xref: /linux/drivers/dma/idxd/init.c (revision 1c4841ccbd2b185587010d6178aac11953f61d4c)
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 int idxd_setup_wqs(struct idxd_device *idxd)
172 {
173 	struct device *dev = &idxd->pdev->dev;
174 	struct idxd_wq *wq;
175 	int i, rc;
176 
177 	idxd->wqs = kcalloc_node(idxd->max_wqs, sizeof(struct idxd_wq *),
178 				 GFP_KERNEL, dev_to_node(dev));
179 	if (!idxd->wqs)
180 		return -ENOMEM;
181 
182 	for (i = 0; i < idxd->max_wqs; i++) {
183 		wq = kzalloc_node(sizeof(*wq), GFP_KERNEL, dev_to_node(dev));
184 		if (!wq) {
185 			rc = -ENOMEM;
186 			goto err;
187 		}
188 
189 		wq->id = i;
190 		wq->idxd = idxd;
191 		device_initialize(&wq->conf_dev);
192 		wq->conf_dev.parent = &idxd->conf_dev;
193 		wq->conf_dev.bus = &dsa_bus_type;
194 		wq->conf_dev.type = &idxd_wq_device_type;
195 		rc = dev_set_name(&wq->conf_dev, "wq%d.%d", idxd->id, wq->id);
196 		if (rc < 0) {
197 			put_device(&wq->conf_dev);
198 			goto err;
199 		}
200 
201 		mutex_init(&wq->wq_lock);
202 		init_waitqueue_head(&wq->err_queue);
203 		init_completion(&wq->wq_dead);
204 		wq->max_xfer_bytes = idxd->max_xfer_bytes;
205 		wq->max_batch_size = idxd->max_batch_size;
206 		wq->wqcfg = kzalloc_node(idxd->wqcfg_size, GFP_KERNEL, dev_to_node(dev));
207 		if (!wq->wqcfg) {
208 			put_device(&wq->conf_dev);
209 			rc = -ENOMEM;
210 			goto err;
211 		}
212 		idxd->wqs[i] = wq;
213 	}
214 
215 	return 0;
216 
217  err:
218 	while (--i >= 0)
219 		put_device(&idxd->wqs[i]->conf_dev);
220 	return rc;
221 }
222 
223 static int idxd_setup_engines(struct idxd_device *idxd)
224 {
225 	struct idxd_engine *engine;
226 	struct device *dev = &idxd->pdev->dev;
227 	int i, rc;
228 
229 	idxd->engines = kcalloc_node(idxd->max_engines, sizeof(struct idxd_engine *),
230 				     GFP_KERNEL, dev_to_node(dev));
231 	if (!idxd->engines)
232 		return -ENOMEM;
233 
234 	for (i = 0; i < idxd->max_engines; i++) {
235 		engine = kzalloc_node(sizeof(*engine), GFP_KERNEL, dev_to_node(dev));
236 		if (!engine) {
237 			rc = -ENOMEM;
238 			goto err;
239 		}
240 
241 		engine->id = i;
242 		engine->idxd = idxd;
243 		device_initialize(&engine->conf_dev);
244 		engine->conf_dev.parent = &idxd->conf_dev;
245 		engine->conf_dev.bus = &dsa_bus_type;
246 		engine->conf_dev.type = &idxd_engine_device_type;
247 		rc = dev_set_name(&engine->conf_dev, "engine%d.%d", idxd->id, engine->id);
248 		if (rc < 0) {
249 			put_device(&engine->conf_dev);
250 			goto err;
251 		}
252 
253 		idxd->engines[i] = engine;
254 	}
255 
256 	return 0;
257 
258  err:
259 	while (--i >= 0)
260 		put_device(&idxd->engines[i]->conf_dev);
261 	return rc;
262 }
263 
264 static int idxd_setup_groups(struct idxd_device *idxd)
265 {
266 	struct device *dev = &idxd->pdev->dev;
267 	struct idxd_group *group;
268 	int i, rc;
269 
270 	idxd->groups = kcalloc_node(idxd->max_groups, sizeof(struct idxd_group *),
271 				    GFP_KERNEL, dev_to_node(dev));
272 	if (!idxd->groups)
273 		return -ENOMEM;
274 
275 	for (i = 0; i < idxd->max_groups; i++) {
276 		group = kzalloc_node(sizeof(*group), GFP_KERNEL, dev_to_node(dev));
277 		if (!group) {
278 			rc = -ENOMEM;
279 			goto err;
280 		}
281 
282 		group->id = i;
283 		group->idxd = idxd;
284 		device_initialize(&group->conf_dev);
285 		group->conf_dev.parent = &idxd->conf_dev;
286 		group->conf_dev.bus = &dsa_bus_type;
287 		group->conf_dev.type = &idxd_group_device_type;
288 		rc = dev_set_name(&group->conf_dev, "group%d.%d", idxd->id, group->id);
289 		if (rc < 0) {
290 			put_device(&group->conf_dev);
291 			goto err;
292 		}
293 
294 		idxd->groups[i] = group;
295 		group->tc_a = -1;
296 		group->tc_b = -1;
297 	}
298 
299 	return 0;
300 
301  err:
302 	while (--i >= 0)
303 		put_device(&idxd->groups[i]->conf_dev);
304 	return rc;
305 }
306 
307 static int idxd_setup_internals(struct idxd_device *idxd)
308 {
309 	struct device *dev = &idxd->pdev->dev;
310 	int rc, i;
311 
312 	init_waitqueue_head(&idxd->cmd_waitq);
313 
314 	if (idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE)) {
315 		idxd->int_handles = devm_kcalloc(dev, idxd->max_wqs, sizeof(int), GFP_KERNEL);
316 		if (!idxd->int_handles)
317 			return -ENOMEM;
318 	}
319 
320 	rc = idxd_setup_wqs(idxd);
321 	if (rc < 0)
322 		goto err_wqs;
323 
324 	rc = idxd_setup_engines(idxd);
325 	if (rc < 0)
326 		goto err_engine;
327 
328 	rc = idxd_setup_groups(idxd);
329 	if (rc < 0)
330 		goto err_group;
331 
332 	idxd->wq = create_workqueue(dev_name(dev));
333 	if (!idxd->wq) {
334 		rc = -ENOMEM;
335 		goto err_wkq_create;
336 	}
337 
338 	return 0;
339 
340  err_wkq_create:
341 	for (i = 0; i < idxd->max_groups; i++)
342 		put_device(&idxd->groups[i]->conf_dev);
343  err_group:
344 	for (i = 0; i < idxd->max_engines; i++)
345 		put_device(&idxd->engines[i]->conf_dev);
346  err_engine:
347 	for (i = 0; i < idxd->max_wqs; i++)
348 		put_device(&idxd->wqs[i]->conf_dev);
349  err_wqs:
350 	kfree(idxd->int_handles);
351 	return rc;
352 }
353 
354 static void idxd_read_table_offsets(struct idxd_device *idxd)
355 {
356 	union offsets_reg offsets;
357 	struct device *dev = &idxd->pdev->dev;
358 
359 	offsets.bits[0] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET);
360 	offsets.bits[1] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET + sizeof(u64));
361 	idxd->grpcfg_offset = offsets.grpcfg * IDXD_TABLE_MULT;
362 	dev_dbg(dev, "IDXD Group Config Offset: %#x\n", idxd->grpcfg_offset);
363 	idxd->wqcfg_offset = offsets.wqcfg * IDXD_TABLE_MULT;
364 	dev_dbg(dev, "IDXD Work Queue Config Offset: %#x\n", idxd->wqcfg_offset);
365 	idxd->msix_perm_offset = offsets.msix_perm * IDXD_TABLE_MULT;
366 	dev_dbg(dev, "IDXD MSIX Permission Offset: %#x\n", idxd->msix_perm_offset);
367 	idxd->perfmon_offset = offsets.perfmon * IDXD_TABLE_MULT;
368 	dev_dbg(dev, "IDXD Perfmon Offset: %#x\n", idxd->perfmon_offset);
369 }
370 
371 static void idxd_read_caps(struct idxd_device *idxd)
372 {
373 	struct device *dev = &idxd->pdev->dev;
374 	int i;
375 
376 	/* reading generic capabilities */
377 	idxd->hw.gen_cap.bits = ioread64(idxd->reg_base + IDXD_GENCAP_OFFSET);
378 	dev_dbg(dev, "gen_cap: %#llx\n", idxd->hw.gen_cap.bits);
379 
380 	if (idxd->hw.gen_cap.cmd_cap) {
381 		idxd->hw.cmd_cap = ioread32(idxd->reg_base + IDXD_CMDCAP_OFFSET);
382 		dev_dbg(dev, "cmd_cap: %#x\n", idxd->hw.cmd_cap);
383 	}
384 
385 	idxd->max_xfer_bytes = 1ULL << idxd->hw.gen_cap.max_xfer_shift;
386 	dev_dbg(dev, "max xfer size: %llu bytes\n", idxd->max_xfer_bytes);
387 	idxd->max_batch_size = 1U << idxd->hw.gen_cap.max_batch_shift;
388 	dev_dbg(dev, "max batch size: %u\n", idxd->max_batch_size);
389 	if (idxd->hw.gen_cap.config_en)
390 		set_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags);
391 
392 	/* reading group capabilities */
393 	idxd->hw.group_cap.bits =
394 		ioread64(idxd->reg_base + IDXD_GRPCAP_OFFSET);
395 	dev_dbg(dev, "group_cap: %#llx\n", idxd->hw.group_cap.bits);
396 	idxd->max_groups = idxd->hw.group_cap.num_groups;
397 	dev_dbg(dev, "max groups: %u\n", idxd->max_groups);
398 	idxd->max_tokens = idxd->hw.group_cap.total_tokens;
399 	dev_dbg(dev, "max tokens: %u\n", idxd->max_tokens);
400 	idxd->nr_tokens = idxd->max_tokens;
401 
402 	/* read engine capabilities */
403 	idxd->hw.engine_cap.bits =
404 		ioread64(idxd->reg_base + IDXD_ENGCAP_OFFSET);
405 	dev_dbg(dev, "engine_cap: %#llx\n", idxd->hw.engine_cap.bits);
406 	idxd->max_engines = idxd->hw.engine_cap.num_engines;
407 	dev_dbg(dev, "max engines: %u\n", idxd->max_engines);
408 
409 	/* read workqueue capabilities */
410 	idxd->hw.wq_cap.bits = ioread64(idxd->reg_base + IDXD_WQCAP_OFFSET);
411 	dev_dbg(dev, "wq_cap: %#llx\n", idxd->hw.wq_cap.bits);
412 	idxd->max_wq_size = idxd->hw.wq_cap.total_wq_size;
413 	dev_dbg(dev, "total workqueue size: %u\n", idxd->max_wq_size);
414 	idxd->max_wqs = idxd->hw.wq_cap.num_wqs;
415 	dev_dbg(dev, "max workqueues: %u\n", idxd->max_wqs);
416 	idxd->wqcfg_size = 1 << (idxd->hw.wq_cap.wqcfg_size + IDXD_WQCFG_MIN);
417 	dev_dbg(dev, "wqcfg size: %u\n", idxd->wqcfg_size);
418 
419 	/* reading operation capabilities */
420 	for (i = 0; i < 4; i++) {
421 		idxd->hw.opcap.bits[i] = ioread64(idxd->reg_base +
422 				IDXD_OPCAP_OFFSET + i * sizeof(u64));
423 		dev_dbg(dev, "opcap[%d]: %#llx\n", i, idxd->hw.opcap.bits[i]);
424 	}
425 }
426 
427 static struct idxd_device *idxd_alloc(struct pci_dev *pdev, struct idxd_driver_data *data)
428 {
429 	struct device *dev = &pdev->dev;
430 	struct idxd_device *idxd;
431 	int rc;
432 
433 	idxd = kzalloc_node(sizeof(*idxd), GFP_KERNEL, dev_to_node(dev));
434 	if (!idxd)
435 		return NULL;
436 
437 	idxd->pdev = pdev;
438 	idxd->data = data;
439 	idxd->id = ida_alloc(&idxd_ida, GFP_KERNEL);
440 	if (idxd->id < 0)
441 		return NULL;
442 
443 	device_initialize(&idxd->conf_dev);
444 	idxd->conf_dev.parent = dev;
445 	idxd->conf_dev.bus = &dsa_bus_type;
446 	idxd->conf_dev.type = idxd->data->dev_type;
447 	rc = dev_set_name(&idxd->conf_dev, "%s%d", idxd->data->name_prefix, idxd->id);
448 	if (rc < 0) {
449 		put_device(&idxd->conf_dev);
450 		return NULL;
451 	}
452 
453 	spin_lock_init(&idxd->dev_lock);
454 	spin_lock_init(&idxd->cmd_lock);
455 
456 	return idxd;
457 }
458 
459 static int idxd_enable_system_pasid(struct idxd_device *idxd)
460 {
461 	int flags;
462 	unsigned int pasid;
463 	struct iommu_sva *sva;
464 
465 	flags = SVM_FLAG_SUPERVISOR_MODE;
466 
467 	sva = iommu_sva_bind_device(&idxd->pdev->dev, NULL, &flags);
468 	if (IS_ERR(sva)) {
469 		dev_warn(&idxd->pdev->dev,
470 			 "iommu sva bind failed: %ld\n", PTR_ERR(sva));
471 		return PTR_ERR(sva);
472 	}
473 
474 	pasid = iommu_sva_get_pasid(sva);
475 	if (pasid == IOMMU_PASID_INVALID) {
476 		iommu_sva_unbind_device(sva);
477 		return -ENODEV;
478 	}
479 
480 	idxd->sva = sva;
481 	idxd->pasid = pasid;
482 	dev_dbg(&idxd->pdev->dev, "system pasid: %u\n", pasid);
483 	return 0;
484 }
485 
486 static void idxd_disable_system_pasid(struct idxd_device *idxd)
487 {
488 
489 	iommu_sva_unbind_device(idxd->sva);
490 	idxd->sva = NULL;
491 }
492 
493 static int idxd_probe(struct idxd_device *idxd)
494 {
495 	struct pci_dev *pdev = idxd->pdev;
496 	struct device *dev = &pdev->dev;
497 	int rc;
498 
499 	dev_dbg(dev, "%s entered and resetting device\n", __func__);
500 	rc = idxd_device_init_reset(idxd);
501 	if (rc < 0)
502 		return rc;
503 
504 	dev_dbg(dev, "IDXD reset complete\n");
505 
506 	if (IS_ENABLED(CONFIG_INTEL_IDXD_SVM) && sva) {
507 		rc = iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA);
508 		if (rc == 0) {
509 			rc = idxd_enable_system_pasid(idxd);
510 			if (rc < 0) {
511 				iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
512 				dev_warn(dev, "Failed to enable PASID. No SVA support: %d\n", rc);
513 			} else {
514 				set_bit(IDXD_FLAG_PASID_ENABLED, &idxd->flags);
515 			}
516 		} else {
517 			dev_warn(dev, "Unable to turn on SVA feature.\n");
518 		}
519 	} else if (!sva) {
520 		dev_warn(dev, "User forced SVA off via module param.\n");
521 	}
522 
523 	idxd_read_caps(idxd);
524 	idxd_read_table_offsets(idxd);
525 
526 	rc = idxd_setup_internals(idxd);
527 	if (rc)
528 		goto err;
529 
530 	/* If the configs are readonly, then load them from device */
531 	if (!test_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags)) {
532 		dev_dbg(dev, "Loading RO device config\n");
533 		rc = idxd_device_load_config(idxd);
534 		if (rc < 0)
535 			goto err;
536 	}
537 
538 	rc = idxd_setup_interrupts(idxd);
539 	if (rc)
540 		goto err;
541 
542 	dev_dbg(dev, "IDXD interrupt setup complete.\n");
543 
544 	idxd->major = idxd_cdev_get_major(idxd);
545 
546 	rc = perfmon_pmu_init(idxd);
547 	if (rc < 0)
548 		dev_warn(dev, "Failed to initialize perfmon. No PMU support: %d\n", rc);
549 
550 	dev_dbg(dev, "IDXD device %d probed successfully\n", idxd->id);
551 	return 0;
552 
553  err:
554 	if (device_pasid_enabled(idxd))
555 		idxd_disable_system_pasid(idxd);
556 	iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
557 	return rc;
558 }
559 
560 static int idxd_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
561 {
562 	struct device *dev = &pdev->dev;
563 	struct idxd_device *idxd;
564 	struct idxd_driver_data *data = (struct idxd_driver_data *)id->driver_data;
565 	int rc;
566 
567 	rc = pci_enable_device(pdev);
568 	if (rc)
569 		return rc;
570 
571 	dev_dbg(dev, "Alloc IDXD context\n");
572 	idxd = idxd_alloc(pdev, data);
573 	if (!idxd) {
574 		rc = -ENOMEM;
575 		goto err_idxd_alloc;
576 	}
577 
578 	dev_dbg(dev, "Mapping BARs\n");
579 	idxd->reg_base = pci_iomap(pdev, IDXD_MMIO_BAR, 0);
580 	if (!idxd->reg_base) {
581 		rc = -ENOMEM;
582 		goto err_iomap;
583 	}
584 
585 	dev_dbg(dev, "Set DMA masks\n");
586 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
587 	if (rc)
588 		rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
589 	if (rc)
590 		goto err;
591 
592 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
593 	if (rc)
594 		rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
595 	if (rc)
596 		goto err;
597 
598 	dev_dbg(dev, "Set PCI master\n");
599 	pci_set_master(pdev);
600 	pci_set_drvdata(pdev, idxd);
601 
602 	idxd->hw.version = ioread32(idxd->reg_base + IDXD_VER_OFFSET);
603 	rc = idxd_probe(idxd);
604 	if (rc) {
605 		dev_err(dev, "Intel(R) IDXD DMA Engine init failed\n");
606 		goto err;
607 	}
608 
609 	rc = idxd_register_devices(idxd);
610 	if (rc) {
611 		dev_err(dev, "IDXD sysfs setup failed\n");
612 		goto err;
613 	}
614 
615 	idxd->state = IDXD_DEV_CONF_READY;
616 
617 	dev_info(&pdev->dev, "Intel(R) Accelerator Device (v%x)\n",
618 		 idxd->hw.version);
619 
620 	return 0;
621 
622  err:
623 	pci_iounmap(pdev, idxd->reg_base);
624  err_iomap:
625 	put_device(&idxd->conf_dev);
626  err_idxd_alloc:
627 	pci_disable_device(pdev);
628 	return rc;
629 }
630 
631 static void idxd_flush_pending_llist(struct idxd_irq_entry *ie)
632 {
633 	struct idxd_desc *desc, *itr;
634 	struct llist_node *head;
635 
636 	head = llist_del_all(&ie->pending_llist);
637 	if (!head)
638 		return;
639 
640 	llist_for_each_entry_safe(desc, itr, head, llnode) {
641 		idxd_dma_complete_txd(desc, IDXD_COMPLETE_ABORT);
642 		idxd_free_desc(desc->wq, desc);
643 	}
644 }
645 
646 static void idxd_flush_work_list(struct idxd_irq_entry *ie)
647 {
648 	struct idxd_desc *desc, *iter;
649 
650 	list_for_each_entry_safe(desc, iter, &ie->work_list, list) {
651 		list_del(&desc->list);
652 		idxd_dma_complete_txd(desc, IDXD_COMPLETE_ABORT);
653 		idxd_free_desc(desc->wq, desc);
654 	}
655 }
656 
657 void idxd_wqs_quiesce(struct idxd_device *idxd)
658 {
659 	struct idxd_wq *wq;
660 	int i;
661 
662 	for (i = 0; i < idxd->max_wqs; i++) {
663 		wq = idxd->wqs[i];
664 		if (wq->state == IDXD_WQ_ENABLED && wq->type == IDXD_WQT_KERNEL)
665 			idxd_wq_quiesce(wq);
666 	}
667 }
668 
669 static void idxd_release_int_handles(struct idxd_device *idxd)
670 {
671 	struct device *dev = &idxd->pdev->dev;
672 	int i, rc;
673 
674 	for (i = 0; i < idxd->num_wq_irqs; i++) {
675 		if (idxd->hw.cmd_cap & BIT(IDXD_CMD_RELEASE_INT_HANDLE)) {
676 			rc = idxd_device_release_int_handle(idxd, idxd->int_handles[i],
677 							    IDXD_IRQ_MSIX);
678 			if (rc < 0)
679 				dev_warn(dev, "irq handle %d release failed\n",
680 					 idxd->int_handles[i]);
681 			else
682 				dev_dbg(dev, "int handle requested: %u\n", idxd->int_handles[i]);
683 		}
684 	}
685 }
686 
687 static void idxd_shutdown(struct pci_dev *pdev)
688 {
689 	struct idxd_device *idxd = pci_get_drvdata(pdev);
690 	int rc, i;
691 	struct idxd_irq_entry *irq_entry;
692 	int msixcnt = pci_msix_vec_count(pdev);
693 
694 	rc = idxd_device_disable(idxd);
695 	if (rc)
696 		dev_err(&pdev->dev, "Disabling device failed\n");
697 
698 	dev_dbg(&pdev->dev, "%s called\n", __func__);
699 	idxd_mask_msix_vectors(idxd);
700 	idxd_mask_error_interrupts(idxd);
701 
702 	for (i = 0; i < msixcnt; i++) {
703 		irq_entry = &idxd->irq_entries[i];
704 		synchronize_irq(irq_entry->vector);
705 		free_irq(irq_entry->vector, irq_entry);
706 		if (i == 0)
707 			continue;
708 		idxd_flush_pending_llist(irq_entry);
709 		idxd_flush_work_list(irq_entry);
710 	}
711 
712 	idxd_msix_perm_clear(idxd);
713 	idxd_release_int_handles(idxd);
714 	pci_free_irq_vectors(pdev);
715 	pci_iounmap(pdev, idxd->reg_base);
716 	pci_disable_device(pdev);
717 	destroy_workqueue(idxd->wq);
718 }
719 
720 static void idxd_remove(struct pci_dev *pdev)
721 {
722 	struct idxd_device *idxd = pci_get_drvdata(pdev);
723 
724 	dev_dbg(&pdev->dev, "%s called\n", __func__);
725 	idxd_shutdown(pdev);
726 	if (device_pasid_enabled(idxd))
727 		idxd_disable_system_pasid(idxd);
728 	idxd_unregister_devices(idxd);
729 	perfmon_pmu_remove(idxd);
730 	iommu_dev_disable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA);
731 }
732 
733 static struct pci_driver idxd_pci_driver = {
734 	.name		= DRV_NAME,
735 	.id_table	= idxd_pci_tbl,
736 	.probe		= idxd_pci_probe,
737 	.remove		= idxd_remove,
738 	.shutdown	= idxd_shutdown,
739 };
740 
741 static int __init idxd_init_module(void)
742 {
743 	int err;
744 
745 	/*
746 	 * If the CPU does not support MOVDIR64B or ENQCMDS, there's no point in
747 	 * enumerating the device. We can not utilize it.
748 	 */
749 	if (!boot_cpu_has(X86_FEATURE_MOVDIR64B)) {
750 		pr_warn("idxd driver failed to load without MOVDIR64B.\n");
751 		return -ENODEV;
752 	}
753 
754 	if (!boot_cpu_has(X86_FEATURE_ENQCMD))
755 		pr_warn("Platform does not have ENQCMD(S) support.\n");
756 	else
757 		support_enqcmd = true;
758 
759 	perfmon_init();
760 
761 	err = idxd_register_bus_type();
762 	if (err < 0)
763 		return err;
764 
765 	err = idxd_register_driver();
766 	if (err < 0)
767 		goto err_idxd_driver_register;
768 
769 	err = idxd_cdev_register();
770 	if (err)
771 		goto err_cdev_register;
772 
773 	err = pci_register_driver(&idxd_pci_driver);
774 	if (err)
775 		goto err_pci_register;
776 
777 	return 0;
778 
779 err_pci_register:
780 	idxd_cdev_remove();
781 err_cdev_register:
782 	idxd_unregister_driver();
783 err_idxd_driver_register:
784 	idxd_unregister_bus_type();
785 	return err;
786 }
787 module_init(idxd_init_module);
788 
789 static void __exit idxd_exit_module(void)
790 {
791 	pci_unregister_driver(&idxd_pci_driver);
792 	idxd_cdev_remove();
793 	idxd_unregister_bus_type();
794 	perfmon_exit();
795 }
796 module_exit(idxd_exit_module);
797