xref: /openbmc/linux/drivers/dma/idxd/init.c (revision 0661cb2a)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2019 Intel Corporation. All rights rsvd. */
3 #include <linux/init.h>
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/slab.h>
7 #include <linux/pci.h>
8 #include <linux/interrupt.h>
9 #include <linux/delay.h>
10 #include <linux/dma-mapping.h>
11 #include <linux/workqueue.h>
12 #include <linux/aer.h>
13 #include <linux/fs.h>
14 #include <linux/io-64-nonatomic-lo-hi.h>
15 #include <linux/device.h>
16 #include <linux/idr.h>
17 #include <linux/intel-svm.h>
18 #include <linux/iommu.h>
19 #include <uapi/linux/idxd.h>
20 #include <linux/dmaengine.h>
21 #include "../dmaengine.h"
22 #include "registers.h"
23 #include "idxd.h"
24 #include "perfmon.h"
25 
26 MODULE_VERSION(IDXD_DRIVER_VERSION);
27 MODULE_LICENSE("GPL v2");
28 MODULE_AUTHOR("Intel Corporation");
29 
30 static bool sva = true;
31 module_param(sva, bool, 0644);
32 MODULE_PARM_DESC(sva, "Toggle SVA support on/off");
33 
34 #define DRV_NAME "idxd"
35 
36 bool support_enqcmd;
37 DEFINE_IDA(idxd_ida);
38 
39 static struct idxd_driver_data idxd_driver_data[] = {
40 	[IDXD_TYPE_DSA] = {
41 		.name_prefix = "dsa",
42 		.type = IDXD_TYPE_DSA,
43 		.compl_size = sizeof(struct dsa_completion_record),
44 		.align = 32,
45 		.dev_type = &dsa_device_type,
46 	},
47 	[IDXD_TYPE_IAX] = {
48 		.name_prefix = "iax",
49 		.type = IDXD_TYPE_IAX,
50 		.compl_size = sizeof(struct iax_completion_record),
51 		.align = 64,
52 		.dev_type = &iax_device_type,
53 	},
54 };
55 
56 static struct pci_device_id idxd_pci_tbl[] = {
57 	/* DSA ver 1.0 platforms */
58 	{ PCI_DEVICE_DATA(INTEL, DSA_SPR0, &idxd_driver_data[IDXD_TYPE_DSA]) },
59 
60 	/* IAX ver 1.0 platforms */
61 	{ PCI_DEVICE_DATA(INTEL, IAX_SPR0, &idxd_driver_data[IDXD_TYPE_IAX]) },
62 	{ 0, }
63 };
64 MODULE_DEVICE_TABLE(pci, idxd_pci_tbl);
65 
66 static int idxd_setup_interrupts(struct idxd_device *idxd)
67 {
68 	struct pci_dev *pdev = idxd->pdev;
69 	struct device *dev = &pdev->dev;
70 	struct idxd_irq_entry *irq_entry;
71 	int i, msixcnt;
72 	int rc = 0;
73 
74 	msixcnt = pci_msix_vec_count(pdev);
75 	if (msixcnt < 0) {
76 		dev_err(dev, "Not MSI-X interrupt capable.\n");
77 		return -ENOSPC;
78 	}
79 
80 	rc = pci_alloc_irq_vectors(pdev, msixcnt, msixcnt, PCI_IRQ_MSIX);
81 	if (rc != msixcnt) {
82 		dev_err(dev, "Failed enabling %d MSIX entries: %d\n", msixcnt, rc);
83 		return -ENOSPC;
84 	}
85 	dev_dbg(dev, "Enabled %d msix vectors\n", msixcnt);
86 
87 	/*
88 	 * We implement 1 completion list per MSI-X entry except for
89 	 * entry 0, which is for errors and others.
90 	 */
91 	idxd->irq_entries = kcalloc_node(msixcnt, sizeof(struct idxd_irq_entry),
92 					 GFP_KERNEL, dev_to_node(dev));
93 	if (!idxd->irq_entries) {
94 		rc = -ENOMEM;
95 		goto err_irq_entries;
96 	}
97 
98 	for (i = 0; i < msixcnt; i++) {
99 		idxd->irq_entries[i].id = i;
100 		idxd->irq_entries[i].idxd = idxd;
101 		idxd->irq_entries[i].vector = pci_irq_vector(pdev, i);
102 		spin_lock_init(&idxd->irq_entries[i].list_lock);
103 	}
104 
105 	irq_entry = &idxd->irq_entries[0];
106 	rc = request_threaded_irq(irq_entry->vector, NULL, idxd_misc_thread,
107 				  0, "idxd-misc", irq_entry);
108 	if (rc < 0) {
109 		dev_err(dev, "Failed to allocate misc interrupt.\n");
110 		goto err_misc_irq;
111 	}
112 
113 	dev_dbg(dev, "Allocated idxd-misc handler on msix vector %d\n", irq_entry->vector);
114 
115 	/* first MSI-X entry is not for wq interrupts */
116 	idxd->num_wq_irqs = msixcnt - 1;
117 
118 	for (i = 1; i < msixcnt; i++) {
119 		irq_entry = &idxd->irq_entries[i];
120 
121 		init_llist_head(&idxd->irq_entries[i].pending_llist);
122 		INIT_LIST_HEAD(&idxd->irq_entries[i].work_list);
123 		rc = request_threaded_irq(irq_entry->vector, NULL,
124 					  idxd_wq_thread, 0, "idxd-portal", irq_entry);
125 		if (rc < 0) {
126 			dev_err(dev, "Failed to allocate irq %d.\n", irq_entry->vector);
127 			goto err_wq_irqs;
128 		}
129 
130 		dev_dbg(dev, "Allocated idxd-msix %d for vector %d\n", i, irq_entry->vector);
131 		if (idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE)) {
132 			/*
133 			 * The MSIX vector enumeration starts at 1 with vector 0 being the
134 			 * misc interrupt that handles non I/O completion events. The
135 			 * interrupt handles are for IMS enumeration on guest. The misc
136 			 * interrupt vector does not require a handle and therefore we start
137 			 * the int_handles at index 0. Since 'i' starts at 1, the first
138 			 * int_handles index will be 0.
139 			 */
140 			rc = idxd_device_request_int_handle(idxd, i, &idxd->int_handles[i - 1],
141 							    IDXD_IRQ_MSIX);
142 			if (rc < 0) {
143 				free_irq(irq_entry->vector, irq_entry);
144 				goto err_wq_irqs;
145 			}
146 			dev_dbg(dev, "int handle requested: %u\n", idxd->int_handles[i - 1]);
147 		}
148 	}
149 
150 	idxd_unmask_error_interrupts(idxd);
151 	idxd_msix_perm_setup(idxd);
152 	return 0;
153 
154  err_wq_irqs:
155 	while (--i >= 0) {
156 		irq_entry = &idxd->irq_entries[i];
157 		free_irq(irq_entry->vector, irq_entry);
158 		if (i != 0)
159 			idxd_device_release_int_handle(idxd,
160 						       idxd->int_handles[i], IDXD_IRQ_MSIX);
161 	}
162  err_misc_irq:
163 	/* Disable error interrupt generation */
164 	idxd_mask_error_interrupts(idxd);
165  err_irq_entries:
166 	pci_free_irq_vectors(pdev);
167 	dev_err(dev, "No usable interrupts\n");
168 	return rc;
169 }
170 
171 static void idxd_cleanup_interrupts(struct idxd_device *idxd)
172 {
173 	struct pci_dev *pdev = idxd->pdev;
174 	struct idxd_irq_entry *irq_entry;
175 	int i, msixcnt;
176 
177 	msixcnt = pci_msix_vec_count(pdev);
178 	if (msixcnt <= 0)
179 		return;
180 
181 	irq_entry = &idxd->irq_entries[0];
182 	free_irq(irq_entry->vector, irq_entry);
183 
184 	for (i = 1; i < msixcnt; i++) {
185 
186 		irq_entry = &idxd->irq_entries[i];
187 		if (idxd->hw.cmd_cap & BIT(IDXD_CMD_RELEASE_INT_HANDLE))
188 			idxd_device_release_int_handle(idxd, idxd->int_handles[i],
189 						       IDXD_IRQ_MSIX);
190 		free_irq(irq_entry->vector, irq_entry);
191 	}
192 
193 	idxd_mask_error_interrupts(idxd);
194 	pci_free_irq_vectors(pdev);
195 }
196 
197 static int idxd_setup_wqs(struct idxd_device *idxd)
198 {
199 	struct device *dev = &idxd->pdev->dev;
200 	struct idxd_wq *wq;
201 	int i, rc;
202 
203 	idxd->wqs = kcalloc_node(idxd->max_wqs, sizeof(struct idxd_wq *),
204 				 GFP_KERNEL, dev_to_node(dev));
205 	if (!idxd->wqs)
206 		return -ENOMEM;
207 
208 	for (i = 0; i < idxd->max_wqs; i++) {
209 		wq = kzalloc_node(sizeof(*wq), GFP_KERNEL, dev_to_node(dev));
210 		if (!wq) {
211 			rc = -ENOMEM;
212 			goto err;
213 		}
214 
215 		wq->id = i;
216 		wq->idxd = idxd;
217 		device_initialize(&wq->conf_dev);
218 		wq->conf_dev.parent = &idxd->conf_dev;
219 		wq->conf_dev.bus = &dsa_bus_type;
220 		wq->conf_dev.type = &idxd_wq_device_type;
221 		rc = dev_set_name(&wq->conf_dev, "wq%d.%d", idxd->id, wq->id);
222 		if (rc < 0) {
223 			put_device(&wq->conf_dev);
224 			goto err;
225 		}
226 
227 		mutex_init(&wq->wq_lock);
228 		init_waitqueue_head(&wq->err_queue);
229 		init_completion(&wq->wq_dead);
230 		wq->max_xfer_bytes = idxd->max_xfer_bytes;
231 		wq->max_batch_size = idxd->max_batch_size;
232 		wq->wqcfg = kzalloc_node(idxd->wqcfg_size, GFP_KERNEL, dev_to_node(dev));
233 		if (!wq->wqcfg) {
234 			put_device(&wq->conf_dev);
235 			rc = -ENOMEM;
236 			goto err;
237 		}
238 		idxd->wqs[i] = wq;
239 	}
240 
241 	return 0;
242 
243  err:
244 	while (--i >= 0)
245 		put_device(&idxd->wqs[i]->conf_dev);
246 	return rc;
247 }
248 
249 static int idxd_setup_engines(struct idxd_device *idxd)
250 {
251 	struct idxd_engine *engine;
252 	struct device *dev = &idxd->pdev->dev;
253 	int i, rc;
254 
255 	idxd->engines = kcalloc_node(idxd->max_engines, sizeof(struct idxd_engine *),
256 				     GFP_KERNEL, dev_to_node(dev));
257 	if (!idxd->engines)
258 		return -ENOMEM;
259 
260 	for (i = 0; i < idxd->max_engines; i++) {
261 		engine = kzalloc_node(sizeof(*engine), GFP_KERNEL, dev_to_node(dev));
262 		if (!engine) {
263 			rc = -ENOMEM;
264 			goto err;
265 		}
266 
267 		engine->id = i;
268 		engine->idxd = idxd;
269 		device_initialize(&engine->conf_dev);
270 		engine->conf_dev.parent = &idxd->conf_dev;
271 		engine->conf_dev.bus = &dsa_bus_type;
272 		engine->conf_dev.type = &idxd_engine_device_type;
273 		rc = dev_set_name(&engine->conf_dev, "engine%d.%d", idxd->id, engine->id);
274 		if (rc < 0) {
275 			put_device(&engine->conf_dev);
276 			goto err;
277 		}
278 
279 		idxd->engines[i] = engine;
280 	}
281 
282 	return 0;
283 
284  err:
285 	while (--i >= 0)
286 		put_device(&idxd->engines[i]->conf_dev);
287 	return rc;
288 }
289 
290 static int idxd_setup_groups(struct idxd_device *idxd)
291 {
292 	struct device *dev = &idxd->pdev->dev;
293 	struct idxd_group *group;
294 	int i, rc;
295 
296 	idxd->groups = kcalloc_node(idxd->max_groups, sizeof(struct idxd_group *),
297 				    GFP_KERNEL, dev_to_node(dev));
298 	if (!idxd->groups)
299 		return -ENOMEM;
300 
301 	for (i = 0; i < idxd->max_groups; i++) {
302 		group = kzalloc_node(sizeof(*group), GFP_KERNEL, dev_to_node(dev));
303 		if (!group) {
304 			rc = -ENOMEM;
305 			goto err;
306 		}
307 
308 		group->id = i;
309 		group->idxd = idxd;
310 		device_initialize(&group->conf_dev);
311 		group->conf_dev.parent = &idxd->conf_dev;
312 		group->conf_dev.bus = &dsa_bus_type;
313 		group->conf_dev.type = &idxd_group_device_type;
314 		rc = dev_set_name(&group->conf_dev, "group%d.%d", idxd->id, group->id);
315 		if (rc < 0) {
316 			put_device(&group->conf_dev);
317 			goto err;
318 		}
319 
320 		idxd->groups[i] = group;
321 		group->tc_a = -1;
322 		group->tc_b = -1;
323 	}
324 
325 	return 0;
326 
327  err:
328 	while (--i >= 0)
329 		put_device(&idxd->groups[i]->conf_dev);
330 	return rc;
331 }
332 
333 static void idxd_cleanup_internals(struct idxd_device *idxd)
334 {
335 	int i;
336 
337 	for (i = 0; i < idxd->max_groups; i++)
338 		put_device(&idxd->groups[i]->conf_dev);
339 	for (i = 0; i < idxd->max_engines; i++)
340 		put_device(&idxd->engines[i]->conf_dev);
341 	for (i = 0; i < idxd->max_wqs; i++)
342 		put_device(&idxd->wqs[i]->conf_dev);
343 	destroy_workqueue(idxd->wq);
344 }
345 
346 static int idxd_setup_internals(struct idxd_device *idxd)
347 {
348 	struct device *dev = &idxd->pdev->dev;
349 	int rc, i;
350 
351 	init_waitqueue_head(&idxd->cmd_waitq);
352 
353 	if (idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE)) {
354 		idxd->int_handles = kcalloc_node(idxd->max_wqs, sizeof(int), GFP_KERNEL,
355 						 dev_to_node(dev));
356 		if (!idxd->int_handles)
357 			return -ENOMEM;
358 	}
359 
360 	rc = idxd_setup_wqs(idxd);
361 	if (rc < 0)
362 		goto err_wqs;
363 
364 	rc = idxd_setup_engines(idxd);
365 	if (rc < 0)
366 		goto err_engine;
367 
368 	rc = idxd_setup_groups(idxd);
369 	if (rc < 0)
370 		goto err_group;
371 
372 	idxd->wq = create_workqueue(dev_name(dev));
373 	if (!idxd->wq) {
374 		rc = -ENOMEM;
375 		goto err_wkq_create;
376 	}
377 
378 	return 0;
379 
380  err_wkq_create:
381 	for (i = 0; i < idxd->max_groups; i++)
382 		put_device(&idxd->groups[i]->conf_dev);
383  err_group:
384 	for (i = 0; i < idxd->max_engines; i++)
385 		put_device(&idxd->engines[i]->conf_dev);
386  err_engine:
387 	for (i = 0; i < idxd->max_wqs; i++)
388 		put_device(&idxd->wqs[i]->conf_dev);
389  err_wqs:
390 	kfree(idxd->int_handles);
391 	return rc;
392 }
393 
394 static void idxd_read_table_offsets(struct idxd_device *idxd)
395 {
396 	union offsets_reg offsets;
397 	struct device *dev = &idxd->pdev->dev;
398 
399 	offsets.bits[0] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET);
400 	offsets.bits[1] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET + sizeof(u64));
401 	idxd->grpcfg_offset = offsets.grpcfg * IDXD_TABLE_MULT;
402 	dev_dbg(dev, "IDXD Group Config Offset: %#x\n", idxd->grpcfg_offset);
403 	idxd->wqcfg_offset = offsets.wqcfg * IDXD_TABLE_MULT;
404 	dev_dbg(dev, "IDXD Work Queue Config Offset: %#x\n", idxd->wqcfg_offset);
405 	idxd->msix_perm_offset = offsets.msix_perm * IDXD_TABLE_MULT;
406 	dev_dbg(dev, "IDXD MSIX Permission Offset: %#x\n", idxd->msix_perm_offset);
407 	idxd->perfmon_offset = offsets.perfmon * IDXD_TABLE_MULT;
408 	dev_dbg(dev, "IDXD Perfmon Offset: %#x\n", idxd->perfmon_offset);
409 }
410 
411 static void idxd_read_caps(struct idxd_device *idxd)
412 {
413 	struct device *dev = &idxd->pdev->dev;
414 	int i;
415 
416 	/* reading generic capabilities */
417 	idxd->hw.gen_cap.bits = ioread64(idxd->reg_base + IDXD_GENCAP_OFFSET);
418 	dev_dbg(dev, "gen_cap: %#llx\n", idxd->hw.gen_cap.bits);
419 
420 	if (idxd->hw.gen_cap.cmd_cap) {
421 		idxd->hw.cmd_cap = ioread32(idxd->reg_base + IDXD_CMDCAP_OFFSET);
422 		dev_dbg(dev, "cmd_cap: %#x\n", idxd->hw.cmd_cap);
423 	}
424 
425 	idxd->max_xfer_bytes = 1ULL << idxd->hw.gen_cap.max_xfer_shift;
426 	dev_dbg(dev, "max xfer size: %llu bytes\n", idxd->max_xfer_bytes);
427 	idxd->max_batch_size = 1U << idxd->hw.gen_cap.max_batch_shift;
428 	dev_dbg(dev, "max batch size: %u\n", idxd->max_batch_size);
429 	if (idxd->hw.gen_cap.config_en)
430 		set_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags);
431 
432 	/* reading group capabilities */
433 	idxd->hw.group_cap.bits =
434 		ioread64(idxd->reg_base + IDXD_GRPCAP_OFFSET);
435 	dev_dbg(dev, "group_cap: %#llx\n", idxd->hw.group_cap.bits);
436 	idxd->max_groups = idxd->hw.group_cap.num_groups;
437 	dev_dbg(dev, "max groups: %u\n", idxd->max_groups);
438 	idxd->max_tokens = idxd->hw.group_cap.total_tokens;
439 	dev_dbg(dev, "max tokens: %u\n", idxd->max_tokens);
440 	idxd->nr_tokens = idxd->max_tokens;
441 
442 	/* read engine capabilities */
443 	idxd->hw.engine_cap.bits =
444 		ioread64(idxd->reg_base + IDXD_ENGCAP_OFFSET);
445 	dev_dbg(dev, "engine_cap: %#llx\n", idxd->hw.engine_cap.bits);
446 	idxd->max_engines = idxd->hw.engine_cap.num_engines;
447 	dev_dbg(dev, "max engines: %u\n", idxd->max_engines);
448 
449 	/* read workqueue capabilities */
450 	idxd->hw.wq_cap.bits = ioread64(idxd->reg_base + IDXD_WQCAP_OFFSET);
451 	dev_dbg(dev, "wq_cap: %#llx\n", idxd->hw.wq_cap.bits);
452 	idxd->max_wq_size = idxd->hw.wq_cap.total_wq_size;
453 	dev_dbg(dev, "total workqueue size: %u\n", idxd->max_wq_size);
454 	idxd->max_wqs = idxd->hw.wq_cap.num_wqs;
455 	dev_dbg(dev, "max workqueues: %u\n", idxd->max_wqs);
456 	idxd->wqcfg_size = 1 << (idxd->hw.wq_cap.wqcfg_size + IDXD_WQCFG_MIN);
457 	dev_dbg(dev, "wqcfg size: %u\n", idxd->wqcfg_size);
458 
459 	/* reading operation capabilities */
460 	for (i = 0; i < 4; i++) {
461 		idxd->hw.opcap.bits[i] = ioread64(idxd->reg_base +
462 				IDXD_OPCAP_OFFSET + i * sizeof(u64));
463 		dev_dbg(dev, "opcap[%d]: %#llx\n", i, idxd->hw.opcap.bits[i]);
464 	}
465 }
466 
467 static struct idxd_device *idxd_alloc(struct pci_dev *pdev, struct idxd_driver_data *data)
468 {
469 	struct device *dev = &pdev->dev;
470 	struct idxd_device *idxd;
471 	int rc;
472 
473 	idxd = kzalloc_node(sizeof(*idxd), GFP_KERNEL, dev_to_node(dev));
474 	if (!idxd)
475 		return NULL;
476 
477 	idxd->pdev = pdev;
478 	idxd->data = data;
479 	idxd->id = ida_alloc(&idxd_ida, GFP_KERNEL);
480 	if (idxd->id < 0)
481 		return NULL;
482 
483 	device_initialize(&idxd->conf_dev);
484 	idxd->conf_dev.parent = dev;
485 	idxd->conf_dev.bus = &dsa_bus_type;
486 	idxd->conf_dev.type = idxd->data->dev_type;
487 	rc = dev_set_name(&idxd->conf_dev, "%s%d", idxd->data->name_prefix, idxd->id);
488 	if (rc < 0) {
489 		put_device(&idxd->conf_dev);
490 		return NULL;
491 	}
492 
493 	spin_lock_init(&idxd->dev_lock);
494 	spin_lock_init(&idxd->cmd_lock);
495 
496 	return idxd;
497 }
498 
499 static int idxd_enable_system_pasid(struct idxd_device *idxd)
500 {
501 	int flags;
502 	unsigned int pasid;
503 	struct iommu_sva *sva;
504 
505 	flags = SVM_FLAG_SUPERVISOR_MODE;
506 
507 	sva = iommu_sva_bind_device(&idxd->pdev->dev, NULL, &flags);
508 	if (IS_ERR(sva)) {
509 		dev_warn(&idxd->pdev->dev,
510 			 "iommu sva bind failed: %ld\n", PTR_ERR(sva));
511 		return PTR_ERR(sva);
512 	}
513 
514 	pasid = iommu_sva_get_pasid(sva);
515 	if (pasid == IOMMU_PASID_INVALID) {
516 		iommu_sva_unbind_device(sva);
517 		return -ENODEV;
518 	}
519 
520 	idxd->sva = sva;
521 	idxd->pasid = pasid;
522 	dev_dbg(&idxd->pdev->dev, "system pasid: %u\n", pasid);
523 	return 0;
524 }
525 
526 static void idxd_disable_system_pasid(struct idxd_device *idxd)
527 {
528 
529 	iommu_sva_unbind_device(idxd->sva);
530 	idxd->sva = NULL;
531 }
532 
533 static int idxd_probe(struct idxd_device *idxd)
534 {
535 	struct pci_dev *pdev = idxd->pdev;
536 	struct device *dev = &pdev->dev;
537 	int rc;
538 
539 	dev_dbg(dev, "%s entered and resetting device\n", __func__);
540 	rc = idxd_device_init_reset(idxd);
541 	if (rc < 0)
542 		return rc;
543 
544 	dev_dbg(dev, "IDXD reset complete\n");
545 
546 	if (IS_ENABLED(CONFIG_INTEL_IDXD_SVM) && sva) {
547 		rc = iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA);
548 		if (rc == 0) {
549 			rc = idxd_enable_system_pasid(idxd);
550 			if (rc < 0) {
551 				iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
552 				dev_warn(dev, "Failed to enable PASID. No SVA support: %d\n", rc);
553 			} else {
554 				set_bit(IDXD_FLAG_PASID_ENABLED, &idxd->flags);
555 			}
556 		} else {
557 			dev_warn(dev, "Unable to turn on SVA feature.\n");
558 		}
559 	} else if (!sva) {
560 		dev_warn(dev, "User forced SVA off via module param.\n");
561 	}
562 
563 	idxd_read_caps(idxd);
564 	idxd_read_table_offsets(idxd);
565 
566 	rc = idxd_setup_internals(idxd);
567 	if (rc)
568 		goto err;
569 
570 	/* If the configs are readonly, then load them from device */
571 	if (!test_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags)) {
572 		dev_dbg(dev, "Loading RO device config\n");
573 		rc = idxd_device_load_config(idxd);
574 		if (rc < 0)
575 			goto err_config;
576 	}
577 
578 	rc = idxd_setup_interrupts(idxd);
579 	if (rc)
580 		goto err_config;
581 
582 	dev_dbg(dev, "IDXD interrupt setup complete.\n");
583 
584 	idxd->major = idxd_cdev_get_major(idxd);
585 
586 	rc = perfmon_pmu_init(idxd);
587 	if (rc < 0)
588 		dev_warn(dev, "Failed to initialize perfmon. No PMU support: %d\n", rc);
589 
590 	dev_dbg(dev, "IDXD device %d probed successfully\n", idxd->id);
591 	return 0;
592 
593  err_config:
594 	idxd_cleanup_internals(idxd);
595  err:
596 	if (device_pasid_enabled(idxd))
597 		idxd_disable_system_pasid(idxd);
598 	iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
599 	return rc;
600 }
601 
602 static void idxd_cleanup(struct idxd_device *idxd)
603 {
604 	struct device *dev = &idxd->pdev->dev;
605 
606 	perfmon_pmu_remove(idxd);
607 	idxd_cleanup_interrupts(idxd);
608 	idxd_cleanup_internals(idxd);
609 	if (device_pasid_enabled(idxd))
610 		idxd_disable_system_pasid(idxd);
611 	iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
612 }
613 
614 static int idxd_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
615 {
616 	struct device *dev = &pdev->dev;
617 	struct idxd_device *idxd;
618 	struct idxd_driver_data *data = (struct idxd_driver_data *)id->driver_data;
619 	int rc;
620 
621 	rc = pci_enable_device(pdev);
622 	if (rc)
623 		return rc;
624 
625 	dev_dbg(dev, "Alloc IDXD context\n");
626 	idxd = idxd_alloc(pdev, data);
627 	if (!idxd) {
628 		rc = -ENOMEM;
629 		goto err_idxd_alloc;
630 	}
631 
632 	dev_dbg(dev, "Mapping BARs\n");
633 	idxd->reg_base = pci_iomap(pdev, IDXD_MMIO_BAR, 0);
634 	if (!idxd->reg_base) {
635 		rc = -ENOMEM;
636 		goto err_iomap;
637 	}
638 
639 	dev_dbg(dev, "Set DMA masks\n");
640 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
641 	if (rc)
642 		rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
643 	if (rc)
644 		goto err;
645 
646 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
647 	if (rc)
648 		rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
649 	if (rc)
650 		goto err;
651 
652 	dev_dbg(dev, "Set PCI master\n");
653 	pci_set_master(pdev);
654 	pci_set_drvdata(pdev, idxd);
655 
656 	idxd->hw.version = ioread32(idxd->reg_base + IDXD_VER_OFFSET);
657 	rc = idxd_probe(idxd);
658 	if (rc) {
659 		dev_err(dev, "Intel(R) IDXD DMA Engine init failed\n");
660 		goto err;
661 	}
662 
663 	rc = idxd_register_devices(idxd);
664 	if (rc) {
665 		dev_err(dev, "IDXD sysfs setup failed\n");
666 		goto err_dev_register;
667 	}
668 
669 	idxd->state = IDXD_DEV_CONF_READY;
670 
671 	dev_info(&pdev->dev, "Intel(R) Accelerator Device (v%x)\n",
672 		 idxd->hw.version);
673 
674 	return 0;
675 
676  err_dev_register:
677 	idxd_cleanup(idxd);
678  err:
679 	pci_iounmap(pdev, idxd->reg_base);
680  err_iomap:
681 	put_device(&idxd->conf_dev);
682  err_idxd_alloc:
683 	pci_disable_device(pdev);
684 	return rc;
685 }
686 
687 static void idxd_flush_pending_llist(struct idxd_irq_entry *ie)
688 {
689 	struct idxd_desc *desc, *itr;
690 	struct llist_node *head;
691 
692 	head = llist_del_all(&ie->pending_llist);
693 	if (!head)
694 		return;
695 
696 	llist_for_each_entry_safe(desc, itr, head, llnode) {
697 		idxd_dma_complete_txd(desc, IDXD_COMPLETE_ABORT);
698 		idxd_free_desc(desc->wq, desc);
699 	}
700 }
701 
702 static void idxd_flush_work_list(struct idxd_irq_entry *ie)
703 {
704 	struct idxd_desc *desc, *iter;
705 
706 	list_for_each_entry_safe(desc, iter, &ie->work_list, list) {
707 		list_del(&desc->list);
708 		idxd_dma_complete_txd(desc, IDXD_COMPLETE_ABORT);
709 		idxd_free_desc(desc->wq, desc);
710 	}
711 }
712 
713 void idxd_wqs_quiesce(struct idxd_device *idxd)
714 {
715 	struct idxd_wq *wq;
716 	int i;
717 
718 	for (i = 0; i < idxd->max_wqs; i++) {
719 		wq = idxd->wqs[i];
720 		if (wq->state == IDXD_WQ_ENABLED && wq->type == IDXD_WQT_KERNEL)
721 			idxd_wq_quiesce(wq);
722 	}
723 }
724 
725 static void idxd_release_int_handles(struct idxd_device *idxd)
726 {
727 	struct device *dev = &idxd->pdev->dev;
728 	int i, rc;
729 
730 	for (i = 0; i < idxd->num_wq_irqs; i++) {
731 		if (idxd->hw.cmd_cap & BIT(IDXD_CMD_RELEASE_INT_HANDLE)) {
732 			rc = idxd_device_release_int_handle(idxd, idxd->int_handles[i],
733 							    IDXD_IRQ_MSIX);
734 			if (rc < 0)
735 				dev_warn(dev, "irq handle %d release failed\n",
736 					 idxd->int_handles[i]);
737 			else
738 				dev_dbg(dev, "int handle requested: %u\n", idxd->int_handles[i]);
739 		}
740 	}
741 }
742 
743 static void idxd_shutdown(struct pci_dev *pdev)
744 {
745 	struct idxd_device *idxd = pci_get_drvdata(pdev);
746 	int rc, i;
747 	struct idxd_irq_entry *irq_entry;
748 	int msixcnt = pci_msix_vec_count(pdev);
749 
750 	rc = idxd_device_disable(idxd);
751 	if (rc)
752 		dev_err(&pdev->dev, "Disabling device failed\n");
753 
754 	dev_dbg(&pdev->dev, "%s called\n", __func__);
755 	idxd_mask_msix_vectors(idxd);
756 	idxd_mask_error_interrupts(idxd);
757 
758 	for (i = 0; i < msixcnt; i++) {
759 		irq_entry = &idxd->irq_entries[i];
760 		synchronize_irq(irq_entry->vector);
761 		free_irq(irq_entry->vector, irq_entry);
762 		if (i == 0)
763 			continue;
764 		idxd_flush_pending_llist(irq_entry);
765 		idxd_flush_work_list(irq_entry);
766 	}
767 
768 	idxd_msix_perm_clear(idxd);
769 	idxd_release_int_handles(idxd);
770 	pci_free_irq_vectors(pdev);
771 	pci_iounmap(pdev, idxd->reg_base);
772 	pci_disable_device(pdev);
773 	destroy_workqueue(idxd->wq);
774 }
775 
776 static void idxd_remove(struct pci_dev *pdev)
777 {
778 	struct idxd_device *idxd = pci_get_drvdata(pdev);
779 
780 	dev_dbg(&pdev->dev, "%s called\n", __func__);
781 	idxd_shutdown(pdev);
782 	if (device_pasid_enabled(idxd))
783 		idxd_disable_system_pasid(idxd);
784 	idxd_unregister_devices(idxd);
785 	perfmon_pmu_remove(idxd);
786 	iommu_dev_disable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA);
787 }
788 
789 static struct pci_driver idxd_pci_driver = {
790 	.name		= DRV_NAME,
791 	.id_table	= idxd_pci_tbl,
792 	.probe		= idxd_pci_probe,
793 	.remove		= idxd_remove,
794 	.shutdown	= idxd_shutdown,
795 };
796 
797 static int __init idxd_init_module(void)
798 {
799 	int err;
800 
801 	/*
802 	 * If the CPU does not support MOVDIR64B or ENQCMDS, there's no point in
803 	 * enumerating the device. We can not utilize it.
804 	 */
805 	if (!cpu_feature_enabled(X86_FEATURE_MOVDIR64B)) {
806 		pr_warn("idxd driver failed to load without MOVDIR64B.\n");
807 		return -ENODEV;
808 	}
809 
810 	if (!cpu_feature_enabled(X86_FEATURE_ENQCMD))
811 		pr_warn("Platform does not have ENQCMD(S) support.\n");
812 	else
813 		support_enqcmd = true;
814 
815 	perfmon_init();
816 
817 	err = idxd_register_bus_type();
818 	if (err < 0)
819 		return err;
820 
821 	err = idxd_register_driver();
822 	if (err < 0)
823 		goto err_idxd_driver_register;
824 
825 	err = idxd_cdev_register();
826 	if (err)
827 		goto err_cdev_register;
828 
829 	err = pci_register_driver(&idxd_pci_driver);
830 	if (err)
831 		goto err_pci_register;
832 
833 	return 0;
834 
835 err_pci_register:
836 	idxd_cdev_remove();
837 err_cdev_register:
838 	idxd_unregister_driver();
839 err_idxd_driver_register:
840 	idxd_unregister_bus_type();
841 	return err;
842 }
843 module_init(idxd_init_module);
844 
845 static void __exit idxd_exit_module(void)
846 {
847 	idxd_unregister_driver();
848 	pci_unregister_driver(&idxd_pci_driver);
849 	idxd_cdev_remove();
850 	idxd_unregister_bus_type();
851 	perfmon_exit();
852 }
853 module_exit(idxd_exit_module);
854