	if (err) {
134 		dev_err(&dev->pdev->dev, "Failed to request irq %s (%d)\n",
135 			irq->name, err);
136 		return err;
137 	}
138 
139 	irq_set_affinity_hint(irq->irqn, &irq->affinity_hint_mask);
140 
141 	return 0;
142 }
143 
efa_setup_comp_irq(struct efa_dev * dev,struct efa_eq * eq,u32 vector)144 static void efa_setup_comp_irq(struct efa_dev *dev, struct efa_eq *eq, u32 vector)
145 {
146 	u32 cpu;
147 
148 	cpu = vector - EFA_COMP_EQS_VEC_BASE;
149 	snprintf(eq->irq.name, EFA_IRQNAME_SIZE, "efa-comp%d@pci:%s", cpu,
150 		 pci_name(dev->pdev));
151 	eq->irq.handler = efa_intr_msix_comp;
152 	eq->irq.data = eq;
153 	eq->irq.vector = vector;
154 	eq->irq.irqn = pci_irq_vector(dev->pdev, vector);
155 	cpumask_set_cpu(cpu, &eq->irq.affinity_hint_mask);
156 }
157 
efa_free_irq(struct efa_dev * dev,struct efa_irq * irq)158 static void efa_free_irq(struct efa_dev *dev, struct efa_irq *irq)
159 {
160 	irq_set_affinity_hint(irq->irqn, NULL);
161 	free_irq(irq->irqn, irq->data);
162 }
163 
efa_setup_mgmnt_irq(struct efa_dev * dev)164 static void efa_setup_mgmnt_irq(struct efa_dev *dev)
165 {
166 	u32 cpu;
167 
168 	snprintf(dev->admin_irq.name, EFA_IRQNAME_SIZE,
169 		 "efa-mgmnt@pci:%s", pci_name(dev->pdev));
170 	dev->admin_irq.handler = efa_intr_msix_mgmnt;
171 	dev->admin_irq.data = dev;
172 	dev->admin_irq.vector = dev->admin_msix_vector_idx;
173 	dev->admin_irq.irqn = pci_irq_vector(dev->pdev,
174 					     dev->admin_msix_vector_idx);
175 	cpu = cpumask_first(cpu_online_mask);
176 	cpumask_set_cpu(cpu,
177 			&dev->admin_irq.affinity_hint_mask);
178 	dev_info(&dev->pdev->dev, "Setup irq:%d name:%s\n",
179 		 dev->admin_irq.irqn,
180 		 dev->admin_irq.name);
181 }
182 
efa_set_mgmnt_irq(struct efa_dev * dev)183 static int efa_set_mgmnt_irq(struct efa_dev *dev)
184 {
185 	efa_setup_mgmnt_irq(dev);
186 
187 	return efa_request_irq(dev, &dev->admin_irq);
188 }
189 
efa_request_doorbell_bar(struct efa_dev * dev)190 static int efa_request_doorbell_bar(struct efa_dev *dev)
191 {
192 	u8 db_bar_idx = dev->dev_attr.db_bar;
193 	struct pci_dev *pdev = dev->pdev;
194 	int pci_mem_bars;
195 	int db_bar;
196 	int err;
197 
198 	db_bar = BIT(db_bar_idx);
199 	if (!(db_bar & EFA_BASE_BAR_MASK)) {
200 		pci_mem_bars = pci_select_bars(pdev, IORESOURCE_MEM);
201 		if (db_bar & ~pci_mem_bars) {
202 			dev_err(&pdev->dev,
203 				"Doorbells BAR unavailable. Requested %#x, available %#x\n",
204 				db_bar, pci_mem_bars);
205 			return -ENODEV;
206 		}
207 
208 		err = pci_request_selected_regions(pdev, db_bar, DRV_MODULE_NAME);
209 		if (err) {
210 			dev_err(&pdev->dev,
211 				"pci_request_selected_regions for bar %d failed %d\n",
212 				db_bar_idx, err);
213 			return err;
214 		}
215 	}
216 
217 	dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx);
218 	dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx);
219 
220 	return 0;
221 }
222 
efa_release_doorbell_bar(struct efa_dev * dev)223 static void efa_release_doorbell_bar(struct efa_dev *dev)
224 {
225 	if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK))
226 		efa_release_bars(dev, BIT(dev->dev_attr.db_bar));
227 }
228 
efa_update_hw_hints(struct efa_dev * dev,struct efa_com_get_hw_hints_result * hw_hints)229 static void efa_update_hw_hints(struct efa_dev *dev,
230 				struct efa_com_get_hw_hints_result *hw_hints)
231 {
232 	struct efa_com_dev *edev = &dev->edev;
233 
234 	if (hw_hints->mmio_read_timeout)
235 		edev->mmio_read.mmio_read_timeout =
236 			hw_hints->mmio_read_timeout * 1000;
237 
238 	if (hw_hints->poll_interval)
239 		edev->aq.poll_interval = hw_hints->poll_interval;
240 
241 	if (hw_hints->admin_completion_timeout)
242 		edev->aq.completion_timeout =
243 			hw_hints->admin_completion_timeout;
244 }
245 
efa_stats_init(struct efa_dev * dev)246 static void efa_stats_init(struct efa_dev *dev)
247 {
248 	atomic64_t *s = (atomic64_t *)&dev->stats;
249 	int i;
250 
251 	for (i = 0; i < sizeof(dev->stats) / sizeof(*s); i++, s++)
252 		atomic64_set(s, 0);
253 }
254 
efa_set_host_info(struct efa_dev * dev)255 static void efa_set_host_info(struct efa_dev *dev)
256 {
257 	struct efa_admin_set_feature_resp resp = {};
258 	struct efa_admin_set_feature_cmd cmd = {};
259 	struct efa_admin_host_info *hinf;
260 	u32 bufsz = sizeof(*hinf);
261 	dma_addr_t hinf_dma;
262 
263 	if (!efa_com_check_supported_feature_id(&dev->edev,
264 						EFA_ADMIN_HOST_INFO))
265 		return;
266 
267 	/* Failures in host info set shall not disturb probe */
268 	hinf = dma_alloc_coherent(&dev->pdev->dev, bufsz, &hinf_dma,
269 				  GFP_KERNEL);
270 	if (!hinf)
271 		return;
272 
273 	strscpy(hinf->os_dist_str, utsname()->release,
274 		sizeof(hinf->os_dist_str));
275 	hinf->os_type = EFA_ADMIN_OS_LINUX;
276 	strscpy(hinf->kernel_ver_str, utsname()->version,
277 		sizeof(hinf->kernel_ver_str));
278 	hinf->kernel_ver = LINUX_VERSION_CODE;
279 	EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MAJOR, 0);
280 	EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MINOR, 0);
281 	EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_SUB_MINOR, 0);
282 	EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MODULE_TYPE, 0);
283 	EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_BUS, dev->pdev->bus->number);
284 	EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_DEVICE,
285 		PCI_SLOT(dev->pdev->devfn));
286 	EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_FUNCTION,
287 		PCI_FUNC(dev->pdev->devfn));
288 	EFA_SET(&hinf->spec_ver, EFA_ADMIN_HOST_INFO_SPEC_MAJOR,
289 		EFA_COMMON_SPEC_VERSION_MAJOR);
290 	EFA_SET(&hinf->spec_ver, EFA_ADMIN_HOST_INFO_SPEC_MINOR,
291 		EFA_COMMON_SPEC_VERSION_MINOR);
292 	EFA_SET(&hinf->flags, EFA_ADMIN_HOST_INFO_INTREE, 1);
293 	EFA_SET(&hinf->flags, EFA_ADMIN_HOST_INFO_GDR, 0);
294 
295 	efa_com_set_feature_ex(&dev->edev, &resp, &cmd, EFA_ADMIN_HOST_INFO,
296 			       hinf_dma, bufsz);
297 
298 	dma_free_coherent(&dev->pdev->dev, bufsz, hinf, hinf_dma);
299 }
300 
efa_destroy_eq(struct efa_dev * dev,struct efa_eq * eq)301 static void efa_destroy_eq(struct efa_dev *dev, struct efa_eq *eq)
302 {
303 	efa_com_eq_destroy(&dev->edev, &eq->eeq);
304 	efa_free_irq(dev, &eq->irq);
305 }
306 
efa_create_eq(struct efa_dev * dev,struct efa_eq * eq,u32 msix_vec)307 static int efa_create_eq(struct efa_dev *dev, struct efa_eq *eq, u32 msix_vec)
308 {
309 	int err;
310 
311 	efa_setup_comp_irq(dev, eq, msix_vec);
312 	err = efa_request_irq(dev, &eq->irq);
313 	if (err)
314 		return err;
315 
316 	err = efa_com_eq_init(&dev->edev, &eq->eeq, efa_process_eqe,
317 			      dev->dev_attr.max_eq_depth, msix_vec);
318 	if (err)
319 		goto err_free_comp_irq;
320 
321 	return 0;
322 
323 err_free_comp_irq:
324 	efa_free_irq(dev, &eq->irq);
325 	return err;
326 }
327 
efa_create_eqs(struct efa_dev * dev)328 static int efa_create_eqs(struct efa_dev *dev)
329 {
330 	u32 neqs = dev->dev_attr.max_eq;
331 	int err, i;
332 
333 	neqs = min_t(u32, neqs, dev->num_irq_vectors - EFA_COMP_EQS_VEC_BASE);
334 	dev->neqs = neqs;
335 	dev->eqs = kcalloc(neqs, sizeof(*dev->eqs), GFP_KERNEL);
336 	if (!dev->eqs)
337 		return -ENOMEM;
338 
339 	for (i = 0; i < neqs; i++) {
340 		err = efa_create_eq(dev, &dev->eqs[i], i + EFA_COMP_EQS_VEC_BASE);
341 		if (err)
342 			goto err_destroy_eqs;
343 	}
344 
345 	return 0;
346 
347 err_destroy_eqs:
348 	for (i--; i >= 0; i--)
349 		efa_destroy_eq(dev, &dev->eqs[i]);
350 	kfree(dev->eqs);
351 
352 	return err;
353 }
354 
efa_destroy_eqs(struct efa_dev * dev)355 static void efa_destroy_eqs(struct efa_dev *dev)
356 {
357 	int i;
358 
359 	for (i = 0; i < dev->neqs; i++)
360 		efa_destroy_eq(dev, &dev->eqs[i]);
361 
362 	kfree(dev->eqs);
363 }
364 
365 static const struct ib_device_ops efa_dev_ops = {
366 	.owner = THIS_MODULE,
367 	.driver_id = RDMA_DRIVER_EFA,
368 	.uverbs_abi_ver = EFA_UVERBS_ABI_VERSION,
369 
370 	.alloc_hw_port_stats = efa_alloc_hw_port_stats,
371 	.alloc_hw_device_stats = efa_alloc_hw_device_stats,
372 	.alloc_pd = efa_alloc_pd,
373 	.alloc_ucontext = efa_alloc_ucontext,
374 	.create_cq = efa_create_cq,
375 	.create_qp = efa_create_qp,
376 	.create_user_ah = efa_create_ah,
377 	.dealloc_pd = efa_dealloc_pd,
378 	.dealloc_ucontext = efa_dealloc_ucontext,
379 	.dereg_mr = efa_dereg_mr,
380 	.destroy_ah = efa_destroy_ah,
381 	.destroy_cq = efa_destroy_cq,
382 	.destroy_qp = efa_destroy_qp,
383 	.get_hw_stats = efa_get_hw_stats,
384 	.get_link_layer = efa_port_link_layer,
385 	.get_port_immutable = efa_get_port_immutable,
386 	.mmap = efa_mmap,
387 	.mmap_free = efa_mmap_free,
388 	.modify_qp = efa_modify_qp,
389 	.query_device = efa_query_device,
390 	.query_gid = efa_query_gid,
391 	.query_pkey = efa_query_pkey,
392 	.query_port = efa_query_port,
393 	.query_qp = efa_query_qp,
394 	.reg_user_mr = efa_reg_mr,
395 	.reg_user_mr_dmabuf = efa_reg_user_mr_dmabuf,
396 
397 	INIT_RDMA_OBJ_SIZE(ib_ah, efa_ah, ibah),
398 	INIT_RDMA_OBJ_SIZE(ib_cq, efa_cq, ibcq),
399 	INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd),
400 	INIT_RDMA_OBJ_SIZE(ib_qp, efa_qp, ibqp),
401 	INIT_RDMA_OBJ_SIZE(ib_ucontext, efa_ucontext, ibucontext),
402 };
403 
efa_ib_device_add(struct efa_dev * dev)404 static int efa_ib_device_add(struct efa_dev *dev)
405 {
406 	struct efa_com_get_hw_hints_result hw_hints;
407 	struct pci_dev *pdev = dev->pdev;
408 	int err;
409 
410 	efa_stats_init(dev);
411 
412 	err = efa_com_get_device_attr(&dev->edev, &dev->dev_attr);
413 	if (err)
414 		return err;
415 
416 	dev_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n", dev->dev_attr.db_bar);
417 	err = efa_request_doorbell_bar(dev);
418 	if (err)
419 		return err;
420 
421 	err = efa_com_get_hw_hints(&dev->edev, &hw_hints);
422 	if (err)
423 		goto err_release_doorbell_bar;
424 
425 	efa_update_hw_hints(dev, &hw_hints);
426 
427 	/* Try to enable all the available aenq groups */
428 	err = efa_com_set_aenq_config(&dev->edev, EFA_AENQ_ENABLED_GROUPS);
429 	if (err)
430 		goto err_release_doorbell_bar;
431 
432 	err = efa_create_eqs(dev);
433 	if (err)
434 		goto err_release_doorbell_bar;
435 
436 	efa_set_host_info(dev);
437 
438 	dev->ibdev.node_type = RDMA_NODE_UNSPECIFIED;
439 	dev->ibdev.node_guid = dev->dev_attr.guid;
440 	dev->ibdev.phys_port_cnt = 1;
441 	dev->ibdev.num_comp_vectors = dev->neqs ?: 1;
442 	dev->ibdev.dev.parent = &pdev->dev;
443 
444 	ib_set_device_ops(&dev->ibdev, &efa_dev_ops);
445 
446 	dev->ibdev.driver_def = efa_uapi_defs;
447 
448 	err = ib_register_device(&dev->ibdev, "efa_%d", &pdev->dev);
449 	if (err)
450 		goto err_destroy_eqs;
451 
452 	ibdev_info(&dev->ibdev, "IB device registered\n");
453 
454 	return 0;
455 
456 err_destroy_eqs:
457 	efa_destroy_eqs(dev);
458 err_release_doorbell_bar:
459 	efa_release_doorbell_bar(dev);
460 	return err;
461 }
462 
efa_ib_device_remove(struct efa_dev * dev)463 static void efa_ib_device_remove(struct efa_dev *dev)
464 {
465 	ibdev_info(&dev->ibdev, "Unregister ib device\n");
466 	ib_unregister_device(&dev->ibdev);
467 	efa_destroy_eqs(dev);
468 	efa_release_doorbell_bar(dev);
469 }
470 
efa_disable_msix(struct efa_dev * dev)471 static void efa_disable_msix(struct efa_dev *dev)
472 {
473 	pci_free_irq_vectors(dev->pdev);
474 }
475 
efa_enable_msix(struct efa_dev * dev)476 static int efa_enable_msix(struct efa_dev *dev)
477 {
478 	int max_vecs, num_vecs;
479 
480 	/*
481 	 * Reserve the max msix vectors we might need, one vector is reserved
482 	 * for admin.
483 	 */
484 	max_vecs = min_t(int, pci_msix_vec_count(dev->pdev),
485 			 num_online_cpus() + 1);
486 	dev_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n",
487 		max_vecs);
488 
489 	dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX;
490 	num_vecs = pci_alloc_irq_vectors(dev->pdev, 1,
491 					 max_vecs, PCI_IRQ_MSIX);
492 
493 	if (num_vecs < 0) {
494 		dev_err(&dev->pdev->dev, "Failed to enable MSI-X. error %d\n",
495 			num_vecs);
496 		return -ENOSPC;
497 	}
498 
499 	dev_dbg(&dev->pdev->dev, "Allocated %d MSI-X vectors\n", num_vecs);
500 
501 	dev->num_irq_vectors = num_vecs;
502 
503 	return 0;
504 }
505 
efa_device_init(struct efa_com_dev * edev,struct pci_dev * pdev)506 static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev)
507 {
508 	int dma_width;
509 	int err;
510 
511 	err = efa_com_dev_reset(edev, EFA_REGS_RESET_NORMAL);
512 	if (err)
513 		return err;
514 
515 	err = efa_com_validate_version(edev);
516 	if (err)
517 		return err;
518 
519 	dma_width = efa_com_get_dma_width(edev);
520 	if (dma_width < 0) {
521 		err = dma_width;
522 		return err;
523 	}
524 
525 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(dma_width));
526 	if (err) {
527 		dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", err);
528 		return err;
529 	}
530 
531 	dma_set_max_seg_size(&pdev->dev, UINT_MAX);
532 	return 0;
533 }
534 
efa_probe_device(struct pci_dev * pdev)535 static struct efa_dev *efa_probe_device(struct pci_dev *pdev)
536 {
537 	struct efa_com_dev *edev;
538 	struct efa_dev *dev;
539 	int pci_mem_bars;
540 	int err;
541 
542 	err = pci_enable_device_mem(pdev);
543 	if (err) {
544 		dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
545 		return ERR_PTR(err);
546 	}
547 
548 	pci_set_master(pdev);
549 
550 	dev = ib_alloc_device(efa_dev, ibdev);
551 	if (!dev) {
552 		dev_err(&pdev->dev, "Device alloc failed\n");
553 		err = -ENOMEM;
554 		goto err_disable_device;
555 	}
556 
557 	pci_set_drvdata(pdev, dev);
558 	edev = &dev->edev;
559 	edev->efa_dev = dev;
560 	edev->dmadev = &pdev->dev;
561 	dev->pdev = pdev;
562 	xa_init(&dev->cqs_xa);
563 
564 	pci_mem_bars = pci_select_bars(pdev, IORESOURCE_MEM);
565 	if (EFA_BASE_BAR_MASK & ~pci_mem_bars) {
566 		dev_err(&pdev->dev, "BARs unavailable. Requested %#x, available %#x\n",
567 			(int)EFA_BASE_BAR_MASK, pci_mem_bars);
568 		err = -ENODEV;
569 		goto err_ibdev_destroy;
570 	}
571 	err = pci_request_selected_regions(pdev, EFA_BASE_BAR_MASK, DRV_MODULE_NAME);
572 	if (err) {
573 		dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
574 			err);
575 		goto err_ibdev_destroy;
576 	}
577 
578 	dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR);
579 	dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR);
580 	dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR);
581 	dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR);
582 
583 	edev->reg_bar = devm_ioremap(&pdev->dev,
584 				     dev->reg_bar_addr,
585 				     dev->reg_bar_len);
586 	if (!edev->reg_bar) {
587 		dev_err(&pdev->dev, "Failed to remap register bar\n");
588 		err = -EFAULT;
589 		goto err_release_bars;
590 	}
591 
592 	err = efa_com_mmio_reg_read_init(edev);
593 	if (err) {
594 		dev_err(&pdev->dev, "Failed to init readless MMIO\n");
595 		goto err_iounmap;
596 	}
597 
598 	err = efa_device_init(edev, pdev);
599 	if (err) {
600 		dev_err(&pdev->dev, "EFA device init failed\n");
601 		if (err == -ETIME)
602 			err = -EPROBE_DEFER;
603 		goto err_reg_read_destroy;
604 	}
605 
606 	err = efa_enable_msix(dev);
607 	if (err)
608 		goto err_reg_read_destroy;
609 
610 	edev->aq.msix_vector_idx = dev->admin_msix_vector_idx;
611 	edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx;
612 
613 	err = efa_set_mgmnt_irq(dev);
614 	if (err)
615 		goto err_disable_msix;
616 
617 	err = efa_com_admin_init(edev, &aenq_handlers);
618 	if (err)
619 		goto err_free_mgmnt_irq;
620 
621 	return dev;
622 
623 err_free_mgmnt_irq:
624 	efa_free_irq(dev, &dev->admin_irq);
625 err_disable_msix:
626 	efa_disable_msix(dev);
627 err_reg_read_destroy:
628 	efa_com_mmio_reg_read_destroy(edev);
629 err_iounmap:
630 	devm_iounmap(&pdev->dev, edev->reg_bar);
631 err_release_bars:
632 	efa_release_bars(dev, EFA_BASE_BAR_MASK);
633 err_ibdev_destroy:
634 	ib_dealloc_device(&dev->ibdev);
635 err_disable_device:
636 	pci_disable_device(pdev);
637 	return ERR_PTR(err);
638 }
639 
efa_remove_device(struct pci_dev * pdev,enum efa_regs_reset_reason_types reset_reason)640 static void efa_remove_device(struct pci_dev *pdev,
641 			      enum efa_regs_reset_reason_types reset_reason)
642 {
643 	struct efa_dev *dev = pci_get_drvdata(pdev);
644 	struct efa_com_dev *edev;
645 
646 	edev = &dev->edev;
647 	efa_com_dev_reset(edev, reset_reason);
648 	efa_com_admin_destroy(edev);
649 	efa_free_irq(dev, &dev->admin_irq);
650 	efa_disable_msix(dev);
651 	efa_com_mmio_reg_read_destroy(edev);
652 	devm_iounmap(&pdev->dev, edev->reg_bar);
653 	efa_release_bars(dev, EFA_BASE_BAR_MASK);
654 	xa_destroy(&dev->cqs_xa);
655 	ib_dealloc_device(&dev->ibdev);
656 	pci_disable_device(pdev);
657 }
658 
efa_probe(struct pci_dev * pdev,const struct pci_device_id * ent)659 static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
660 {
661 	struct efa_dev *dev;
662 	int err;
663 
664 	dev = efa_probe_device(pdev);
665 	if (IS_ERR(dev))
666 		return PTR_ERR(dev);
667 
668 	err = efa_ib_device_add(dev);
669 	if (err)
670 		goto err_remove_device;
671 
672 	return 0;
673 
674 err_remove_device:
675 	efa_remove_device(pdev, EFA_REGS_RESET_INIT_ERR);
676 	return err;
677 }
678 
efa_remove(struct pci_dev * pdev)679 static void efa_remove(struct pci_dev *pdev)
680 {
681 	struct efa_dev *dev = pci_get_drvdata(pdev);
682 
683 	efa_ib_device_remove(dev);
684 	efa_remove_device(pdev, EFA_REGS_RESET_NORMAL);
685 }
686 
efa_shutdown(struct pci_dev * pdev)687 static void efa_shutdown(struct pci_dev *pdev)
688 {
689 	struct efa_dev *dev = pci_get_drvdata(pdev);
690 
691 	efa_destroy_eqs(dev);
692 	efa_com_dev_reset(&dev->edev, EFA_REGS_RESET_SHUTDOWN);
693 	efa_free_irq(dev, &dev->admin_irq);
694 	efa_disable_msix(dev);
695 }
696 
697 static struct pci_driver efa_pci_driver = {
698 	.name           = DRV_MODULE_NAME,
699 	.id_table       = efa_pci_tbl,
700 	.probe          = efa_probe,
701 	.remove         = efa_remove,
702 	.shutdown       = efa_shutdown,
703 };
704 
705 module_pci_driver(efa_pci_driver);
706