xref: /openbmc/linux/drivers/dma/idxd/init.c (revision ac73d4bf)
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 MODULE_IMPORT_NS(IDXD);
30 
31 static bool sva = true;
32 module_param(sva, bool, 0644);
33 MODULE_PARM_DESC(sva, "Toggle SVA support on/off");
34 
35 bool tc_override;
36 module_param(tc_override, bool, 0644);
37 MODULE_PARM_DESC(tc_override, "Override traffic class defaults");
38 
39 #define DRV_NAME "idxd"
40 
41 bool support_enqcmd;
42 DEFINE_IDA(idxd_ida);
43 
44 static struct idxd_driver_data idxd_driver_data[] = {
45 	[IDXD_TYPE_DSA] = {
46 		.name_prefix = "dsa",
47 		.type = IDXD_TYPE_DSA,
48 		.compl_size = sizeof(struct dsa_completion_record),
49 		.align = 32,
50 		.dev_type = &dsa_device_type,
51 	},
52 	[IDXD_TYPE_IAX] = {
53 		.name_prefix = "iax",
54 		.type = IDXD_TYPE_IAX,
55 		.compl_size = sizeof(struct iax_completion_record),
56 		.align = 64,
57 		.dev_type = &iax_device_type,
58 	},
59 };
60 
61 static struct pci_device_id idxd_pci_tbl[] = {
62 	/* DSA ver 1.0 platforms */
63 	{ PCI_DEVICE_DATA(INTEL, DSA_SPR0, &idxd_driver_data[IDXD_TYPE_DSA]) },
64 
65 	/* IAX ver 1.0 platforms */
66 	{ PCI_DEVICE_DATA(INTEL, IAX_SPR0, &idxd_driver_data[IDXD_TYPE_IAX]) },
67 	{ 0, }
68 };
69 MODULE_DEVICE_TABLE(pci, idxd_pci_tbl);
70 
71 static int idxd_setup_interrupts(struct idxd_device *idxd)
72 {
73 	struct pci_dev *pdev = idxd->pdev;
74 	struct device *dev = &pdev->dev;
75 	struct idxd_irq_entry *ie;
76 	int i, msixcnt;
77 	int rc = 0;
78 
79 	msixcnt = pci_msix_vec_count(pdev);
80 	if (msixcnt < 0) {
81 		dev_err(dev, "Not MSI-X interrupt capable.\n");
82 		return -ENOSPC;
83 	}
84 	idxd->irq_cnt = msixcnt;
85 
86 	rc = pci_alloc_irq_vectors(pdev, msixcnt, msixcnt, PCI_IRQ_MSIX);
87 	if (rc != msixcnt) {
88 		dev_err(dev, "Failed enabling %d MSIX entries: %d\n", msixcnt, rc);
89 		return -ENOSPC;
90 	}
91 	dev_dbg(dev, "Enabled %d msix vectors\n", msixcnt);
92 
93 
94 	ie = idxd_get_ie(idxd, 0);
95 	ie->vector = pci_irq_vector(pdev, 0);
96 	rc = request_threaded_irq(ie->vector, NULL, idxd_misc_thread, 0, "idxd-misc", ie);
97 	if (rc < 0) {
98 		dev_err(dev, "Failed to allocate misc interrupt.\n");
99 		goto err_misc_irq;
100 	}
101 	dev_dbg(dev, "Requested idxd-misc handler on msix vector %d\n", ie->vector);
102 
103 	for (i = 0; i < idxd->max_wqs; i++) {
104 		int msix_idx = i + 1;
105 
106 		ie = idxd_get_ie(idxd, msix_idx);
107 		ie->id = msix_idx;
108 		ie->int_handle = INVALID_INT_HANDLE;
109 		ie->pasid = INVALID_IOASID;
110 
111 		spin_lock_init(&ie->list_lock);
112 		init_llist_head(&ie->pending_llist);
113 		INIT_LIST_HEAD(&ie->work_list);
114 	}
115 
116 	idxd_unmask_error_interrupts(idxd);
117 	return 0;
118 
119  err_misc_irq:
120 	idxd_mask_error_interrupts(idxd);
121 	pci_free_irq_vectors(pdev);
122 	dev_err(dev, "No usable interrupts\n");
123 	return rc;
124 }
125 
126 static void idxd_cleanup_interrupts(struct idxd_device *idxd)
127 {
128 	struct pci_dev *pdev = idxd->pdev;
129 	struct idxd_irq_entry *ie;
130 	int msixcnt;
131 
132 	msixcnt = pci_msix_vec_count(pdev);
133 	if (msixcnt <= 0)
134 		return;
135 
136 	ie = idxd_get_ie(idxd, 0);
137 	idxd_mask_error_interrupts(idxd);
138 	free_irq(ie->vector, ie);
139 	pci_free_irq_vectors(pdev);
140 }
141 
142 static int idxd_setup_wqs(struct idxd_device *idxd)
143 {
144 	struct device *dev = &idxd->pdev->dev;
145 	struct idxd_wq *wq;
146 	struct device *conf_dev;
147 	int i, rc;
148 
149 	idxd->wqs = kcalloc_node(idxd->max_wqs, sizeof(struct idxd_wq *),
150 				 GFP_KERNEL, dev_to_node(dev));
151 	if (!idxd->wqs)
152 		return -ENOMEM;
153 
154 	idxd->wq_enable_map = bitmap_zalloc_node(idxd->max_wqs, GFP_KERNEL, dev_to_node(dev));
155 	if (!idxd->wq_enable_map) {
156 		kfree(idxd->wqs);
157 		return -ENOMEM;
158 	}
159 
160 	for (i = 0; i < idxd->max_wqs; i++) {
161 		wq = kzalloc_node(sizeof(*wq), GFP_KERNEL, dev_to_node(dev));
162 		if (!wq) {
163 			rc = -ENOMEM;
164 			goto err;
165 		}
166 
167 		idxd_dev_set_type(&wq->idxd_dev, IDXD_DEV_WQ);
168 		conf_dev = wq_confdev(wq);
169 		wq->id = i;
170 		wq->idxd = idxd;
171 		device_initialize(wq_confdev(wq));
172 		conf_dev->parent = idxd_confdev(idxd);
173 		conf_dev->bus = &dsa_bus_type;
174 		conf_dev->type = &idxd_wq_device_type;
175 		rc = dev_set_name(conf_dev, "wq%d.%d", idxd->id, wq->id);
176 		if (rc < 0) {
177 			put_device(conf_dev);
178 			goto err;
179 		}
180 
181 		mutex_init(&wq->wq_lock);
182 		init_waitqueue_head(&wq->err_queue);
183 		init_completion(&wq->wq_dead);
184 		init_completion(&wq->wq_resurrect);
185 		wq->max_xfer_bytes = WQ_DEFAULT_MAX_XFER;
186 		wq->max_batch_size = WQ_DEFAULT_MAX_BATCH;
187 		wq->enqcmds_retries = IDXD_ENQCMDS_RETRIES;
188 		wq->wqcfg = kzalloc_node(idxd->wqcfg_size, GFP_KERNEL, dev_to_node(dev));
189 		if (!wq->wqcfg) {
190 			put_device(conf_dev);
191 			rc = -ENOMEM;
192 			goto err;
193 		}
194 
195 		if (idxd->hw.wq_cap.op_config) {
196 			wq->opcap_bmap = bitmap_zalloc(IDXD_MAX_OPCAP_BITS, GFP_KERNEL);
197 			if (!wq->opcap_bmap) {
198 				put_device(conf_dev);
199 				rc = -ENOMEM;
200 				goto err;
201 			}
202 			bitmap_copy(wq->opcap_bmap, idxd->opcap_bmap, IDXD_MAX_OPCAP_BITS);
203 		}
204 		idxd->wqs[i] = wq;
205 	}
206 
207 	return 0;
208 
209  err:
210 	while (--i >= 0) {
211 		wq = idxd->wqs[i];
212 		conf_dev = wq_confdev(wq);
213 		put_device(conf_dev);
214 	}
215 	return rc;
216 }
217 
218 static int idxd_setup_engines(struct idxd_device *idxd)
219 {
220 	struct idxd_engine *engine;
221 	struct device *dev = &idxd->pdev->dev;
222 	struct device *conf_dev;
223 	int i, rc;
224 
225 	idxd->engines = kcalloc_node(idxd->max_engines, sizeof(struct idxd_engine *),
226 				     GFP_KERNEL, dev_to_node(dev));
227 	if (!idxd->engines)
228 		return -ENOMEM;
229 
230 	for (i = 0; i < idxd->max_engines; i++) {
231 		engine = kzalloc_node(sizeof(*engine), GFP_KERNEL, dev_to_node(dev));
232 		if (!engine) {
233 			rc = -ENOMEM;
234 			goto err;
235 		}
236 
237 		idxd_dev_set_type(&engine->idxd_dev, IDXD_DEV_ENGINE);
238 		conf_dev = engine_confdev(engine);
239 		engine->id = i;
240 		engine->idxd = idxd;
241 		device_initialize(conf_dev);
242 		conf_dev->parent = idxd_confdev(idxd);
243 		conf_dev->bus = &dsa_bus_type;
244 		conf_dev->type = &idxd_engine_device_type;
245 		rc = dev_set_name(conf_dev, "engine%d.%d", idxd->id, engine->id);
246 		if (rc < 0) {
247 			put_device(conf_dev);
248 			goto err;
249 		}
250 
251 		idxd->engines[i] = engine;
252 	}
253 
254 	return 0;
255 
256  err:
257 	while (--i >= 0) {
258 		engine = idxd->engines[i];
259 		conf_dev = engine_confdev(engine);
260 		put_device(conf_dev);
261 	}
262 	return rc;
263 }
264 
265 static int idxd_setup_groups(struct idxd_device *idxd)
266 {
267 	struct device *dev = &idxd->pdev->dev;
268 	struct device *conf_dev;
269 	struct idxd_group *group;
270 	int i, rc;
271 
272 	idxd->groups = kcalloc_node(idxd->max_groups, sizeof(struct idxd_group *),
273 				    GFP_KERNEL, dev_to_node(dev));
274 	if (!idxd->groups)
275 		return -ENOMEM;
276 
277 	for (i = 0; i < idxd->max_groups; i++) {
278 		group = kzalloc_node(sizeof(*group), GFP_KERNEL, dev_to_node(dev));
279 		if (!group) {
280 			rc = -ENOMEM;
281 			goto err;
282 		}
283 
284 		idxd_dev_set_type(&group->idxd_dev, IDXD_DEV_GROUP);
285 		conf_dev = group_confdev(group);
286 		group->id = i;
287 		group->idxd = idxd;
288 		device_initialize(conf_dev);
289 		conf_dev->parent = idxd_confdev(idxd);
290 		conf_dev->bus = &dsa_bus_type;
291 		conf_dev->type = &idxd_group_device_type;
292 		rc = dev_set_name(conf_dev, "group%d.%d", idxd->id, group->id);
293 		if (rc < 0) {
294 			put_device(conf_dev);
295 			goto err;
296 		}
297 
298 		idxd->groups[i] = group;
299 		if (idxd->hw.version < DEVICE_VERSION_2 && !tc_override) {
300 			group->tc_a = 1;
301 			group->tc_b = 1;
302 		} else {
303 			group->tc_a = -1;
304 			group->tc_b = -1;
305 		}
306 	}
307 
308 	return 0;
309 
310  err:
311 	while (--i >= 0) {
312 		group = idxd->groups[i];
313 		put_device(group_confdev(group));
314 	}
315 	return rc;
316 }
317 
318 static void idxd_cleanup_internals(struct idxd_device *idxd)
319 {
320 	int i;
321 
322 	for (i = 0; i < idxd->max_groups; i++)
323 		put_device(group_confdev(idxd->groups[i]));
324 	for (i = 0; i < idxd->max_engines; i++)
325 		put_device(engine_confdev(idxd->engines[i]));
326 	for (i = 0; i < idxd->max_wqs; i++)
327 		put_device(wq_confdev(idxd->wqs[i]));
328 	destroy_workqueue(idxd->wq);
329 }
330 
331 static int idxd_setup_internals(struct idxd_device *idxd)
332 {
333 	struct device *dev = &idxd->pdev->dev;
334 	int rc, i;
335 
336 	init_waitqueue_head(&idxd->cmd_waitq);
337 
338 	rc = idxd_setup_wqs(idxd);
339 	if (rc < 0)
340 		goto err_wqs;
341 
342 	rc = idxd_setup_engines(idxd);
343 	if (rc < 0)
344 		goto err_engine;
345 
346 	rc = idxd_setup_groups(idxd);
347 	if (rc < 0)
348 		goto err_group;
349 
350 	idxd->wq = create_workqueue(dev_name(dev));
351 	if (!idxd->wq) {
352 		rc = -ENOMEM;
353 		goto err_wkq_create;
354 	}
355 
356 	return 0;
357 
358  err_wkq_create:
359 	for (i = 0; i < idxd->max_groups; i++)
360 		put_device(group_confdev(idxd->groups[i]));
361  err_group:
362 	for (i = 0; i < idxd->max_engines; i++)
363 		put_device(engine_confdev(idxd->engines[i]));
364  err_engine:
365 	for (i = 0; i < idxd->max_wqs; i++)
366 		put_device(wq_confdev(idxd->wqs[i]));
367  err_wqs:
368 	return rc;
369 }
370 
371 static void idxd_read_table_offsets(struct idxd_device *idxd)
372 {
373 	union offsets_reg offsets;
374 	struct device *dev = &idxd->pdev->dev;
375 
376 	offsets.bits[0] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET);
377 	offsets.bits[1] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET + sizeof(u64));
378 	idxd->grpcfg_offset = offsets.grpcfg * IDXD_TABLE_MULT;
379 	dev_dbg(dev, "IDXD Group Config Offset: %#x\n", idxd->grpcfg_offset);
380 	idxd->wqcfg_offset = offsets.wqcfg * IDXD_TABLE_MULT;
381 	dev_dbg(dev, "IDXD Work Queue Config Offset: %#x\n", idxd->wqcfg_offset);
382 	idxd->msix_perm_offset = offsets.msix_perm * IDXD_TABLE_MULT;
383 	dev_dbg(dev, "IDXD MSIX Permission Offset: %#x\n", idxd->msix_perm_offset);
384 	idxd->perfmon_offset = offsets.perfmon * IDXD_TABLE_MULT;
385 	dev_dbg(dev, "IDXD Perfmon Offset: %#x\n", idxd->perfmon_offset);
386 }
387 
388 static void multi_u64_to_bmap(unsigned long *bmap, u64 *val, int count)
389 {
390 	int i, j, nr;
391 
392 	for (i = 0, nr = 0; i < count; i++) {
393 		for (j = 0; j < BITS_PER_LONG_LONG; j++) {
394 			if (val[i] & BIT(j))
395 				set_bit(nr, bmap);
396 			nr++;
397 		}
398 	}
399 }
400 
401 static void idxd_read_caps(struct idxd_device *idxd)
402 {
403 	struct device *dev = &idxd->pdev->dev;
404 	int i;
405 
406 	/* reading generic capabilities */
407 	idxd->hw.gen_cap.bits = ioread64(idxd->reg_base + IDXD_GENCAP_OFFSET);
408 	dev_dbg(dev, "gen_cap: %#llx\n", idxd->hw.gen_cap.bits);
409 
410 	if (idxd->hw.gen_cap.cmd_cap) {
411 		idxd->hw.cmd_cap = ioread32(idxd->reg_base + IDXD_CMDCAP_OFFSET);
412 		dev_dbg(dev, "cmd_cap: %#x\n", idxd->hw.cmd_cap);
413 	}
414 
415 	/* reading command capabilities */
416 	if (idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE))
417 		idxd->request_int_handles = true;
418 
419 	idxd->max_xfer_bytes = 1ULL << idxd->hw.gen_cap.max_xfer_shift;
420 	dev_dbg(dev, "max xfer size: %llu bytes\n", idxd->max_xfer_bytes);
421 	idxd->max_batch_size = 1U << idxd->hw.gen_cap.max_batch_shift;
422 	dev_dbg(dev, "max batch size: %u\n", idxd->max_batch_size);
423 	if (idxd->hw.gen_cap.config_en)
424 		set_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags);
425 
426 	/* reading group capabilities */
427 	idxd->hw.group_cap.bits =
428 		ioread64(idxd->reg_base + IDXD_GRPCAP_OFFSET);
429 	dev_dbg(dev, "group_cap: %#llx\n", idxd->hw.group_cap.bits);
430 	idxd->max_groups = idxd->hw.group_cap.num_groups;
431 	dev_dbg(dev, "max groups: %u\n", idxd->max_groups);
432 	idxd->max_rdbufs = idxd->hw.group_cap.total_rdbufs;
433 	dev_dbg(dev, "max read buffers: %u\n", idxd->max_rdbufs);
434 	idxd->nr_rdbufs = idxd->max_rdbufs;
435 
436 	/* read engine capabilities */
437 	idxd->hw.engine_cap.bits =
438 		ioread64(idxd->reg_base + IDXD_ENGCAP_OFFSET);
439 	dev_dbg(dev, "engine_cap: %#llx\n", idxd->hw.engine_cap.bits);
440 	idxd->max_engines = idxd->hw.engine_cap.num_engines;
441 	dev_dbg(dev, "max engines: %u\n", idxd->max_engines);
442 
443 	/* read workqueue capabilities */
444 	idxd->hw.wq_cap.bits = ioread64(idxd->reg_base + IDXD_WQCAP_OFFSET);
445 	dev_dbg(dev, "wq_cap: %#llx\n", idxd->hw.wq_cap.bits);
446 	idxd->max_wq_size = idxd->hw.wq_cap.total_wq_size;
447 	dev_dbg(dev, "total workqueue size: %u\n", idxd->max_wq_size);
448 	idxd->max_wqs = idxd->hw.wq_cap.num_wqs;
449 	dev_dbg(dev, "max workqueues: %u\n", idxd->max_wqs);
450 	idxd->wqcfg_size = 1 << (idxd->hw.wq_cap.wqcfg_size + IDXD_WQCFG_MIN);
451 	dev_dbg(dev, "wqcfg size: %u\n", idxd->wqcfg_size);
452 
453 	/* reading operation capabilities */
454 	for (i = 0; i < 4; i++) {
455 		idxd->hw.opcap.bits[i] = ioread64(idxd->reg_base +
456 				IDXD_OPCAP_OFFSET + i * sizeof(u64));
457 		dev_dbg(dev, "opcap[%d]: %#llx\n", i, idxd->hw.opcap.bits[i]);
458 	}
459 	multi_u64_to_bmap(idxd->opcap_bmap, &idxd->hw.opcap.bits[0], 4);
460 }
461 
462 static struct idxd_device *idxd_alloc(struct pci_dev *pdev, struct idxd_driver_data *data)
463 {
464 	struct device *dev = &pdev->dev;
465 	struct device *conf_dev;
466 	struct idxd_device *idxd;
467 	int rc;
468 
469 	idxd = kzalloc_node(sizeof(*idxd), GFP_KERNEL, dev_to_node(dev));
470 	if (!idxd)
471 		return NULL;
472 
473 	conf_dev = idxd_confdev(idxd);
474 	idxd->pdev = pdev;
475 	idxd->data = data;
476 	idxd_dev_set_type(&idxd->idxd_dev, idxd->data->type);
477 	idxd->id = ida_alloc(&idxd_ida, GFP_KERNEL);
478 	if (idxd->id < 0)
479 		return NULL;
480 
481 	idxd->opcap_bmap = bitmap_zalloc_node(IDXD_MAX_OPCAP_BITS, GFP_KERNEL, dev_to_node(dev));
482 	if (!idxd->opcap_bmap) {
483 		ida_free(&idxd_ida, idxd->id);
484 		return NULL;
485 	}
486 
487 	device_initialize(conf_dev);
488 	conf_dev->parent = dev;
489 	conf_dev->bus = &dsa_bus_type;
490 	conf_dev->type = idxd->data->dev_type;
491 	rc = dev_set_name(conf_dev, "%s%d", idxd->data->name_prefix, idxd->id);
492 	if (rc < 0) {
493 		put_device(conf_dev);
494 		return NULL;
495 	}
496 
497 	spin_lock_init(&idxd->dev_lock);
498 	spin_lock_init(&idxd->cmd_lock);
499 
500 	return idxd;
501 }
502 
503 static int idxd_enable_system_pasid(struct idxd_device *idxd)
504 {
505 	int flags;
506 	unsigned int pasid;
507 	struct iommu_sva *sva;
508 
509 	flags = SVM_FLAG_SUPERVISOR_MODE;
510 
511 	sva = iommu_sva_bind_device(&idxd->pdev->dev, NULL, &flags);
512 	if (IS_ERR(sva)) {
513 		dev_warn(&idxd->pdev->dev,
514 			 "iommu sva bind failed: %ld\n", PTR_ERR(sva));
515 		return PTR_ERR(sva);
516 	}
517 
518 	pasid = iommu_sva_get_pasid(sva);
519 	if (pasid == IOMMU_PASID_INVALID) {
520 		iommu_sva_unbind_device(sva);
521 		return -ENODEV;
522 	}
523 
524 	idxd->sva = sva;
525 	idxd->pasid = pasid;
526 	dev_dbg(&idxd->pdev->dev, "system pasid: %u\n", pasid);
527 	return 0;
528 }
529 
530 static void idxd_disable_system_pasid(struct idxd_device *idxd)
531 {
532 
533 	iommu_sva_unbind_device(idxd->sva);
534 	idxd->sva = NULL;
535 }
536 
537 static int idxd_probe(struct idxd_device *idxd)
538 {
539 	struct pci_dev *pdev = idxd->pdev;
540 	struct device *dev = &pdev->dev;
541 	int rc;
542 
543 	dev_dbg(dev, "%s entered and resetting device\n", __func__);
544 	rc = idxd_device_init_reset(idxd);
545 	if (rc < 0)
546 		return rc;
547 
548 	dev_dbg(dev, "IDXD reset complete\n");
549 
550 	if (IS_ENABLED(CONFIG_INTEL_IDXD_SVM) && sva) {
551 		if (iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA)) {
552 			dev_warn(dev, "Unable to turn on user SVA feature.\n");
553 		} else {
554 			set_bit(IDXD_FLAG_USER_PASID_ENABLED, &idxd->flags);
555 
556 			if (idxd_enable_system_pasid(idxd))
557 				dev_warn(dev, "No in-kernel DMA with PASID.\n");
558 			else
559 				set_bit(IDXD_FLAG_PASID_ENABLED, &idxd->flags);
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 	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 	if (device_user_pasid_enabled(idxd))
599 		iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
600 	return rc;
601 }
602 
603 static void idxd_cleanup(struct idxd_device *idxd)
604 {
605 	struct device *dev = &idxd->pdev->dev;
606 
607 	perfmon_pmu_remove(idxd);
608 	idxd_cleanup_interrupts(idxd);
609 	idxd_cleanup_internals(idxd);
610 	if (device_pasid_enabled(idxd))
611 		idxd_disable_system_pasid(idxd);
612 	if (device_user_pasid_enabled(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 = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
643 	if (rc)
644 		goto err;
645 
646 	dev_dbg(dev, "Set PCI master\n");
647 	pci_set_master(pdev);
648 	pci_set_drvdata(pdev, idxd);
649 
650 	idxd->hw.version = ioread32(idxd->reg_base + IDXD_VER_OFFSET);
651 	rc = idxd_probe(idxd);
652 	if (rc) {
653 		dev_err(dev, "Intel(R) IDXD DMA Engine init failed\n");
654 		goto err;
655 	}
656 
657 	rc = idxd_register_devices(idxd);
658 	if (rc) {
659 		dev_err(dev, "IDXD sysfs setup failed\n");
660 		goto err_dev_register;
661 	}
662 
663 	dev_info(&pdev->dev, "Intel(R) Accelerator Device (v%x)\n",
664 		 idxd->hw.version);
665 
666 	return 0;
667 
668  err_dev_register:
669 	idxd_cleanup(idxd);
670  err:
671 	pci_iounmap(pdev, idxd->reg_base);
672  err_iomap:
673 	put_device(idxd_confdev(idxd));
674  err_idxd_alloc:
675 	pci_disable_device(pdev);
676 	return rc;
677 }
678 
679 void idxd_wqs_quiesce(struct idxd_device *idxd)
680 {
681 	struct idxd_wq *wq;
682 	int i;
683 
684 	for (i = 0; i < idxd->max_wqs; i++) {
685 		wq = idxd->wqs[i];
686 		if (wq->state == IDXD_WQ_ENABLED && wq->type == IDXD_WQT_KERNEL)
687 			idxd_wq_quiesce(wq);
688 	}
689 }
690 
691 static void idxd_shutdown(struct pci_dev *pdev)
692 {
693 	struct idxd_device *idxd = pci_get_drvdata(pdev);
694 	struct idxd_irq_entry *irq_entry;
695 	int rc;
696 
697 	rc = idxd_device_disable(idxd);
698 	if (rc)
699 		dev_err(&pdev->dev, "Disabling device failed\n");
700 
701 	irq_entry = &idxd->ie;
702 	synchronize_irq(irq_entry->vector);
703 	idxd_mask_error_interrupts(idxd);
704 	flush_workqueue(idxd->wq);
705 }
706 
707 static void idxd_remove(struct pci_dev *pdev)
708 {
709 	struct idxd_device *idxd = pci_get_drvdata(pdev);
710 	struct idxd_irq_entry *irq_entry;
711 
712 	idxd_unregister_devices(idxd);
713 	/*
714 	 * When ->release() is called for the idxd->conf_dev, it frees all the memory related
715 	 * to the idxd context. The driver still needs those bits in order to do the rest of
716 	 * the cleanup. However, we do need to unbound the idxd sub-driver. So take a ref
717 	 * on the device here to hold off the freeing while allowing the idxd sub-driver
718 	 * to unbind.
719 	 */
720 	get_device(idxd_confdev(idxd));
721 	device_unregister(idxd_confdev(idxd));
722 	idxd_shutdown(pdev);
723 	if (device_pasid_enabled(idxd))
724 		idxd_disable_system_pasid(idxd);
725 
726 	irq_entry = idxd_get_ie(idxd, 0);
727 	free_irq(irq_entry->vector, irq_entry);
728 	pci_free_irq_vectors(pdev);
729 	pci_iounmap(pdev, idxd->reg_base);
730 	if (device_user_pasid_enabled(idxd))
731 		iommu_dev_disable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA);
732 	pci_disable_device(pdev);
733 	destroy_workqueue(idxd->wq);
734 	perfmon_pmu_remove(idxd);
735 	put_device(idxd_confdev(idxd));
736 }
737 
738 static struct pci_driver idxd_pci_driver = {
739 	.name		= DRV_NAME,
740 	.id_table	= idxd_pci_tbl,
741 	.probe		= idxd_pci_probe,
742 	.remove		= idxd_remove,
743 	.shutdown	= idxd_shutdown,
744 };
745 
746 static int __init idxd_init_module(void)
747 {
748 	int err;
749 
750 	/*
751 	 * If the CPU does not support MOVDIR64B or ENQCMDS, there's no point in
752 	 * enumerating the device. We can not utilize it.
753 	 */
754 	if (!cpu_feature_enabled(X86_FEATURE_MOVDIR64B)) {
755 		pr_warn("idxd driver failed to load without MOVDIR64B.\n");
756 		return -ENODEV;
757 	}
758 
759 	if (!cpu_feature_enabled(X86_FEATURE_ENQCMD))
760 		pr_warn("Platform does not have ENQCMD(S) support.\n");
761 	else
762 		support_enqcmd = true;
763 
764 	perfmon_init();
765 
766 	err = idxd_driver_register(&idxd_drv);
767 	if (err < 0)
768 		goto err_idxd_driver_register;
769 
770 	err = idxd_driver_register(&idxd_dmaengine_drv);
771 	if (err < 0)
772 		goto err_idxd_dmaengine_driver_register;
773 
774 	err = idxd_driver_register(&idxd_user_drv);
775 	if (err < 0)
776 		goto err_idxd_user_driver_register;
777 
778 	err = idxd_cdev_register();
779 	if (err)
780 		goto err_cdev_register;
781 
782 	err = pci_register_driver(&idxd_pci_driver);
783 	if (err)
784 		goto err_pci_register;
785 
786 	return 0;
787 
788 err_pci_register:
789 	idxd_cdev_remove();
790 err_cdev_register:
791 	idxd_driver_unregister(&idxd_user_drv);
792 err_idxd_user_driver_register:
793 	idxd_driver_unregister(&idxd_dmaengine_drv);
794 err_idxd_dmaengine_driver_register:
795 	idxd_driver_unregister(&idxd_drv);
796 err_idxd_driver_register:
797 	return err;
798 }
799 module_init(idxd_init_module);
800 
801 static void __exit idxd_exit_module(void)
802 {
803 	idxd_driver_unregister(&idxd_user_drv);
804 	idxd_driver_unregister(&idxd_dmaengine_drv);
805 	idxd_driver_unregister(&idxd_drv);
806 	pci_unregister_driver(&idxd_pci_driver);
807 	idxd_cdev_remove();
808 	perfmon_exit();
809 }
810 module_exit(idxd_exit_module);
811