xref: /linux/drivers/accel/qaic/qaic_drv.c (revision 260f6f4fda93c8485c8037865c941b42b9cba5d2) !
1 // SPDX-License-Identifier: GPL-2.0-only
2 
3 /* Copyright (c) 2019-2021, The Linux Foundation. All rights reserved. */
4 /* Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved. */
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_timesync.h"
34 #include "sahara.h"
35 
36 MODULE_IMPORT_NS("DMA_BUF");
37 
38 #define PCI_DEVICE_ID_QCOM_AIC080	0xa080
39 #define PCI_DEVICE_ID_QCOM_AIC100	0xa100
40 #define PCI_DEVICE_ID_QCOM_AIC200	0xa110
41 #define QAIC_NAME			"qaic"
42 #define QAIC_DESC			"Qualcomm Cloud AI Accelerators"
43 #define CNTL_MAJOR			5
44 #define CNTL_MINOR			0
45 
46 struct qaic_device_config {
47 	/* Indicates the AIC family the device belongs to */
48 	int family;
49 	/* A bitmask representing the available BARs */
50 	int bar_mask;
51 	/* An index value used to identify the MHI controller BAR */
52 	unsigned int mhi_bar_idx;
53 	/* An index value used to identify the DBCs BAR */
54 	unsigned int dbc_bar_idx;
55 };
56 
57 static const struct qaic_device_config aic080_config = {
58 	.family = FAMILY_AIC100,
59 	.bar_mask = BIT(0) | BIT(2) | BIT(4),
60 	.mhi_bar_idx = 0,
61 	.dbc_bar_idx = 2,
62 };
63 
64 static const struct qaic_device_config aic100_config = {
65 	.family = FAMILY_AIC100,
66 	.bar_mask = BIT(0) | BIT(2) | BIT(4),
67 	.mhi_bar_idx = 0,
68 	.dbc_bar_idx = 2,
69 };
70 
71 static const struct qaic_device_config aic200_config = {
72 	.family = FAMILY_AIC200,
73 	.bar_mask = BIT(0) | BIT(1) | BIT(2) | BIT(4),
74 	.mhi_bar_idx = 1,
75 	.dbc_bar_idx = 2,
76 };
77 
78 bool datapath_polling;
79 module_param(datapath_polling, bool, 0400);
80 MODULE_PARM_DESC(datapath_polling, "Operate the datapath in polling mode");
81 static bool link_up;
82 static DEFINE_IDA(qaic_usrs);
83 
qaicm_wq_release(struct drm_device * dev,void * res)84 static void qaicm_wq_release(struct drm_device *dev, void *res)
85 {
86 	struct workqueue_struct *wq = res;
87 
88 	destroy_workqueue(wq);
89 }
90 
qaicm_wq_init(struct drm_device * dev,const char * name)91 static struct workqueue_struct *qaicm_wq_init(struct drm_device *dev, const char *name)
92 {
93 	struct workqueue_struct *wq;
94 	int ret;
95 
96 	wq = alloc_workqueue("%s", WQ_UNBOUND, 0, name);
97 	if (!wq)
98 		return ERR_PTR(-ENOMEM);
99 	ret = drmm_add_action_or_reset(dev, qaicm_wq_release, wq);
100 	if (ret)
101 		return ERR_PTR(ret);
102 
103 	return wq;
104 }
105 
qaicm_srcu_release(struct drm_device * dev,void * res)106 static void qaicm_srcu_release(struct drm_device *dev, void *res)
107 {
108 	struct srcu_struct *lock = res;
109 
110 	cleanup_srcu_struct(lock);
111 }
112 
qaicm_srcu_init(struct drm_device * dev,struct srcu_struct * lock)113 static int qaicm_srcu_init(struct drm_device *dev, struct srcu_struct *lock)
114 {
115 	int ret;
116 
117 	ret = init_srcu_struct(lock);
118 	if (ret)
119 		return ret;
120 
121 	return drmm_add_action_or_reset(dev, qaicm_srcu_release, lock);
122 }
123 
qaicm_pci_release(struct drm_device * dev,void * res)124 static void qaicm_pci_release(struct drm_device *dev, void *res)
125 {
126 	struct qaic_device *qdev = to_qaic_device(dev);
127 
128 	pci_set_drvdata(qdev->pdev, NULL);
129 }
130 
free_usr(struct kref * kref)131 static void free_usr(struct kref *kref)
132 {
133 	struct qaic_user *usr = container_of(kref, struct qaic_user, ref_count);
134 
135 	cleanup_srcu_struct(&usr->qddev_lock);
136 	ida_free(&qaic_usrs, usr->handle);
137 	kfree(usr);
138 }
139 
qaic_open(struct drm_device * dev,struct drm_file * file)140 static int qaic_open(struct drm_device *dev, struct drm_file *file)
141 {
142 	struct qaic_drm_device *qddev = to_qaic_drm_device(dev);
143 	struct qaic_device *qdev = qddev->qdev;
144 	struct qaic_user *usr;
145 	int rcu_id;
146 	int ret;
147 
148 	rcu_id = srcu_read_lock(&qdev->dev_lock);
149 	if (qdev->dev_state != QAIC_ONLINE) {
150 		ret = -ENODEV;
151 		goto dev_unlock;
152 	}
153 
154 	usr = kmalloc(sizeof(*usr), GFP_KERNEL);
155 	if (!usr) {
156 		ret = -ENOMEM;
157 		goto dev_unlock;
158 	}
159 
160 	usr->handle = ida_alloc(&qaic_usrs, GFP_KERNEL);
161 	if (usr->handle < 0) {
162 		ret = usr->handle;
163 		goto free_usr;
164 	}
165 	usr->qddev = qddev;
166 	atomic_set(&usr->chunk_id, 0);
167 	init_srcu_struct(&usr->qddev_lock);
168 	kref_init(&usr->ref_count);
169 
170 	ret = mutex_lock_interruptible(&qddev->users_mutex);
171 	if (ret)
172 		goto cleanup_usr;
173 
174 	list_add(&usr->node, &qddev->users);
175 	mutex_unlock(&qddev->users_mutex);
176 
177 	file->driver_priv = usr;
178 
179 	srcu_read_unlock(&qdev->dev_lock, rcu_id);
180 	return 0;
181 
182 cleanup_usr:
183 	cleanup_srcu_struct(&usr->qddev_lock);
184 	ida_free(&qaic_usrs, usr->handle);
185 free_usr:
186 	kfree(usr);
187 dev_unlock:
188 	srcu_read_unlock(&qdev->dev_lock, rcu_id);
189 	return ret;
190 }
191 
qaic_postclose(struct drm_device * dev,struct drm_file * file)192 static void qaic_postclose(struct drm_device *dev, struct drm_file *file)
193 {
194 	struct qaic_user *usr = file->driver_priv;
195 	struct qaic_drm_device *qddev;
196 	struct qaic_device *qdev;
197 	int qdev_rcu_id;
198 	int usr_rcu_id;
199 	int i;
200 
201 	qddev = usr->qddev;
202 	usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
203 	if (qddev) {
204 		qdev = qddev->qdev;
205 		qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
206 		if (qdev->dev_state == QAIC_ONLINE) {
207 			qaic_release_usr(qdev, usr);
208 			for (i = 0; i < qdev->num_dbc; ++i)
209 				if (qdev->dbc[i].usr && qdev->dbc[i].usr->handle == usr->handle)
210 					release_dbc(qdev, i);
211 		}
212 		srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
213 
214 		mutex_lock(&qddev->users_mutex);
215 		if (!list_empty(&usr->node))
216 			list_del_init(&usr->node);
217 		mutex_unlock(&qddev->users_mutex);
218 	}
219 
220 	srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
221 	kref_put(&usr->ref_count, free_usr);
222 
223 	file->driver_priv = NULL;
224 }
225 
226 DEFINE_DRM_ACCEL_FOPS(qaic_accel_fops);
227 
228 static const struct drm_ioctl_desc qaic_drm_ioctls[] = {
229 	DRM_IOCTL_DEF_DRV(QAIC_MANAGE, qaic_manage_ioctl, 0),
230 	DRM_IOCTL_DEF_DRV(QAIC_CREATE_BO, qaic_create_bo_ioctl, 0),
231 	DRM_IOCTL_DEF_DRV(QAIC_MMAP_BO, qaic_mmap_bo_ioctl, 0),
232 	DRM_IOCTL_DEF_DRV(QAIC_ATTACH_SLICE_BO, qaic_attach_slice_bo_ioctl, 0),
233 	DRM_IOCTL_DEF_DRV(QAIC_EXECUTE_BO, qaic_execute_bo_ioctl, 0),
234 	DRM_IOCTL_DEF_DRV(QAIC_PARTIAL_EXECUTE_BO, qaic_partial_execute_bo_ioctl, 0),
235 	DRM_IOCTL_DEF_DRV(QAIC_WAIT_BO, qaic_wait_bo_ioctl, 0),
236 	DRM_IOCTL_DEF_DRV(QAIC_PERF_STATS_BO, qaic_perf_stats_bo_ioctl, 0),
237 	DRM_IOCTL_DEF_DRV(QAIC_DETACH_SLICE_BO, qaic_detach_slice_bo_ioctl, 0),
238 };
239 
240 static const struct drm_driver qaic_accel_driver = {
241 	.driver_features	= DRIVER_GEM | DRIVER_COMPUTE_ACCEL,
242 
243 	.name			= QAIC_NAME,
244 	.desc			= QAIC_DESC,
245 
246 	.fops			= &qaic_accel_fops,
247 	.open			= qaic_open,
248 	.postclose		= qaic_postclose,
249 
250 	.ioctls			= qaic_drm_ioctls,
251 	.num_ioctls		= ARRAY_SIZE(qaic_drm_ioctls),
252 	.gem_prime_import	= qaic_gem_prime_import,
253 };
254 
qaic_create_drm_device(struct qaic_device * qdev,s32 partition_id)255 static int qaic_create_drm_device(struct qaic_device *qdev, s32 partition_id)
256 {
257 	struct qaic_drm_device *qddev = qdev->qddev;
258 	struct drm_device *drm = to_drm(qddev);
259 	int ret;
260 
261 	/* Hold off implementing partitions until the uapi is determined */
262 	if (partition_id != QAIC_NO_PARTITION)
263 		return -EINVAL;
264 
265 	qddev->partition_id = partition_id;
266 
267 	ret = drm_dev_register(drm, 0);
268 	if (ret) {
269 		pci_dbg(qdev->pdev, "drm_dev_register failed %d\n", ret);
270 		return ret;
271 	}
272 
273 	qaic_debugfs_init(qddev);
274 
275 	return ret;
276 }
277 
qaic_destroy_drm_device(struct qaic_device * qdev,s32 partition_id)278 static void qaic_destroy_drm_device(struct qaic_device *qdev, s32 partition_id)
279 {
280 	struct qaic_drm_device *qddev = qdev->qddev;
281 	struct drm_device *drm = to_drm(qddev);
282 	struct qaic_user *usr;
283 
284 	drm_dev_unregister(drm);
285 	qddev->partition_id = 0;
286 	/*
287 	 * Existing users get unresolvable errors till they close FDs.
288 	 * Need to sync carefully with users calling close(). The
289 	 * list of users can be modified elsewhere when the lock isn't
290 	 * held here, but the sync'ing the srcu with the mutex held
291 	 * could deadlock. Grab the mutex so that the list will be
292 	 * unmodified. The user we get will exist as long as the
293 	 * lock is held. Signal that the qcdev is going away, and
294 	 * grab a reference to the user so they don't go away for
295 	 * synchronize_srcu(). Then release the mutex to avoid
296 	 * deadlock and make sure the user has observed the signal.
297 	 * With the lock released, we cannot maintain any state of the
298 	 * user list.
299 	 */
300 	mutex_lock(&qddev->users_mutex);
301 	while (!list_empty(&qddev->users)) {
302 		usr = list_first_entry(&qddev->users, struct qaic_user, node);
303 		list_del_init(&usr->node);
304 		kref_get(&usr->ref_count);
305 		usr->qddev = NULL;
306 		mutex_unlock(&qddev->users_mutex);
307 		synchronize_srcu(&usr->qddev_lock);
308 		kref_put(&usr->ref_count, free_usr);
309 		mutex_lock(&qddev->users_mutex);
310 	}
311 	mutex_unlock(&qddev->users_mutex);
312 }
313 
qaic_mhi_probe(struct mhi_device * mhi_dev,const struct mhi_device_id * id)314 static int qaic_mhi_probe(struct mhi_device *mhi_dev, const struct mhi_device_id *id)
315 {
316 	u16 major = -1, minor = -1;
317 	struct qaic_device *qdev;
318 	int ret;
319 
320 	/*
321 	 * Invoking this function indicates that the control channel to the
322 	 * device is available. We use that as a signal to indicate that
323 	 * the device side firmware has booted. The device side firmware
324 	 * manages the device resources, so we need to communicate with it
325 	 * via the control channel in order to utilize the device. Therefore
326 	 * we wait until this signal to create the drm dev that userspace will
327 	 * use to control the device, because without the device side firmware,
328 	 * userspace can't do anything useful.
329 	 */
330 
331 	qdev = pci_get_drvdata(to_pci_dev(mhi_dev->mhi_cntrl->cntrl_dev));
332 
333 	dev_set_drvdata(&mhi_dev->dev, qdev);
334 	qdev->cntl_ch = mhi_dev;
335 
336 	ret = qaic_control_open(qdev);
337 	if (ret) {
338 		pci_dbg(qdev->pdev, "%s: control_open failed %d\n", __func__, ret);
339 		return ret;
340 	}
341 
342 	qdev->dev_state = QAIC_BOOT;
343 	ret = get_cntl_version(qdev, NULL, &major, &minor);
344 	if (ret || major != CNTL_MAJOR || minor > CNTL_MINOR) {
345 		pci_err(qdev->pdev, "%s: Control protocol version (%d.%d) not supported. Supported version is (%d.%d). Ret: %d\n",
346 			__func__, major, minor, CNTL_MAJOR, CNTL_MINOR, ret);
347 		ret = -EINVAL;
348 		goto close_control;
349 	}
350 	qdev->dev_state = QAIC_ONLINE;
351 	kobject_uevent(&(to_accel_kdev(qdev->qddev))->kobj, KOBJ_ONLINE);
352 
353 	return ret;
354 
355 close_control:
356 	qaic_control_close(qdev);
357 	return ret;
358 }
359 
qaic_mhi_remove(struct mhi_device * mhi_dev)360 static void qaic_mhi_remove(struct mhi_device *mhi_dev)
361 {
362 /* This is redundant since we have already observed the device crash */
363 }
364 
qaic_notify_reset(struct qaic_device * qdev)365 static void qaic_notify_reset(struct qaic_device *qdev)
366 {
367 	int i;
368 
369 	kobject_uevent(&(to_accel_kdev(qdev->qddev))->kobj, KOBJ_OFFLINE);
370 	qdev->dev_state = QAIC_OFFLINE;
371 	/* wake up any waiters to avoid waiting for timeouts at sync */
372 	wake_all_cntl(qdev);
373 	for (i = 0; i < qdev->num_dbc; ++i)
374 		wakeup_dbc(qdev, i);
375 	synchronize_srcu(&qdev->dev_lock);
376 }
377 
qaic_dev_reset_clean_local_state(struct qaic_device * qdev)378 void qaic_dev_reset_clean_local_state(struct qaic_device *qdev)
379 {
380 	int i;
381 
382 	qaic_notify_reset(qdev);
383 
384 	/* start tearing things down */
385 	for (i = 0; i < qdev->num_dbc; ++i)
386 		release_dbc(qdev, i);
387 }
388 
create_qdev(struct pci_dev * pdev,const struct qaic_device_config * config)389 static struct qaic_device *create_qdev(struct pci_dev *pdev,
390 				       const struct qaic_device_config *config)
391 {
392 	struct device *dev = &pdev->dev;
393 	struct qaic_drm_device *qddev;
394 	struct qaic_device *qdev;
395 	struct drm_device *drm;
396 	int i, ret;
397 
398 	qdev = devm_kzalloc(dev, sizeof(*qdev), GFP_KERNEL);
399 	if (!qdev)
400 		return NULL;
401 
402 	qdev->dev_state = QAIC_OFFLINE;
403 	qdev->num_dbc = 16;
404 	qdev->dbc = devm_kcalloc(dev, qdev->num_dbc, sizeof(*qdev->dbc), GFP_KERNEL);
405 	if (!qdev->dbc)
406 		return NULL;
407 
408 	qddev = devm_drm_dev_alloc(&pdev->dev, &qaic_accel_driver, struct qaic_drm_device, drm);
409 	if (IS_ERR(qddev))
410 		return NULL;
411 
412 	drm = to_drm(qddev);
413 	pci_set_drvdata(pdev, qdev);
414 
415 	ret = drmm_mutex_init(drm, &qddev->users_mutex);
416 	if (ret)
417 		return NULL;
418 	ret = drmm_add_action_or_reset(drm, qaicm_pci_release, NULL);
419 	if (ret)
420 		return NULL;
421 	ret = drmm_mutex_init(drm, &qdev->cntl_mutex);
422 	if (ret)
423 		return NULL;
424 	ret = drmm_mutex_init(drm, &qdev->bootlog_mutex);
425 	if (ret)
426 		return NULL;
427 
428 	qdev->cntl_wq = qaicm_wq_init(drm, "qaic_cntl");
429 	if (IS_ERR(qdev->cntl_wq))
430 		return NULL;
431 	qdev->qts_wq = qaicm_wq_init(drm, "qaic_ts");
432 	if (IS_ERR(qdev->qts_wq))
433 		return NULL;
434 
435 	ret = qaicm_srcu_init(drm, &qdev->dev_lock);
436 	if (ret)
437 		return NULL;
438 
439 	qdev->qddev = qddev;
440 	qdev->pdev = pdev;
441 	qddev->qdev = qdev;
442 
443 	INIT_LIST_HEAD(&qdev->cntl_xfer_list);
444 	INIT_LIST_HEAD(&qdev->bootlog);
445 	INIT_LIST_HEAD(&qddev->users);
446 
447 	for (i = 0; i < qdev->num_dbc; ++i) {
448 		spin_lock_init(&qdev->dbc[i].xfer_lock);
449 		qdev->dbc[i].qdev = qdev;
450 		qdev->dbc[i].id = i;
451 		INIT_LIST_HEAD(&qdev->dbc[i].xfer_list);
452 		ret = qaicm_srcu_init(drm, &qdev->dbc[i].ch_lock);
453 		if (ret)
454 			return NULL;
455 		init_waitqueue_head(&qdev->dbc[i].dbc_release);
456 		INIT_LIST_HEAD(&qdev->dbc[i].bo_lists);
457 	}
458 
459 	return qdev;
460 }
461 
init_pci(struct qaic_device * qdev,struct pci_dev * pdev,const struct qaic_device_config * config)462 static int init_pci(struct qaic_device *qdev, struct pci_dev *pdev,
463 		    const struct qaic_device_config *config)
464 {
465 	int bars;
466 	int ret;
467 
468 	bars = pci_select_bars(pdev, IORESOURCE_MEM) & 0x3f;
469 
470 	/* make sure the device has the expected BARs */
471 	if (bars != config->bar_mask) {
472 		pci_dbg(pdev, "%s: expected BARs %#x not found in device. Found %#x\n",
473 			__func__, config->bar_mask, bars);
474 		return -EINVAL;
475 	}
476 
477 	ret = pcim_enable_device(pdev);
478 	if (ret)
479 		return ret;
480 
481 	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
482 	if (ret)
483 		return ret;
484 	dma_set_max_seg_size(&pdev->dev, UINT_MAX);
485 
486 	qdev->bar_mhi = devm_ioremap_resource(&pdev->dev, &pdev->resource[config->mhi_bar_idx]);
487 	if (IS_ERR(qdev->bar_mhi))
488 		return PTR_ERR(qdev->bar_mhi);
489 
490 	qdev->bar_dbc = devm_ioremap_resource(&pdev->dev, &pdev->resource[config->dbc_bar_idx]);
491 	if (IS_ERR(qdev->bar_dbc))
492 		return PTR_ERR(qdev->bar_dbc);
493 
494 	/* Managed release since we use pcim_enable_device above */
495 	pci_set_master(pdev);
496 
497 	return 0;
498 }
499 
init_msi(struct qaic_device * qdev,struct pci_dev * pdev)500 static int init_msi(struct qaic_device *qdev, struct pci_dev *pdev)
501 {
502 	int irq_count = qdev->num_dbc + 1;
503 	int mhi_irq;
504 	int ret;
505 	int i;
506 
507 	/* Managed release since we use pcim_enable_device */
508 	ret = pci_alloc_irq_vectors(pdev, irq_count, irq_count, PCI_IRQ_MSI | PCI_IRQ_MSIX);
509 	if (ret == -ENOSPC) {
510 		ret = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI | PCI_IRQ_MSIX);
511 		if (ret < 0)
512 			return ret;
513 
514 		/*
515 		 * Operate in one MSI mode. All interrupts will be directed to
516 		 * MSI0; every interrupt will wake up all the interrupt handlers
517 		 * (MHI and DBC[0-15]). Since the interrupt is now shared, it is
518 		 * not disabled during DBC threaded handler, but only one thread
519 		 * will be allowed to run per DBC, so while it can be
520 		 * interrupted, it shouldn't race with itself.
521 		 */
522 		qdev->single_msi = true;
523 		pci_info(pdev, "Allocating %d MSIs failed, operating in 1 MSI mode. Performance may be impacted.\n",
524 			 irq_count);
525 	} else if (ret < 0) {
526 		return ret;
527 	}
528 
529 	mhi_irq = pci_irq_vector(pdev, 0);
530 	if (mhi_irq < 0)
531 		return mhi_irq;
532 
533 	for (i = 0; i < qdev->num_dbc; ++i) {
534 		ret = devm_request_threaded_irq(&pdev->dev,
535 						pci_irq_vector(pdev, qdev->single_msi ? 0 : i + 1),
536 						dbc_irq_handler, dbc_irq_threaded_fn, IRQF_SHARED,
537 						"qaic_dbc", &qdev->dbc[i]);
538 		if (ret)
539 			return ret;
540 
541 		if (datapath_polling) {
542 			qdev->dbc[i].irq = pci_irq_vector(pdev, qdev->single_msi ? 0 : i + 1);
543 			if (!qdev->single_msi)
544 				disable_irq_nosync(qdev->dbc[i].irq);
545 			INIT_WORK(&qdev->dbc[i].poll_work, irq_polling_work);
546 		}
547 	}
548 
549 	return mhi_irq;
550 }
551 
qaic_pci_probe(struct pci_dev * pdev,const struct pci_device_id * id)552 static int qaic_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
553 {
554 	struct qaic_device_config *config = (struct qaic_device_config *)id->driver_data;
555 	struct qaic_device *qdev;
556 	int mhi_irq;
557 	int ret;
558 	int i;
559 
560 	qdev = create_qdev(pdev, config);
561 	if (!qdev)
562 		return -ENOMEM;
563 
564 	ret = init_pci(qdev, pdev, config);
565 	if (ret)
566 		return ret;
567 
568 	for (i = 0; i < qdev->num_dbc; ++i)
569 		qdev->dbc[i].dbc_base = qdev->bar_dbc + QAIC_DBC_OFF(i);
570 
571 	mhi_irq = init_msi(qdev, pdev);
572 	if (mhi_irq < 0)
573 		return mhi_irq;
574 
575 	ret = qaic_create_drm_device(qdev, QAIC_NO_PARTITION);
576 	if (ret)
577 		return ret;
578 
579 	qdev->mhi_cntrl = qaic_mhi_register_controller(pdev, qdev->bar_mhi, mhi_irq,
580 						       qdev->single_msi, config->family);
581 	if (IS_ERR(qdev->mhi_cntrl)) {
582 		ret = PTR_ERR(qdev->mhi_cntrl);
583 		qaic_destroy_drm_device(qdev, QAIC_NO_PARTITION);
584 		return ret;
585 	}
586 
587 	return 0;
588 }
589 
qaic_pci_remove(struct pci_dev * pdev)590 static void qaic_pci_remove(struct pci_dev *pdev)
591 {
592 	struct qaic_device *qdev = pci_get_drvdata(pdev);
593 
594 	if (!qdev)
595 		return;
596 
597 	qaic_dev_reset_clean_local_state(qdev);
598 	qaic_mhi_free_controller(qdev->mhi_cntrl, link_up);
599 	qaic_destroy_drm_device(qdev, QAIC_NO_PARTITION);
600 }
601 
qaic_pci_shutdown(struct pci_dev * pdev)602 static void qaic_pci_shutdown(struct pci_dev *pdev)
603 {
604 	/* see qaic_exit for what link_up is doing */
605 	link_up = true;
606 	qaic_pci_remove(pdev);
607 }
608 
qaic_pci_error_detected(struct pci_dev * pdev,pci_channel_state_t error)609 static pci_ers_result_t qaic_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t error)
610 {
611 	return PCI_ERS_RESULT_NEED_RESET;
612 }
613 
qaic_pci_reset_prepare(struct pci_dev * pdev)614 static void qaic_pci_reset_prepare(struct pci_dev *pdev)
615 {
616 	struct qaic_device *qdev = pci_get_drvdata(pdev);
617 
618 	qaic_notify_reset(qdev);
619 	qaic_mhi_start_reset(qdev->mhi_cntrl);
620 	qaic_dev_reset_clean_local_state(qdev);
621 }
622 
qaic_pci_reset_done(struct pci_dev * pdev)623 static void qaic_pci_reset_done(struct pci_dev *pdev)
624 {
625 	struct qaic_device *qdev = pci_get_drvdata(pdev);
626 
627 	qaic_mhi_reset_done(qdev->mhi_cntrl);
628 }
629 
630 static const struct mhi_device_id qaic_mhi_match_table[] = {
631 	{ .chan = "QAIC_CONTROL", },
632 	{},
633 };
634 
635 static struct mhi_driver qaic_mhi_driver = {
636 	.id_table = qaic_mhi_match_table,
637 	.remove = qaic_mhi_remove,
638 	.probe = qaic_mhi_probe,
639 	.ul_xfer_cb = qaic_mhi_ul_xfer_cb,
640 	.dl_xfer_cb = qaic_mhi_dl_xfer_cb,
641 	.driver = {
642 		.name = "qaic_mhi",
643 	},
644 };
645 
646 static const struct pci_device_id qaic_ids[] = {
647 	{ PCI_DEVICE_DATA(QCOM, AIC080, (kernel_ulong_t)&aic080_config), },
648 	{ PCI_DEVICE_DATA(QCOM, AIC100, (kernel_ulong_t)&aic100_config), },
649 	{ PCI_DEVICE_DATA(QCOM, AIC200, (kernel_ulong_t)&aic200_config), },
650 	{ }
651 };
652 MODULE_DEVICE_TABLE(pci, qaic_ids);
653 
654 static const struct pci_error_handlers qaic_pci_err_handler = {
655 	.error_detected = qaic_pci_error_detected,
656 	.reset_prepare = qaic_pci_reset_prepare,
657 	.reset_done = qaic_pci_reset_done,
658 };
659 
660 static struct pci_driver qaic_pci_driver = {
661 	.name = QAIC_NAME,
662 	.id_table = qaic_ids,
663 	.probe = qaic_pci_probe,
664 	.remove = qaic_pci_remove,
665 	.shutdown = qaic_pci_shutdown,
666 	.err_handler = &qaic_pci_err_handler,
667 };
668 
qaic_init(void)669 static int __init qaic_init(void)
670 {
671 	int ret;
672 
673 	ret = pci_register_driver(&qaic_pci_driver);
674 	if (ret) {
675 		pr_debug("qaic: pci_register_driver failed %d\n", ret);
676 		return ret;
677 	}
678 
679 	ret = mhi_driver_register(&qaic_mhi_driver);
680 	if (ret) {
681 		pr_debug("qaic: mhi_driver_register failed %d\n", ret);
682 		goto free_pci;
683 	}
684 
685 	ret = sahara_register();
686 	if (ret) {
687 		pr_debug("qaic: sahara_register failed %d\n", ret);
688 		goto free_mhi;
689 	}
690 
691 	ret = qaic_timesync_init();
692 	if (ret)
693 		pr_debug("qaic: qaic_timesync_init failed %d\n", ret);
694 
695 	ret = qaic_bootlog_register();
696 	if (ret)
697 		pr_debug("qaic: qaic_bootlog_register failed %d\n", ret);
698 
699 	ret = qaic_ras_register();
700 	if (ret)
701 		pr_debug("qaic: qaic_ras_register failed %d\n", ret);
702 
703 	return 0;
704 
705 free_mhi:
706 	mhi_driver_unregister(&qaic_mhi_driver);
707 free_pci:
708 	pci_unregister_driver(&qaic_pci_driver);
709 	return ret;
710 }
711 
qaic_exit(void)712 static void __exit qaic_exit(void)
713 {
714 	/*
715 	 * We assume that qaic_pci_remove() is called due to a hotplug event
716 	 * which would mean that the link is down, and thus
717 	 * qaic_mhi_free_controller() should not try to access the device during
718 	 * cleanup.
719 	 * We call pci_unregister_driver() below, which also triggers
720 	 * qaic_pci_remove(), but since this is module exit, we expect the link
721 	 * to the device to be up, in which case qaic_mhi_free_controller()
722 	 * should try to access the device during cleanup to put the device in
723 	 * a sane state.
724 	 * For that reason, we set link_up here to let qaic_mhi_free_controller
725 	 * know the expected link state. Since the module is going to be
726 	 * removed at the end of this, we don't need to worry about
727 	 * reinitializing the link_up state after the cleanup is done.
728 	 */
729 	link_up = true;
730 	qaic_ras_unregister();
731 	qaic_bootlog_unregister();
732 	qaic_timesync_deinit();
733 	sahara_unregister();
734 	mhi_driver_unregister(&qaic_mhi_driver);
735 	pci_unregister_driver(&qaic_pci_driver);
736 }
737 
738 module_init(qaic_init);
739 module_exit(qaic_exit);
740 
741 MODULE_AUTHOR(QAIC_DESC " Kernel Driver Team");
742 MODULE_DESCRIPTION(QAIC_DESC " Accel Driver");
743 MODULE_LICENSE("GPL");
744