xref: /openbmc/linux/drivers/misc/ocxl/file.c (revision 151f4e2b)
1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright 2017 IBM Corp.
3 #include <linux/fs.h>
4 #include <linux/poll.h>
5 #include <linux/sched/signal.h>
6 #include <linux/eventfd.h>
7 #include <linux/uaccess.h>
8 #include <uapi/misc/ocxl.h>
9 #include <asm/reg.h>
10 #include <asm/switch_to.h>
11 #include "ocxl_internal.h"
12 
13 
14 #define OCXL_NUM_MINORS 256 /* Total to reserve */
15 
16 static dev_t ocxl_dev;
17 static struct class *ocxl_class;
18 static struct mutex minors_idr_lock;
19 static struct idr minors_idr;
20 
21 static struct ocxl_file_info *find_file_info(dev_t devno)
22 {
23 	struct ocxl_file_info *info;
24 
25 	/*
26 	 * We don't declare an RCU critical section here, as our AFU
27 	 * is protected by a reference counter on the device. By the time the
28 	 * info reference is removed from the idr, the ref count of
29 	 * the device is already at 0, so no user API will access that AFU and
30 	 * this function can't return it.
31 	 */
32 	info = idr_find(&minors_idr, MINOR(devno));
33 	return info;
34 }
35 
36 static int allocate_minor(struct ocxl_file_info *info)
37 {
38 	int minor;
39 
40 	mutex_lock(&minors_idr_lock);
41 	minor = idr_alloc(&minors_idr, info, 0, OCXL_NUM_MINORS, GFP_KERNEL);
42 	mutex_unlock(&minors_idr_lock);
43 	return minor;
44 }
45 
46 static void free_minor(struct ocxl_file_info *info)
47 {
48 	mutex_lock(&minors_idr_lock);
49 	idr_remove(&minors_idr, MINOR(info->dev.devt));
50 	mutex_unlock(&minors_idr_lock);
51 }
52 
53 static int afu_open(struct inode *inode, struct file *file)
54 {
55 	struct ocxl_file_info *info;
56 	struct ocxl_context *ctx;
57 	int rc;
58 
59 	pr_debug("%s for device %x\n", __func__, inode->i_rdev);
60 
61 	info = find_file_info(inode->i_rdev);
62 	if (!info)
63 		return -ENODEV;
64 
65 	rc = ocxl_context_alloc(&ctx, info->afu, inode->i_mapping);
66 	if (rc)
67 		return rc;
68 
69 	file->private_data = ctx;
70 	return 0;
71 }
72 
73 static long afu_ioctl_attach(struct ocxl_context *ctx,
74 			struct ocxl_ioctl_attach __user *uarg)
75 {
76 	struct ocxl_ioctl_attach arg;
77 	u64 amr = 0;
78 	int rc;
79 
80 	pr_debug("%s for context %d\n", __func__, ctx->pasid);
81 
82 	if (copy_from_user(&arg, uarg, sizeof(arg)))
83 		return -EFAULT;
84 
85 	/* Make sure reserved fields are not set for forward compatibility */
86 	if (arg.reserved1 || arg.reserved2 || arg.reserved3)
87 		return -EINVAL;
88 
89 	amr = arg.amr & mfspr(SPRN_UAMOR);
90 	rc = ocxl_context_attach(ctx, amr, current->mm);
91 	return rc;
92 }
93 
94 static long afu_ioctl_get_metadata(struct ocxl_context *ctx,
95 		struct ocxl_ioctl_metadata __user *uarg)
96 {
97 	struct ocxl_ioctl_metadata arg;
98 
99 	memset(&arg, 0, sizeof(arg));
100 
101 	arg.version = 0;
102 
103 	arg.afu_version_major = ctx->afu->config.version_major;
104 	arg.afu_version_minor = ctx->afu->config.version_minor;
105 	arg.pasid = ctx->pasid;
106 	arg.pp_mmio_size = ctx->afu->config.pp_mmio_stride;
107 	arg.global_mmio_size = ctx->afu->config.global_mmio_size;
108 
109 	if (copy_to_user(uarg, &arg, sizeof(arg)))
110 		return -EFAULT;
111 
112 	return 0;
113 }
114 
115 #ifdef CONFIG_PPC64
116 static long afu_ioctl_enable_p9_wait(struct ocxl_context *ctx,
117 		struct ocxl_ioctl_p9_wait __user *uarg)
118 {
119 	struct ocxl_ioctl_p9_wait arg;
120 
121 	memset(&arg, 0, sizeof(arg));
122 
123 	if (cpu_has_feature(CPU_FTR_P9_TIDR)) {
124 		enum ocxl_context_status status;
125 
126 		// Locks both status & tidr
127 		mutex_lock(&ctx->status_mutex);
128 		if (!ctx->tidr) {
129 			if (set_thread_tidr(current)) {
130 				mutex_unlock(&ctx->status_mutex);
131 				return -ENOENT;
132 			}
133 
134 			ctx->tidr = current->thread.tidr;
135 		}
136 
137 		status = ctx->status;
138 		mutex_unlock(&ctx->status_mutex);
139 
140 		if (status == ATTACHED) {
141 			int rc = ocxl_link_update_pe(ctx->afu->fn->link,
142 				ctx->pasid, ctx->tidr);
143 
144 			if (rc)
145 				return rc;
146 		}
147 
148 		arg.thread_id = ctx->tidr;
149 	} else
150 		return -ENOENT;
151 
152 	if (copy_to_user(uarg, &arg, sizeof(arg)))
153 		return -EFAULT;
154 
155 	return 0;
156 }
157 #endif
158 
159 
160 static long afu_ioctl_get_features(struct ocxl_context *ctx,
161 		struct ocxl_ioctl_features __user *uarg)
162 {
163 	struct ocxl_ioctl_features arg;
164 
165 	memset(&arg, 0, sizeof(arg));
166 
167 #ifdef CONFIG_PPC64
168 	if (cpu_has_feature(CPU_FTR_P9_TIDR))
169 		arg.flags[0] |= OCXL_IOCTL_FEATURES_FLAGS0_P9_WAIT;
170 #endif
171 
172 	if (copy_to_user(uarg, &arg, sizeof(arg)))
173 		return -EFAULT;
174 
175 	return 0;
176 }
177 
178 #define CMD_STR(x) (x == OCXL_IOCTL_ATTACH ? "ATTACH" :			\
179 			x == OCXL_IOCTL_IRQ_ALLOC ? "IRQ_ALLOC" :	\
180 			x == OCXL_IOCTL_IRQ_FREE ? "IRQ_FREE" :		\
181 			x == OCXL_IOCTL_IRQ_SET_FD ? "IRQ_SET_FD" :	\
182 			x == OCXL_IOCTL_GET_METADATA ? "GET_METADATA" :	\
183 			x == OCXL_IOCTL_ENABLE_P9_WAIT ? "ENABLE_P9_WAIT" :	\
184 			x == OCXL_IOCTL_GET_FEATURES ? "GET_FEATURES" :	\
185 			"UNKNOWN")
186 
187 static irqreturn_t irq_handler(void *private)
188 {
189 	struct eventfd_ctx *ev_ctx = private;
190 
191 	eventfd_signal(ev_ctx, 1);
192 	return IRQ_HANDLED;
193 }
194 
195 static void irq_free(void *private)
196 {
197 	struct eventfd_ctx *ev_ctx = private;
198 
199 	eventfd_ctx_put(ev_ctx);
200 }
201 
202 static long afu_ioctl(struct file *file, unsigned int cmd,
203 		unsigned long args)
204 {
205 	struct ocxl_context *ctx = file->private_data;
206 	struct ocxl_ioctl_irq_fd irq_fd;
207 	struct eventfd_ctx *ev_ctx;
208 	int irq_id;
209 	u64 irq_offset;
210 	long rc;
211 	bool closed;
212 
213 	pr_debug("%s for context %d, command %s\n", __func__, ctx->pasid,
214 		CMD_STR(cmd));
215 
216 	mutex_lock(&ctx->status_mutex);
217 	closed = (ctx->status == CLOSED);
218 	mutex_unlock(&ctx->status_mutex);
219 
220 	if (closed)
221 		return -EIO;
222 
223 	switch (cmd) {
224 	case OCXL_IOCTL_ATTACH:
225 		rc = afu_ioctl_attach(ctx,
226 				(struct ocxl_ioctl_attach __user *) args);
227 		break;
228 
229 	case OCXL_IOCTL_IRQ_ALLOC:
230 		rc = ocxl_afu_irq_alloc(ctx, &irq_id);
231 		if (!rc) {
232 			irq_offset = ocxl_irq_id_to_offset(ctx, irq_id);
233 			rc = copy_to_user((u64 __user *) args, &irq_offset,
234 					sizeof(irq_offset));
235 			if (rc) {
236 				ocxl_afu_irq_free(ctx, irq_id);
237 				return -EFAULT;
238 			}
239 		}
240 		break;
241 
242 	case OCXL_IOCTL_IRQ_FREE:
243 		rc = copy_from_user(&irq_offset, (u64 __user *) args,
244 				sizeof(irq_offset));
245 		if (rc)
246 			return -EFAULT;
247 		irq_id = ocxl_irq_offset_to_id(ctx, irq_offset);
248 		rc = ocxl_afu_irq_free(ctx, irq_id);
249 		break;
250 
251 	case OCXL_IOCTL_IRQ_SET_FD:
252 		rc = copy_from_user(&irq_fd, (u64 __user *) args,
253 				sizeof(irq_fd));
254 		if (rc)
255 			return -EFAULT;
256 		if (irq_fd.reserved)
257 			return -EINVAL;
258 		irq_id = ocxl_irq_offset_to_id(ctx, irq_fd.irq_offset);
259 		ev_ctx = eventfd_ctx_fdget(irq_fd.eventfd);
260 		if (IS_ERR(ev_ctx))
261 			return PTR_ERR(ev_ctx);
262 		rc = ocxl_irq_set_handler(ctx, irq_id, irq_handler, irq_free, ev_ctx);
263 		break;
264 
265 	case OCXL_IOCTL_GET_METADATA:
266 		rc = afu_ioctl_get_metadata(ctx,
267 				(struct ocxl_ioctl_metadata __user *) args);
268 		break;
269 
270 #ifdef CONFIG_PPC64
271 	case OCXL_IOCTL_ENABLE_P9_WAIT:
272 		rc = afu_ioctl_enable_p9_wait(ctx,
273 				(struct ocxl_ioctl_p9_wait __user *) args);
274 		break;
275 #endif
276 
277 	case OCXL_IOCTL_GET_FEATURES:
278 		rc = afu_ioctl_get_features(ctx,
279 				(struct ocxl_ioctl_features __user *) args);
280 		break;
281 
282 	default:
283 		rc = -EINVAL;
284 	}
285 	return rc;
286 }
287 
288 static long afu_compat_ioctl(struct file *file, unsigned int cmd,
289 			unsigned long args)
290 {
291 	return afu_ioctl(file, cmd, args);
292 }
293 
294 static int afu_mmap(struct file *file, struct vm_area_struct *vma)
295 {
296 	struct ocxl_context *ctx = file->private_data;
297 
298 	pr_debug("%s for context %d\n", __func__, ctx->pasid);
299 	return ocxl_context_mmap(ctx, vma);
300 }
301 
302 static bool has_xsl_error(struct ocxl_context *ctx)
303 {
304 	bool ret;
305 
306 	mutex_lock(&ctx->xsl_error_lock);
307 	ret = !!ctx->xsl_error.addr;
308 	mutex_unlock(&ctx->xsl_error_lock);
309 
310 	return ret;
311 }
312 
313 /*
314  * Are there any events pending on the AFU
315  * ctx: The AFU context
316  * Returns: true if there are events pending
317  */
318 static bool afu_events_pending(struct ocxl_context *ctx)
319 {
320 	if (has_xsl_error(ctx))
321 		return true;
322 	return false;
323 }
324 
325 static unsigned int afu_poll(struct file *file, struct poll_table_struct *wait)
326 {
327 	struct ocxl_context *ctx = file->private_data;
328 	unsigned int mask = 0;
329 	bool closed;
330 
331 	pr_debug("%s for context %d\n", __func__, ctx->pasid);
332 
333 	poll_wait(file, &ctx->events_wq, wait);
334 
335 	mutex_lock(&ctx->status_mutex);
336 	closed = (ctx->status == CLOSED);
337 	mutex_unlock(&ctx->status_mutex);
338 
339 	if (afu_events_pending(ctx))
340 		mask = EPOLLIN | EPOLLRDNORM;
341 	else if (closed)
342 		mask = EPOLLERR;
343 
344 	return mask;
345 }
346 
347 /*
348  * Populate the supplied buffer with a single XSL error
349  * ctx:	The AFU context to report the error from
350  * header: the event header to populate
351  * buf: The buffer to write the body into (should be at least
352  *      AFU_EVENT_BODY_XSL_ERROR_SIZE)
353  * Return: the amount of buffer that was populated
354  */
355 static ssize_t append_xsl_error(struct ocxl_context *ctx,
356 				struct ocxl_kernel_event_header *header,
357 				char __user *buf)
358 {
359 	struct ocxl_kernel_event_xsl_fault_error body;
360 
361 	memset(&body, 0, sizeof(body));
362 
363 	mutex_lock(&ctx->xsl_error_lock);
364 	if (!ctx->xsl_error.addr) {
365 		mutex_unlock(&ctx->xsl_error_lock);
366 		return 0;
367 	}
368 
369 	body.addr = ctx->xsl_error.addr;
370 	body.dsisr = ctx->xsl_error.dsisr;
371 	body.count = ctx->xsl_error.count;
372 
373 	ctx->xsl_error.addr = 0;
374 	ctx->xsl_error.dsisr = 0;
375 	ctx->xsl_error.count = 0;
376 
377 	mutex_unlock(&ctx->xsl_error_lock);
378 
379 	header->type = OCXL_AFU_EVENT_XSL_FAULT_ERROR;
380 
381 	if (copy_to_user(buf, &body, sizeof(body)))
382 		return -EFAULT;
383 
384 	return sizeof(body);
385 }
386 
387 #define AFU_EVENT_BODY_MAX_SIZE sizeof(struct ocxl_kernel_event_xsl_fault_error)
388 
389 /*
390  * Reports events on the AFU
391  * Format:
392  *	Header (struct ocxl_kernel_event_header)
393  *	Body (struct ocxl_kernel_event_*)
394  *	Header...
395  */
396 static ssize_t afu_read(struct file *file, char __user *buf, size_t count,
397 			loff_t *off)
398 {
399 	struct ocxl_context *ctx = file->private_data;
400 	struct ocxl_kernel_event_header header;
401 	ssize_t rc;
402 	ssize_t used = 0;
403 	DEFINE_WAIT(event_wait);
404 
405 	memset(&header, 0, sizeof(header));
406 
407 	/* Require offset to be 0 */
408 	if (*off != 0)
409 		return -EINVAL;
410 
411 	if (count < (sizeof(struct ocxl_kernel_event_header) +
412 			AFU_EVENT_BODY_MAX_SIZE))
413 		return -EINVAL;
414 
415 	for (;;) {
416 		prepare_to_wait(&ctx->events_wq, &event_wait,
417 				TASK_INTERRUPTIBLE);
418 
419 		if (afu_events_pending(ctx))
420 			break;
421 
422 		if (ctx->status == CLOSED)
423 			break;
424 
425 		if (file->f_flags & O_NONBLOCK) {
426 			finish_wait(&ctx->events_wq, &event_wait);
427 			return -EAGAIN;
428 		}
429 
430 		if (signal_pending(current)) {
431 			finish_wait(&ctx->events_wq, &event_wait);
432 			return -ERESTARTSYS;
433 		}
434 
435 		schedule();
436 	}
437 
438 	finish_wait(&ctx->events_wq, &event_wait);
439 
440 	if (has_xsl_error(ctx)) {
441 		used = append_xsl_error(ctx, &header, buf + sizeof(header));
442 		if (used < 0)
443 			return used;
444 	}
445 
446 	if (!afu_events_pending(ctx))
447 		header.flags |= OCXL_KERNEL_EVENT_FLAG_LAST;
448 
449 	if (copy_to_user(buf, &header, sizeof(header)))
450 		return -EFAULT;
451 
452 	used += sizeof(header);
453 
454 	rc = used;
455 	return rc;
456 }
457 
458 static int afu_release(struct inode *inode, struct file *file)
459 {
460 	struct ocxl_context *ctx = file->private_data;
461 	int rc;
462 
463 	pr_debug("%s for device %x\n", __func__, inode->i_rdev);
464 	rc = ocxl_context_detach(ctx);
465 	mutex_lock(&ctx->mapping_lock);
466 	ctx->mapping = NULL;
467 	mutex_unlock(&ctx->mapping_lock);
468 	wake_up_all(&ctx->events_wq);
469 	if (rc != -EBUSY)
470 		ocxl_context_free(ctx);
471 	return 0;
472 }
473 
474 static const struct file_operations ocxl_afu_fops = {
475 	.owner		= THIS_MODULE,
476 	.open           = afu_open,
477 	.unlocked_ioctl = afu_ioctl,
478 	.compat_ioctl   = afu_compat_ioctl,
479 	.mmap           = afu_mmap,
480 	.poll           = afu_poll,
481 	.read           = afu_read,
482 	.release        = afu_release,
483 };
484 
485 // Free the info struct
486 static void info_release(struct device *dev)
487 {
488 	struct ocxl_file_info *info = container_of(dev, struct ocxl_file_info, dev);
489 
490 	free_minor(info);
491 	ocxl_afu_put(info->afu);
492 	kfree(info);
493 }
494 
495 static int ocxl_file_make_visible(struct ocxl_file_info *info)
496 {
497 	int rc;
498 
499 	cdev_init(&info->cdev, &ocxl_afu_fops);
500 	rc = cdev_add(&info->cdev, info->dev.devt, 1);
501 	if (rc) {
502 		dev_err(&info->dev, "Unable to add afu char device: %d\n", rc);
503 		return rc;
504 	}
505 
506 	return 0;
507 }
508 
509 static void ocxl_file_make_invisible(struct ocxl_file_info *info)
510 {
511 	cdev_del(&info->cdev);
512 }
513 
514 int ocxl_file_register_afu(struct ocxl_afu *afu)
515 {
516 	int minor;
517 	int rc;
518 	struct ocxl_file_info *info;
519 	struct ocxl_fn *fn = afu->fn;
520 	struct pci_dev *pci_dev = to_pci_dev(fn->dev.parent);
521 
522 	info = kzalloc(sizeof(*info), GFP_KERNEL);
523 	if (info == NULL)
524 		return -ENOMEM;
525 
526 	minor = allocate_minor(info);
527 	if (minor < 0) {
528 		kfree(info);
529 		return minor;
530 	}
531 
532 	info->dev.parent = &fn->dev;
533 	info->dev.devt = MKDEV(MAJOR(ocxl_dev), minor);
534 	info->dev.class = ocxl_class;
535 	info->dev.release = info_release;
536 
537 	info->afu = afu;
538 	ocxl_afu_get(afu);
539 
540 	rc = dev_set_name(&info->dev, "%s.%s.%hhu",
541 		afu->config.name, dev_name(&pci_dev->dev), afu->config.idx);
542 	if (rc)
543 		goto err_put;
544 
545 	rc = device_register(&info->dev);
546 	if (rc)
547 		goto err_put;
548 
549 	rc = ocxl_sysfs_register_afu(info);
550 	if (rc)
551 		goto err_unregister;
552 
553 	rc = ocxl_file_make_visible(info);
554 	if (rc)
555 		goto err_unregister;
556 
557 	ocxl_afu_set_private(afu, info);
558 
559 	return 0;
560 
561 err_unregister:
562 	ocxl_sysfs_unregister_afu(info); // safe to call even if register failed
563 	device_unregister(&info->dev);
564 err_put:
565 	ocxl_afu_put(afu);
566 	free_minor(info);
567 	kfree(info);
568 	return rc;
569 }
570 
571 void ocxl_file_unregister_afu(struct ocxl_afu *afu)
572 {
573 	struct ocxl_file_info *info = ocxl_afu_get_private(afu);
574 
575 	if (!info)
576 		return;
577 
578 	ocxl_file_make_invisible(info);
579 	ocxl_sysfs_unregister_afu(info);
580 	device_unregister(&info->dev);
581 }
582 
583 static char *ocxl_devnode(struct device *dev, umode_t *mode)
584 {
585 	return kasprintf(GFP_KERNEL, "ocxl/%s", dev_name(dev));
586 }
587 
588 int ocxl_file_init(void)
589 {
590 	int rc;
591 
592 	mutex_init(&minors_idr_lock);
593 	idr_init(&minors_idr);
594 
595 	rc = alloc_chrdev_region(&ocxl_dev, 0, OCXL_NUM_MINORS, "ocxl");
596 	if (rc) {
597 		pr_err("Unable to allocate ocxl major number: %d\n", rc);
598 		return rc;
599 	}
600 
601 	ocxl_class = class_create(THIS_MODULE, "ocxl");
602 	if (IS_ERR(ocxl_class)) {
603 		pr_err("Unable to create ocxl class\n");
604 		unregister_chrdev_region(ocxl_dev, OCXL_NUM_MINORS);
605 		return PTR_ERR(ocxl_class);
606 	}
607 
608 	ocxl_class->devnode = ocxl_devnode;
609 	return 0;
610 }
611 
612 void ocxl_file_exit(void)
613 {
614 	class_destroy(ocxl_class);
615 	unregister_chrdev_region(ocxl_dev, OCXL_NUM_MINORS);
616 	idr_destroy(&minors_idr);
617 }
618