xref: /openbmc/linux/drivers/dma/idxd/init.c (revision e368cd72)
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 	idxd_msix_perm_setup(idxd);
106 
107 	irq_entry = &idxd->irq_entries[0];
108 	rc = request_threaded_irq(irq_entry->vector, NULL, idxd_misc_thread,
109 				  0, "idxd-misc", irq_entry);
110 	if (rc < 0) {
111 		dev_err(dev, "Failed to allocate misc interrupt.\n");
112 		goto err_misc_irq;
113 	}
114 
115 	dev_dbg(dev, "Allocated idxd-misc handler on msix vector %d\n", irq_entry->vector);
116 
117 	/* first MSI-X entry is not for wq interrupts */
118 	idxd->num_wq_irqs = msixcnt - 1;
119 
120 	for (i = 1; i < msixcnt; i++) {
121 		irq_entry = &idxd->irq_entries[i];
122 
123 		init_llist_head(&idxd->irq_entries[i].pending_llist);
124 		INIT_LIST_HEAD(&idxd->irq_entries[i].work_list);
125 		rc = request_threaded_irq(irq_entry->vector, NULL,
126 					  idxd_wq_thread, 0, "idxd-portal", irq_entry);
127 		if (rc < 0) {
128 			dev_err(dev, "Failed to allocate irq %d.\n", irq_entry->vector);
129 			goto err_wq_irqs;
130 		}
131 
132 		dev_dbg(dev, "Allocated idxd-msix %d for vector %d\n", i, irq_entry->vector);
133 		if (idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE)) {
134 			/*
135 			 * The MSIX vector enumeration starts at 1 with vector 0 being the
136 			 * misc interrupt that handles non I/O completion events. The
137 			 * interrupt handles are for IMS enumeration on guest. The misc
138 			 * interrupt vector does not require a handle and therefore we start
139 			 * the int_handles at index 0. Since 'i' starts at 1, the first
140 			 * int_handles index will be 0.
141 			 */
142 			rc = idxd_device_request_int_handle(idxd, i, &idxd->int_handles[i - 1],
143 							    IDXD_IRQ_MSIX);
144 			if (rc < 0) {
145 				free_irq(irq_entry->vector, irq_entry);
146 				goto err_wq_irqs;
147 			}
148 			dev_dbg(dev, "int handle requested: %u\n", idxd->int_handles[i - 1]);
149 		}
150 	}
151 
152 	idxd_unmask_error_interrupts(idxd);
153 	return 0;
154 
155  err_wq_irqs:
156 	while (--i >= 0) {
157 		irq_entry = &idxd->irq_entries[i];
158 		free_irq(irq_entry->vector, irq_entry);
159 		if (i != 0)
160 			idxd_device_release_int_handle(idxd,
161 						       idxd->int_handles[i], IDXD_IRQ_MSIX);
162 	}
163  err_misc_irq:
164 	/* Disable error interrupt generation */
165 	idxd_mask_error_interrupts(idxd);
166 	idxd_msix_perm_clear(idxd);
167  err_irq_entries:
168 	pci_free_irq_vectors(pdev);
169 	dev_err(dev, "No usable interrupts\n");
170 	return rc;
171 }
172 
173 static void idxd_cleanup_interrupts(struct idxd_device *idxd)
174 {
175 	struct pci_dev *pdev = idxd->pdev;
176 	struct idxd_irq_entry *irq_entry;
177 	int i, msixcnt;
178 
179 	msixcnt = pci_msix_vec_count(pdev);
180 	if (msixcnt <= 0)
181 		return;
182 
183 	irq_entry = &idxd->irq_entries[0];
184 	free_irq(irq_entry->vector, irq_entry);
185 
186 	for (i = 1; i < msixcnt; i++) {
187 
188 		irq_entry = &idxd->irq_entries[i];
189 		if (idxd->hw.cmd_cap & BIT(IDXD_CMD_RELEASE_INT_HANDLE))
190 			idxd_device_release_int_handle(idxd, idxd->int_handles[i],
191 						       IDXD_IRQ_MSIX);
192 		free_irq(irq_entry->vector, irq_entry);
193 	}
194 
195 	idxd_mask_error_interrupts(idxd);
196 	pci_free_irq_vectors(pdev);
197 }
198 
199 static int idxd_setup_wqs(struct idxd_device *idxd)
200 {
201 	struct device *dev = &idxd->pdev->dev;
202 	struct idxd_wq *wq;
203 	int i, rc;
204 
205 	idxd->wqs = kcalloc_node(idxd->max_wqs, sizeof(struct idxd_wq *),
206 				 GFP_KERNEL, dev_to_node(dev));
207 	if (!idxd->wqs)
208 		return -ENOMEM;
209 
210 	for (i = 0; i < idxd->max_wqs; i++) {
211 		wq = kzalloc_node(sizeof(*wq), GFP_KERNEL, dev_to_node(dev));
212 		if (!wq) {
213 			rc = -ENOMEM;
214 			goto err;
215 		}
216 
217 		wq->id = i;
218 		wq->idxd = idxd;
219 		device_initialize(&wq->conf_dev);
220 		wq->conf_dev.parent = &idxd->conf_dev;
221 		wq->conf_dev.bus = &dsa_bus_type;
222 		wq->conf_dev.type = &idxd_wq_device_type;
223 		rc = dev_set_name(&wq->conf_dev, "wq%d.%d", idxd->id, wq->id);
224 		if (rc < 0) {
225 			put_device(&wq->conf_dev);
226 			goto err;
227 		}
228 
229 		mutex_init(&wq->wq_lock);
230 		init_waitqueue_head(&wq->err_queue);
231 		init_completion(&wq->wq_dead);
232 		wq->max_xfer_bytes = idxd->max_xfer_bytes;
233 		wq->max_batch_size = idxd->max_batch_size;
234 		wq->wqcfg = kzalloc_node(idxd->wqcfg_size, GFP_KERNEL, dev_to_node(dev));
235 		if (!wq->wqcfg) {
236 			put_device(&wq->conf_dev);
237 			rc = -ENOMEM;
238 			goto err;
239 		}
240 		idxd->wqs[i] = wq;
241 	}
242 
243 	return 0;
244 
245  err:
246 	while (--i >= 0)
247 		put_device(&idxd->wqs[i]->conf_dev);
248 	return rc;
249 }
250 
251 static int idxd_setup_engines(struct idxd_device *idxd)
252 {
253 	struct idxd_engine *engine;
254 	struct device *dev = &idxd->pdev->dev;
255 	int i, rc;
256 
257 	idxd->engines = kcalloc_node(idxd->max_engines, sizeof(struct idxd_engine *),
258 				     GFP_KERNEL, dev_to_node(dev));
259 	if (!idxd->engines)
260 		return -ENOMEM;
261 
262 	for (i = 0; i < idxd->max_engines; i++) {
263 		engine = kzalloc_node(sizeof(*engine), GFP_KERNEL, dev_to_node(dev));
264 		if (!engine) {
265 			rc = -ENOMEM;
266 			goto err;
267 		}
268 
269 		engine->id = i;
270 		engine->idxd = idxd;
271 		device_initialize(&engine->conf_dev);
272 		engine->conf_dev.parent = &idxd->conf_dev;
273 		engine->conf_dev.bus = &dsa_bus_type;
274 		engine->conf_dev.type = &idxd_engine_device_type;
275 		rc = dev_set_name(&engine->conf_dev, "engine%d.%d", idxd->id, engine->id);
276 		if (rc < 0) {
277 			put_device(&engine->conf_dev);
278 			goto err;
279 		}
280 
281 		idxd->engines[i] = engine;
282 	}
283 
284 	return 0;
285 
286  err:
287 	while (--i >= 0)
288 		put_device(&idxd->engines[i]->conf_dev);
289 	return rc;
290 }
291 
292 static int idxd_setup_groups(struct idxd_device *idxd)
293 {
294 	struct device *dev = &idxd->pdev->dev;
295 	struct idxd_group *group;
296 	int i, rc;
297 
298 	idxd->groups = kcalloc_node(idxd->max_groups, sizeof(struct idxd_group *),
299 				    GFP_KERNEL, dev_to_node(dev));
300 	if (!idxd->groups)
301 		return -ENOMEM;
302 
303 	for (i = 0; i < idxd->max_groups; i++) {
304 		group = kzalloc_node(sizeof(*group), GFP_KERNEL, dev_to_node(dev));
305 		if (!group) {
306 			rc = -ENOMEM;
307 			goto err;
308 		}
309 
310 		group->id = i;
311 		group->idxd = idxd;
312 		device_initialize(&group->conf_dev);
313 		group->conf_dev.parent = &idxd->conf_dev;
314 		group->conf_dev.bus = &dsa_bus_type;
315 		group->conf_dev.type = &idxd_group_device_type;
316 		rc = dev_set_name(&group->conf_dev, "group%d.%d", idxd->id, group->id);
317 		if (rc < 0) {
318 			put_device(&group->conf_dev);
319 			goto err;
320 		}
321 
322 		idxd->groups[i] = group;
323 		group->tc_a = -1;
324 		group->tc_b = -1;
325 	}
326 
327 	return 0;
328 
329  err:
330 	while (--i >= 0)
331 		put_device(&idxd->groups[i]->conf_dev);
332 	return rc;
333 }
334 
335 static void idxd_cleanup_internals(struct idxd_device *idxd)
336 {
337 	int i;
338 
339 	for (i = 0; i < idxd->max_groups; i++)
340 		put_device(&idxd->groups[i]->conf_dev);
341 	for (i = 0; i < idxd->max_engines; i++)
342 		put_device(&idxd->engines[i]->conf_dev);
343 	for (i = 0; i < idxd->max_wqs; i++)
344 		put_device(&idxd->wqs[i]->conf_dev);
345 	destroy_workqueue(idxd->wq);
346 }
347 
348 static int idxd_setup_internals(struct idxd_device *idxd)
349 {
350 	struct device *dev = &idxd->pdev->dev;
351 	int rc, i;
352 
353 	init_waitqueue_head(&idxd->cmd_waitq);
354 
355 	if (idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE)) {
356 		idxd->int_handles = kcalloc_node(idxd->max_wqs, sizeof(int), GFP_KERNEL,
357 						 dev_to_node(dev));
358 		if (!idxd->int_handles)
359 			return -ENOMEM;
360 	}
361 
362 	rc = idxd_setup_wqs(idxd);
363 	if (rc < 0)
364 		goto err_wqs;
365 
366 	rc = idxd_setup_engines(idxd);
367 	if (rc < 0)
368 		goto err_engine;
369 
370 	rc = idxd_setup_groups(idxd);
371 	if (rc < 0)
372 		goto err_group;
373 
374 	idxd->wq = create_workqueue(dev_name(dev));
375 	if (!idxd->wq) {
376 		rc = -ENOMEM;
377 		goto err_wkq_create;
378 	}
379 
380 	return 0;
381 
382  err_wkq_create:
383 	for (i = 0; i < idxd->max_groups; i++)
384 		put_device(&idxd->groups[i]->conf_dev);
385  err_group:
386 	for (i = 0; i < idxd->max_engines; i++)
387 		put_device(&idxd->engines[i]->conf_dev);
388  err_engine:
389 	for (i = 0; i < idxd->max_wqs; i++)
390 		put_device(&idxd->wqs[i]->conf_dev);
391  err_wqs:
392 	kfree(idxd->int_handles);
393 	return rc;
394 }
395 
396 static void idxd_read_table_offsets(struct idxd_device *idxd)
397 {
398 	union offsets_reg offsets;
399 	struct device *dev = &idxd->pdev->dev;
400 
401 	offsets.bits[0] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET);
402 	offsets.bits[1] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET + sizeof(u64));
403 	idxd->grpcfg_offset = offsets.grpcfg * IDXD_TABLE_MULT;
404 	dev_dbg(dev, "IDXD Group Config Offset: %#x\n", idxd->grpcfg_offset);
405 	idxd->wqcfg_offset = offsets.wqcfg * IDXD_TABLE_MULT;
406 	dev_dbg(dev, "IDXD Work Queue Config Offset: %#x\n", idxd->wqcfg_offset);
407 	idxd->msix_perm_offset = offsets.msix_perm * IDXD_TABLE_MULT;
408 	dev_dbg(dev, "IDXD MSIX Permission Offset: %#x\n", idxd->msix_perm_offset);
409 	idxd->perfmon_offset = offsets.perfmon * IDXD_TABLE_MULT;
410 	dev_dbg(dev, "IDXD Perfmon Offset: %#x\n", idxd->perfmon_offset);
411 }
412 
413 static void idxd_read_caps(struct idxd_device *idxd)
414 {
415 	struct device *dev = &idxd->pdev->dev;
416 	int i;
417 
418 	/* reading generic capabilities */
419 	idxd->hw.gen_cap.bits = ioread64(idxd->reg_base + IDXD_GENCAP_OFFSET);
420 	dev_dbg(dev, "gen_cap: %#llx\n", idxd->hw.gen_cap.bits);
421 
422 	if (idxd->hw.gen_cap.cmd_cap) {
423 		idxd->hw.cmd_cap = ioread32(idxd->reg_base + IDXD_CMDCAP_OFFSET);
424 		dev_dbg(dev, "cmd_cap: %#x\n", idxd->hw.cmd_cap);
425 	}
426 
427 	idxd->max_xfer_bytes = 1ULL << idxd->hw.gen_cap.max_xfer_shift;
428 	dev_dbg(dev, "max xfer size: %llu bytes\n", idxd->max_xfer_bytes);
429 	idxd->max_batch_size = 1U << idxd->hw.gen_cap.max_batch_shift;
430 	dev_dbg(dev, "max batch size: %u\n", idxd->max_batch_size);
431 	if (idxd->hw.gen_cap.config_en)
432 		set_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags);
433 
434 	/* reading group capabilities */
435 	idxd->hw.group_cap.bits =
436 		ioread64(idxd->reg_base + IDXD_GRPCAP_OFFSET);
437 	dev_dbg(dev, "group_cap: %#llx\n", idxd->hw.group_cap.bits);
438 	idxd->max_groups = idxd->hw.group_cap.num_groups;
439 	dev_dbg(dev, "max groups: %u\n", idxd->max_groups);
440 	idxd->max_tokens = idxd->hw.group_cap.total_tokens;
441 	dev_dbg(dev, "max tokens: %u\n", idxd->max_tokens);
442 	idxd->nr_tokens = idxd->max_tokens;
443 
444 	/* read engine capabilities */
445 	idxd->hw.engine_cap.bits =
446 		ioread64(idxd->reg_base + IDXD_ENGCAP_OFFSET);
447 	dev_dbg(dev, "engine_cap: %#llx\n", idxd->hw.engine_cap.bits);
448 	idxd->max_engines = idxd->hw.engine_cap.num_engines;
449 	dev_dbg(dev, "max engines: %u\n", idxd->max_engines);
450 
451 	/* read workqueue capabilities */
452 	idxd->hw.wq_cap.bits = ioread64(idxd->reg_base + IDXD_WQCAP_OFFSET);
453 	dev_dbg(dev, "wq_cap: %#llx\n", idxd->hw.wq_cap.bits);
454 	idxd->max_wq_size = idxd->hw.wq_cap.total_wq_size;
455 	dev_dbg(dev, "total workqueue size: %u\n", idxd->max_wq_size);
456 	idxd->max_wqs = idxd->hw.wq_cap.num_wqs;
457 	dev_dbg(dev, "max workqueues: %u\n", idxd->max_wqs);
458 	idxd->wqcfg_size = 1 << (idxd->hw.wq_cap.wqcfg_size + IDXD_WQCFG_MIN);
459 	dev_dbg(dev, "wqcfg size: %u\n", idxd->wqcfg_size);
460 
461 	/* reading operation capabilities */
462 	for (i = 0; i < 4; i++) {
463 		idxd->hw.opcap.bits[i] = ioread64(idxd->reg_base +
464 				IDXD_OPCAP_OFFSET + i * sizeof(u64));
465 		dev_dbg(dev, "opcap[%d]: %#llx\n", i, idxd->hw.opcap.bits[i]);
466 	}
467 }
468 
469 static struct idxd_device *idxd_alloc(struct pci_dev *pdev, struct idxd_driver_data *data)
470 {
471 	struct device *dev = &pdev->dev;
472 	struct idxd_device *idxd;
473 	int rc;
474 
475 	idxd = kzalloc_node(sizeof(*idxd), GFP_KERNEL, dev_to_node(dev));
476 	if (!idxd)
477 		return NULL;
478 
479 	idxd->pdev = pdev;
480 	idxd->data = data;
481 	idxd->id = ida_alloc(&idxd_ida, GFP_KERNEL);
482 	if (idxd->id < 0)
483 		return NULL;
484 
485 	device_initialize(&idxd->conf_dev);
486 	idxd->conf_dev.parent = dev;
487 	idxd->conf_dev.bus = &dsa_bus_type;
488 	idxd->conf_dev.type = idxd->data->dev_type;
489 	rc = dev_set_name(&idxd->conf_dev, "%s%d", idxd->data->name_prefix, idxd->id);
490 	if (rc < 0) {
491 		put_device(&idxd->conf_dev);
492 		return NULL;
493 	}
494 
495 	spin_lock_init(&idxd->dev_lock);
496 	spin_lock_init(&idxd->cmd_lock);
497 
498 	return idxd;
499 }
500 
501 static int idxd_enable_system_pasid(struct idxd_device *idxd)
502 {
503 	int flags;
504 	unsigned int pasid;
505 	struct iommu_sva *sva;
506 
507 	flags = SVM_FLAG_SUPERVISOR_MODE;
508 
509 	sva = iommu_sva_bind_device(&idxd->pdev->dev, NULL, &flags);
510 	if (IS_ERR(sva)) {
511 		dev_warn(&idxd->pdev->dev,
512 			 "iommu sva bind failed: %ld\n", PTR_ERR(sva));
513 		return PTR_ERR(sva);
514 	}
515 
516 	pasid = iommu_sva_get_pasid(sva);
517 	if (pasid == IOMMU_PASID_INVALID) {
518 		iommu_sva_unbind_device(sva);
519 		return -ENODEV;
520 	}
521 
522 	idxd->sva = sva;
523 	idxd->pasid = pasid;
524 	dev_dbg(&idxd->pdev->dev, "system pasid: %u\n", pasid);
525 	return 0;
526 }
527 
528 static void idxd_disable_system_pasid(struct idxd_device *idxd)
529 {
530 
531 	iommu_sva_unbind_device(idxd->sva);
532 	idxd->sva = NULL;
533 }
534 
535 static int idxd_probe(struct idxd_device *idxd)
536 {
537 	struct pci_dev *pdev = idxd->pdev;
538 	struct device *dev = &pdev->dev;
539 	int rc;
540 
541 	dev_dbg(dev, "%s entered and resetting device\n", __func__);
542 	rc = idxd_device_init_reset(idxd);
543 	if (rc < 0)
544 		return rc;
545 
546 	dev_dbg(dev, "IDXD reset complete\n");
547 
548 	if (IS_ENABLED(CONFIG_INTEL_IDXD_SVM) && sva) {
549 		rc = iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA);
550 		if (rc == 0) {
551 			rc = idxd_enable_system_pasid(idxd);
552 			if (rc < 0) {
553 				iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
554 				dev_warn(dev, "Failed to enable PASID. No SVA support: %d\n", rc);
555 			} else {
556 				set_bit(IDXD_FLAG_PASID_ENABLED, &idxd->flags);
557 			}
558 		} else {
559 			dev_warn(dev, "Unable to turn on SVA feature.\n");
560 		}
561 	} else if (!sva) {
562 		dev_warn(dev, "User forced SVA off via module param.\n");
563 	}
564 
565 	idxd_read_caps(idxd);
566 	idxd_read_table_offsets(idxd);
567 
568 	rc = idxd_setup_internals(idxd);
569 	if (rc)
570 		goto err;
571 
572 	/* If the configs are readonly, then load them from device */
573 	if (!test_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags)) {
574 		dev_dbg(dev, "Loading RO device config\n");
575 		rc = idxd_device_load_config(idxd);
576 		if (rc < 0)
577 			goto err_config;
578 	}
579 
580 	rc = idxd_setup_interrupts(idxd);
581 	if (rc)
582 		goto err_config;
583 
584 	dev_dbg(dev, "IDXD interrupt setup complete.\n");
585 
586 	idxd->major = idxd_cdev_get_major(idxd);
587 
588 	rc = perfmon_pmu_init(idxd);
589 	if (rc < 0)
590 		dev_warn(dev, "Failed to initialize perfmon. No PMU support: %d\n", rc);
591 
592 	dev_dbg(dev, "IDXD device %d probed successfully\n", idxd->id);
593 	return 0;
594 
595  err_config:
596 	idxd_cleanup_internals(idxd);
597  err:
598 	if (device_pasid_enabled(idxd))
599 		idxd_disable_system_pasid(idxd);
600 	iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
601 	return rc;
602 }
603 
604 static void idxd_cleanup(struct idxd_device *idxd)
605 {
606 	struct device *dev = &idxd->pdev->dev;
607 
608 	perfmon_pmu_remove(idxd);
609 	idxd_cleanup_interrupts(idxd);
610 	idxd_cleanup_internals(idxd);
611 	if (device_pasid_enabled(idxd))
612 		idxd_disable_system_pasid(idxd);
613 	iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
614 }
615 
616 static int idxd_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
617 {
618 	struct device *dev = &pdev->dev;
619 	struct idxd_device *idxd;
620 	struct idxd_driver_data *data = (struct idxd_driver_data *)id->driver_data;
621 	int rc;
622 
623 	rc = pci_enable_device(pdev);
624 	if (rc)
625 		return rc;
626 
627 	dev_dbg(dev, "Alloc IDXD context\n");
628 	idxd = idxd_alloc(pdev, data);
629 	if (!idxd) {
630 		rc = -ENOMEM;
631 		goto err_idxd_alloc;
632 	}
633 
634 	dev_dbg(dev, "Mapping BARs\n");
635 	idxd->reg_base = pci_iomap(pdev, IDXD_MMIO_BAR, 0);
636 	if (!idxd->reg_base) {
637 		rc = -ENOMEM;
638 		goto err_iomap;
639 	}
640 
641 	dev_dbg(dev, "Set DMA masks\n");
642 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
643 	if (rc)
644 		rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
645 	if (rc)
646 		goto err;
647 
648 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
649 	if (rc)
650 		rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
651 	if (rc)
652 		goto err;
653 
654 	dev_dbg(dev, "Set PCI master\n");
655 	pci_set_master(pdev);
656 	pci_set_drvdata(pdev, idxd);
657 
658 	idxd->hw.version = ioread32(idxd->reg_base + IDXD_VER_OFFSET);
659 	rc = idxd_probe(idxd);
660 	if (rc) {
661 		dev_err(dev, "Intel(R) IDXD DMA Engine init failed\n");
662 		goto err;
663 	}
664 
665 	rc = idxd_register_devices(idxd);
666 	if (rc) {
667 		dev_err(dev, "IDXD sysfs setup failed\n");
668 		goto err_dev_register;
669 	}
670 
671 	idxd->state = IDXD_DEV_CONF_READY;
672 
673 	dev_info(&pdev->dev, "Intel(R) Accelerator Device (v%x)\n",
674 		 idxd->hw.version);
675 
676 	return 0;
677 
678  err_dev_register:
679 	idxd_cleanup(idxd);
680  err:
681 	pci_iounmap(pdev, idxd->reg_base);
682  err_iomap:
683 	put_device(&idxd->conf_dev);
684  err_idxd_alloc:
685 	pci_disable_device(pdev);
686 	return rc;
687 }
688 
689 static void idxd_flush_pending_llist(struct idxd_irq_entry *ie)
690 {
691 	struct idxd_desc *desc, *itr;
692 	struct llist_node *head;
693 
694 	head = llist_del_all(&ie->pending_llist);
695 	if (!head)
696 		return;
697 
698 	llist_for_each_entry_safe(desc, itr, head, llnode) {
699 		idxd_dma_complete_txd(desc, IDXD_COMPLETE_ABORT);
700 		idxd_free_desc(desc->wq, desc);
701 	}
702 }
703 
704 static void idxd_flush_work_list(struct idxd_irq_entry *ie)
705 {
706 	struct idxd_desc *desc, *iter;
707 
708 	list_for_each_entry_safe(desc, iter, &ie->work_list, list) {
709 		list_del(&desc->list);
710 		idxd_dma_complete_txd(desc, IDXD_COMPLETE_ABORT);
711 		idxd_free_desc(desc->wq, desc);
712 	}
713 }
714 
715 void idxd_wqs_quiesce(struct idxd_device *idxd)
716 {
717 	struct idxd_wq *wq;
718 	int i;
719 
720 	for (i = 0; i < idxd->max_wqs; i++) {
721 		wq = idxd->wqs[i];
722 		if (wq->state == IDXD_WQ_ENABLED && wq->type == IDXD_WQT_KERNEL)
723 			idxd_wq_quiesce(wq);
724 	}
725 }
726 
727 static void idxd_release_int_handles(struct idxd_device *idxd)
728 {
729 	struct device *dev = &idxd->pdev->dev;
730 	int i, rc;
731 
732 	for (i = 0; i < idxd->num_wq_irqs; i++) {
733 		if (idxd->hw.cmd_cap & BIT(IDXD_CMD_RELEASE_INT_HANDLE)) {
734 			rc = idxd_device_release_int_handle(idxd, idxd->int_handles[i],
735 							    IDXD_IRQ_MSIX);
736 			if (rc < 0)
737 				dev_warn(dev, "irq handle %d release failed\n",
738 					 idxd->int_handles[i]);
739 			else
740 				dev_dbg(dev, "int handle requested: %u\n", idxd->int_handles[i]);
741 		}
742 	}
743 }
744 
745 static void idxd_shutdown(struct pci_dev *pdev)
746 {
747 	struct idxd_device *idxd = pci_get_drvdata(pdev);
748 	int rc, i;
749 	struct idxd_irq_entry *irq_entry;
750 	int msixcnt = pci_msix_vec_count(pdev);
751 
752 	rc = idxd_device_disable(idxd);
753 	if (rc)
754 		dev_err(&pdev->dev, "Disabling device failed\n");
755 
756 	dev_dbg(&pdev->dev, "%s called\n", __func__);
757 	idxd_mask_msix_vectors(idxd);
758 	idxd_mask_error_interrupts(idxd);
759 
760 	for (i = 0; i < msixcnt; i++) {
761 		irq_entry = &idxd->irq_entries[i];
762 		synchronize_irq(irq_entry->vector);
763 		if (i == 0)
764 			continue;
765 		idxd_flush_pending_llist(irq_entry);
766 		idxd_flush_work_list(irq_entry);
767 	}
768 	flush_workqueue(idxd->wq);
769 }
770 
771 static void idxd_remove(struct pci_dev *pdev)
772 {
773 	struct idxd_device *idxd = pci_get_drvdata(pdev);
774 	struct idxd_irq_entry *irq_entry;
775 	int msixcnt = pci_msix_vec_count(pdev);
776 	int i;
777 
778 	dev_dbg(&pdev->dev, "%s called\n", __func__);
779 	idxd_shutdown(pdev);
780 	if (device_pasid_enabled(idxd))
781 		idxd_disable_system_pasid(idxd);
782 	idxd_unregister_devices(idxd);
783 
784 	for (i = 0; i < msixcnt; i++) {
785 		irq_entry = &idxd->irq_entries[i];
786 		free_irq(irq_entry->vector, irq_entry);
787 	}
788 	idxd_msix_perm_clear(idxd);
789 	idxd_release_int_handles(idxd);
790 	pci_free_irq_vectors(pdev);
791 	pci_iounmap(pdev, idxd->reg_base);
792 	iommu_dev_disable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA);
793 	pci_disable_device(pdev);
794 	destroy_workqueue(idxd->wq);
795 	perfmon_pmu_remove(idxd);
796 	device_unregister(&idxd->conf_dev);
797 }
798 
799 static struct pci_driver idxd_pci_driver = {
800 	.name		= DRV_NAME,
801 	.id_table	= idxd_pci_tbl,
802 	.probe		= idxd_pci_probe,
803 	.remove		= idxd_remove,
804 	.shutdown	= idxd_shutdown,
805 };
806 
807 static int __init idxd_init_module(void)
808 {
809 	int err;
810 
811 	/*
812 	 * If the CPU does not support MOVDIR64B or ENQCMDS, there's no point in
813 	 * enumerating the device. We can not utilize it.
814 	 */
815 	if (!cpu_feature_enabled(X86_FEATURE_MOVDIR64B)) {
816 		pr_warn("idxd driver failed to load without MOVDIR64B.\n");
817 		return -ENODEV;
818 	}
819 
820 	if (!cpu_feature_enabled(X86_FEATURE_ENQCMD))
821 		pr_warn("Platform does not have ENQCMD(S) support.\n");
822 	else
823 		support_enqcmd = true;
824 
825 	perfmon_init();
826 
827 	err = idxd_register_bus_type();
828 	if (err < 0)
829 		return err;
830 
831 	err = idxd_register_driver();
832 	if (err < 0)
833 		goto err_idxd_driver_register;
834 
835 	err = idxd_cdev_register();
836 	if (err)
837 		goto err_cdev_register;
838 
839 	err = pci_register_driver(&idxd_pci_driver);
840 	if (err)
841 		goto err_pci_register;
842 
843 	return 0;
844 
845 err_pci_register:
846 	idxd_cdev_remove();
847 err_cdev_register:
848 	idxd_unregister_driver();
849 err_idxd_driver_register:
850 	idxd_unregister_bus_type();
851 	return err;
852 }
853 module_init(idxd_init_module);
854 
855 static void __exit idxd_exit_module(void)
856 {
857 	idxd_unregister_driver();
858 	pci_unregister_driver(&idxd_pci_driver);
859 	idxd_cdev_remove();
860 	idxd_unregister_bus_type();
861 	perfmon_exit();
862 }
863 module_exit(idxd_exit_module);
864