xref: /linux/drivers/misc/ocxl/file.c (revision 60684c2bd35064043360e6f716d1b7c20e967b7d)
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 DEFINE_MUTEX(minors_idr_lock);
19 static struct idr minors_idr;
20 
21 static struct ocxl_file_info *find_and_get_file_info(dev_t devno)
22 {
23 	struct ocxl_file_info *info;
24 
25 	mutex_lock(&minors_idr_lock);
26 	info = idr_find(&minors_idr, MINOR(devno));
27 	if (info)
28 		get_device(&info->dev);
29 	mutex_unlock(&minors_idr_lock);
30 	return info;
31 }
32 
33 static int allocate_minor(struct ocxl_file_info *info)
34 {
35 	int minor;
36 
37 	mutex_lock(&minors_idr_lock);
38 	minor = idr_alloc(&minors_idr, info, 0, OCXL_NUM_MINORS, GFP_KERNEL);
39 	mutex_unlock(&minors_idr_lock);
40 	return minor;
41 }
42 
43 static void free_minor(struct ocxl_file_info *info)
44 {
45 	mutex_lock(&minors_idr_lock);
46 	idr_remove(&minors_idr, MINOR(info->dev.devt));
47 	mutex_unlock(&minors_idr_lock);
48 }
49 
50 static int afu_open(struct inode *inode, struct file *file)
51 {
52 	struct ocxl_file_info *info;
53 	struct ocxl_context *ctx;
54 	int rc;
55 
56 	pr_debug("%s for device %x\n", __func__, inode->i_rdev);
57 
58 	info = find_and_get_file_info(inode->i_rdev);
59 	if (!info)
60 		return -ENODEV;
61 
62 	rc = ocxl_context_alloc(&ctx, info->afu, inode->i_mapping);
63 	if (rc) {
64 		put_device(&info->dev);
65 		return rc;
66 	}
67 	put_device(&info->dev);
68 	file->private_data = ctx;
69 	return 0;
70 }
71 
72 static long afu_ioctl_attach(struct ocxl_context *ctx,
73 			struct ocxl_ioctl_attach __user *uarg)
74 {
75 	struct ocxl_ioctl_attach arg;
76 	u64 amr = 0;
77 
78 	pr_debug("%s for context %d\n", __func__, ctx->pasid);
79 
80 	if (copy_from_user(&arg, uarg, sizeof(arg)))
81 		return -EFAULT;
82 
83 	/* Make sure reserved fields are not set for forward compatibility */
84 	if (arg.reserved1 || arg.reserved2 || arg.reserved3)
85 		return -EINVAL;
86 
87 	amr = arg.amr & mfspr(SPRN_UAMOR);
88 	return ocxl_context_attach(ctx, amr, current->mm);
89 }
90 
91 static long afu_ioctl_get_metadata(struct ocxl_context *ctx,
92 		struct ocxl_ioctl_metadata __user *uarg)
93 {
94 	struct ocxl_ioctl_metadata arg;
95 
96 	memset(&arg, 0, sizeof(arg));
97 
98 	arg.version = 0;
99 
100 	arg.afu_version_major = ctx->afu->config.version_major;
101 	arg.afu_version_minor = ctx->afu->config.version_minor;
102 	arg.pasid = ctx->pasid;
103 	arg.pp_mmio_size = ctx->afu->config.pp_mmio_stride;
104 	arg.global_mmio_size = ctx->afu->config.global_mmio_size;
105 
106 	if (copy_to_user(uarg, &arg, sizeof(arg)))
107 		return -EFAULT;
108 
109 	return 0;
110 }
111 
112 #ifdef CONFIG_PPC64
113 static long afu_ioctl_enable_p9_wait(struct ocxl_context *ctx,
114 		struct ocxl_ioctl_p9_wait __user *uarg)
115 {
116 	struct ocxl_ioctl_p9_wait arg;
117 
118 	memset(&arg, 0, sizeof(arg));
119 
120 	if (cpu_has_feature(CPU_FTR_P9_TIDR)) {
121 		enum ocxl_context_status status;
122 
123 		// Locks both status & tidr
124 		mutex_lock(&ctx->status_mutex);
125 		if (!ctx->tidr) {
126 			if (set_thread_tidr(current)) {
127 				mutex_unlock(&ctx->status_mutex);
128 				return -ENOENT;
129 			}
130 
131 			ctx->tidr = current->thread.tidr;
132 		}
133 
134 		status = ctx->status;
135 		mutex_unlock(&ctx->status_mutex);
136 
137 		if (status == ATTACHED) {
138 			int rc = ocxl_link_update_pe(ctx->afu->fn->link,
139 				ctx->pasid, ctx->tidr);
140 
141 			if (rc)
142 				return rc;
143 		}
144 
145 		arg.thread_id = ctx->tidr;
146 	} else
147 		return -ENOENT;
148 
149 	if (copy_to_user(uarg, &arg, sizeof(arg)))
150 		return -EFAULT;
151 
152 	return 0;
153 }
154 #endif
155 
156 
157 static long afu_ioctl_get_features(struct ocxl_context *ctx,
158 		struct ocxl_ioctl_features __user *uarg)
159 {
160 	struct ocxl_ioctl_features arg;
161 
162 	memset(&arg, 0, sizeof(arg));
163 
164 #ifdef CONFIG_PPC64
165 	if (cpu_has_feature(CPU_FTR_P9_TIDR))
166 		arg.flags[0] |= OCXL_IOCTL_FEATURES_FLAGS0_P9_WAIT;
167 #endif
168 
169 	if (copy_to_user(uarg, &arg, sizeof(arg)))
170 		return -EFAULT;
171 
172 	return 0;
173 }
174 
175 #define CMD_STR(x) (x == OCXL_IOCTL_ATTACH ? "ATTACH" :			\
176 			x == OCXL_IOCTL_IRQ_ALLOC ? "IRQ_ALLOC" :	\
177 			x == OCXL_IOCTL_IRQ_FREE ? "IRQ_FREE" :		\
178 			x == OCXL_IOCTL_IRQ_SET_FD ? "IRQ_SET_FD" :	\
179 			x == OCXL_IOCTL_GET_METADATA ? "GET_METADATA" :	\
180 			x == OCXL_IOCTL_ENABLE_P9_WAIT ? "ENABLE_P9_WAIT" :	\
181 			x == OCXL_IOCTL_GET_FEATURES ? "GET_FEATURES" :	\
182 			"UNKNOWN")
183 
184 static irqreturn_t irq_handler(void *private)
185 {
186 	struct eventfd_ctx *ev_ctx = private;
187 
188 	eventfd_signal(ev_ctx, 1);
189 	return IRQ_HANDLED;
190 }
191 
192 static void irq_free(void *private)
193 {
194 	struct eventfd_ctx *ev_ctx = private;
195 
196 	eventfd_ctx_put(ev_ctx);
197 }
198 
199 static long afu_ioctl(struct file *file, unsigned int cmd,
200 		unsigned long args)
201 {
202 	struct ocxl_context *ctx = file->private_data;
203 	struct ocxl_ioctl_irq_fd irq_fd;
204 	struct eventfd_ctx *ev_ctx;
205 	int irq_id;
206 	u64 irq_offset;
207 	long rc;
208 	bool closed;
209 
210 	pr_debug("%s for context %d, command %s\n", __func__, ctx->pasid,
211 		CMD_STR(cmd));
212 
213 	mutex_lock(&ctx->status_mutex);
214 	closed = (ctx->status == CLOSED);
215 	mutex_unlock(&ctx->status_mutex);
216 
217 	if (closed)
218 		return -EIO;
219 
220 	switch (cmd) {
221 	case OCXL_IOCTL_ATTACH:
222 		rc = afu_ioctl_attach(ctx,
223 				(struct ocxl_ioctl_attach __user *) args);
224 		break;
225 
226 	case OCXL_IOCTL_IRQ_ALLOC:
227 		rc = ocxl_afu_irq_alloc(ctx, &irq_id);
228 		if (!rc) {
229 			irq_offset = ocxl_irq_id_to_offset(ctx, irq_id);
230 			rc = copy_to_user((u64 __user *) args, &irq_offset,
231 					sizeof(irq_offset));
232 			if (rc) {
233 				ocxl_afu_irq_free(ctx, irq_id);
234 				return -EFAULT;
235 			}
236 		}
237 		break;
238 
239 	case OCXL_IOCTL_IRQ_FREE:
240 		rc = copy_from_user(&irq_offset, (u64 __user *) args,
241 				sizeof(irq_offset));
242 		if (rc)
243 			return -EFAULT;
244 		irq_id = ocxl_irq_offset_to_id(ctx, irq_offset);
245 		rc = ocxl_afu_irq_free(ctx, irq_id);
246 		break;
247 
248 	case OCXL_IOCTL_IRQ_SET_FD:
249 		rc = copy_from_user(&irq_fd, (u64 __user *) args,
250 				sizeof(irq_fd));
251 		if (rc)
252 			return -EFAULT;
253 		if (irq_fd.reserved)
254 			return -EINVAL;
255 		irq_id = ocxl_irq_offset_to_id(ctx, irq_fd.irq_offset);
256 		ev_ctx = eventfd_ctx_fdget(irq_fd.eventfd);
257 		if (IS_ERR(ev_ctx))
258 			return PTR_ERR(ev_ctx);
259 		rc = ocxl_irq_set_handler(ctx, irq_id, irq_handler, irq_free, ev_ctx);
260 		if (rc)
261 			eventfd_ctx_put(ev_ctx);
262 		break;
263 
264 	case OCXL_IOCTL_GET_METADATA:
265 		rc = afu_ioctl_get_metadata(ctx,
266 				(struct ocxl_ioctl_metadata __user *) args);
267 		break;
268 
269 #ifdef CONFIG_PPC64
270 	case OCXL_IOCTL_ENABLE_P9_WAIT:
271 		rc = afu_ioctl_enable_p9_wait(ctx,
272 				(struct ocxl_ioctl_p9_wait __user *) args);
273 		break;
274 #endif
275 
276 	case OCXL_IOCTL_GET_FEATURES:
277 		rc = afu_ioctl_get_features(ctx,
278 				(struct ocxl_ioctl_features __user *) args);
279 		break;
280 
281 	default:
282 		rc = -EINVAL;
283 	}
284 	return rc;
285 }
286 
287 static long afu_compat_ioctl(struct file *file, unsigned int cmd,
288 			unsigned long args)
289 {
290 	return afu_ioctl(file, cmd, args);
291 }
292 
293 static int afu_mmap(struct file *file, struct vm_area_struct *vma)
294 {
295 	struct ocxl_context *ctx = file->private_data;
296 
297 	pr_debug("%s for context %d\n", __func__, ctx->pasid);
298 	return ocxl_context_mmap(ctx, vma);
299 }
300 
301 static bool has_xsl_error(struct ocxl_context *ctx)
302 {
303 	bool ret;
304 
305 	mutex_lock(&ctx->xsl_error_lock);
306 	ret = !!ctx->xsl_error.addr;
307 	mutex_unlock(&ctx->xsl_error_lock);
308 
309 	return ret;
310 }
311 
312 /*
313  * Are there any events pending on the AFU
314  * ctx: The AFU context
315  * Returns: true if there are events pending
316  */
317 static bool afu_events_pending(struct ocxl_context *ctx)
318 {
319 	if (has_xsl_error(ctx))
320 		return true;
321 	return false;
322 }
323 
324 static unsigned int afu_poll(struct file *file, struct poll_table_struct *wait)
325 {
326 	struct ocxl_context *ctx = file->private_data;
327 	unsigned int mask = 0;
328 	bool closed;
329 
330 	pr_debug("%s for context %d\n", __func__, ctx->pasid);
331 
332 	poll_wait(file, &ctx->events_wq, wait);
333 
334 	mutex_lock(&ctx->status_mutex);
335 	closed = (ctx->status == CLOSED);
336 	mutex_unlock(&ctx->status_mutex);
337 
338 	if (afu_events_pending(ctx))
339 		mask = EPOLLIN | EPOLLRDNORM;
340 	else if (closed)
341 		mask = EPOLLERR;
342 
343 	return mask;
344 }
345 
346 /*
347  * Populate the supplied buffer with a single XSL error
348  * ctx:	The AFU context to report the error from
349  * header: the event header to populate
350  * buf: The buffer to write the body into (should be at least
351  *      AFU_EVENT_BODY_XSL_ERROR_SIZE)
352  * Return: the amount of buffer that was populated
353  */
354 static ssize_t append_xsl_error(struct ocxl_context *ctx,
355 				struct ocxl_kernel_event_header *header,
356 				char __user *buf)
357 {
358 	struct ocxl_kernel_event_xsl_fault_error body;
359 
360 	memset(&body, 0, sizeof(body));
361 
362 	mutex_lock(&ctx->xsl_error_lock);
363 	if (!ctx->xsl_error.addr) {
364 		mutex_unlock(&ctx->xsl_error_lock);
365 		return 0;
366 	}
367 
368 	body.addr = ctx->xsl_error.addr;
369 	body.dsisr = ctx->xsl_error.dsisr;
370 	body.count = ctx->xsl_error.count;
371 
372 	ctx->xsl_error.addr = 0;
373 	ctx->xsl_error.dsisr = 0;
374 	ctx->xsl_error.count = 0;
375 
376 	mutex_unlock(&ctx->xsl_error_lock);
377 
378 	header->type = OCXL_AFU_EVENT_XSL_FAULT_ERROR;
379 
380 	if (copy_to_user(buf, &body, sizeof(body)))
381 		return -EFAULT;
382 
383 	return sizeof(body);
384 }
385 
386 #define AFU_EVENT_BODY_MAX_SIZE sizeof(struct ocxl_kernel_event_xsl_fault_error)
387 
388 /*
389  * Reports events on the AFU
390  * Format:
391  *	Header (struct ocxl_kernel_event_header)
392  *	Body (struct ocxl_kernel_event_*)
393  *	Header...
394  */
395 static ssize_t afu_read(struct file *file, char __user *buf, size_t count,
396 			loff_t *off)
397 {
398 	struct ocxl_context *ctx = file->private_data;
399 	struct ocxl_kernel_event_header header;
400 	ssize_t rc;
401 	ssize_t used = 0;
402 	DEFINE_WAIT(event_wait);
403 
404 	memset(&header, 0, sizeof(header));
405 
406 	/* Require offset to be 0 */
407 	if (*off != 0)
408 		return -EINVAL;
409 
410 	if (count < (sizeof(struct ocxl_kernel_event_header) +
411 			AFU_EVENT_BODY_MAX_SIZE))
412 		return -EINVAL;
413 
414 	for (;;) {
415 		prepare_to_wait(&ctx->events_wq, &event_wait,
416 				TASK_INTERRUPTIBLE);
417 
418 		if (afu_events_pending(ctx))
419 			break;
420 
421 		if (ctx->status == CLOSED)
422 			break;
423 
424 		if (file->f_flags & O_NONBLOCK) {
425 			finish_wait(&ctx->events_wq, &event_wait);
426 			return -EAGAIN;
427 		}
428 
429 		if (signal_pending(current)) {
430 			finish_wait(&ctx->events_wq, &event_wait);
431 			return -ERESTARTSYS;
432 		}
433 
434 		schedule();
435 	}
436 
437 	finish_wait(&ctx->events_wq, &event_wait);
438 
439 	if (has_xsl_error(ctx)) {
440 		used = append_xsl_error(ctx, &header, buf + sizeof(header));
441 		if (used < 0)
442 			return used;
443 	}
444 
445 	if (!afu_events_pending(ctx))
446 		header.flags |= OCXL_KERNEL_EVENT_FLAG_LAST;
447 
448 	if (copy_to_user(buf, &header, sizeof(header)))
449 		return -EFAULT;
450 
451 	used += sizeof(header);
452 
453 	rc = used;
454 	return rc;
455 }
456 
457 static int afu_release(struct inode *inode, struct file *file)
458 {
459 	struct ocxl_context *ctx = file->private_data;
460 	int rc;
461 
462 	pr_debug("%s for device %x\n", __func__, inode->i_rdev);
463 	rc = ocxl_context_detach(ctx);
464 	mutex_lock(&ctx->mapping_lock);
465 	ctx->mapping = NULL;
466 	mutex_unlock(&ctx->mapping_lock);
467 	wake_up_all(&ctx->events_wq);
468 	if (rc != -EBUSY)
469 		ocxl_context_free(ctx);
470 	return 0;
471 }
472 
473 static const struct file_operations ocxl_afu_fops = {
474 	.owner		= THIS_MODULE,
475 	.open           = afu_open,
476 	.unlocked_ioctl = afu_ioctl,
477 	.compat_ioctl   = afu_compat_ioctl,
478 	.mmap           = afu_mmap,
479 	.poll           = afu_poll,
480 	.read           = afu_read,
481 	.release        = afu_release,
482 };
483 
484 // Free the info struct
485 static void info_release(struct device *dev)
486 {
487 	struct ocxl_file_info *info = container_of(dev, struct ocxl_file_info, dev);
488 
489 	ocxl_afu_put(info->afu);
490 	kfree(info);
491 }
492 
493 static int ocxl_file_make_visible(struct ocxl_file_info *info)
494 {
495 	int rc;
496 
497 	cdev_init(&info->cdev, &ocxl_afu_fops);
498 	rc = cdev_add(&info->cdev, info->dev.devt, 1);
499 	if (rc) {
500 		dev_err(&info->dev, "Unable to add afu char device: %d\n", rc);
501 		return rc;
502 	}
503 
504 	return 0;
505 }
506 
507 static void ocxl_file_make_invisible(struct ocxl_file_info *info)
508 {
509 	cdev_del(&info->cdev);
510 }
511 
512 int ocxl_file_register_afu(struct ocxl_afu *afu)
513 {
514 	int minor;
515 	int rc;
516 	struct ocxl_file_info *info;
517 	struct ocxl_fn *fn = afu->fn;
518 	struct pci_dev *pci_dev = to_pci_dev(fn->dev.parent);
519 
520 	info = kzalloc(sizeof(*info), GFP_KERNEL);
521 	if (info == NULL)
522 		return -ENOMEM;
523 
524 	minor = allocate_minor(info);
525 	if (minor < 0) {
526 		kfree(info);
527 		return minor;
528 	}
529 
530 	info->dev.parent = &fn->dev;
531 	info->dev.devt = MKDEV(MAJOR(ocxl_dev), minor);
532 	info->dev.class = ocxl_class;
533 	info->dev.release = info_release;
534 
535 	info->afu = afu;
536 	ocxl_afu_get(afu);
537 
538 	rc = dev_set_name(&info->dev, "%s.%s.%hhu",
539 		afu->config.name, dev_name(&pci_dev->dev), afu->config.idx);
540 	if (rc)
541 		goto err_put;
542 
543 	rc = device_register(&info->dev);
544 	if (rc) {
545 		free_minor(info);
546 		put_device(&info->dev);
547 		return rc;
548 	}
549 
550 	rc = ocxl_sysfs_register_afu(info);
551 	if (rc)
552 		goto err_unregister;
553 
554 	rc = ocxl_file_make_visible(info);
555 	if (rc)
556 		goto err_unregister;
557 
558 	ocxl_afu_set_private(afu, info);
559 
560 	return 0;
561 
562 err_unregister:
563 	ocxl_sysfs_unregister_afu(info); // safe to call even if register failed
564 	free_minor(info);
565 	device_unregister(&info->dev);
566 	return rc;
567 err_put:
568 	ocxl_afu_put(afu);
569 	free_minor(info);
570 	kfree(info);
571 	return rc;
572 }
573 
574 void ocxl_file_unregister_afu(struct ocxl_afu *afu)
575 {
576 	struct ocxl_file_info *info = ocxl_afu_get_private(afu);
577 
578 	if (!info)
579 		return;
580 
581 	ocxl_file_make_invisible(info);
582 	ocxl_sysfs_unregister_afu(info);
583 	free_minor(info);
584 	device_unregister(&info->dev);
585 }
586 
587 static char *ocxl_devnode(const struct device *dev, umode_t *mode)
588 {
589 	return kasprintf(GFP_KERNEL, "ocxl/%s", dev_name(dev));
590 }
591 
592 int ocxl_file_init(void)
593 {
594 	int rc;
595 
596 	idr_init(&minors_idr);
597 
598 	rc = alloc_chrdev_region(&ocxl_dev, 0, OCXL_NUM_MINORS, "ocxl");
599 	if (rc) {
600 		pr_err("Unable to allocate ocxl major number: %d\n", rc);
601 		return rc;
602 	}
603 
604 	ocxl_class = class_create(THIS_MODULE, "ocxl");
605 	if (IS_ERR(ocxl_class)) {
606 		pr_err("Unable to create ocxl class\n");
607 		unregister_chrdev_region(ocxl_dev, OCXL_NUM_MINORS);
608 		return PTR_ERR(ocxl_class);
609 	}
610 
611 	ocxl_class->devnode = ocxl_devnode;
612 	return 0;
613 }
614 
615 void ocxl_file_exit(void)
616 {
617 	class_destroy(ocxl_class);
618 	unregister_chrdev_region(ocxl_dev, OCXL_NUM_MINORS);
619 	idr_destroy(&minors_idr);
620 }
621