xref: /linux/drivers/dma/idxd/init.c (revision 497e6b37b0099dc415578488287fd84fb74433eb)
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/iommu.h>
18 #include <uapi/linux/idxd.h>
19 #include <linux/dmaengine.h>
20 #include "../dmaengine.h"
21 #include "registers.h"
22 #include "idxd.h"
23 #include "perfmon.h"
24 
25 MODULE_VERSION(IDXD_DRIVER_VERSION);
26 MODULE_LICENSE("GPL v2");
27 MODULE_AUTHOR("Intel Corporation");
28 MODULE_IMPORT_NS(IDXD);
29 
30 static bool sva = true;
31 module_param(sva, bool, 0644);
32 MODULE_PARM_DESC(sva, "Toggle SVA support on/off");
33 
34 bool tc_override;
35 module_param(tc_override, bool, 0644);
36 MODULE_PARM_DESC(tc_override, "Override traffic class defaults");
37 
38 #define DRV_NAME "idxd"
39 
40 bool support_enqcmd;
41 DEFINE_IDA(idxd_ida);
42 
43 static struct idxd_driver_data idxd_driver_data[] = {
44 	[IDXD_TYPE_DSA] = {
45 		.name_prefix = "dsa",
46 		.type = IDXD_TYPE_DSA,
47 		.compl_size = sizeof(struct dsa_completion_record),
48 		.align = 32,
49 		.dev_type = &dsa_device_type,
50 	},
51 	[IDXD_TYPE_IAX] = {
52 		.name_prefix = "iax",
53 		.type = IDXD_TYPE_IAX,
54 		.compl_size = sizeof(struct iax_completion_record),
55 		.align = 64,
56 		.dev_type = &iax_device_type,
57 	},
58 };
59 
60 static struct pci_device_id idxd_pci_tbl[] = {
61 	/* DSA ver 1.0 platforms */
62 	{ PCI_DEVICE_DATA(INTEL, DSA_SPR0, &idxd_driver_data[IDXD_TYPE_DSA]) },
63 
64 	/* IAX ver 1.0 platforms */
65 	{ PCI_DEVICE_DATA(INTEL, IAX_SPR0, &idxd_driver_data[IDXD_TYPE_IAX]) },
66 	{ 0, }
67 };
68 MODULE_DEVICE_TABLE(pci, idxd_pci_tbl);
69 
70 static int idxd_setup_interrupts(struct idxd_device *idxd)
71 {
72 	struct pci_dev *pdev = idxd->pdev;
73 	struct device *dev = &pdev->dev;
74 	struct idxd_irq_entry *ie;
75 	int i, msixcnt;
76 	int rc = 0;
77 
78 	msixcnt = pci_msix_vec_count(pdev);
79 	if (msixcnt < 0) {
80 		dev_err(dev, "Not MSI-X interrupt capable.\n");
81 		return -ENOSPC;
82 	}
83 	idxd->irq_cnt = msixcnt;
84 
85 	rc = pci_alloc_irq_vectors(pdev, msixcnt, msixcnt, PCI_IRQ_MSIX);
86 	if (rc != msixcnt) {
87 		dev_err(dev, "Failed enabling %d MSIX entries: %d\n", msixcnt, rc);
88 		return -ENOSPC;
89 	}
90 	dev_dbg(dev, "Enabled %d msix vectors\n", msixcnt);
91 
92 
93 	ie = idxd_get_ie(idxd, 0);
94 	ie->vector = pci_irq_vector(pdev, 0);
95 	rc = request_threaded_irq(ie->vector, NULL, idxd_misc_thread, 0, "idxd-misc", ie);
96 	if (rc < 0) {
97 		dev_err(dev, "Failed to allocate misc interrupt.\n");
98 		goto err_misc_irq;
99 	}
100 	dev_dbg(dev, "Requested idxd-misc handler on msix vector %d\n", ie->vector);
101 
102 	for (i = 0; i < idxd->max_wqs; i++) {
103 		int msix_idx = i + 1;
104 
105 		ie = idxd_get_ie(idxd, msix_idx);
106 		ie->id = msix_idx;
107 		ie->int_handle = INVALID_INT_HANDLE;
108 		ie->pasid = INVALID_IOASID;
109 
110 		spin_lock_init(&ie->list_lock);
111 		init_llist_head(&ie->pending_llist);
112 		INIT_LIST_HEAD(&ie->work_list);
113 	}
114 
115 	idxd_unmask_error_interrupts(idxd);
116 	return 0;
117 
118  err_misc_irq:
119 	idxd_mask_error_interrupts(idxd);
120 	pci_free_irq_vectors(pdev);
121 	dev_err(dev, "No usable interrupts\n");
122 	return rc;
123 }
124 
125 static void idxd_cleanup_interrupts(struct idxd_device *idxd)
126 {
127 	struct pci_dev *pdev = idxd->pdev;
128 	struct idxd_irq_entry *ie;
129 	int msixcnt;
130 
131 	msixcnt = pci_msix_vec_count(pdev);
132 	if (msixcnt <= 0)
133 		return;
134 
135 	ie = idxd_get_ie(idxd, 0);
136 	idxd_mask_error_interrupts(idxd);
137 	free_irq(ie->vector, ie);
138 	pci_free_irq_vectors(pdev);
139 }
140 
141 static int idxd_setup_wqs(struct idxd_device *idxd)
142 {
143 	struct device *dev = &idxd->pdev->dev;
144 	struct idxd_wq *wq;
145 	struct device *conf_dev;
146 	int i, rc;
147 
148 	idxd->wqs = kcalloc_node(idxd->max_wqs, sizeof(struct idxd_wq *),
149 				 GFP_KERNEL, dev_to_node(dev));
150 	if (!idxd->wqs)
151 		return -ENOMEM;
152 
153 	idxd->wq_enable_map = bitmap_zalloc_node(idxd->max_wqs, GFP_KERNEL, dev_to_node(dev));
154 	if (!idxd->wq_enable_map) {
155 		kfree(idxd->wqs);
156 		return -ENOMEM;
157 	}
158 
159 	for (i = 0; i < idxd->max_wqs; i++) {
160 		wq = kzalloc_node(sizeof(*wq), GFP_KERNEL, dev_to_node(dev));
161 		if (!wq) {
162 			rc = -ENOMEM;
163 			goto err;
164 		}
165 
166 		idxd_dev_set_type(&wq->idxd_dev, IDXD_DEV_WQ);
167 		conf_dev = wq_confdev(wq);
168 		wq->id = i;
169 		wq->idxd = idxd;
170 		device_initialize(wq_confdev(wq));
171 		conf_dev->parent = idxd_confdev(idxd);
172 		conf_dev->bus = &dsa_bus_type;
173 		conf_dev->type = &idxd_wq_device_type;
174 		rc = dev_set_name(conf_dev, "wq%d.%d", idxd->id, wq->id);
175 		if (rc < 0) {
176 			put_device(conf_dev);
177 			goto err;
178 		}
179 
180 		mutex_init(&wq->wq_lock);
181 		init_waitqueue_head(&wq->err_queue);
182 		init_completion(&wq->wq_dead);
183 		init_completion(&wq->wq_resurrect);
184 		wq->max_xfer_bytes = WQ_DEFAULT_MAX_XFER;
185 		idxd_wq_set_max_batch_size(idxd->data->type, wq, WQ_DEFAULT_MAX_BATCH);
186 		wq->enqcmds_retries = IDXD_ENQCMDS_RETRIES;
187 		wq->wqcfg = kzalloc_node(idxd->wqcfg_size, GFP_KERNEL, dev_to_node(dev));
188 		if (!wq->wqcfg) {
189 			put_device(conf_dev);
190 			rc = -ENOMEM;
191 			goto err;
192 		}
193 
194 		if (idxd->hw.wq_cap.op_config) {
195 			wq->opcap_bmap = bitmap_zalloc(IDXD_MAX_OPCAP_BITS, GFP_KERNEL);
196 			if (!wq->opcap_bmap) {
197 				put_device(conf_dev);
198 				rc = -ENOMEM;
199 				goto err;
200 			}
201 			bitmap_copy(wq->opcap_bmap, idxd->opcap_bmap, IDXD_MAX_OPCAP_BITS);
202 		}
203 		idxd->wqs[i] = wq;
204 	}
205 
206 	return 0;
207 
208  err:
209 	while (--i >= 0) {
210 		wq = idxd->wqs[i];
211 		conf_dev = wq_confdev(wq);
212 		put_device(conf_dev);
213 	}
214 	return rc;
215 }
216 
217 static int idxd_setup_engines(struct idxd_device *idxd)
218 {
219 	struct idxd_engine *engine;
220 	struct device *dev = &idxd->pdev->dev;
221 	struct device *conf_dev;
222 	int i, rc;
223 
224 	idxd->engines = kcalloc_node(idxd->max_engines, sizeof(struct idxd_engine *),
225 				     GFP_KERNEL, dev_to_node(dev));
226 	if (!idxd->engines)
227 		return -ENOMEM;
228 
229 	for (i = 0; i < idxd->max_engines; i++) {
230 		engine = kzalloc_node(sizeof(*engine), GFP_KERNEL, dev_to_node(dev));
231 		if (!engine) {
232 			rc = -ENOMEM;
233 			goto err;
234 		}
235 
236 		idxd_dev_set_type(&engine->idxd_dev, IDXD_DEV_ENGINE);
237 		conf_dev = engine_confdev(engine);
238 		engine->id = i;
239 		engine->idxd = idxd;
240 		device_initialize(conf_dev);
241 		conf_dev->parent = idxd_confdev(idxd);
242 		conf_dev->bus = &dsa_bus_type;
243 		conf_dev->type = &idxd_engine_device_type;
244 		rc = dev_set_name(conf_dev, "engine%d.%d", idxd->id, engine->id);
245 		if (rc < 0) {
246 			put_device(conf_dev);
247 			goto err;
248 		}
249 
250 		idxd->engines[i] = engine;
251 	}
252 
253 	return 0;
254 
255  err:
256 	while (--i >= 0) {
257 		engine = idxd->engines[i];
258 		conf_dev = engine_confdev(engine);
259 		put_device(conf_dev);
260 	}
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 device *conf_dev;
268 	struct idxd_group *group;
269 	int i, rc;
270 
271 	idxd->groups = kcalloc_node(idxd->max_groups, sizeof(struct idxd_group *),
272 				    GFP_KERNEL, dev_to_node(dev));
273 	if (!idxd->groups)
274 		return -ENOMEM;
275 
276 	for (i = 0; i < idxd->max_groups; i++) {
277 		group = kzalloc_node(sizeof(*group), GFP_KERNEL, dev_to_node(dev));
278 		if (!group) {
279 			rc = -ENOMEM;
280 			goto err;
281 		}
282 
283 		idxd_dev_set_type(&group->idxd_dev, IDXD_DEV_GROUP);
284 		conf_dev = group_confdev(group);
285 		group->id = i;
286 		group->idxd = idxd;
287 		device_initialize(conf_dev);
288 		conf_dev->parent = idxd_confdev(idxd);
289 		conf_dev->bus = &dsa_bus_type;
290 		conf_dev->type = &idxd_group_device_type;
291 		rc = dev_set_name(conf_dev, "group%d.%d", idxd->id, group->id);
292 		if (rc < 0) {
293 			put_device(conf_dev);
294 			goto err;
295 		}
296 
297 		idxd->groups[i] = group;
298 		if (idxd->hw.version < DEVICE_VERSION_2 && !tc_override) {
299 			group->tc_a = 1;
300 			group->tc_b = 1;
301 		} else {
302 			group->tc_a = -1;
303 			group->tc_b = -1;
304 		}
305 	}
306 
307 	return 0;
308 
309  err:
310 	while (--i >= 0) {
311 		group = idxd->groups[i];
312 		put_device(group_confdev(group));
313 	}
314 	return rc;
315 }
316 
317 static void idxd_cleanup_internals(struct idxd_device *idxd)
318 {
319 	int i;
320 
321 	for (i = 0; i < idxd->max_groups; i++)
322 		put_device(group_confdev(idxd->groups[i]));
323 	for (i = 0; i < idxd->max_engines; i++)
324 		put_device(engine_confdev(idxd->engines[i]));
325 	for (i = 0; i < idxd->max_wqs; i++)
326 		put_device(wq_confdev(idxd->wqs[i]));
327 	destroy_workqueue(idxd->wq);
328 }
329 
330 static int idxd_setup_internals(struct idxd_device *idxd)
331 {
332 	struct device *dev = &idxd->pdev->dev;
333 	int rc, i;
334 
335 	init_waitqueue_head(&idxd->cmd_waitq);
336 
337 	rc = idxd_setup_wqs(idxd);
338 	if (rc < 0)
339 		goto err_wqs;
340 
341 	rc = idxd_setup_engines(idxd);
342 	if (rc < 0)
343 		goto err_engine;
344 
345 	rc = idxd_setup_groups(idxd);
346 	if (rc < 0)
347 		goto err_group;
348 
349 	idxd->wq = create_workqueue(dev_name(dev));
350 	if (!idxd->wq) {
351 		rc = -ENOMEM;
352 		goto err_wkq_create;
353 	}
354 
355 	return 0;
356 
357  err_wkq_create:
358 	for (i = 0; i < idxd->max_groups; i++)
359 		put_device(group_confdev(idxd->groups[i]));
360  err_group:
361 	for (i = 0; i < idxd->max_engines; i++)
362 		put_device(engine_confdev(idxd->engines[i]));
363  err_engine:
364 	for (i = 0; i < idxd->max_wqs; i++)
365 		put_device(wq_confdev(idxd->wqs[i]));
366  err_wqs:
367 	return rc;
368 }
369 
370 static void idxd_read_table_offsets(struct idxd_device *idxd)
371 {
372 	union offsets_reg offsets;
373 	struct device *dev = &idxd->pdev->dev;
374 
375 	offsets.bits[0] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET);
376 	offsets.bits[1] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET + sizeof(u64));
377 	idxd->grpcfg_offset = offsets.grpcfg * IDXD_TABLE_MULT;
378 	dev_dbg(dev, "IDXD Group Config Offset: %#x\n", idxd->grpcfg_offset);
379 	idxd->wqcfg_offset = offsets.wqcfg * IDXD_TABLE_MULT;
380 	dev_dbg(dev, "IDXD Work Queue Config Offset: %#x\n", idxd->wqcfg_offset);
381 	idxd->msix_perm_offset = offsets.msix_perm * IDXD_TABLE_MULT;
382 	dev_dbg(dev, "IDXD MSIX Permission Offset: %#x\n", idxd->msix_perm_offset);
383 	idxd->perfmon_offset = offsets.perfmon * IDXD_TABLE_MULT;
384 	dev_dbg(dev, "IDXD Perfmon Offset: %#x\n", idxd->perfmon_offset);
385 }
386 
387 static void multi_u64_to_bmap(unsigned long *bmap, u64 *val, int count)
388 {
389 	int i, j, nr;
390 
391 	for (i = 0, nr = 0; i < count; i++) {
392 		for (j = 0; j < BITS_PER_LONG_LONG; j++) {
393 			if (val[i] & BIT(j))
394 				set_bit(nr, bmap);
395 			nr++;
396 		}
397 	}
398 }
399 
400 static void idxd_read_caps(struct idxd_device *idxd)
401 {
402 	struct device *dev = &idxd->pdev->dev;
403 	int i;
404 
405 	/* reading generic capabilities */
406 	idxd->hw.gen_cap.bits = ioread64(idxd->reg_base + IDXD_GENCAP_OFFSET);
407 	dev_dbg(dev, "gen_cap: %#llx\n", idxd->hw.gen_cap.bits);
408 
409 	if (idxd->hw.gen_cap.cmd_cap) {
410 		idxd->hw.cmd_cap = ioread32(idxd->reg_base + IDXD_CMDCAP_OFFSET);
411 		dev_dbg(dev, "cmd_cap: %#x\n", idxd->hw.cmd_cap);
412 	}
413 
414 	/* reading command capabilities */
415 	if (idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE))
416 		idxd->request_int_handles = true;
417 
418 	idxd->max_xfer_bytes = 1ULL << idxd->hw.gen_cap.max_xfer_shift;
419 	dev_dbg(dev, "max xfer size: %llu bytes\n", idxd->max_xfer_bytes);
420 	idxd_set_max_batch_size(idxd->data->type, idxd, 1U << idxd->hw.gen_cap.max_batch_shift);
421 	dev_dbg(dev, "max batch size: %u\n", idxd->max_batch_size);
422 	if (idxd->hw.gen_cap.config_en)
423 		set_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags);
424 
425 	/* reading group capabilities */
426 	idxd->hw.group_cap.bits =
427 		ioread64(idxd->reg_base + IDXD_GRPCAP_OFFSET);
428 	dev_dbg(dev, "group_cap: %#llx\n", idxd->hw.group_cap.bits);
429 	idxd->max_groups = idxd->hw.group_cap.num_groups;
430 	dev_dbg(dev, "max groups: %u\n", idxd->max_groups);
431 	idxd->max_rdbufs = idxd->hw.group_cap.total_rdbufs;
432 	dev_dbg(dev, "max read buffers: %u\n", idxd->max_rdbufs);
433 	idxd->nr_rdbufs = idxd->max_rdbufs;
434 
435 	/* read engine capabilities */
436 	idxd->hw.engine_cap.bits =
437 		ioread64(idxd->reg_base + IDXD_ENGCAP_OFFSET);
438 	dev_dbg(dev, "engine_cap: %#llx\n", idxd->hw.engine_cap.bits);
439 	idxd->max_engines = idxd->hw.engine_cap.num_engines;
440 	dev_dbg(dev, "max engines: %u\n", idxd->max_engines);
441 
442 	/* read workqueue capabilities */
443 	idxd->hw.wq_cap.bits = ioread64(idxd->reg_base + IDXD_WQCAP_OFFSET);
444 	dev_dbg(dev, "wq_cap: %#llx\n", idxd->hw.wq_cap.bits);
445 	idxd->max_wq_size = idxd->hw.wq_cap.total_wq_size;
446 	dev_dbg(dev, "total workqueue size: %u\n", idxd->max_wq_size);
447 	idxd->max_wqs = idxd->hw.wq_cap.num_wqs;
448 	dev_dbg(dev, "max workqueues: %u\n", idxd->max_wqs);
449 	idxd->wqcfg_size = 1 << (idxd->hw.wq_cap.wqcfg_size + IDXD_WQCFG_MIN);
450 	dev_dbg(dev, "wqcfg size: %u\n", idxd->wqcfg_size);
451 
452 	/* reading operation capabilities */
453 	for (i = 0; i < 4; i++) {
454 		idxd->hw.opcap.bits[i] = ioread64(idxd->reg_base +
455 				IDXD_OPCAP_OFFSET + i * sizeof(u64));
456 		dev_dbg(dev, "opcap[%d]: %#llx\n", i, idxd->hw.opcap.bits[i]);
457 	}
458 	multi_u64_to_bmap(idxd->opcap_bmap, &idxd->hw.opcap.bits[0], 4);
459 }
460 
461 static struct idxd_device *idxd_alloc(struct pci_dev *pdev, struct idxd_driver_data *data)
462 {
463 	struct device *dev = &pdev->dev;
464 	struct device *conf_dev;
465 	struct idxd_device *idxd;
466 	int rc;
467 
468 	idxd = kzalloc_node(sizeof(*idxd), GFP_KERNEL, dev_to_node(dev));
469 	if (!idxd)
470 		return NULL;
471 
472 	conf_dev = idxd_confdev(idxd);
473 	idxd->pdev = pdev;
474 	idxd->data = data;
475 	idxd_dev_set_type(&idxd->idxd_dev, idxd->data->type);
476 	idxd->id = ida_alloc(&idxd_ida, GFP_KERNEL);
477 	if (idxd->id < 0)
478 		return NULL;
479 
480 	idxd->opcap_bmap = bitmap_zalloc_node(IDXD_MAX_OPCAP_BITS, GFP_KERNEL, dev_to_node(dev));
481 	if (!idxd->opcap_bmap) {
482 		ida_free(&idxd_ida, idxd->id);
483 		return NULL;
484 	}
485 
486 	device_initialize(conf_dev);
487 	conf_dev->parent = dev;
488 	conf_dev->bus = &dsa_bus_type;
489 	conf_dev->type = idxd->data->dev_type;
490 	rc = dev_set_name(conf_dev, "%s%d", idxd->data->name_prefix, idxd->id);
491 	if (rc < 0) {
492 		put_device(conf_dev);
493 		return NULL;
494 	}
495 
496 	spin_lock_init(&idxd->dev_lock);
497 	spin_lock_init(&idxd->cmd_lock);
498 
499 	return idxd;
500 }
501 
502 static int idxd_enable_system_pasid(struct idxd_device *idxd)
503 {
504 	return -EOPNOTSUPP;
505 }
506 
507 static void idxd_disable_system_pasid(struct idxd_device *idxd)
508 {
509 
510 	iommu_sva_unbind_device(idxd->sva);
511 	idxd->sva = NULL;
512 }
513 
514 static int idxd_probe(struct idxd_device *idxd)
515 {
516 	struct pci_dev *pdev = idxd->pdev;
517 	struct device *dev = &pdev->dev;
518 	int rc;
519 
520 	dev_dbg(dev, "%s entered and resetting device\n", __func__);
521 	rc = idxd_device_init_reset(idxd);
522 	if (rc < 0)
523 		return rc;
524 
525 	dev_dbg(dev, "IDXD reset complete\n");
526 
527 	if (IS_ENABLED(CONFIG_INTEL_IDXD_SVM) && sva) {
528 		if (iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA)) {
529 			dev_warn(dev, "Unable to turn on user SVA feature.\n");
530 		} else {
531 			set_bit(IDXD_FLAG_USER_PASID_ENABLED, &idxd->flags);
532 
533 			if (idxd_enable_system_pasid(idxd))
534 				dev_warn(dev, "No in-kernel DMA with PASID.\n");
535 			else
536 				set_bit(IDXD_FLAG_PASID_ENABLED, &idxd->flags);
537 		}
538 	} else if (!sva) {
539 		dev_warn(dev, "User forced SVA off via module param.\n");
540 	}
541 
542 	idxd_read_caps(idxd);
543 	idxd_read_table_offsets(idxd);
544 
545 	rc = idxd_setup_internals(idxd);
546 	if (rc)
547 		goto err;
548 
549 	/* If the configs are readonly, then load them from device */
550 	if (!test_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags)) {
551 		dev_dbg(dev, "Loading RO device config\n");
552 		rc = idxd_device_load_config(idxd);
553 		if (rc < 0)
554 			goto err_config;
555 	}
556 
557 	rc = idxd_setup_interrupts(idxd);
558 	if (rc)
559 		goto err_config;
560 
561 	idxd->major = idxd_cdev_get_major(idxd);
562 
563 	rc = perfmon_pmu_init(idxd);
564 	if (rc < 0)
565 		dev_warn(dev, "Failed to initialize perfmon. No PMU support: %d\n", rc);
566 
567 	dev_dbg(dev, "IDXD device %d probed successfully\n", idxd->id);
568 	return 0;
569 
570  err_config:
571 	idxd_cleanup_internals(idxd);
572  err:
573 	if (device_pasid_enabled(idxd))
574 		idxd_disable_system_pasid(idxd);
575 	if (device_user_pasid_enabled(idxd))
576 		iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
577 	return rc;
578 }
579 
580 static void idxd_cleanup(struct idxd_device *idxd)
581 {
582 	struct device *dev = &idxd->pdev->dev;
583 
584 	perfmon_pmu_remove(idxd);
585 	idxd_cleanup_interrupts(idxd);
586 	idxd_cleanup_internals(idxd);
587 	if (device_pasid_enabled(idxd))
588 		idxd_disable_system_pasid(idxd);
589 	if (device_user_pasid_enabled(idxd))
590 		iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
591 }
592 
593 static int idxd_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
594 {
595 	struct device *dev = &pdev->dev;
596 	struct idxd_device *idxd;
597 	struct idxd_driver_data *data = (struct idxd_driver_data *)id->driver_data;
598 	int rc;
599 
600 	rc = pci_enable_device(pdev);
601 	if (rc)
602 		return rc;
603 
604 	dev_dbg(dev, "Alloc IDXD context\n");
605 	idxd = idxd_alloc(pdev, data);
606 	if (!idxd) {
607 		rc = -ENOMEM;
608 		goto err_idxd_alloc;
609 	}
610 
611 	dev_dbg(dev, "Mapping BARs\n");
612 	idxd->reg_base = pci_iomap(pdev, IDXD_MMIO_BAR, 0);
613 	if (!idxd->reg_base) {
614 		rc = -ENOMEM;
615 		goto err_iomap;
616 	}
617 
618 	dev_dbg(dev, "Set DMA masks\n");
619 	rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
620 	if (rc)
621 		goto err;
622 
623 	dev_dbg(dev, "Set PCI master\n");
624 	pci_set_master(pdev);
625 	pci_set_drvdata(pdev, idxd);
626 
627 	idxd->hw.version = ioread32(idxd->reg_base + IDXD_VER_OFFSET);
628 	rc = idxd_probe(idxd);
629 	if (rc) {
630 		dev_err(dev, "Intel(R) IDXD DMA Engine init failed\n");
631 		goto err;
632 	}
633 
634 	rc = idxd_register_devices(idxd);
635 	if (rc) {
636 		dev_err(dev, "IDXD sysfs setup failed\n");
637 		goto err_dev_register;
638 	}
639 
640 	dev_info(&pdev->dev, "Intel(R) Accelerator Device (v%x)\n",
641 		 idxd->hw.version);
642 
643 	return 0;
644 
645  err_dev_register:
646 	idxd_cleanup(idxd);
647  err:
648 	pci_iounmap(pdev, idxd->reg_base);
649  err_iomap:
650 	put_device(idxd_confdev(idxd));
651  err_idxd_alloc:
652 	pci_disable_device(pdev);
653 	return rc;
654 }
655 
656 void idxd_wqs_quiesce(struct idxd_device *idxd)
657 {
658 	struct idxd_wq *wq;
659 	int i;
660 
661 	for (i = 0; i < idxd->max_wqs; i++) {
662 		wq = idxd->wqs[i];
663 		if (wq->state == IDXD_WQ_ENABLED && wq->type == IDXD_WQT_KERNEL)
664 			idxd_wq_quiesce(wq);
665 	}
666 }
667 
668 static void idxd_shutdown(struct pci_dev *pdev)
669 {
670 	struct idxd_device *idxd = pci_get_drvdata(pdev);
671 	struct idxd_irq_entry *irq_entry;
672 	int rc;
673 
674 	rc = idxd_device_disable(idxd);
675 	if (rc)
676 		dev_err(&pdev->dev, "Disabling device failed\n");
677 
678 	irq_entry = &idxd->ie;
679 	synchronize_irq(irq_entry->vector);
680 	idxd_mask_error_interrupts(idxd);
681 	flush_workqueue(idxd->wq);
682 }
683 
684 static void idxd_remove(struct pci_dev *pdev)
685 {
686 	struct idxd_device *idxd = pci_get_drvdata(pdev);
687 	struct idxd_irq_entry *irq_entry;
688 
689 	idxd_unregister_devices(idxd);
690 	/*
691 	 * When ->release() is called for the idxd->conf_dev, it frees all the memory related
692 	 * to the idxd context. The driver still needs those bits in order to do the rest of
693 	 * the cleanup. However, we do need to unbound the idxd sub-driver. So take a ref
694 	 * on the device here to hold off the freeing while allowing the idxd sub-driver
695 	 * to unbind.
696 	 */
697 	get_device(idxd_confdev(idxd));
698 	device_unregister(idxd_confdev(idxd));
699 	idxd_shutdown(pdev);
700 	if (device_pasid_enabled(idxd))
701 		idxd_disable_system_pasid(idxd);
702 
703 	irq_entry = idxd_get_ie(idxd, 0);
704 	free_irq(irq_entry->vector, irq_entry);
705 	pci_free_irq_vectors(pdev);
706 	pci_iounmap(pdev, idxd->reg_base);
707 	if (device_user_pasid_enabled(idxd))
708 		iommu_dev_disable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA);
709 	pci_disable_device(pdev);
710 	destroy_workqueue(idxd->wq);
711 	perfmon_pmu_remove(idxd);
712 	put_device(idxd_confdev(idxd));
713 }
714 
715 static struct pci_driver idxd_pci_driver = {
716 	.name		= DRV_NAME,
717 	.id_table	= idxd_pci_tbl,
718 	.probe		= idxd_pci_probe,
719 	.remove		= idxd_remove,
720 	.shutdown	= idxd_shutdown,
721 };
722 
723 static int __init idxd_init_module(void)
724 {
725 	int err;
726 
727 	/*
728 	 * If the CPU does not support MOVDIR64B or ENQCMDS, there's no point in
729 	 * enumerating the device. We can not utilize it.
730 	 */
731 	if (!cpu_feature_enabled(X86_FEATURE_MOVDIR64B)) {
732 		pr_warn("idxd driver failed to load without MOVDIR64B.\n");
733 		return -ENODEV;
734 	}
735 
736 	if (!cpu_feature_enabled(X86_FEATURE_ENQCMD))
737 		pr_warn("Platform does not have ENQCMD(S) support.\n");
738 	else
739 		support_enqcmd = true;
740 
741 	perfmon_init();
742 
743 	err = idxd_driver_register(&idxd_drv);
744 	if (err < 0)
745 		goto err_idxd_driver_register;
746 
747 	err = idxd_driver_register(&idxd_dmaengine_drv);
748 	if (err < 0)
749 		goto err_idxd_dmaengine_driver_register;
750 
751 	err = idxd_driver_register(&idxd_user_drv);
752 	if (err < 0)
753 		goto err_idxd_user_driver_register;
754 
755 	err = idxd_cdev_register();
756 	if (err)
757 		goto err_cdev_register;
758 
759 	err = pci_register_driver(&idxd_pci_driver);
760 	if (err)
761 		goto err_pci_register;
762 
763 	return 0;
764 
765 err_pci_register:
766 	idxd_cdev_remove();
767 err_cdev_register:
768 	idxd_driver_unregister(&idxd_user_drv);
769 err_idxd_user_driver_register:
770 	idxd_driver_unregister(&idxd_dmaengine_drv);
771 err_idxd_dmaengine_driver_register:
772 	idxd_driver_unregister(&idxd_drv);
773 err_idxd_driver_register:
774 	return err;
775 }
776 module_init(idxd_init_module);
777 
778 static void __exit idxd_exit_module(void)
779 {
780 	idxd_driver_unregister(&idxd_user_drv);
781 	idxd_driver_unregister(&idxd_dmaengine_drv);
782 	idxd_driver_unregister(&idxd_drv);
783 	pci_unregister_driver(&idxd_pci_driver);
784 	idxd_cdev_remove();
785 	perfmon_exit();
786 }
787 module_exit(idxd_exit_module);
788