xref: /openbmc/linux/drivers/dma/idxd/init.c (revision d88e2a2b)
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 		 * The default value is the same as the value of
307 		 * total read buffers in GRPCAP.
308 		 */
309 		group->rdbufs_allowed = idxd->max_rdbufs;
310 	}
311 
312 	return 0;
313 
314  err:
315 	while (--i >= 0) {
316 		group = idxd->groups[i];
317 		put_device(group_confdev(group));
318 	}
319 	return rc;
320 }
321 
322 static void idxd_cleanup_internals(struct idxd_device *idxd)
323 {
324 	int i;
325 
326 	for (i = 0; i < idxd->max_groups; i++)
327 		put_device(group_confdev(idxd->groups[i]));
328 	for (i = 0; i < idxd->max_engines; i++)
329 		put_device(engine_confdev(idxd->engines[i]));
330 	for (i = 0; i < idxd->max_wqs; i++)
331 		put_device(wq_confdev(idxd->wqs[i]));
332 	destroy_workqueue(idxd->wq);
333 }
334 
335 static int idxd_setup_internals(struct idxd_device *idxd)
336 {
337 	struct device *dev = &idxd->pdev->dev;
338 	int rc, i;
339 
340 	init_waitqueue_head(&idxd->cmd_waitq);
341 
342 	rc = idxd_setup_wqs(idxd);
343 	if (rc < 0)
344 		goto err_wqs;
345 
346 	rc = idxd_setup_engines(idxd);
347 	if (rc < 0)
348 		goto err_engine;
349 
350 	rc = idxd_setup_groups(idxd);
351 	if (rc < 0)
352 		goto err_group;
353 
354 	idxd->wq = create_workqueue(dev_name(dev));
355 	if (!idxd->wq) {
356 		rc = -ENOMEM;
357 		goto err_wkq_create;
358 	}
359 
360 	return 0;
361 
362  err_wkq_create:
363 	for (i = 0; i < idxd->max_groups; i++)
364 		put_device(group_confdev(idxd->groups[i]));
365  err_group:
366 	for (i = 0; i < idxd->max_engines; i++)
367 		put_device(engine_confdev(idxd->engines[i]));
368  err_engine:
369 	for (i = 0; i < idxd->max_wqs; i++)
370 		put_device(wq_confdev(idxd->wqs[i]));
371  err_wqs:
372 	return rc;
373 }
374 
375 static void idxd_read_table_offsets(struct idxd_device *idxd)
376 {
377 	union offsets_reg offsets;
378 	struct device *dev = &idxd->pdev->dev;
379 
380 	offsets.bits[0] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET);
381 	offsets.bits[1] = ioread64(idxd->reg_base + IDXD_TABLE_OFFSET + sizeof(u64));
382 	idxd->grpcfg_offset = offsets.grpcfg * IDXD_TABLE_MULT;
383 	dev_dbg(dev, "IDXD Group Config Offset: %#x\n", idxd->grpcfg_offset);
384 	idxd->wqcfg_offset = offsets.wqcfg * IDXD_TABLE_MULT;
385 	dev_dbg(dev, "IDXD Work Queue Config Offset: %#x\n", idxd->wqcfg_offset);
386 	idxd->msix_perm_offset = offsets.msix_perm * IDXD_TABLE_MULT;
387 	dev_dbg(dev, "IDXD MSIX Permission Offset: %#x\n", idxd->msix_perm_offset);
388 	idxd->perfmon_offset = offsets.perfmon * IDXD_TABLE_MULT;
389 	dev_dbg(dev, "IDXD Perfmon Offset: %#x\n", idxd->perfmon_offset);
390 }
391 
392 static void multi_u64_to_bmap(unsigned long *bmap, u64 *val, int count)
393 {
394 	int i, j, nr;
395 
396 	for (i = 0, nr = 0; i < count; i++) {
397 		for (j = 0; j < BITS_PER_LONG_LONG; j++) {
398 			if (val[i] & BIT(j))
399 				set_bit(nr, bmap);
400 			nr++;
401 		}
402 	}
403 }
404 
405 static void idxd_read_caps(struct idxd_device *idxd)
406 {
407 	struct device *dev = &idxd->pdev->dev;
408 	int i;
409 
410 	/* reading generic capabilities */
411 	idxd->hw.gen_cap.bits = ioread64(idxd->reg_base + IDXD_GENCAP_OFFSET);
412 	dev_dbg(dev, "gen_cap: %#llx\n", idxd->hw.gen_cap.bits);
413 
414 	if (idxd->hw.gen_cap.cmd_cap) {
415 		idxd->hw.cmd_cap = ioread32(idxd->reg_base + IDXD_CMDCAP_OFFSET);
416 		dev_dbg(dev, "cmd_cap: %#x\n", idxd->hw.cmd_cap);
417 	}
418 
419 	/* reading command capabilities */
420 	if (idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE))
421 		idxd->request_int_handles = true;
422 
423 	idxd->max_xfer_bytes = 1ULL << idxd->hw.gen_cap.max_xfer_shift;
424 	dev_dbg(dev, "max xfer size: %llu bytes\n", idxd->max_xfer_bytes);
425 	idxd_set_max_batch_size(idxd->data->type, idxd, 1U << idxd->hw.gen_cap.max_batch_shift);
426 	dev_dbg(dev, "max batch size: %u\n", idxd->max_batch_size);
427 	if (idxd->hw.gen_cap.config_en)
428 		set_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags);
429 
430 	/* reading group capabilities */
431 	idxd->hw.group_cap.bits =
432 		ioread64(idxd->reg_base + IDXD_GRPCAP_OFFSET);
433 	dev_dbg(dev, "group_cap: %#llx\n", idxd->hw.group_cap.bits);
434 	idxd->max_groups = idxd->hw.group_cap.num_groups;
435 	dev_dbg(dev, "max groups: %u\n", idxd->max_groups);
436 	idxd->max_rdbufs = idxd->hw.group_cap.total_rdbufs;
437 	dev_dbg(dev, "max read buffers: %u\n", idxd->max_rdbufs);
438 	idxd->nr_rdbufs = idxd->max_rdbufs;
439 
440 	/* read engine capabilities */
441 	idxd->hw.engine_cap.bits =
442 		ioread64(idxd->reg_base + IDXD_ENGCAP_OFFSET);
443 	dev_dbg(dev, "engine_cap: %#llx\n", idxd->hw.engine_cap.bits);
444 	idxd->max_engines = idxd->hw.engine_cap.num_engines;
445 	dev_dbg(dev, "max engines: %u\n", idxd->max_engines);
446 
447 	/* read workqueue capabilities */
448 	idxd->hw.wq_cap.bits = ioread64(idxd->reg_base + IDXD_WQCAP_OFFSET);
449 	dev_dbg(dev, "wq_cap: %#llx\n", idxd->hw.wq_cap.bits);
450 	idxd->max_wq_size = idxd->hw.wq_cap.total_wq_size;
451 	dev_dbg(dev, "total workqueue size: %u\n", idxd->max_wq_size);
452 	idxd->max_wqs = idxd->hw.wq_cap.num_wqs;
453 	dev_dbg(dev, "max workqueues: %u\n", idxd->max_wqs);
454 	idxd->wqcfg_size = 1 << (idxd->hw.wq_cap.wqcfg_size + IDXD_WQCFG_MIN);
455 	dev_dbg(dev, "wqcfg size: %u\n", idxd->wqcfg_size);
456 
457 	/* reading operation capabilities */
458 	for (i = 0; i < 4; i++) {
459 		idxd->hw.opcap.bits[i] = ioread64(idxd->reg_base +
460 				IDXD_OPCAP_OFFSET + i * sizeof(u64));
461 		dev_dbg(dev, "opcap[%d]: %#llx\n", i, idxd->hw.opcap.bits[i]);
462 	}
463 	multi_u64_to_bmap(idxd->opcap_bmap, &idxd->hw.opcap.bits[0], 4);
464 }
465 
466 static struct idxd_device *idxd_alloc(struct pci_dev *pdev, struct idxd_driver_data *data)
467 {
468 	struct device *dev = &pdev->dev;
469 	struct device *conf_dev;
470 	struct idxd_device *idxd;
471 	int rc;
472 
473 	idxd = kzalloc_node(sizeof(*idxd), GFP_KERNEL, dev_to_node(dev));
474 	if (!idxd)
475 		return NULL;
476 
477 	conf_dev = idxd_confdev(idxd);
478 	idxd->pdev = pdev;
479 	idxd->data = data;
480 	idxd_dev_set_type(&idxd->idxd_dev, idxd->data->type);
481 	idxd->id = ida_alloc(&idxd_ida, GFP_KERNEL);
482 	if (idxd->id < 0)
483 		return NULL;
484 
485 	idxd->opcap_bmap = bitmap_zalloc_node(IDXD_MAX_OPCAP_BITS, GFP_KERNEL, dev_to_node(dev));
486 	if (!idxd->opcap_bmap) {
487 		ida_free(&idxd_ida, idxd->id);
488 		return NULL;
489 	}
490 
491 	device_initialize(conf_dev);
492 	conf_dev->parent = dev;
493 	conf_dev->bus = &dsa_bus_type;
494 	conf_dev->type = idxd->data->dev_type;
495 	rc = dev_set_name(conf_dev, "%s%d", idxd->data->name_prefix, idxd->id);
496 	if (rc < 0) {
497 		put_device(conf_dev);
498 		return NULL;
499 	}
500 
501 	spin_lock_init(&idxd->dev_lock);
502 	spin_lock_init(&idxd->cmd_lock);
503 
504 	return idxd;
505 }
506 
507 static int idxd_enable_system_pasid(struct idxd_device *idxd)
508 {
509 	return -EOPNOTSUPP;
510 }
511 
512 static void idxd_disable_system_pasid(struct idxd_device *idxd)
513 {
514 
515 	iommu_sva_unbind_device(idxd->sva);
516 	idxd->sva = NULL;
517 }
518 
519 static int idxd_probe(struct idxd_device *idxd)
520 {
521 	struct pci_dev *pdev = idxd->pdev;
522 	struct device *dev = &pdev->dev;
523 	int rc;
524 
525 	dev_dbg(dev, "%s entered and resetting device\n", __func__);
526 	rc = idxd_device_init_reset(idxd);
527 	if (rc < 0)
528 		return rc;
529 
530 	dev_dbg(dev, "IDXD reset complete\n");
531 
532 	if (IS_ENABLED(CONFIG_INTEL_IDXD_SVM) && sva) {
533 		if (iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA)) {
534 			dev_warn(dev, "Unable to turn on user SVA feature.\n");
535 		} else {
536 			set_bit(IDXD_FLAG_USER_PASID_ENABLED, &idxd->flags);
537 
538 			if (idxd_enable_system_pasid(idxd))
539 				dev_warn(dev, "No in-kernel DMA with PASID.\n");
540 			else
541 				set_bit(IDXD_FLAG_PASID_ENABLED, &idxd->flags);
542 		}
543 	} else if (!sva) {
544 		dev_warn(dev, "User forced SVA off via module param.\n");
545 	}
546 
547 	idxd_read_caps(idxd);
548 	idxd_read_table_offsets(idxd);
549 
550 	rc = idxd_setup_internals(idxd);
551 	if (rc)
552 		goto err;
553 
554 	/* If the configs are readonly, then load them from device */
555 	if (!test_bit(IDXD_FLAG_CONFIGURABLE, &idxd->flags)) {
556 		dev_dbg(dev, "Loading RO device config\n");
557 		rc = idxd_device_load_config(idxd);
558 		if (rc < 0)
559 			goto err_config;
560 	}
561 
562 	rc = idxd_setup_interrupts(idxd);
563 	if (rc)
564 		goto err_config;
565 
566 	idxd->major = idxd_cdev_get_major(idxd);
567 
568 	rc = perfmon_pmu_init(idxd);
569 	if (rc < 0)
570 		dev_warn(dev, "Failed to initialize perfmon. No PMU support: %d\n", rc);
571 
572 	dev_dbg(dev, "IDXD device %d probed successfully\n", idxd->id);
573 	return 0;
574 
575  err_config:
576 	idxd_cleanup_internals(idxd);
577  err:
578 	if (device_pasid_enabled(idxd))
579 		idxd_disable_system_pasid(idxd);
580 	if (device_user_pasid_enabled(idxd))
581 		iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
582 	return rc;
583 }
584 
585 static void idxd_cleanup(struct idxd_device *idxd)
586 {
587 	struct device *dev = &idxd->pdev->dev;
588 
589 	perfmon_pmu_remove(idxd);
590 	idxd_cleanup_interrupts(idxd);
591 	idxd_cleanup_internals(idxd);
592 	if (device_pasid_enabled(idxd))
593 		idxd_disable_system_pasid(idxd);
594 	if (device_user_pasid_enabled(idxd))
595 		iommu_dev_disable_feature(dev, IOMMU_DEV_FEAT_SVA);
596 }
597 
598 static int idxd_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
599 {
600 	struct device *dev = &pdev->dev;
601 	struct idxd_device *idxd;
602 	struct idxd_driver_data *data = (struct idxd_driver_data *)id->driver_data;
603 	int rc;
604 
605 	rc = pci_enable_device(pdev);
606 	if (rc)
607 		return rc;
608 
609 	dev_dbg(dev, "Alloc IDXD context\n");
610 	idxd = idxd_alloc(pdev, data);
611 	if (!idxd) {
612 		rc = -ENOMEM;
613 		goto err_idxd_alloc;
614 	}
615 
616 	dev_dbg(dev, "Mapping BARs\n");
617 	idxd->reg_base = pci_iomap(pdev, IDXD_MMIO_BAR, 0);
618 	if (!idxd->reg_base) {
619 		rc = -ENOMEM;
620 		goto err_iomap;
621 	}
622 
623 	dev_dbg(dev, "Set DMA masks\n");
624 	rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
625 	if (rc)
626 		goto err;
627 
628 	dev_dbg(dev, "Set PCI master\n");
629 	pci_set_master(pdev);
630 	pci_set_drvdata(pdev, idxd);
631 
632 	idxd->hw.version = ioread32(idxd->reg_base + IDXD_VER_OFFSET);
633 	rc = idxd_probe(idxd);
634 	if (rc) {
635 		dev_err(dev, "Intel(R) IDXD DMA Engine init failed\n");
636 		goto err;
637 	}
638 
639 	rc = idxd_register_devices(idxd);
640 	if (rc) {
641 		dev_err(dev, "IDXD sysfs setup failed\n");
642 		goto err_dev_register;
643 	}
644 
645 	dev_info(&pdev->dev, "Intel(R) Accelerator Device (v%x)\n",
646 		 idxd->hw.version);
647 
648 	return 0;
649 
650  err_dev_register:
651 	idxd_cleanup(idxd);
652  err:
653 	pci_iounmap(pdev, idxd->reg_base);
654  err_iomap:
655 	put_device(idxd_confdev(idxd));
656  err_idxd_alloc:
657 	pci_disable_device(pdev);
658 	return rc;
659 }
660 
661 void idxd_wqs_quiesce(struct idxd_device *idxd)
662 {
663 	struct idxd_wq *wq;
664 	int i;
665 
666 	for (i = 0; i < idxd->max_wqs; i++) {
667 		wq = idxd->wqs[i];
668 		if (wq->state == IDXD_WQ_ENABLED && wq->type == IDXD_WQT_KERNEL)
669 			idxd_wq_quiesce(wq);
670 	}
671 }
672 
673 static void idxd_shutdown(struct pci_dev *pdev)
674 {
675 	struct idxd_device *idxd = pci_get_drvdata(pdev);
676 	struct idxd_irq_entry *irq_entry;
677 	int rc;
678 
679 	rc = idxd_device_disable(idxd);
680 	if (rc)
681 		dev_err(&pdev->dev, "Disabling device failed\n");
682 
683 	irq_entry = &idxd->ie;
684 	synchronize_irq(irq_entry->vector);
685 	idxd_mask_error_interrupts(idxd);
686 	flush_workqueue(idxd->wq);
687 }
688 
689 static void idxd_remove(struct pci_dev *pdev)
690 {
691 	struct idxd_device *idxd = pci_get_drvdata(pdev);
692 	struct idxd_irq_entry *irq_entry;
693 
694 	idxd_unregister_devices(idxd);
695 	/*
696 	 * When ->release() is called for the idxd->conf_dev, it frees all the memory related
697 	 * to the idxd context. The driver still needs those bits in order to do the rest of
698 	 * the cleanup. However, we do need to unbound the idxd sub-driver. So take a ref
699 	 * on the device here to hold off the freeing while allowing the idxd sub-driver
700 	 * to unbind.
701 	 */
702 	get_device(idxd_confdev(idxd));
703 	device_unregister(idxd_confdev(idxd));
704 	idxd_shutdown(pdev);
705 	if (device_pasid_enabled(idxd))
706 		idxd_disable_system_pasid(idxd);
707 
708 	irq_entry = idxd_get_ie(idxd, 0);
709 	free_irq(irq_entry->vector, irq_entry);
710 	pci_free_irq_vectors(pdev);
711 	pci_iounmap(pdev, idxd->reg_base);
712 	if (device_user_pasid_enabled(idxd))
713 		iommu_dev_disable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA);
714 	pci_disable_device(pdev);
715 	destroy_workqueue(idxd->wq);
716 	perfmon_pmu_remove(idxd);
717 	put_device(idxd_confdev(idxd));
718 }
719 
720 static struct pci_driver idxd_pci_driver = {
721 	.name		= DRV_NAME,
722 	.id_table	= idxd_pci_tbl,
723 	.probe		= idxd_pci_probe,
724 	.remove		= idxd_remove,
725 	.shutdown	= idxd_shutdown,
726 };
727 
728 static int __init idxd_init_module(void)
729 {
730 	int err;
731 
732 	/*
733 	 * If the CPU does not support MOVDIR64B or ENQCMDS, there's no point in
734 	 * enumerating the device. We can not utilize it.
735 	 */
736 	if (!cpu_feature_enabled(X86_FEATURE_MOVDIR64B)) {
737 		pr_warn("idxd driver failed to load without MOVDIR64B.\n");
738 		return -ENODEV;
739 	}
740 
741 	if (!cpu_feature_enabled(X86_FEATURE_ENQCMD))
742 		pr_warn("Platform does not have ENQCMD(S) support.\n");
743 	else
744 		support_enqcmd = true;
745 
746 	perfmon_init();
747 
748 	err = idxd_driver_register(&idxd_drv);
749 	if (err < 0)
750 		goto err_idxd_driver_register;
751 
752 	err = idxd_driver_register(&idxd_dmaengine_drv);
753 	if (err < 0)
754 		goto err_idxd_dmaengine_driver_register;
755 
756 	err = idxd_driver_register(&idxd_user_drv);
757 	if (err < 0)
758 		goto err_idxd_user_driver_register;
759 
760 	err = idxd_cdev_register();
761 	if (err)
762 		goto err_cdev_register;
763 
764 	err = pci_register_driver(&idxd_pci_driver);
765 	if (err)
766 		goto err_pci_register;
767 
768 	return 0;
769 
770 err_pci_register:
771 	idxd_cdev_remove();
772 err_cdev_register:
773 	idxd_driver_unregister(&idxd_user_drv);
774 err_idxd_user_driver_register:
775 	idxd_driver_unregister(&idxd_dmaengine_drv);
776 err_idxd_dmaengine_driver_register:
777 	idxd_driver_unregister(&idxd_drv);
778 err_idxd_driver_register:
779 	return err;
780 }
781 module_init(idxd_init_module);
782 
783 static void __exit idxd_exit_module(void)
784 {
785 	idxd_driver_unregister(&idxd_user_drv);
786 	idxd_driver_unregister(&idxd_dmaengine_drv);
787 	idxd_driver_unregister(&idxd_drv);
788 	pci_unregister_driver(&idxd_pci_driver);
789 	idxd_cdev_remove();
790 	perfmon_exit();
791 }
792 module_exit(idxd_exit_module);
793