xref: /openbmc/linux/drivers/virt/fsl_hypervisor.c (revision e3d786a3)
1 /*
2  * Freescale Hypervisor Management Driver
3 
4  * Copyright (C) 2008-2011 Freescale Semiconductor, Inc.
5  * Author: Timur Tabi <timur@freescale.com>
6  *
7  * This file is licensed under the terms of the GNU General Public License
8  * version 2.  This program is licensed "as is" without any warranty of any
9  * kind, whether express or implied.
10  *
11  * The Freescale hypervisor management driver provides several services to
12  * drivers and applications related to the Freescale hypervisor:
13  *
14  * 1. An ioctl interface for querying and managing partitions.
15  *
16  * 2. A file interface to reading incoming doorbells.
17  *
18  * 3. An interrupt handler for shutting down the partition upon receiving the
19  *    shutdown doorbell from a manager partition.
20  *
21  * 4. A kernel interface for receiving callbacks when a managed partition
22  *    shuts down.
23  */
24 
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/init.h>
28 #include <linux/types.h>
29 #include <linux/err.h>
30 #include <linux/fs.h>
31 #include <linux/miscdevice.h>
32 #include <linux/mm.h>
33 #include <linux/pagemap.h>
34 #include <linux/slab.h>
35 #include <linux/poll.h>
36 #include <linux/of.h>
37 #include <linux/of_irq.h>
38 #include <linux/reboot.h>
39 #include <linux/uaccess.h>
40 #include <linux/notifier.h>
41 #include <linux/interrupt.h>
42 
43 #include <linux/io.h>
44 #include <asm/fsl_hcalls.h>
45 
46 #include <linux/fsl_hypervisor.h>
47 
48 static BLOCKING_NOTIFIER_HEAD(failover_subscribers);
49 
50 /*
51  * Ioctl interface for FSL_HV_IOCTL_PARTITION_RESTART
52  *
53  * Restart a running partition
54  */
55 static long ioctl_restart(struct fsl_hv_ioctl_restart __user *p)
56 {
57 	struct fsl_hv_ioctl_restart param;
58 
59 	/* Get the parameters from the user */
60 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_restart)))
61 		return -EFAULT;
62 
63 	param.ret = fh_partition_restart(param.partition);
64 
65 	if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
66 		return -EFAULT;
67 
68 	return 0;
69 }
70 
71 /*
72  * Ioctl interface for FSL_HV_IOCTL_PARTITION_STATUS
73  *
74  * Query the status of a partition
75  */
76 static long ioctl_status(struct fsl_hv_ioctl_status __user *p)
77 {
78 	struct fsl_hv_ioctl_status param;
79 	u32 status;
80 
81 	/* Get the parameters from the user */
82 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_status)))
83 		return -EFAULT;
84 
85 	param.ret = fh_partition_get_status(param.partition, &status);
86 	if (!param.ret)
87 		param.status = status;
88 
89 	if (copy_to_user(p, &param, sizeof(struct fsl_hv_ioctl_status)))
90 		return -EFAULT;
91 
92 	return 0;
93 }
94 
95 /*
96  * Ioctl interface for FSL_HV_IOCTL_PARTITION_START
97  *
98  * Start a stopped partition.
99  */
100 static long ioctl_start(struct fsl_hv_ioctl_start __user *p)
101 {
102 	struct fsl_hv_ioctl_start param;
103 
104 	/* Get the parameters from the user */
105 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_start)))
106 		return -EFAULT;
107 
108 	param.ret = fh_partition_start(param.partition, param.entry_point,
109 				       param.load);
110 
111 	if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
112 		return -EFAULT;
113 
114 	return 0;
115 }
116 
117 /*
118  * Ioctl interface for FSL_HV_IOCTL_PARTITION_STOP
119  *
120  * Stop a running partition
121  */
122 static long ioctl_stop(struct fsl_hv_ioctl_stop __user *p)
123 {
124 	struct fsl_hv_ioctl_stop param;
125 
126 	/* Get the parameters from the user */
127 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_stop)))
128 		return -EFAULT;
129 
130 	param.ret = fh_partition_stop(param.partition);
131 
132 	if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
133 		return -EFAULT;
134 
135 	return 0;
136 }
137 
138 /*
139  * Ioctl interface for FSL_HV_IOCTL_MEMCPY
140  *
141  * The FH_MEMCPY hypercall takes an array of address/address/size structures
142  * to represent the data being copied.  As a convenience to the user, this
143  * ioctl takes a user-create buffer and a pointer to a guest physically
144  * contiguous buffer in the remote partition, and creates the
145  * address/address/size array for the hypercall.
146  */
147 static long ioctl_memcpy(struct fsl_hv_ioctl_memcpy __user *p)
148 {
149 	struct fsl_hv_ioctl_memcpy param;
150 
151 	struct page **pages = NULL;
152 	void *sg_list_unaligned = NULL;
153 	struct fh_sg_list *sg_list = NULL;
154 
155 	unsigned int num_pages;
156 	unsigned long lb_offset; /* Offset within a page of the local buffer */
157 
158 	unsigned int i;
159 	long ret = 0;
160 	int num_pinned; /* return value from get_user_pages() */
161 	phys_addr_t remote_paddr; /* The next address in the remote buffer */
162 	uint32_t count; /* The number of bytes left to copy */
163 
164 	/* Get the parameters from the user */
165 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_memcpy)))
166 		return -EFAULT;
167 
168 	/*
169 	 * One partition must be local, the other must be remote.  In other
170 	 * words, if source and target are both -1, or are both not -1, then
171 	 * return an error.
172 	 */
173 	if ((param.source == -1) == (param.target == -1))
174 		return -EINVAL;
175 
176 	/*
177 	 * The array of pages returned by get_user_pages() covers only
178 	 * page-aligned memory.  Since the user buffer is probably not
179 	 * page-aligned, we need to handle the discrepancy.
180 	 *
181 	 * We calculate the offset within a page of the S/G list, and make
182 	 * adjustments accordingly.  This will result in a page list that looks
183 	 * like this:
184 	 *
185 	 *      ----    <-- first page starts before the buffer
186 	 *     |    |
187 	 *     |////|-> ----
188 	 *     |////|  |    |
189 	 *      ----   |    |
190 	 *             |    |
191 	 *      ----   |    |
192 	 *     |////|  |    |
193 	 *     |////|  |    |
194 	 *     |////|  |    |
195 	 *      ----   |    |
196 	 *             |    |
197 	 *      ----   |    |
198 	 *     |////|  |    |
199 	 *     |////|  |    |
200 	 *     |////|  |    |
201 	 *      ----   |    |
202 	 *             |    |
203 	 *      ----   |    |
204 	 *     |////|  |    |
205 	 *     |////|-> ----
206 	 *     |    |   <-- last page ends after the buffer
207 	 *      ----
208 	 *
209 	 * The distance between the start of the first page and the start of the
210 	 * buffer is lb_offset.  The hashed (///) areas are the parts of the
211 	 * page list that contain the actual buffer.
212 	 *
213 	 * The advantage of this approach is that the number of pages is
214 	 * equal to the number of entries in the S/G list that we give to the
215 	 * hypervisor.
216 	 */
217 	lb_offset = param.local_vaddr & (PAGE_SIZE - 1);
218 	num_pages = (param.count + lb_offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
219 
220 	/* Allocate the buffers we need */
221 
222 	/*
223 	 * 'pages' is an array of struct page pointers that's initialized by
224 	 * get_user_pages().
225 	 */
226 	pages = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
227 	if (!pages) {
228 		pr_debug("fsl-hv: could not allocate page list\n");
229 		return -ENOMEM;
230 	}
231 
232 	/*
233 	 * sg_list is the list of fh_sg_list objects that we pass to the
234 	 * hypervisor.
235 	 */
236 	sg_list_unaligned = kmalloc(num_pages * sizeof(struct fh_sg_list) +
237 		sizeof(struct fh_sg_list) - 1, GFP_KERNEL);
238 	if (!sg_list_unaligned) {
239 		pr_debug("fsl-hv: could not allocate S/G list\n");
240 		ret = -ENOMEM;
241 		goto exit;
242 	}
243 	sg_list = PTR_ALIGN(sg_list_unaligned, sizeof(struct fh_sg_list));
244 
245 	/* Get the physical addresses of the source buffer */
246 	num_pinned = get_user_pages_fast(param.local_vaddr - lb_offset,
247 		num_pages, param.source != -1, pages);
248 
249 	if (num_pinned != num_pages) {
250 		/* get_user_pages() failed */
251 		pr_debug("fsl-hv: could not lock source buffer\n");
252 		ret = (num_pinned < 0) ? num_pinned : -EFAULT;
253 		goto exit;
254 	}
255 
256 	/*
257 	 * Build the fh_sg_list[] array.  The first page is special
258 	 * because it's misaligned.
259 	 */
260 	if (param.source == -1) {
261 		sg_list[0].source = page_to_phys(pages[0]) + lb_offset;
262 		sg_list[0].target = param.remote_paddr;
263 	} else {
264 		sg_list[0].source = param.remote_paddr;
265 		sg_list[0].target = page_to_phys(pages[0]) + lb_offset;
266 	}
267 	sg_list[0].size = min_t(uint64_t, param.count, PAGE_SIZE - lb_offset);
268 
269 	remote_paddr = param.remote_paddr + sg_list[0].size;
270 	count = param.count - sg_list[0].size;
271 
272 	for (i = 1; i < num_pages; i++) {
273 		if (param.source == -1) {
274 			/* local to remote */
275 			sg_list[i].source = page_to_phys(pages[i]);
276 			sg_list[i].target = remote_paddr;
277 		} else {
278 			/* remote to local */
279 			sg_list[i].source = remote_paddr;
280 			sg_list[i].target = page_to_phys(pages[i]);
281 		}
282 		sg_list[i].size = min_t(uint64_t, count, PAGE_SIZE);
283 
284 		remote_paddr += sg_list[i].size;
285 		count -= sg_list[i].size;
286 	}
287 
288 	param.ret = fh_partition_memcpy(param.source, param.target,
289 		virt_to_phys(sg_list), num_pages);
290 
291 exit:
292 	if (pages) {
293 		for (i = 0; i < num_pages; i++)
294 			if (pages[i])
295 				put_page(pages[i]);
296 	}
297 
298 	kfree(sg_list_unaligned);
299 	kfree(pages);
300 
301 	if (!ret)
302 		if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
303 			return -EFAULT;
304 
305 	return ret;
306 }
307 
308 /*
309  * Ioctl interface for FSL_HV_IOCTL_DOORBELL
310  *
311  * Ring a doorbell
312  */
313 static long ioctl_doorbell(struct fsl_hv_ioctl_doorbell __user *p)
314 {
315 	struct fsl_hv_ioctl_doorbell param;
316 
317 	/* Get the parameters from the user. */
318 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_doorbell)))
319 		return -EFAULT;
320 
321 	param.ret = ev_doorbell_send(param.doorbell);
322 
323 	if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
324 		return -EFAULT;
325 
326 	return 0;
327 }
328 
329 static long ioctl_dtprop(struct fsl_hv_ioctl_prop __user *p, int set)
330 {
331 	struct fsl_hv_ioctl_prop param;
332 	char __user *upath, *upropname;
333 	void __user *upropval;
334 	char *path = NULL, *propname = NULL;
335 	void *propval = NULL;
336 	int ret = 0;
337 
338 	/* Get the parameters from the user. */
339 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_prop)))
340 		return -EFAULT;
341 
342 	upath = (char __user *)(uintptr_t)param.path;
343 	upropname = (char __user *)(uintptr_t)param.propname;
344 	upropval = (void __user *)(uintptr_t)param.propval;
345 
346 	path = strndup_user(upath, FH_DTPROP_MAX_PATHLEN);
347 	if (IS_ERR(path)) {
348 		ret = PTR_ERR(path);
349 		goto out;
350 	}
351 
352 	propname = strndup_user(upropname, FH_DTPROP_MAX_PATHLEN);
353 	if (IS_ERR(propname)) {
354 		ret = PTR_ERR(propname);
355 		goto out;
356 	}
357 
358 	if (param.proplen > FH_DTPROP_MAX_PROPLEN) {
359 		ret = -EINVAL;
360 		goto out;
361 	}
362 
363 	propval = kmalloc(param.proplen, GFP_KERNEL);
364 	if (!propval) {
365 		ret = -ENOMEM;
366 		goto out;
367 	}
368 
369 	if (set) {
370 		if (copy_from_user(propval, upropval, param.proplen)) {
371 			ret = -EFAULT;
372 			goto out;
373 		}
374 
375 		param.ret = fh_partition_set_dtprop(param.handle,
376 						    virt_to_phys(path),
377 						    virt_to_phys(propname),
378 						    virt_to_phys(propval),
379 						    param.proplen);
380 	} else {
381 		param.ret = fh_partition_get_dtprop(param.handle,
382 						    virt_to_phys(path),
383 						    virt_to_phys(propname),
384 						    virt_to_phys(propval),
385 						    &param.proplen);
386 
387 		if (param.ret == 0) {
388 			if (copy_to_user(upropval, propval, param.proplen) ||
389 			    put_user(param.proplen, &p->proplen)) {
390 				ret = -EFAULT;
391 				goto out;
392 			}
393 		}
394 	}
395 
396 	if (put_user(param.ret, &p->ret))
397 		ret = -EFAULT;
398 
399 out:
400 	kfree(path);
401 	kfree(propval);
402 	kfree(propname);
403 
404 	return ret;
405 }
406 
407 /*
408  * Ioctl main entry point
409  */
410 static long fsl_hv_ioctl(struct file *file, unsigned int cmd,
411 			 unsigned long argaddr)
412 {
413 	void __user *arg = (void __user *)argaddr;
414 	long ret;
415 
416 	switch (cmd) {
417 	case FSL_HV_IOCTL_PARTITION_RESTART:
418 		ret = ioctl_restart(arg);
419 		break;
420 	case FSL_HV_IOCTL_PARTITION_GET_STATUS:
421 		ret = ioctl_status(arg);
422 		break;
423 	case FSL_HV_IOCTL_PARTITION_START:
424 		ret = ioctl_start(arg);
425 		break;
426 	case FSL_HV_IOCTL_PARTITION_STOP:
427 		ret = ioctl_stop(arg);
428 		break;
429 	case FSL_HV_IOCTL_MEMCPY:
430 		ret = ioctl_memcpy(arg);
431 		break;
432 	case FSL_HV_IOCTL_DOORBELL:
433 		ret = ioctl_doorbell(arg);
434 		break;
435 	case FSL_HV_IOCTL_GETPROP:
436 		ret = ioctl_dtprop(arg, 0);
437 		break;
438 	case FSL_HV_IOCTL_SETPROP:
439 		ret = ioctl_dtprop(arg, 1);
440 		break;
441 	default:
442 		pr_debug("fsl-hv: bad ioctl dir=%u type=%u cmd=%u size=%u\n",
443 			 _IOC_DIR(cmd), _IOC_TYPE(cmd), _IOC_NR(cmd),
444 			 _IOC_SIZE(cmd));
445 		return -ENOTTY;
446 	}
447 
448 	return ret;
449 }
450 
451 /* Linked list of processes that have us open */
452 static struct list_head db_list;
453 
454 /* spinlock for db_list */
455 static DEFINE_SPINLOCK(db_list_lock);
456 
457 /* The size of the doorbell event queue.  This must be a power of two. */
458 #define QSIZE	16
459 
460 /* Returns the next head/tail pointer, wrapping around the queue if necessary */
461 #define nextp(x) (((x) + 1) & (QSIZE - 1))
462 
463 /* Per-open data structure */
464 struct doorbell_queue {
465 	struct list_head list;
466 	spinlock_t lock;
467 	wait_queue_head_t wait;
468 	unsigned int head;
469 	unsigned int tail;
470 	uint32_t q[QSIZE];
471 };
472 
473 /* Linked list of ISRs that we registered */
474 struct list_head isr_list;
475 
476 /* Per-ISR data structure */
477 struct doorbell_isr {
478 	struct list_head list;
479 	unsigned int irq;
480 	uint32_t doorbell;	/* The doorbell handle */
481 	uint32_t partition;	/* The partition handle, if used */
482 };
483 
484 /*
485  * Add a doorbell to all of the doorbell queues
486  */
487 static void fsl_hv_queue_doorbell(uint32_t doorbell)
488 {
489 	struct doorbell_queue *dbq;
490 	unsigned long flags;
491 
492 	/* Prevent another core from modifying db_list */
493 	spin_lock_irqsave(&db_list_lock, flags);
494 
495 	list_for_each_entry(dbq, &db_list, list) {
496 		if (dbq->head != nextp(dbq->tail)) {
497 			dbq->q[dbq->tail] = doorbell;
498 			/*
499 			 * This memory barrier eliminates the need to grab
500 			 * the spinlock for dbq.
501 			 */
502 			smp_wmb();
503 			dbq->tail = nextp(dbq->tail);
504 			wake_up_interruptible(&dbq->wait);
505 		}
506 	}
507 
508 	spin_unlock_irqrestore(&db_list_lock, flags);
509 }
510 
511 /*
512  * Interrupt handler for all doorbells
513  *
514  * We use the same interrupt handler for all doorbells.  Whenever a doorbell
515  * is rung, and we receive an interrupt, we just put the handle for that
516  * doorbell (passed to us as *data) into all of the queues.
517  */
518 static irqreturn_t fsl_hv_isr(int irq, void *data)
519 {
520 	fsl_hv_queue_doorbell((uintptr_t) data);
521 
522 	return IRQ_HANDLED;
523 }
524 
525 /*
526  * State change thread function
527  *
528  * The state change notification arrives in an interrupt, but we can't call
529  * blocking_notifier_call_chain() in an interrupt handler.  We could call
530  * atomic_notifier_call_chain(), but that would require the clients' call-back
531  * function to run in interrupt context.  Since we don't want to impose that
532  * restriction on the clients, we use a threaded IRQ to process the
533  * notification in kernel context.
534  */
535 static irqreturn_t fsl_hv_state_change_thread(int irq, void *data)
536 {
537 	struct doorbell_isr *dbisr = data;
538 
539 	blocking_notifier_call_chain(&failover_subscribers, dbisr->partition,
540 				     NULL);
541 
542 	return IRQ_HANDLED;
543 }
544 
545 /*
546  * Interrupt handler for state-change doorbells
547  */
548 static irqreturn_t fsl_hv_state_change_isr(int irq, void *data)
549 {
550 	unsigned int status;
551 	struct doorbell_isr *dbisr = data;
552 	int ret;
553 
554 	/* It's still a doorbell, so add it to all the queues. */
555 	fsl_hv_queue_doorbell(dbisr->doorbell);
556 
557 	/* Determine the new state, and if it's stopped, notify the clients. */
558 	ret = fh_partition_get_status(dbisr->partition, &status);
559 	if (!ret && (status == FH_PARTITION_STOPPED))
560 		return IRQ_WAKE_THREAD;
561 
562 	return IRQ_HANDLED;
563 }
564 
565 /*
566  * Returns a bitmask indicating whether a read will block
567  */
568 static __poll_t fsl_hv_poll(struct file *filp, struct poll_table_struct *p)
569 {
570 	struct doorbell_queue *dbq = filp->private_data;
571 	unsigned long flags;
572 	__poll_t mask;
573 
574 	spin_lock_irqsave(&dbq->lock, flags);
575 
576 	poll_wait(filp, &dbq->wait, p);
577 	mask = (dbq->head == dbq->tail) ? 0 : (EPOLLIN | EPOLLRDNORM);
578 
579 	spin_unlock_irqrestore(&dbq->lock, flags);
580 
581 	return mask;
582 }
583 
584 /*
585  * Return the handles for any incoming doorbells
586  *
587  * If there are doorbell handles in the queue for this open instance, then
588  * return them to the caller as an array of 32-bit integers.  Otherwise,
589  * block until there is at least one handle to return.
590  */
591 static ssize_t fsl_hv_read(struct file *filp, char __user *buf, size_t len,
592 			   loff_t *off)
593 {
594 	struct doorbell_queue *dbq = filp->private_data;
595 	uint32_t __user *p = (uint32_t __user *) buf; /* for put_user() */
596 	unsigned long flags;
597 	ssize_t count = 0;
598 
599 	/* Make sure we stop when the user buffer is full. */
600 	while (len >= sizeof(uint32_t)) {
601 		uint32_t dbell;	/* Local copy of doorbell queue data */
602 
603 		spin_lock_irqsave(&dbq->lock, flags);
604 
605 		/*
606 		 * If the queue is empty, then either we're done or we need
607 		 * to block.  If the application specified O_NONBLOCK, then
608 		 * we return the appropriate error code.
609 		 */
610 		if (dbq->head == dbq->tail) {
611 			spin_unlock_irqrestore(&dbq->lock, flags);
612 			if (count)
613 				break;
614 			if (filp->f_flags & O_NONBLOCK)
615 				return -EAGAIN;
616 			if (wait_event_interruptible(dbq->wait,
617 						     dbq->head != dbq->tail))
618 				return -ERESTARTSYS;
619 			continue;
620 		}
621 
622 		/*
623 		 * Even though we have an smp_wmb() in the ISR, the core
624 		 * might speculatively execute the "dbell = ..." below while
625 		 * it's evaluating the if-statement above.  In that case, the
626 		 * value put into dbell could be stale if the core accepts the
627 		 * speculation. To prevent that, we need a read memory barrier
628 		 * here as well.
629 		 */
630 		smp_rmb();
631 
632 		/* Copy the data to a temporary local buffer, because
633 		 * we can't call copy_to_user() from inside a spinlock
634 		 */
635 		dbell = dbq->q[dbq->head];
636 		dbq->head = nextp(dbq->head);
637 
638 		spin_unlock_irqrestore(&dbq->lock, flags);
639 
640 		if (put_user(dbell, p))
641 			return -EFAULT;
642 		p++;
643 		count += sizeof(uint32_t);
644 		len -= sizeof(uint32_t);
645 	}
646 
647 	return count;
648 }
649 
650 /*
651  * Open the driver and prepare for reading doorbells.
652  *
653  * Every time an application opens the driver, we create a doorbell queue
654  * for that file handle.  This queue is used for any incoming doorbells.
655  */
656 static int fsl_hv_open(struct inode *inode, struct file *filp)
657 {
658 	struct doorbell_queue *dbq;
659 	unsigned long flags;
660 	int ret = 0;
661 
662 	dbq = kzalloc(sizeof(struct doorbell_queue), GFP_KERNEL);
663 	if (!dbq) {
664 		pr_err("fsl-hv: out of memory\n");
665 		return -ENOMEM;
666 	}
667 
668 	spin_lock_init(&dbq->lock);
669 	init_waitqueue_head(&dbq->wait);
670 
671 	spin_lock_irqsave(&db_list_lock, flags);
672 	list_add(&dbq->list, &db_list);
673 	spin_unlock_irqrestore(&db_list_lock, flags);
674 
675 	filp->private_data = dbq;
676 
677 	return ret;
678 }
679 
680 /*
681  * Close the driver
682  */
683 static int fsl_hv_close(struct inode *inode, struct file *filp)
684 {
685 	struct doorbell_queue *dbq = filp->private_data;
686 	unsigned long flags;
687 
688 	int ret = 0;
689 
690 	spin_lock_irqsave(&db_list_lock, flags);
691 	list_del(&dbq->list);
692 	spin_unlock_irqrestore(&db_list_lock, flags);
693 
694 	kfree(dbq);
695 
696 	return ret;
697 }
698 
699 static const struct file_operations fsl_hv_fops = {
700 	.owner = THIS_MODULE,
701 	.open = fsl_hv_open,
702 	.release = fsl_hv_close,
703 	.poll = fsl_hv_poll,
704 	.read = fsl_hv_read,
705 	.unlocked_ioctl = fsl_hv_ioctl,
706 	.compat_ioctl = fsl_hv_ioctl,
707 };
708 
709 static struct miscdevice fsl_hv_misc_dev = {
710 	MISC_DYNAMIC_MINOR,
711 	"fsl-hv",
712 	&fsl_hv_fops
713 };
714 
715 static irqreturn_t fsl_hv_shutdown_isr(int irq, void *data)
716 {
717 	orderly_poweroff(false);
718 
719 	return IRQ_HANDLED;
720 }
721 
722 /*
723  * Returns the handle of the parent of the given node
724  *
725  * The handle is the value of the 'hv-handle' property
726  */
727 static int get_parent_handle(struct device_node *np)
728 {
729 	struct device_node *parent;
730 	const uint32_t *prop;
731 	uint32_t handle;
732 	int len;
733 
734 	parent = of_get_parent(np);
735 	if (!parent)
736 		/* It's not really possible for this to fail */
737 		return -ENODEV;
738 
739 	/*
740 	 * The proper name for the handle property is "hv-handle", but some
741 	 * older versions of the hypervisor used "reg".
742 	 */
743 	prop = of_get_property(parent, "hv-handle", &len);
744 	if (!prop)
745 		prop = of_get_property(parent, "reg", &len);
746 
747 	if (!prop || (len != sizeof(uint32_t))) {
748 		/* This can happen only if the node is malformed */
749 		of_node_put(parent);
750 		return -ENODEV;
751 	}
752 
753 	handle = be32_to_cpup(prop);
754 	of_node_put(parent);
755 
756 	return handle;
757 }
758 
759 /*
760  * Register a callback for failover events
761  *
762  * This function is called by device drivers to register their callback
763  * functions for fail-over events.
764  */
765 int fsl_hv_failover_register(struct notifier_block *nb)
766 {
767 	return blocking_notifier_chain_register(&failover_subscribers, nb);
768 }
769 EXPORT_SYMBOL(fsl_hv_failover_register);
770 
771 /*
772  * Unregister a callback for failover events
773  */
774 int fsl_hv_failover_unregister(struct notifier_block *nb)
775 {
776 	return blocking_notifier_chain_unregister(&failover_subscribers, nb);
777 }
778 EXPORT_SYMBOL(fsl_hv_failover_unregister);
779 
780 /*
781  * Return TRUE if we're running under FSL hypervisor
782  *
783  * This function checks to see if we're running under the Freescale
784  * hypervisor, and returns zero if we're not, or non-zero if we are.
785  *
786  * First, it checks if MSR[GS]==1, which means we're running under some
787  * hypervisor.  Then it checks if there is a hypervisor node in the device
788  * tree.  Currently, that means there needs to be a node in the root called
789  * "hypervisor" and which has a property named "fsl,hv-version".
790  */
791 static int has_fsl_hypervisor(void)
792 {
793 	struct device_node *node;
794 	int ret;
795 
796 	node = of_find_node_by_path("/hypervisor");
797 	if (!node)
798 		return 0;
799 
800 	ret = of_find_property(node, "fsl,hv-version", NULL) != NULL;
801 
802 	of_node_put(node);
803 
804 	return ret;
805 }
806 
807 /*
808  * Freescale hypervisor management driver init
809  *
810  * This function is called when this module is loaded.
811  *
812  * Register ourselves as a miscellaneous driver.  This will register the
813  * fops structure and create the right sysfs entries for udev.
814  */
815 static int __init fsl_hypervisor_init(void)
816 {
817 	struct device_node *np;
818 	struct doorbell_isr *dbisr, *n;
819 	int ret;
820 
821 	pr_info("Freescale hypervisor management driver\n");
822 
823 	if (!has_fsl_hypervisor()) {
824 		pr_info("fsl-hv: no hypervisor found\n");
825 		return -ENODEV;
826 	}
827 
828 	ret = misc_register(&fsl_hv_misc_dev);
829 	if (ret) {
830 		pr_err("fsl-hv: cannot register device\n");
831 		return ret;
832 	}
833 
834 	INIT_LIST_HEAD(&db_list);
835 	INIT_LIST_HEAD(&isr_list);
836 
837 	for_each_compatible_node(np, NULL, "epapr,hv-receive-doorbell") {
838 		unsigned int irq;
839 		const uint32_t *handle;
840 
841 		handle = of_get_property(np, "interrupts", NULL);
842 		irq = irq_of_parse_and_map(np, 0);
843 		if (!handle || (irq == NO_IRQ)) {
844 			pr_err("fsl-hv: no 'interrupts' property in %pOF node\n",
845 				np);
846 			continue;
847 		}
848 
849 		dbisr = kzalloc(sizeof(*dbisr), GFP_KERNEL);
850 		if (!dbisr)
851 			goto out_of_memory;
852 
853 		dbisr->irq = irq;
854 		dbisr->doorbell = be32_to_cpup(handle);
855 
856 		if (of_device_is_compatible(np, "fsl,hv-shutdown-doorbell")) {
857 			/* The shutdown doorbell gets its own ISR */
858 			ret = request_irq(irq, fsl_hv_shutdown_isr, 0,
859 					  np->name, NULL);
860 		} else if (of_device_is_compatible(np,
861 			"fsl,hv-state-change-doorbell")) {
862 			/*
863 			 * The state change doorbell triggers a notification if
864 			 * the state of the managed partition changes to
865 			 * "stopped". We need a separate interrupt handler for
866 			 * that, and we also need to know the handle of the
867 			 * target partition, not just the handle of the
868 			 * doorbell.
869 			 */
870 			dbisr->partition = ret = get_parent_handle(np);
871 			if (ret < 0) {
872 				pr_err("fsl-hv: node %pOF has missing or "
873 				       "malformed parent\n", np);
874 				kfree(dbisr);
875 				continue;
876 			}
877 			ret = request_threaded_irq(irq, fsl_hv_state_change_isr,
878 						   fsl_hv_state_change_thread,
879 						   0, np->name, dbisr);
880 		} else
881 			ret = request_irq(irq, fsl_hv_isr, 0, np->name, dbisr);
882 
883 		if (ret < 0) {
884 			pr_err("fsl-hv: could not request irq %u for node %pOF\n",
885 			       irq, np);
886 			kfree(dbisr);
887 			continue;
888 		}
889 
890 		list_add(&dbisr->list, &isr_list);
891 
892 		pr_info("fsl-hv: registered handler for doorbell %u\n",
893 			dbisr->doorbell);
894 	}
895 
896 	return 0;
897 
898 out_of_memory:
899 	list_for_each_entry_safe(dbisr, n, &isr_list, list) {
900 		free_irq(dbisr->irq, dbisr);
901 		list_del(&dbisr->list);
902 		kfree(dbisr);
903 	}
904 
905 	misc_deregister(&fsl_hv_misc_dev);
906 
907 	return -ENOMEM;
908 }
909 
910 /*
911  * Freescale hypervisor management driver termination
912  *
913  * This function is called when this driver is unloaded.
914  */
915 static void __exit fsl_hypervisor_exit(void)
916 {
917 	struct doorbell_isr *dbisr, *n;
918 
919 	list_for_each_entry_safe(dbisr, n, &isr_list, list) {
920 		free_irq(dbisr->irq, dbisr);
921 		list_del(&dbisr->list);
922 		kfree(dbisr);
923 	}
924 
925 	misc_deregister(&fsl_hv_misc_dev);
926 }
927 
928 module_init(fsl_hypervisor_init);
929 module_exit(fsl_hypervisor_exit);
930 
931 MODULE_AUTHOR("Timur Tabi <timur@freescale.com>");
932 MODULE_DESCRIPTION("Freescale hypervisor management driver");
933 MODULE_LICENSE("GPL v2");
934