xref: /linux/drivers/accel/qaic/qaic_drv.c (revision 6101f78b684895d5860a96322e607e0f46f433ad)
1 // SPDX-License-Identifier: GPL-2.0-only
2 
3 /* Copyright (c) 2019-2021, The Linux Foundation. All rights reserved. */
4 /* Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries. */
5 
6 #include <linux/delay.h>
7 #include <linux/dma-mapping.h>
8 #include <linux/idr.h>
9 #include <linux/interrupt.h>
10 #include <linux/list.h>
11 #include <linux/kobject.h>
12 #include <linux/kref.h>
13 #include <linux/mhi.h>
14 #include <linux/module.h>
15 #include <linux/msi.h>
16 #include <linux/mutex.h>
17 #include <linux/pci.h>
18 #include <linux/spinlock.h>
19 #include <linux/workqueue.h>
20 #include <linux/wait.h>
21 #include <drm/drm_accel.h>
22 #include <drm/drm_drv.h>
23 #include <drm/drm_file.h>
24 #include <drm/drm_gem.h>
25 #include <drm/drm_ioctl.h>
26 #include <drm/drm_managed.h>
27 #include <uapi/drm/qaic_accel.h>
28 
29 #include "mhi_controller.h"
30 #include "qaic.h"
31 #include "qaic_debugfs.h"
32 #include "qaic_ras.h"
33 #include "qaic_ssr.h"
34 #include "qaic_timesync.h"
35 #include "sahara.h"
36 
37 MODULE_IMPORT_NS("DMA_BUF");
38 
39 #define PCI_DEVICE_ID_QCOM_AIC080	0xa080
40 #define PCI_DEVICE_ID_QCOM_AIC100	0xa100
41 #define PCI_DEVICE_ID_QCOM_AIC200	0xa110
42 #define QAIC_NAME			"qaic"
43 #define QAIC_DESC			"Qualcomm Cloud AI Accelerators"
44 #define CNTL_MAJOR			5
45 #define CNTL_MINOR			0
46 #define DBC_NUM				16
47 
48 struct qaic_device_config {
49 	/* Indicates the AIC family the device belongs to */
50 	int family;
51 	/* A bitmask representing the available BARs */
52 	int bar_mask;
53 	/* An index value used to identify the MHI controller BAR */
54 	unsigned int mhi_bar_idx;
55 	/* An index value used to identify the DBCs BAR */
56 	unsigned int dbc_bar_idx;
57 };
58 
59 static const struct qaic_device_config aic080_config = {
60 	.family = FAMILY_AIC100,
61 	.bar_mask = BIT(0) | BIT(2) | BIT(4),
62 	.mhi_bar_idx = 0,
63 	.dbc_bar_idx = 2,
64 };
65 
66 static const struct qaic_device_config aic100_config = {
67 	.family = FAMILY_AIC100,
68 	.bar_mask = BIT(0) | BIT(2) | BIT(4),
69 	.mhi_bar_idx = 0,
70 	.dbc_bar_idx = 2,
71 };
72 
73 static const struct qaic_device_config aic200_config = {
74 	.family = FAMILY_AIC200,
75 	.bar_mask = BIT(0) | BIT(1) | BIT(2) | BIT(4),
76 	.mhi_bar_idx = 1,
77 	.dbc_bar_idx = 2,
78 };
79 
80 bool datapath_polling;
81 module_param(datapath_polling, bool, 0400);
82 MODULE_PARM_DESC(datapath_polling, "Operate the datapath in polling mode");
83 static bool link_up;
84 static DEFINE_IDA(qaic_usrs);
85 
86 static void qaicm_wq_release(struct drm_device *dev, void *res)
87 {
88 	struct workqueue_struct *wq = res;
89 
90 	destroy_workqueue(wq);
91 }
92 
93 static struct workqueue_struct *qaicm_wq_init(struct drm_device *dev, const char *name)
94 {
95 	struct workqueue_struct *wq;
96 	int ret;
97 
98 	wq = alloc_workqueue("%s", WQ_UNBOUND, 0, name);
99 	if (!wq)
100 		return ERR_PTR(-ENOMEM);
101 	ret = drmm_add_action_or_reset(dev, qaicm_wq_release, wq);
102 	if (ret)
103 		return ERR_PTR(ret);
104 
105 	return wq;
106 }
107 
108 static void qaicm_srcu_release(struct drm_device *dev, void *res)
109 {
110 	struct srcu_struct *lock = res;
111 
112 	cleanup_srcu_struct(lock);
113 }
114 
115 static int qaicm_srcu_init(struct drm_device *dev, struct srcu_struct *lock)
116 {
117 	int ret;
118 
119 	ret = init_srcu_struct(lock);
120 	if (ret)
121 		return ret;
122 
123 	return drmm_add_action_or_reset(dev, qaicm_srcu_release, lock);
124 }
125 
126 static void qaicm_pci_release(struct drm_device *dev, void *res)
127 {
128 	struct qaic_device *qdev = to_qaic_device(dev);
129 
130 	pci_set_drvdata(qdev->pdev, NULL);
131 }
132 
133 static void free_usr(struct kref *kref)
134 {
135 	struct qaic_user *usr = container_of(kref, struct qaic_user, ref_count);
136 
137 	cleanup_srcu_struct(&usr->qddev_lock);
138 	ida_free(&qaic_usrs, usr->handle);
139 	kfree(usr);
140 }
141 
142 static int qaic_open(struct drm_device *dev, struct drm_file *file)
143 {
144 	struct qaic_drm_device *qddev = to_qaic_drm_device(dev);
145 	struct qaic_device *qdev = qddev->qdev;
146 	struct qaic_user *usr;
147 	int rcu_id;
148 	int ret;
149 
150 	rcu_id = srcu_read_lock(&qdev->dev_lock);
151 	if (qdev->dev_state != QAIC_ONLINE) {
152 		ret = -ENODEV;
153 		goto dev_unlock;
154 	}
155 
156 	usr = kmalloc_obj(*usr);
157 	if (!usr) {
158 		ret = -ENOMEM;
159 		goto dev_unlock;
160 	}
161 
162 	usr->handle = ida_alloc(&qaic_usrs, GFP_KERNEL);
163 	if (usr->handle < 0) {
164 		ret = usr->handle;
165 		goto free_usr;
166 	}
167 	usr->qddev = qddev;
168 	atomic_set(&usr->chunk_id, 0);
169 	init_srcu_struct(&usr->qddev_lock);
170 	kref_init(&usr->ref_count);
171 
172 	ret = mutex_lock_interruptible(&qddev->users_mutex);
173 	if (ret)
174 		goto cleanup_usr;
175 
176 	list_add(&usr->node, &qddev->users);
177 	mutex_unlock(&qddev->users_mutex);
178 
179 	file->driver_priv = usr;
180 
181 	srcu_read_unlock(&qdev->dev_lock, rcu_id);
182 	return 0;
183 
184 cleanup_usr:
185 	cleanup_srcu_struct(&usr->qddev_lock);
186 	ida_free(&qaic_usrs, usr->handle);
187 free_usr:
188 	kfree(usr);
189 dev_unlock:
190 	srcu_read_unlock(&qdev->dev_lock, rcu_id);
191 	return ret;
192 }
193 
194 static void qaic_postclose(struct drm_device *dev, struct drm_file *file)
195 {
196 	struct qaic_user *usr = file->driver_priv;
197 	struct qaic_drm_device *qddev;
198 	struct qaic_device *qdev;
199 	int qdev_rcu_id;
200 	int usr_rcu_id;
201 	int i;
202 
203 	qddev = usr->qddev;
204 	usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
205 	if (qddev) {
206 		qdev = qddev->qdev;
207 		qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
208 		if (qdev->dev_state == QAIC_ONLINE) {
209 			qaic_release_usr(qdev, usr);
210 			for (i = 0; i < qdev->num_dbc; ++i)
211 				if (qdev->dbc[i].usr && qdev->dbc[i].usr->handle == usr->handle)
212 					release_dbc(qdev, i);
213 		}
214 		srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
215 
216 		mutex_lock(&qddev->users_mutex);
217 		if (!list_empty(&usr->node))
218 			list_del_init(&usr->node);
219 		mutex_unlock(&qddev->users_mutex);
220 	}
221 
222 	srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
223 	kref_put(&usr->ref_count, free_usr);
224 
225 	file->driver_priv = NULL;
226 }
227 
228 DEFINE_DRM_ACCEL_FOPS(qaic_accel_fops);
229 
230 static const struct drm_ioctl_desc qaic_drm_ioctls[] = {
231 	DRM_IOCTL_DEF_DRV(QAIC_MANAGE, qaic_manage_ioctl, 0),
232 	DRM_IOCTL_DEF_DRV(QAIC_CREATE_BO, qaic_create_bo_ioctl, 0),
233 	DRM_IOCTL_DEF_DRV(QAIC_MMAP_BO, qaic_mmap_bo_ioctl, 0),
234 	DRM_IOCTL_DEF_DRV(QAIC_ATTACH_SLICE_BO, qaic_attach_slice_bo_ioctl, 0),
235 	DRM_IOCTL_DEF_DRV(QAIC_EXECUTE_BO, qaic_execute_bo_ioctl, 0),
236 	DRM_IOCTL_DEF_DRV(QAIC_PARTIAL_EXECUTE_BO, qaic_partial_execute_bo_ioctl, 0),
237 	DRM_IOCTL_DEF_DRV(QAIC_WAIT_BO, qaic_wait_bo_ioctl, 0),
238 	DRM_IOCTL_DEF_DRV(QAIC_PERF_STATS_BO, qaic_perf_stats_bo_ioctl, 0),
239 	DRM_IOCTL_DEF_DRV(QAIC_DETACH_SLICE_BO, qaic_detach_slice_bo_ioctl, 0),
240 };
241 
242 static const struct drm_driver qaic_accel_driver = {
243 	.driver_features	= DRIVER_GEM | DRIVER_COMPUTE_ACCEL,
244 
245 	.name			= QAIC_NAME,
246 	.desc			= QAIC_DESC,
247 
248 	.fops			= &qaic_accel_fops,
249 	.open			= qaic_open,
250 	.postclose		= qaic_postclose,
251 
252 	.ioctls			= qaic_drm_ioctls,
253 	.num_ioctls		= ARRAY_SIZE(qaic_drm_ioctls),
254 	.gem_prime_import	= qaic_gem_prime_import,
255 };
256 
257 static int qaic_create_drm_device(struct qaic_device *qdev, s32 partition_id)
258 {
259 	struct qaic_drm_device *qddev = qdev->qddev;
260 	struct drm_device *drm = to_drm(qddev);
261 	int ret;
262 
263 	/* Hold off implementing partitions until the uapi is determined */
264 	if (partition_id != QAIC_NO_PARTITION)
265 		return -EINVAL;
266 
267 	qddev->partition_id = partition_id;
268 
269 	ret = drm_dev_register(drm, 0);
270 	if (ret) {
271 		pci_dbg(qdev->pdev, "drm_dev_register failed %d\n", ret);
272 		return ret;
273 	}
274 
275 	ret = qaic_sysfs_init(qddev);
276 	if (ret) {
277 		drm_dev_unregister(drm);
278 		pci_dbg(qdev->pdev, "qaic_sysfs_init failed %d\n", ret);
279 		return ret;
280 	}
281 
282 	qaic_debugfs_init(qddev);
283 
284 	return ret;
285 }
286 
287 static void qaic_destroy_drm_device(struct qaic_device *qdev, s32 partition_id)
288 {
289 	struct qaic_drm_device *qddev = qdev->qddev;
290 	struct drm_device *drm = to_drm(qddev);
291 	struct qaic_user *usr;
292 
293 	qaic_sysfs_remove(qddev);
294 	drm_dev_unregister(drm);
295 	qddev->partition_id = 0;
296 	/*
297 	 * Existing users get unresolvable errors till they close FDs.
298 	 * Need to sync carefully with users calling close(). The
299 	 * list of users can be modified elsewhere when the lock isn't
300 	 * held here, but the sync'ing the srcu with the mutex held
301 	 * could deadlock. Grab the mutex so that the list will be
302 	 * unmodified. The user we get will exist as long as the
303 	 * lock is held. Signal that the qcdev is going away, and
304 	 * grab a reference to the user so they don't go away for
305 	 * synchronize_srcu(). Then release the mutex to avoid
306 	 * deadlock and make sure the user has observed the signal.
307 	 * With the lock released, we cannot maintain any state of the
308 	 * user list.
309 	 */
310 	mutex_lock(&qddev->users_mutex);
311 	while (!list_empty(&qddev->users)) {
312 		usr = list_first_entry(&qddev->users, struct qaic_user, node);
313 		list_del_init(&usr->node);
314 		kref_get(&usr->ref_count);
315 		usr->qddev = NULL;
316 		mutex_unlock(&qddev->users_mutex);
317 		synchronize_srcu(&usr->qddev_lock);
318 		kref_put(&usr->ref_count, free_usr);
319 		mutex_lock(&qddev->users_mutex);
320 	}
321 	mutex_unlock(&qddev->users_mutex);
322 }
323 
324 static int qaic_mhi_probe(struct mhi_device *mhi_dev, const struct mhi_device_id *id)
325 {
326 	u16 major = -1, minor = -1;
327 	struct qaic_device *qdev;
328 	int ret;
329 
330 	/*
331 	 * Invoking this function indicates that the control channel to the
332 	 * device is available. We use that as a signal to indicate that
333 	 * the device side firmware has booted. The device side firmware
334 	 * manages the device resources, so we need to communicate with it
335 	 * via the control channel in order to utilize the device. Therefore
336 	 * we wait until this signal to create the drm dev that userspace will
337 	 * use to control the device, because without the device side firmware,
338 	 * userspace can't do anything useful.
339 	 */
340 
341 	qdev = pci_get_drvdata(to_pci_dev(mhi_dev->mhi_cntrl->cntrl_dev));
342 
343 	dev_set_drvdata(&mhi_dev->dev, qdev);
344 	qdev->cntl_ch = mhi_dev;
345 
346 	ret = qaic_control_open(qdev);
347 	if (ret) {
348 		pci_dbg(qdev->pdev, "%s: control_open failed %d\n", __func__, ret);
349 		return ret;
350 	}
351 
352 	qdev->dev_state = QAIC_BOOT;
353 	ret = get_cntl_version(qdev, NULL, &major, &minor);
354 	if (ret || major != CNTL_MAJOR || minor > CNTL_MINOR) {
355 		pci_err(qdev->pdev, "%s: Control protocol version (%d.%d) not supported. Supported version is (%d.%d). Ret: %d\n",
356 			__func__, major, minor, CNTL_MAJOR, CNTL_MINOR, ret);
357 		ret = -EINVAL;
358 		goto close_control;
359 	}
360 	qdev->dev_state = QAIC_ONLINE;
361 	kobject_uevent(&(to_accel_kdev(qdev->qddev))->kobj, KOBJ_ONLINE);
362 
363 	return ret;
364 
365 close_control:
366 	qaic_control_close(qdev);
367 	return ret;
368 }
369 
370 static void qaic_mhi_remove(struct mhi_device *mhi_dev)
371 {
372 /* This is redundant since we have already observed the device crash */
373 }
374 
375 static void qaic_notify_reset(struct qaic_device *qdev)
376 {
377 	int i;
378 
379 	kobject_uevent(&(to_accel_kdev(qdev->qddev))->kobj, KOBJ_OFFLINE);
380 	qdev->dev_state = QAIC_OFFLINE;
381 	/* wake up any waiters to avoid waiting for timeouts at sync */
382 	wake_all_cntl(qdev);
383 	for (i = 0; i < qdev->num_dbc; ++i)
384 		wakeup_dbc(qdev, i);
385 	synchronize_srcu(&qdev->dev_lock);
386 }
387 
388 void qaic_dev_reset_clean_local_state(struct qaic_device *qdev)
389 {
390 	int i;
391 
392 	qaic_notify_reset(qdev);
393 
394 	/* start tearing things down */
395 	qaic_clean_up_ssr(qdev);
396 	for (i = 0; i < qdev->num_dbc; ++i)
397 		release_dbc(qdev, i);
398 }
399 
400 static struct qaic_device *create_qdev(struct pci_dev *pdev,
401 				       const struct qaic_device_config *config)
402 {
403 	struct device *dev = &pdev->dev;
404 	struct qaic_drm_device *qddev;
405 	struct qaic_device *qdev;
406 	struct drm_device *drm;
407 	int i, ret;
408 
409 	qdev = devm_kzalloc(dev, struct_size(qdev, dbc, DBC_NUM), GFP_KERNEL);
410 	if (!qdev)
411 		return NULL;
412 
413 	qdev->num_dbc = DBC_NUM;
414 	qdev->dev_state = QAIC_OFFLINE;
415 
416 	qddev = devm_drm_dev_alloc(&pdev->dev, &qaic_accel_driver, struct qaic_drm_device, drm);
417 	if (IS_ERR(qddev))
418 		return NULL;
419 
420 	drm = to_drm(qddev);
421 	pci_set_drvdata(pdev, qdev);
422 
423 	ret = drmm_mutex_init(drm, &qddev->users_mutex);
424 	if (ret)
425 		return NULL;
426 	ret = drmm_add_action_or_reset(drm, qaicm_pci_release, NULL);
427 	if (ret)
428 		return NULL;
429 	ret = drmm_mutex_init(drm, &qdev->cntl_mutex);
430 	if (ret)
431 		return NULL;
432 	ret = drmm_mutex_init(drm, &qdev->bootlog_mutex);
433 	if (ret)
434 		return NULL;
435 
436 	qdev->cntl_wq = qaicm_wq_init(drm, "qaic_cntl");
437 	if (IS_ERR(qdev->cntl_wq))
438 		return NULL;
439 	qdev->qts_wq = qaicm_wq_init(drm, "qaic_ts");
440 	if (IS_ERR(qdev->qts_wq))
441 		return NULL;
442 	qdev->ssr_wq = qaicm_wq_init(drm, "qaic_ssr");
443 	if (IS_ERR(qdev->ssr_wq))
444 		return NULL;
445 
446 	ret = qaicm_srcu_init(drm, &qdev->dev_lock);
447 	if (ret)
448 		return NULL;
449 
450 	ret = qaic_ssr_init(qdev, drm);
451 	if (ret)
452 		pci_info(pdev, "QAIC SSR crashdump collection not supported.\n");
453 
454 	qdev->qddev = qddev;
455 	qdev->pdev = pdev;
456 	qddev->qdev = qdev;
457 
458 	INIT_LIST_HEAD(&qdev->cntl_xfer_list);
459 	INIT_LIST_HEAD(&qdev->bootlog);
460 	INIT_LIST_HEAD(&qddev->users);
461 
462 	for (i = 0; i < qdev->num_dbc; ++i) {
463 		spin_lock_init(&qdev->dbc[i].xfer_lock);
464 		qdev->dbc[i].qdev = qdev;
465 		qdev->dbc[i].id = i;
466 		INIT_LIST_HEAD(&qdev->dbc[i].xfer_list);
467 		ret = qaicm_srcu_init(drm, &qdev->dbc[i].ch_lock);
468 		if (ret)
469 			return NULL;
470 		init_waitqueue_head(&qdev->dbc[i].dbc_release);
471 		INIT_LIST_HEAD(&qdev->dbc[i].bo_lists);
472 		ret = drmm_mutex_init(drm, &qdev->dbc[i].req_lock);
473 		if (ret)
474 			return NULL;
475 	}
476 
477 	return qdev;
478 }
479 
480 static int init_pci(struct qaic_device *qdev, struct pci_dev *pdev,
481 		    const struct qaic_device_config *config)
482 {
483 	int bars;
484 	int ret;
485 
486 	bars = pci_select_bars(pdev, IORESOURCE_MEM) & 0x3f;
487 
488 	/* make sure the device has the expected BARs */
489 	if (bars != config->bar_mask) {
490 		pci_dbg(pdev, "%s: expected BARs %#x not found in device. Found %#x\n",
491 			__func__, config->bar_mask, bars);
492 		return -EINVAL;
493 	}
494 
495 	ret = pcim_enable_device(pdev);
496 	if (ret)
497 		return ret;
498 
499 	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
500 	if (ret)
501 		return ret;
502 	dma_set_max_seg_size(&pdev->dev, UINT_MAX);
503 
504 	qdev->bar_mhi = devm_ioremap_resource(&pdev->dev, &pdev->resource[config->mhi_bar_idx]);
505 	if (IS_ERR(qdev->bar_mhi))
506 		return PTR_ERR(qdev->bar_mhi);
507 
508 	qdev->bar_dbc = devm_ioremap_resource(&pdev->dev, &pdev->resource[config->dbc_bar_idx]);
509 	if (IS_ERR(qdev->bar_dbc))
510 		return PTR_ERR(qdev->bar_dbc);
511 
512 	/* Managed release since we use pcim_enable_device above */
513 	pci_set_master(pdev);
514 
515 	return 0;
516 }
517 
518 static int init_msi(struct qaic_device *qdev, struct pci_dev *pdev)
519 {
520 	int irq_count = qdev->num_dbc + 1;
521 	int mhi_irq;
522 	int ret;
523 	int i;
524 
525 	/* Managed release since we use pcim_enable_device */
526 	ret = pci_alloc_irq_vectors(pdev, irq_count, irq_count, PCI_IRQ_MSI | PCI_IRQ_MSIX);
527 	if (ret == -ENOSPC) {
528 		ret = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI | PCI_IRQ_MSIX);
529 		if (ret < 0)
530 			return ret;
531 
532 		/*
533 		 * Operate in one MSI mode. All interrupts will be directed to
534 		 * MSI0; every interrupt will wake up all the interrupt handlers
535 		 * (MHI and DBC[0-15]). Since the interrupt is now shared, it is
536 		 * not disabled during DBC threaded handler, but only one thread
537 		 * will be allowed to run per DBC, so while it can be
538 		 * interrupted, it shouldn't race with itself.
539 		 */
540 		qdev->single_msi = true;
541 		pci_info(pdev, "Allocating %d MSIs failed, operating in 1 MSI mode. Performance may be impacted.\n",
542 			 irq_count);
543 	} else if (ret < 0) {
544 		return ret;
545 	}
546 
547 	mhi_irq = pci_irq_vector(pdev, 0);
548 	if (mhi_irq < 0)
549 		return mhi_irq;
550 
551 	for (i = 0; i < qdev->num_dbc; ++i) {
552 		ret = devm_request_threaded_irq(&pdev->dev,
553 						pci_irq_vector(pdev, qdev->single_msi ? 0 : i + 1),
554 						dbc_irq_handler, dbc_irq_threaded_fn, IRQF_SHARED,
555 						"qaic_dbc", &qdev->dbc[i]);
556 		if (ret)
557 			return ret;
558 
559 		if (datapath_polling) {
560 			qdev->dbc[i].irq = pci_irq_vector(pdev, qdev->single_msi ? 0 : i + 1);
561 			if (!qdev->single_msi)
562 				disable_irq_nosync(qdev->dbc[i].irq);
563 			INIT_WORK(&qdev->dbc[i].poll_work, qaic_irq_polling_work);
564 		}
565 	}
566 
567 	return mhi_irq;
568 }
569 
570 static int qaic_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
571 {
572 	struct qaic_device_config *config = (struct qaic_device_config *)id->driver_data;
573 	struct qaic_device *qdev;
574 	int mhi_irq;
575 	int ret;
576 	int i;
577 
578 	qdev = create_qdev(pdev, config);
579 	if (!qdev)
580 		return -ENOMEM;
581 
582 	ret = init_pci(qdev, pdev, config);
583 	if (ret)
584 		return ret;
585 
586 	for (i = 0; i < qdev->num_dbc; ++i)
587 		qdev->dbc[i].dbc_base = qdev->bar_dbc + QAIC_DBC_OFF(i);
588 
589 	mhi_irq = init_msi(qdev, pdev);
590 	if (mhi_irq < 0)
591 		return mhi_irq;
592 
593 	ret = qaic_create_drm_device(qdev, QAIC_NO_PARTITION);
594 	if (ret)
595 		return ret;
596 
597 	qdev->mhi_cntrl = qaic_mhi_register_controller(pdev, qdev->bar_mhi, mhi_irq,
598 						       qdev->single_msi, config->family);
599 	if (IS_ERR(qdev->mhi_cntrl)) {
600 		ret = PTR_ERR(qdev->mhi_cntrl);
601 		qaic_destroy_drm_device(qdev, QAIC_NO_PARTITION);
602 		return ret;
603 	}
604 
605 	return 0;
606 }
607 
608 static void qaic_pci_remove(struct pci_dev *pdev)
609 {
610 	struct qaic_device *qdev = pci_get_drvdata(pdev);
611 
612 	if (!qdev)
613 		return;
614 
615 	qaic_dev_reset_clean_local_state(qdev);
616 	qaic_mhi_free_controller(qdev->mhi_cntrl, link_up);
617 	qaic_destroy_drm_device(qdev, QAIC_NO_PARTITION);
618 }
619 
620 static void qaic_pci_shutdown(struct pci_dev *pdev)
621 {
622 	/* see qaic_exit for what link_up is doing */
623 	link_up = true;
624 	qaic_pci_remove(pdev);
625 }
626 
627 static pci_ers_result_t qaic_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t error)
628 {
629 	return PCI_ERS_RESULT_NEED_RESET;
630 }
631 
632 static void qaic_pci_reset_prepare(struct pci_dev *pdev)
633 {
634 	struct qaic_device *qdev = pci_get_drvdata(pdev);
635 
636 	qaic_notify_reset(qdev);
637 	qaic_mhi_start_reset(qdev->mhi_cntrl);
638 	qaic_dev_reset_clean_local_state(qdev);
639 }
640 
641 static void qaic_pci_reset_done(struct pci_dev *pdev)
642 {
643 	struct qaic_device *qdev = pci_get_drvdata(pdev);
644 
645 	qaic_mhi_reset_done(qdev->mhi_cntrl);
646 }
647 
648 static const struct mhi_device_id qaic_mhi_match_table[] = {
649 	{ .chan = "QAIC_CONTROL", },
650 	{},
651 };
652 
653 static struct mhi_driver qaic_mhi_driver = {
654 	.id_table = qaic_mhi_match_table,
655 	.remove = qaic_mhi_remove,
656 	.probe = qaic_mhi_probe,
657 	.ul_xfer_cb = qaic_mhi_ul_xfer_cb,
658 	.dl_xfer_cb = qaic_mhi_dl_xfer_cb,
659 	.driver = {
660 		.name = "qaic_mhi",
661 	},
662 };
663 
664 static const struct pci_device_id qaic_ids[] = {
665 	{ PCI_DEVICE_DATA(QCOM, AIC080, (kernel_ulong_t)&aic080_config), },
666 	{ PCI_DEVICE_DATA(QCOM, AIC100, (kernel_ulong_t)&aic100_config), },
667 	{ PCI_DEVICE_DATA(QCOM, AIC200, (kernel_ulong_t)&aic200_config), },
668 	{ }
669 };
670 MODULE_DEVICE_TABLE(pci, qaic_ids);
671 
672 static const struct pci_error_handlers qaic_pci_err_handler = {
673 	.error_detected = qaic_pci_error_detected,
674 	.reset_prepare = qaic_pci_reset_prepare,
675 	.reset_done = qaic_pci_reset_done,
676 };
677 
678 static bool qaic_is_under_reset(struct qaic_device *qdev)
679 {
680 	int rcu_id;
681 	bool ret;
682 
683 	rcu_id = srcu_read_lock(&qdev->dev_lock);
684 	ret = qdev->dev_state != QAIC_ONLINE;
685 	srcu_read_unlock(&qdev->dev_lock, rcu_id);
686 	return ret;
687 }
688 
689 static bool qaic_data_path_busy(struct qaic_device *qdev)
690 {
691 	bool ret = false;
692 	int dev_rcu_id;
693 	int i;
694 
695 	dev_rcu_id = srcu_read_lock(&qdev->dev_lock);
696 	if (qdev->dev_state != QAIC_ONLINE) {
697 		srcu_read_unlock(&qdev->dev_lock, dev_rcu_id);
698 		return false;
699 	}
700 	for (i = 0; i < qdev->num_dbc; i++) {
701 		struct dma_bridge_chan *dbc = &qdev->dbc[i];
702 		unsigned long flags;
703 		int ch_rcu_id;
704 
705 		ch_rcu_id = srcu_read_lock(&dbc->ch_lock);
706 		if (!dbc->usr || !dbc->in_use) {
707 			srcu_read_unlock(&dbc->ch_lock, ch_rcu_id);
708 			continue;
709 		}
710 		spin_lock_irqsave(&dbc->xfer_lock, flags);
711 		ret = !list_empty(&dbc->xfer_list);
712 		spin_unlock_irqrestore(&dbc->xfer_lock, flags);
713 		srcu_read_unlock(&dbc->ch_lock, ch_rcu_id);
714 		if (ret)
715 			break;
716 	}
717 	srcu_read_unlock(&qdev->dev_lock, dev_rcu_id);
718 	return ret;
719 }
720 
721 static int qaic_pm_suspend(struct device *dev)
722 {
723 	struct qaic_device *qdev = pci_get_drvdata(to_pci_dev(dev));
724 
725 	dev_dbg(dev, "Suspending..\n");
726 	if (qaic_data_path_busy(qdev)) {
727 		dev_dbg(dev, "Device's datapath is busy. Aborting suspend..\n");
728 		return -EBUSY;
729 	}
730 	if (qaic_is_under_reset(qdev)) {
731 		dev_dbg(dev, "Device is under reset. Aborting suspend..\n");
732 		return -EBUSY;
733 	}
734 	qaic_mqts_ch_stop_timer(qdev->mqts_ch);
735 	qaic_pci_reset_prepare(qdev->pdev);
736 	pci_save_state(qdev->pdev);
737 	pci_disable_device(qdev->pdev);
738 	pci_set_power_state(qdev->pdev, PCI_D3hot);
739 	return 0;
740 }
741 
742 static int qaic_pm_resume(struct device *dev)
743 {
744 	struct qaic_device *qdev = pci_get_drvdata(to_pci_dev(dev));
745 	int ret;
746 
747 	dev_dbg(dev, "Resuming..\n");
748 	pci_set_power_state(qdev->pdev, PCI_D0);
749 	pci_restore_state(qdev->pdev);
750 	ret = pci_enable_device(qdev->pdev);
751 	if (ret) {
752 		dev_err(dev, "pci_enable_device failed on resume %d\n", ret);
753 		return ret;
754 	}
755 	pci_set_master(qdev->pdev);
756 	qaic_pci_reset_done(qdev->pdev);
757 	return 0;
758 }
759 
760 static const struct dev_pm_ops qaic_pm_ops = {
761 	SYSTEM_SLEEP_PM_OPS(qaic_pm_suspend, qaic_pm_resume)
762 };
763 
764 static struct pci_driver qaic_pci_driver = {
765 	.name = QAIC_NAME,
766 	.id_table = qaic_ids,
767 	.probe = qaic_pci_probe,
768 	.remove = qaic_pci_remove,
769 	.shutdown = qaic_pci_shutdown,
770 	.err_handler = &qaic_pci_err_handler,
771 	.driver = {
772 		.pm = pm_sleep_ptr(&qaic_pm_ops),
773 	},
774 };
775 
776 static int __init qaic_init(void)
777 {
778 	int ret;
779 
780 	ret = pci_register_driver(&qaic_pci_driver);
781 	if (ret) {
782 		pr_debug("qaic: pci_register_driver failed %d\n", ret);
783 		return ret;
784 	}
785 
786 	ret = mhi_driver_register(&qaic_mhi_driver);
787 	if (ret) {
788 		pr_debug("qaic: mhi_driver_register failed %d\n", ret);
789 		goto free_pci;
790 	}
791 
792 	ret = sahara_register();
793 	if (ret) {
794 		pr_debug("qaic: sahara_register failed %d\n", ret);
795 		goto free_mhi;
796 	}
797 
798 	ret = qaic_timesync_init();
799 	if (ret)
800 		pr_debug("qaic: qaic_timesync_init failed %d\n", ret);
801 
802 	ret = qaic_bootlog_register();
803 	if (ret)
804 		pr_debug("qaic: qaic_bootlog_register failed %d\n", ret);
805 
806 	ret = qaic_ras_register();
807 	if (ret)
808 		pr_debug("qaic: qaic_ras_register failed %d\n", ret);
809 	ret = qaic_ssr_register();
810 	if (ret) {
811 		pr_debug("qaic: qaic_ssr_register failed %d\n", ret);
812 		goto free_bootlog;
813 	}
814 
815 	return 0;
816 
817 free_bootlog:
818 	qaic_bootlog_unregister();
819 free_mhi:
820 	mhi_driver_unregister(&qaic_mhi_driver);
821 free_pci:
822 	pci_unregister_driver(&qaic_pci_driver);
823 	return ret;
824 }
825 
826 static void __exit qaic_exit(void)
827 {
828 	/*
829 	 * We assume that qaic_pci_remove() is called due to a hotplug event
830 	 * which would mean that the link is down, and thus
831 	 * qaic_mhi_free_controller() should not try to access the device during
832 	 * cleanup.
833 	 * We call pci_unregister_driver() below, which also triggers
834 	 * qaic_pci_remove(), but since this is module exit, we expect the link
835 	 * to the device to be up, in which case qaic_mhi_free_controller()
836 	 * should try to access the device during cleanup to put the device in
837 	 * a sane state.
838 	 * For that reason, we set link_up here to let qaic_mhi_free_controller
839 	 * know the expected link state. Since the module is going to be
840 	 * removed at the end of this, we don't need to worry about
841 	 * reinitializing the link_up state after the cleanup is done.
842 	 */
843 	link_up = true;
844 	qaic_ssr_unregister();
845 	qaic_ras_unregister();
846 	qaic_bootlog_unregister();
847 	qaic_timesync_deinit();
848 	sahara_unregister();
849 	mhi_driver_unregister(&qaic_mhi_driver);
850 	pci_unregister_driver(&qaic_pci_driver);
851 }
852 
853 module_init(qaic_init);
854 module_exit(qaic_exit);
855 
856 MODULE_AUTHOR(QAIC_DESC " Kernel Driver Team");
857 MODULE_DESCRIPTION(QAIC_DESC " Accel Driver");
858 MODULE_LICENSE("GPL");
859