xref: /openbmc/linux/drivers/virt/fsl_hypervisor.c (revision c819e2cf)
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 = kzalloc(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 	down_read(&current->mm->mmap_sem);
247 	num_pinned = get_user_pages(current, current->mm,
248 		param.local_vaddr - lb_offset, num_pages,
249 		(param.source == -1) ? READ : WRITE,
250 		0, pages, NULL);
251 	up_read(&current->mm->mmap_sem);
252 
253 	if (num_pinned != num_pages) {
254 		/* get_user_pages() failed */
255 		pr_debug("fsl-hv: could not lock source buffer\n");
256 		ret = (num_pinned < 0) ? num_pinned : -EFAULT;
257 		goto exit;
258 	}
259 
260 	/*
261 	 * Build the fh_sg_list[] array.  The first page is special
262 	 * because it's misaligned.
263 	 */
264 	if (param.source == -1) {
265 		sg_list[0].source = page_to_phys(pages[0]) + lb_offset;
266 		sg_list[0].target = param.remote_paddr;
267 	} else {
268 		sg_list[0].source = param.remote_paddr;
269 		sg_list[0].target = page_to_phys(pages[0]) + lb_offset;
270 	}
271 	sg_list[0].size = min_t(uint64_t, param.count, PAGE_SIZE - lb_offset);
272 
273 	remote_paddr = param.remote_paddr + sg_list[0].size;
274 	count = param.count - sg_list[0].size;
275 
276 	for (i = 1; i < num_pages; i++) {
277 		if (param.source == -1) {
278 			/* local to remote */
279 			sg_list[i].source = page_to_phys(pages[i]);
280 			sg_list[i].target = remote_paddr;
281 		} else {
282 			/* remote to local */
283 			sg_list[i].source = remote_paddr;
284 			sg_list[i].target = page_to_phys(pages[i]);
285 		}
286 		sg_list[i].size = min_t(uint64_t, count, PAGE_SIZE);
287 
288 		remote_paddr += sg_list[i].size;
289 		count -= sg_list[i].size;
290 	}
291 
292 	param.ret = fh_partition_memcpy(param.source, param.target,
293 		virt_to_phys(sg_list), num_pages);
294 
295 exit:
296 	if (pages) {
297 		for (i = 0; i < num_pages; i++)
298 			if (pages[i])
299 				put_page(pages[i]);
300 	}
301 
302 	kfree(sg_list_unaligned);
303 	kfree(pages);
304 
305 	if (!ret)
306 		if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
307 			return -EFAULT;
308 
309 	return ret;
310 }
311 
312 /*
313  * Ioctl interface for FSL_HV_IOCTL_DOORBELL
314  *
315  * Ring a doorbell
316  */
317 static long ioctl_doorbell(struct fsl_hv_ioctl_doorbell __user *p)
318 {
319 	struct fsl_hv_ioctl_doorbell param;
320 
321 	/* Get the parameters from the user. */
322 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_doorbell)))
323 		return -EFAULT;
324 
325 	param.ret = ev_doorbell_send(param.doorbell);
326 
327 	if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
328 		return -EFAULT;
329 
330 	return 0;
331 }
332 
333 static long ioctl_dtprop(struct fsl_hv_ioctl_prop __user *p, int set)
334 {
335 	struct fsl_hv_ioctl_prop param;
336 	char __user *upath, *upropname;
337 	void __user *upropval;
338 	char *path = NULL, *propname = NULL;
339 	void *propval = NULL;
340 	int ret = 0;
341 
342 	/* Get the parameters from the user. */
343 	if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_prop)))
344 		return -EFAULT;
345 
346 	upath = (char __user *)(uintptr_t)param.path;
347 	upropname = (char __user *)(uintptr_t)param.propname;
348 	upropval = (void __user *)(uintptr_t)param.propval;
349 
350 	path = strndup_user(upath, FH_DTPROP_MAX_PATHLEN);
351 	if (IS_ERR(path)) {
352 		ret = PTR_ERR(path);
353 		goto out;
354 	}
355 
356 	propname = strndup_user(upropname, FH_DTPROP_MAX_PATHLEN);
357 	if (IS_ERR(propname)) {
358 		ret = PTR_ERR(propname);
359 		goto out;
360 	}
361 
362 	if (param.proplen > FH_DTPROP_MAX_PROPLEN) {
363 		ret = -EINVAL;
364 		goto out;
365 	}
366 
367 	propval = kmalloc(param.proplen, GFP_KERNEL);
368 	if (!propval) {
369 		ret = -ENOMEM;
370 		goto out;
371 	}
372 
373 	if (set) {
374 		if (copy_from_user(propval, upropval, param.proplen)) {
375 			ret = -EFAULT;
376 			goto out;
377 		}
378 
379 		param.ret = fh_partition_set_dtprop(param.handle,
380 						    virt_to_phys(path),
381 						    virt_to_phys(propname),
382 						    virt_to_phys(propval),
383 						    param.proplen);
384 	} else {
385 		param.ret = fh_partition_get_dtprop(param.handle,
386 						    virt_to_phys(path),
387 						    virt_to_phys(propname),
388 						    virt_to_phys(propval),
389 						    &param.proplen);
390 
391 		if (param.ret == 0) {
392 			if (copy_to_user(upropval, propval, param.proplen) ||
393 			    put_user(param.proplen, &p->proplen)) {
394 				ret = -EFAULT;
395 				goto out;
396 			}
397 		}
398 	}
399 
400 	if (put_user(param.ret, &p->ret))
401 		ret = -EFAULT;
402 
403 out:
404 	kfree(path);
405 	kfree(propval);
406 	kfree(propname);
407 
408 	return ret;
409 }
410 
411 /*
412  * Ioctl main entry point
413  */
414 static long fsl_hv_ioctl(struct file *file, unsigned int cmd,
415 			 unsigned long argaddr)
416 {
417 	void __user *arg = (void __user *)argaddr;
418 	long ret;
419 
420 	switch (cmd) {
421 	case FSL_HV_IOCTL_PARTITION_RESTART:
422 		ret = ioctl_restart(arg);
423 		break;
424 	case FSL_HV_IOCTL_PARTITION_GET_STATUS:
425 		ret = ioctl_status(arg);
426 		break;
427 	case FSL_HV_IOCTL_PARTITION_START:
428 		ret = ioctl_start(arg);
429 		break;
430 	case FSL_HV_IOCTL_PARTITION_STOP:
431 		ret = ioctl_stop(arg);
432 		break;
433 	case FSL_HV_IOCTL_MEMCPY:
434 		ret = ioctl_memcpy(arg);
435 		break;
436 	case FSL_HV_IOCTL_DOORBELL:
437 		ret = ioctl_doorbell(arg);
438 		break;
439 	case FSL_HV_IOCTL_GETPROP:
440 		ret = ioctl_dtprop(arg, 0);
441 		break;
442 	case FSL_HV_IOCTL_SETPROP:
443 		ret = ioctl_dtprop(arg, 1);
444 		break;
445 	default:
446 		pr_debug("fsl-hv: bad ioctl dir=%u type=%u cmd=%u size=%u\n",
447 			 _IOC_DIR(cmd), _IOC_TYPE(cmd), _IOC_NR(cmd),
448 			 _IOC_SIZE(cmd));
449 		return -ENOTTY;
450 	}
451 
452 	return ret;
453 }
454 
455 /* Linked list of processes that have us open */
456 static struct list_head db_list;
457 
458 /* spinlock for db_list */
459 static DEFINE_SPINLOCK(db_list_lock);
460 
461 /* The size of the doorbell event queue.  This must be a power of two. */
462 #define QSIZE	16
463 
464 /* Returns the next head/tail pointer, wrapping around the queue if necessary */
465 #define nextp(x) (((x) + 1) & (QSIZE - 1))
466 
467 /* Per-open data structure */
468 struct doorbell_queue {
469 	struct list_head list;
470 	spinlock_t lock;
471 	wait_queue_head_t wait;
472 	unsigned int head;
473 	unsigned int tail;
474 	uint32_t q[QSIZE];
475 };
476 
477 /* Linked list of ISRs that we registered */
478 struct list_head isr_list;
479 
480 /* Per-ISR data structure */
481 struct doorbell_isr {
482 	struct list_head list;
483 	unsigned int irq;
484 	uint32_t doorbell;	/* The doorbell handle */
485 	uint32_t partition;	/* The partition handle, if used */
486 };
487 
488 /*
489  * Add a doorbell to all of the doorbell queues
490  */
491 static void fsl_hv_queue_doorbell(uint32_t doorbell)
492 {
493 	struct doorbell_queue *dbq;
494 	unsigned long flags;
495 
496 	/* Prevent another core from modifying db_list */
497 	spin_lock_irqsave(&db_list_lock, flags);
498 
499 	list_for_each_entry(dbq, &db_list, list) {
500 		if (dbq->head != nextp(dbq->tail)) {
501 			dbq->q[dbq->tail] = doorbell;
502 			/*
503 			 * This memory barrier eliminates the need to grab
504 			 * the spinlock for dbq.
505 			 */
506 			smp_wmb();
507 			dbq->tail = nextp(dbq->tail);
508 			wake_up_interruptible(&dbq->wait);
509 		}
510 	}
511 
512 	spin_unlock_irqrestore(&db_list_lock, flags);
513 }
514 
515 /*
516  * Interrupt handler for all doorbells
517  *
518  * We use the same interrupt handler for all doorbells.  Whenever a doorbell
519  * is rung, and we receive an interrupt, we just put the handle for that
520  * doorbell (passed to us as *data) into all of the queues.
521  */
522 static irqreturn_t fsl_hv_isr(int irq, void *data)
523 {
524 	fsl_hv_queue_doorbell((uintptr_t) data);
525 
526 	return IRQ_HANDLED;
527 }
528 
529 /*
530  * State change thread function
531  *
532  * The state change notification arrives in an interrupt, but we can't call
533  * blocking_notifier_call_chain() in an interrupt handler.  We could call
534  * atomic_notifier_call_chain(), but that would require the clients' call-back
535  * function to run in interrupt context.  Since we don't want to impose that
536  * restriction on the clients, we use a threaded IRQ to process the
537  * notification in kernel context.
538  */
539 static irqreturn_t fsl_hv_state_change_thread(int irq, void *data)
540 {
541 	struct doorbell_isr *dbisr = data;
542 
543 	blocking_notifier_call_chain(&failover_subscribers, dbisr->partition,
544 				     NULL);
545 
546 	return IRQ_HANDLED;
547 }
548 
549 /*
550  * Interrupt handler for state-change doorbells
551  */
552 static irqreturn_t fsl_hv_state_change_isr(int irq, void *data)
553 {
554 	unsigned int status;
555 	struct doorbell_isr *dbisr = data;
556 	int ret;
557 
558 	/* It's still a doorbell, so add it to all the queues. */
559 	fsl_hv_queue_doorbell(dbisr->doorbell);
560 
561 	/* Determine the new state, and if it's stopped, notify the clients. */
562 	ret = fh_partition_get_status(dbisr->partition, &status);
563 	if (!ret && (status == FH_PARTITION_STOPPED))
564 		return IRQ_WAKE_THREAD;
565 
566 	return IRQ_HANDLED;
567 }
568 
569 /*
570  * Returns a bitmask indicating whether a read will block
571  */
572 static unsigned int fsl_hv_poll(struct file *filp, struct poll_table_struct *p)
573 {
574 	struct doorbell_queue *dbq = filp->private_data;
575 	unsigned long flags;
576 	unsigned int mask;
577 
578 	spin_lock_irqsave(&dbq->lock, flags);
579 
580 	poll_wait(filp, &dbq->wait, p);
581 	mask = (dbq->head == dbq->tail) ? 0 : (POLLIN | POLLRDNORM);
582 
583 	spin_unlock_irqrestore(&dbq->lock, flags);
584 
585 	return mask;
586 }
587 
588 /*
589  * Return the handles for any incoming doorbells
590  *
591  * If there are doorbell handles in the queue for this open instance, then
592  * return them to the caller as an array of 32-bit integers.  Otherwise,
593  * block until there is at least one handle to return.
594  */
595 static ssize_t fsl_hv_read(struct file *filp, char __user *buf, size_t len,
596 			   loff_t *off)
597 {
598 	struct doorbell_queue *dbq = filp->private_data;
599 	uint32_t __user *p = (uint32_t __user *) buf; /* for put_user() */
600 	unsigned long flags;
601 	ssize_t count = 0;
602 
603 	/* Make sure we stop when the user buffer is full. */
604 	while (len >= sizeof(uint32_t)) {
605 		uint32_t dbell;	/* Local copy of doorbell queue data */
606 
607 		spin_lock_irqsave(&dbq->lock, flags);
608 
609 		/*
610 		 * If the queue is empty, then either we're done or we need
611 		 * to block.  If the application specified O_NONBLOCK, then
612 		 * we return the appropriate error code.
613 		 */
614 		if (dbq->head == dbq->tail) {
615 			spin_unlock_irqrestore(&dbq->lock, flags);
616 			if (count)
617 				break;
618 			if (filp->f_flags & O_NONBLOCK)
619 				return -EAGAIN;
620 			if (wait_event_interruptible(dbq->wait,
621 						     dbq->head != dbq->tail))
622 				return -ERESTARTSYS;
623 			continue;
624 		}
625 
626 		/*
627 		 * Even though we have an smp_wmb() in the ISR, the core
628 		 * might speculatively execute the "dbell = ..." below while
629 		 * it's evaluating the if-statement above.  In that case, the
630 		 * value put into dbell could be stale if the core accepts the
631 		 * speculation. To prevent that, we need a read memory barrier
632 		 * here as well.
633 		 */
634 		smp_rmb();
635 
636 		/* Copy the data to a temporary local buffer, because
637 		 * we can't call copy_to_user() from inside a spinlock
638 		 */
639 		dbell = dbq->q[dbq->head];
640 		dbq->head = nextp(dbq->head);
641 
642 		spin_unlock_irqrestore(&dbq->lock, flags);
643 
644 		if (put_user(dbell, p))
645 			return -EFAULT;
646 		p++;
647 		count += sizeof(uint32_t);
648 		len -= sizeof(uint32_t);
649 	}
650 
651 	return count;
652 }
653 
654 /*
655  * Open the driver and prepare for reading doorbells.
656  *
657  * Every time an application opens the driver, we create a doorbell queue
658  * for that file handle.  This queue is used for any incoming doorbells.
659  */
660 static int fsl_hv_open(struct inode *inode, struct file *filp)
661 {
662 	struct doorbell_queue *dbq;
663 	unsigned long flags;
664 	int ret = 0;
665 
666 	dbq = kzalloc(sizeof(struct doorbell_queue), GFP_KERNEL);
667 	if (!dbq) {
668 		pr_err("fsl-hv: out of memory\n");
669 		return -ENOMEM;
670 	}
671 
672 	spin_lock_init(&dbq->lock);
673 	init_waitqueue_head(&dbq->wait);
674 
675 	spin_lock_irqsave(&db_list_lock, flags);
676 	list_add(&dbq->list, &db_list);
677 	spin_unlock_irqrestore(&db_list_lock, flags);
678 
679 	filp->private_data = dbq;
680 
681 	return ret;
682 }
683 
684 /*
685  * Close the driver
686  */
687 static int fsl_hv_close(struct inode *inode, struct file *filp)
688 {
689 	struct doorbell_queue *dbq = filp->private_data;
690 	unsigned long flags;
691 
692 	int ret = 0;
693 
694 	spin_lock_irqsave(&db_list_lock, flags);
695 	list_del(&dbq->list);
696 	spin_unlock_irqrestore(&db_list_lock, flags);
697 
698 	kfree(dbq);
699 
700 	return ret;
701 }
702 
703 static const struct file_operations fsl_hv_fops = {
704 	.owner = THIS_MODULE,
705 	.open = fsl_hv_open,
706 	.release = fsl_hv_close,
707 	.poll = fsl_hv_poll,
708 	.read = fsl_hv_read,
709 	.unlocked_ioctl = fsl_hv_ioctl,
710 	.compat_ioctl = fsl_hv_ioctl,
711 };
712 
713 static struct miscdevice fsl_hv_misc_dev = {
714 	MISC_DYNAMIC_MINOR,
715 	"fsl-hv",
716 	&fsl_hv_fops
717 };
718 
719 static irqreturn_t fsl_hv_shutdown_isr(int irq, void *data)
720 {
721 	orderly_poweroff(false);
722 
723 	return IRQ_HANDLED;
724 }
725 
726 /*
727  * Returns the handle of the parent of the given node
728  *
729  * The handle is the value of the 'hv-handle' property
730  */
731 static int get_parent_handle(struct device_node *np)
732 {
733 	struct device_node *parent;
734 	const uint32_t *prop;
735 	uint32_t handle;
736 	int len;
737 
738 	parent = of_get_parent(np);
739 	if (!parent)
740 		/* It's not really possible for this to fail */
741 		return -ENODEV;
742 
743 	/*
744 	 * The proper name for the handle property is "hv-handle", but some
745 	 * older versions of the hypervisor used "reg".
746 	 */
747 	prop = of_get_property(parent, "hv-handle", &len);
748 	if (!prop)
749 		prop = of_get_property(parent, "reg", &len);
750 
751 	if (!prop || (len != sizeof(uint32_t))) {
752 		/* This can happen only if the node is malformed */
753 		of_node_put(parent);
754 		return -ENODEV;
755 	}
756 
757 	handle = be32_to_cpup(prop);
758 	of_node_put(parent);
759 
760 	return handle;
761 }
762 
763 /*
764  * Register a callback for failover events
765  *
766  * This function is called by device drivers to register their callback
767  * functions for fail-over events.
768  */
769 int fsl_hv_failover_register(struct notifier_block *nb)
770 {
771 	return blocking_notifier_chain_register(&failover_subscribers, nb);
772 }
773 EXPORT_SYMBOL(fsl_hv_failover_register);
774 
775 /*
776  * Unregister a callback for failover events
777  */
778 int fsl_hv_failover_unregister(struct notifier_block *nb)
779 {
780 	return blocking_notifier_chain_unregister(&failover_subscribers, nb);
781 }
782 EXPORT_SYMBOL(fsl_hv_failover_unregister);
783 
784 /*
785  * Return TRUE if we're running under FSL hypervisor
786  *
787  * This function checks to see if we're running under the Freescale
788  * hypervisor, and returns zero if we're not, or non-zero if we are.
789  *
790  * First, it checks if MSR[GS]==1, which means we're running under some
791  * hypervisor.  Then it checks if there is a hypervisor node in the device
792  * tree.  Currently, that means there needs to be a node in the root called
793  * "hypervisor" and which has a property named "fsl,hv-version".
794  */
795 static int has_fsl_hypervisor(void)
796 {
797 	struct device_node *node;
798 	int ret;
799 
800 	node = of_find_node_by_path("/hypervisor");
801 	if (!node)
802 		return 0;
803 
804 	ret = of_find_property(node, "fsl,hv-version", NULL) != NULL;
805 
806 	of_node_put(node);
807 
808 	return ret;
809 }
810 
811 /*
812  * Freescale hypervisor management driver init
813  *
814  * This function is called when this module is loaded.
815  *
816  * Register ourselves as a miscellaneous driver.  This will register the
817  * fops structure and create the right sysfs entries for udev.
818  */
819 static int __init fsl_hypervisor_init(void)
820 {
821 	struct device_node *np;
822 	struct doorbell_isr *dbisr, *n;
823 	int ret;
824 
825 	pr_info("Freescale hypervisor management driver\n");
826 
827 	if (!has_fsl_hypervisor()) {
828 		pr_info("fsl-hv: no hypervisor found\n");
829 		return -ENODEV;
830 	}
831 
832 	ret = misc_register(&fsl_hv_misc_dev);
833 	if (ret) {
834 		pr_err("fsl-hv: cannot register device\n");
835 		return ret;
836 	}
837 
838 	INIT_LIST_HEAD(&db_list);
839 	INIT_LIST_HEAD(&isr_list);
840 
841 	for_each_compatible_node(np, NULL, "epapr,hv-receive-doorbell") {
842 		unsigned int irq;
843 		const uint32_t *handle;
844 
845 		handle = of_get_property(np, "interrupts", NULL);
846 		irq = irq_of_parse_and_map(np, 0);
847 		if (!handle || (irq == NO_IRQ)) {
848 			pr_err("fsl-hv: no 'interrupts' property in %s node\n",
849 				np->full_name);
850 			continue;
851 		}
852 
853 		dbisr = kzalloc(sizeof(*dbisr), GFP_KERNEL);
854 		if (!dbisr)
855 			goto out_of_memory;
856 
857 		dbisr->irq = irq;
858 		dbisr->doorbell = be32_to_cpup(handle);
859 
860 		if (of_device_is_compatible(np, "fsl,hv-shutdown-doorbell")) {
861 			/* The shutdown doorbell gets its own ISR */
862 			ret = request_irq(irq, fsl_hv_shutdown_isr, 0,
863 					  np->name, NULL);
864 		} else if (of_device_is_compatible(np,
865 			"fsl,hv-state-change-doorbell")) {
866 			/*
867 			 * The state change doorbell triggers a notification if
868 			 * the state of the managed partition changes to
869 			 * "stopped". We need a separate interrupt handler for
870 			 * that, and we also need to know the handle of the
871 			 * target partition, not just the handle of the
872 			 * doorbell.
873 			 */
874 			dbisr->partition = ret = get_parent_handle(np);
875 			if (ret < 0) {
876 				pr_err("fsl-hv: node %s has missing or "
877 				       "malformed parent\n", np->full_name);
878 				kfree(dbisr);
879 				continue;
880 			}
881 			ret = request_threaded_irq(irq, fsl_hv_state_change_isr,
882 						   fsl_hv_state_change_thread,
883 						   0, np->name, dbisr);
884 		} else
885 			ret = request_irq(irq, fsl_hv_isr, 0, np->name, dbisr);
886 
887 		if (ret < 0) {
888 			pr_err("fsl-hv: could not request irq %u for node %s\n",
889 			       irq, np->full_name);
890 			kfree(dbisr);
891 			continue;
892 		}
893 
894 		list_add(&dbisr->list, &isr_list);
895 
896 		pr_info("fsl-hv: registered handler for doorbell %u\n",
897 			dbisr->doorbell);
898 	}
899 
900 	return 0;
901 
902 out_of_memory:
903 	list_for_each_entry_safe(dbisr, n, &isr_list, list) {
904 		free_irq(dbisr->irq, dbisr);
905 		list_del(&dbisr->list);
906 		kfree(dbisr);
907 	}
908 
909 	misc_deregister(&fsl_hv_misc_dev);
910 
911 	return -ENOMEM;
912 }
913 
914 /*
915  * Freescale hypervisor management driver termination
916  *
917  * This function is called when this driver is unloaded.
918  */
919 static void __exit fsl_hypervisor_exit(void)
920 {
921 	struct doorbell_isr *dbisr, *n;
922 
923 	list_for_each_entry_safe(dbisr, n, &isr_list, list) {
924 		free_irq(dbisr->irq, dbisr);
925 		list_del(&dbisr->list);
926 		kfree(dbisr);
927 	}
928 
929 	misc_deregister(&fsl_hv_misc_dev);
930 }
931 
932 module_init(fsl_hypervisor_init);
933 module_exit(fsl_hypervisor_exit);
934 
935 MODULE_AUTHOR("Timur Tabi <timur@freescale.com>");
936 MODULE_DESCRIPTION("Freescale hypervisor management driver");
937 MODULE_LICENSE("GPL v2");
938