blob: 1d3b8ebb31414f730787b998d1646380613a97cf [file] [log] [blame]
Timur Tabi6db71992011-06-09 15:52:06 -05001/*
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/reboot.h>
38#include <linux/uaccess.h>
39#include <linux/notifier.h>
40
41#include <linux/io.h>
42#include <asm/fsl_hcalls.h>
43
44#include <linux/fsl_hypervisor.h>
45
46static BLOCKING_NOTIFIER_HEAD(failover_subscribers);
47
48/*
49 * Ioctl interface for FSL_HV_IOCTL_PARTITION_RESTART
50 *
51 * Restart a running partition
52 */
53static long ioctl_restart(struct fsl_hv_ioctl_restart __user *p)
54{
55 struct fsl_hv_ioctl_restart param;
56
57 /* Get the parameters from the user */
58 if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_restart)))
59 return -EFAULT;
60
61 param.ret = fh_partition_restart(param.partition);
62
63 if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
64 return -EFAULT;
65
66 return 0;
67}
68
69/*
70 * Ioctl interface for FSL_HV_IOCTL_PARTITION_STATUS
71 *
72 * Query the status of a partition
73 */
74static long ioctl_status(struct fsl_hv_ioctl_status __user *p)
75{
76 struct fsl_hv_ioctl_status param;
77 u32 status;
78
79 /* Get the parameters from the user */
80 if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_status)))
81 return -EFAULT;
82
83 param.ret = fh_partition_get_status(param.partition, &status);
84 if (!param.ret)
85 param.status = status;
86
87 if (copy_to_user(p, &param, sizeof(struct fsl_hv_ioctl_status)))
88 return -EFAULT;
89
90 return 0;
91}
92
93/*
94 * Ioctl interface for FSL_HV_IOCTL_PARTITION_START
95 *
96 * Start a stopped partition.
97 */
98static long ioctl_start(struct fsl_hv_ioctl_start __user *p)
99{
100 struct fsl_hv_ioctl_start param;
101
102 /* Get the parameters from the user */
103 if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_start)))
104 return -EFAULT;
105
106 param.ret = fh_partition_start(param.partition, param.entry_point,
107 param.load);
108
109 if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
110 return -EFAULT;
111
112 return 0;
113}
114
115/*
116 * Ioctl interface for FSL_HV_IOCTL_PARTITION_STOP
117 *
118 * Stop a running partition
119 */
120static long ioctl_stop(struct fsl_hv_ioctl_stop __user *p)
121{
122 struct fsl_hv_ioctl_stop param;
123
124 /* Get the parameters from the user */
125 if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_stop)))
126 return -EFAULT;
127
128 param.ret = fh_partition_stop(param.partition);
129
130 if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
131 return -EFAULT;
132
133 return 0;
134}
135
136/*
137 * Ioctl interface for FSL_HV_IOCTL_MEMCPY
138 *
139 * The FH_MEMCPY hypercall takes an array of address/address/size structures
140 * to represent the data being copied. As a convenience to the user, this
141 * ioctl takes a user-create buffer and a pointer to a guest physically
142 * contiguous buffer in the remote partition, and creates the
143 * address/address/size array for the hypercall.
144 */
145static long ioctl_memcpy(struct fsl_hv_ioctl_memcpy __user *p)
146{
147 struct fsl_hv_ioctl_memcpy param;
148
149 struct page **pages = NULL;
150 void *sg_list_unaligned = NULL;
151 struct fh_sg_list *sg_list = NULL;
152
153 unsigned int num_pages;
154 unsigned long lb_offset; /* Offset within a page of the local buffer */
155
156 unsigned int i;
157 long ret = 0;
158 int num_pinned; /* return value from get_user_pages() */
159 phys_addr_t remote_paddr; /* The next address in the remote buffer */
160 uint32_t count; /* The number of bytes left to copy */
161
162 /* Get the parameters from the user */
163 if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_memcpy)))
164 return -EFAULT;
165
166 /*
167 * One partition must be local, the other must be remote. In other
168 * words, if source and target are both -1, or are both not -1, then
169 * return an error.
170 */
171 if ((param.source == -1) == (param.target == -1))
172 return -EINVAL;
173
174 /*
175 * The array of pages returned by get_user_pages() covers only
176 * page-aligned memory. Since the user buffer is probably not
177 * page-aligned, we need to handle the discrepancy.
178 *
179 * We calculate the offset within a page of the S/G list, and make
180 * adjustments accordingly. This will result in a page list that looks
181 * like this:
182 *
183 * ---- <-- first page starts before the buffer
184 * | |
185 * |////|-> ----
186 * |////| | |
187 * ---- | |
188 * | |
189 * ---- | |
190 * |////| | |
191 * |////| | |
192 * |////| | |
193 * ---- | |
194 * | |
195 * ---- | |
196 * |////| | |
197 * |////| | |
198 * |////| | |
199 * ---- | |
200 * | |
201 * ---- | |
202 * |////| | |
203 * |////|-> ----
204 * | | <-- last page ends after the buffer
205 * ----
206 *
207 * The distance between the start of the first page and the start of the
208 * buffer is lb_offset. The hashed (///) areas are the parts of the
209 * page list that contain the actual buffer.
210 *
211 * The advantage of this approach is that the number of pages is
212 * equal to the number of entries in the S/G list that we give to the
213 * hypervisor.
214 */
215 lb_offset = param.local_vaddr & (PAGE_SIZE - 1);
216 num_pages = (param.count + lb_offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
217
218 /* Allocate the buffers we need */
219
220 /*
221 * 'pages' is an array of struct page pointers that's initialized by
222 * get_user_pages().
223 */
224 pages = kzalloc(num_pages * sizeof(struct page *), GFP_KERNEL);
225 if (!pages) {
226 pr_debug("fsl-hv: could not allocate page list\n");
227 return -ENOMEM;
228 }
229
230 /*
231 * sg_list is the list of fh_sg_list objects that we pass to the
232 * hypervisor.
233 */
234 sg_list_unaligned = kmalloc(num_pages * sizeof(struct fh_sg_list) +
235 sizeof(struct fh_sg_list) - 1, GFP_KERNEL);
236 if (!sg_list_unaligned) {
237 pr_debug("fsl-hv: could not allocate S/G list\n");
238 ret = -ENOMEM;
239 goto exit;
240 }
241 sg_list = PTR_ALIGN(sg_list_unaligned, sizeof(struct fh_sg_list));
242
243 /* Get the physical addresses of the source buffer */
244 down_read(&current->mm->mmap_sem);
245 num_pinned = get_user_pages(current, current->mm,
246 param.local_vaddr - lb_offset, num_pages,
247 (param.source == -1) ? READ : WRITE,
248 0, pages, NULL);
249 up_read(&current->mm->mmap_sem);
250
251 if (num_pinned != num_pages) {
252 /* get_user_pages() failed */
253 pr_debug("fsl-hv: could not lock source buffer\n");
254 ret = (num_pinned < 0) ? num_pinned : -EFAULT;
255 goto exit;
256 }
257
258 /*
259 * Build the fh_sg_list[] array. The first page is special
260 * because it's misaligned.
261 */
262 if (param.source == -1) {
263 sg_list[0].source = page_to_phys(pages[0]) + lb_offset;
264 sg_list[0].target = param.remote_paddr;
265 } else {
266 sg_list[0].source = param.remote_paddr;
267 sg_list[0].target = page_to_phys(pages[0]) + lb_offset;
268 }
269 sg_list[0].size = min_t(uint64_t, param.count, PAGE_SIZE - lb_offset);
270
271 remote_paddr = param.remote_paddr + sg_list[0].size;
272 count = param.count - sg_list[0].size;
273
274 for (i = 1; i < num_pages; i++) {
275 if (param.source == -1) {
276 /* local to remote */
277 sg_list[i].source = page_to_phys(pages[i]);
278 sg_list[i].target = remote_paddr;
279 } else {
280 /* remote to local */
281 sg_list[i].source = remote_paddr;
282 sg_list[i].target = page_to_phys(pages[i]);
283 }
284 sg_list[i].size = min_t(uint64_t, count, PAGE_SIZE);
285
286 remote_paddr += sg_list[i].size;
287 count -= sg_list[i].size;
288 }
289
290 param.ret = fh_partition_memcpy(param.source, param.target,
291 virt_to_phys(sg_list), num_pages);
292
293exit:
294 if (pages) {
295 for (i = 0; i < num_pages; i++)
296 if (pages[i])
297 put_page(pages[i]);
298 }
299
300 kfree(sg_list_unaligned);
301 kfree(pages);
302
303 if (!ret)
304 if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
305 return -EFAULT;
306
307 return ret;
308}
309
310/*
311 * Ioctl interface for FSL_HV_IOCTL_DOORBELL
312 *
313 * Ring a doorbell
314 */
315static long ioctl_doorbell(struct fsl_hv_ioctl_doorbell __user *p)
316{
317 struct fsl_hv_ioctl_doorbell param;
318
319 /* Get the parameters from the user. */
320 if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_doorbell)))
321 return -EFAULT;
322
323 param.ret = ev_doorbell_send(param.doorbell);
324
325 if (copy_to_user(&p->ret, &param.ret, sizeof(__u32)))
326 return -EFAULT;
327
328 return 0;
329}
330
331static long ioctl_dtprop(struct fsl_hv_ioctl_prop __user *p, int set)
332{
333 struct fsl_hv_ioctl_prop param;
334 char __user *upath, *upropname;
335 void __user *upropval;
336 char *path = NULL, *propname = NULL;
337 void *propval = NULL;
338 int ret = 0;
339
340 /* Get the parameters from the user. */
341 if (copy_from_user(&param, p, sizeof(struct fsl_hv_ioctl_prop)))
342 return -EFAULT;
343
344 upath = (char __user *)(uintptr_t)param.path;
345 upropname = (char __user *)(uintptr_t)param.propname;
346 upropval = (void __user *)(uintptr_t)param.propval;
347
348 path = strndup_user(upath, FH_DTPROP_MAX_PATHLEN);
349 if (IS_ERR(path)) {
350 ret = PTR_ERR(path);
351 goto out;
352 }
353
354 propname = strndup_user(upropname, FH_DTPROP_MAX_PATHLEN);
355 if (IS_ERR(propname)) {
356 ret = PTR_ERR(propname);
357 goto out;
358 }
359
360 if (param.proplen > FH_DTPROP_MAX_PROPLEN) {
361 ret = -EINVAL;
362 goto out;
363 }
364
365 propval = kmalloc(param.proplen, GFP_KERNEL);
366 if (!propval) {
367 ret = -ENOMEM;
368 goto out;
369 }
370
371 if (set) {
372 if (copy_from_user(propval, upropval, param.proplen)) {
373 ret = -EFAULT;
374 goto out;
375 }
376
377 param.ret = fh_partition_set_dtprop(param.handle,
378 virt_to_phys(path),
379 virt_to_phys(propname),
380 virt_to_phys(propval),
381 param.proplen);
382 } else {
383 param.ret = fh_partition_get_dtprop(param.handle,
384 virt_to_phys(path),
385 virt_to_phys(propname),
386 virt_to_phys(propval),
387 &param.proplen);
388
389 if (param.ret == 0) {
390 if (copy_to_user(upropval, propval, param.proplen) ||
391 put_user(param.proplen, &p->proplen)) {
392 ret = -EFAULT;
393 goto out;
394 }
395 }
396 }
397
398 if (put_user(param.ret, &p->ret))
399 ret = -EFAULT;
400
401out:
402 kfree(path);
403 kfree(propval);
404 kfree(propname);
405
406 return ret;
407}
408
409/*
410 * Ioctl main entry point
411 */
412static long fsl_hv_ioctl(struct file *file, unsigned int cmd,
413 unsigned long argaddr)
414{
415 void __user *arg = (void __user *)argaddr;
416 long ret;
417
418 switch (cmd) {
419 case FSL_HV_IOCTL_PARTITION_RESTART:
420 ret = ioctl_restart(arg);
421 break;
422 case FSL_HV_IOCTL_PARTITION_GET_STATUS:
423 ret = ioctl_status(arg);
424 break;
425 case FSL_HV_IOCTL_PARTITION_START:
426 ret = ioctl_start(arg);
427 break;
428 case FSL_HV_IOCTL_PARTITION_STOP:
429 ret = ioctl_stop(arg);
430 break;
431 case FSL_HV_IOCTL_MEMCPY:
432 ret = ioctl_memcpy(arg);
433 break;
434 case FSL_HV_IOCTL_DOORBELL:
435 ret = ioctl_doorbell(arg);
436 break;
437 case FSL_HV_IOCTL_GETPROP:
438 ret = ioctl_dtprop(arg, 0);
439 break;
440 case FSL_HV_IOCTL_SETPROP:
441 ret = ioctl_dtprop(arg, 1);
442 break;
443 default:
444 pr_debug("fsl-hv: bad ioctl dir=%u type=%u cmd=%u size=%u\n",
445 _IOC_DIR(cmd), _IOC_TYPE(cmd), _IOC_NR(cmd),
446 _IOC_SIZE(cmd));
447 return -ENOTTY;
448 }
449
450 return ret;
451}
452
453/* Linked list of processes that have us open */
454static struct list_head db_list;
455
456/* spinlock for db_list */
457static DEFINE_SPINLOCK(db_list_lock);
458
459/* The size of the doorbell event queue. This must be a power of two. */
460#define QSIZE 16
461
462/* Returns the next head/tail pointer, wrapping around the queue if necessary */
463#define nextp(x) (((x) + 1) & (QSIZE - 1))
464
465/* Per-open data structure */
466struct doorbell_queue {
467 struct list_head list;
468 spinlock_t lock;
469 wait_queue_head_t wait;
470 unsigned int head;
471 unsigned int tail;
472 uint32_t q[QSIZE];
473};
474
475/* Linked list of ISRs that we registered */
476struct list_head isr_list;
477
478/* Per-ISR data structure */
479struct doorbell_isr {
480 struct list_head list;
481 unsigned int irq;
482 uint32_t doorbell; /* The doorbell handle */
483 uint32_t partition; /* The partition handle, if used */
484};
485
486/*
487 * Add a doorbell to all of the doorbell queues
488 */
489static void fsl_hv_queue_doorbell(uint32_t doorbell)
490{
491 struct doorbell_queue *dbq;
492 unsigned long flags;
493
494 /* Prevent another core from modifying db_list */
495 spin_lock_irqsave(&db_list_lock, flags);
496
497 list_for_each_entry(dbq, &db_list, list) {
498 if (dbq->head != nextp(dbq->tail)) {
499 dbq->q[dbq->tail] = doorbell;
500 /*
501 * This memory barrier eliminates the need to grab
502 * the spinlock for dbq.
503 */
504 smp_wmb();
505 dbq->tail = nextp(dbq->tail);
506 wake_up_interruptible(&dbq->wait);
507 }
508 }
509
510 spin_unlock_irqrestore(&db_list_lock, flags);
511}
512
513/*
514 * Interrupt handler for all doorbells
515 *
516 * We use the same interrupt handler for all doorbells. Whenever a doorbell
517 * is rung, and we receive an interrupt, we just put the handle for that
518 * doorbell (passed to us as *data) into all of the queues.
519 */
520static irqreturn_t fsl_hv_isr(int irq, void *data)
521{
522 fsl_hv_queue_doorbell((uintptr_t) data);
523
524 return IRQ_HANDLED;
525}
526
527/*
528 * State change thread function
529 *
530 * The state change notification arrives in an interrupt, but we can't call
531 * blocking_notifier_call_chain() in an interrupt handler. We could call
532 * atomic_notifier_call_chain(), but that would require the clients' call-back
533 * function to run in interrupt context. Since we don't want to impose that
534 * restriction on the clients, we use a threaded IRQ to process the
535 * notification in kernel context.
536 */
537static irqreturn_t fsl_hv_state_change_thread(int irq, void *data)
538{
539 struct doorbell_isr *dbisr = data;
540
541 blocking_notifier_call_chain(&failover_subscribers, dbisr->partition,
542 NULL);
543
544 return IRQ_HANDLED;
545}
546
547/*
548 * Interrupt handler for state-change doorbells
549 */
550static irqreturn_t fsl_hv_state_change_isr(int irq, void *data)
551{
552 unsigned int status;
553 struct doorbell_isr *dbisr = data;
554 int ret;
555
556 /* It's still a doorbell, so add it to all the queues. */
557 fsl_hv_queue_doorbell(dbisr->doorbell);
558
559 /* Determine the new state, and if it's stopped, notify the clients. */
560 ret = fh_partition_get_status(dbisr->partition, &status);
561 if (!ret && (status == FH_PARTITION_STOPPED))
562 return IRQ_WAKE_THREAD;
563
564 return IRQ_HANDLED;
565}
566
567/*
568 * Returns a bitmask indicating whether a read will block
569 */
570static unsigned int fsl_hv_poll(struct file *filp, struct poll_table_struct *p)
571{
572 struct doorbell_queue *dbq = filp->private_data;
573 unsigned long flags;
574 unsigned int mask;
575
576 spin_lock_irqsave(&dbq->lock, flags);
577
578 poll_wait(filp, &dbq->wait, p);
579 mask = (dbq->head == dbq->tail) ? 0 : (POLLIN | POLLRDNORM);
580
581 spin_unlock_irqrestore(&dbq->lock, flags);
582
583 return mask;
584}
585
586/*
587 * Return the handles for any incoming doorbells
588 *
589 * If there are doorbell handles in the queue for this open instance, then
590 * return them to the caller as an array of 32-bit integers. Otherwise,
591 * block until there is at least one handle to return.
592 */
593static ssize_t fsl_hv_read(struct file *filp, char __user *buf, size_t len,
594 loff_t *off)
595{
596 struct doorbell_queue *dbq = filp->private_data;
597 uint32_t __user *p = (uint32_t __user *) buf; /* for put_user() */
598 unsigned long flags;
599 ssize_t count = 0;
600
601 /* Make sure we stop when the user buffer is full. */
602 while (len >= sizeof(uint32_t)) {
603 uint32_t dbell; /* Local copy of doorbell queue data */
604
605 spin_lock_irqsave(&dbq->lock, flags);
606
607 /*
608 * If the queue is empty, then either we're done or we need
609 * to block. If the application specified O_NONBLOCK, then
610 * we return the appropriate error code.
611 */
612 if (dbq->head == dbq->tail) {
613 spin_unlock_irqrestore(&dbq->lock, flags);
614 if (count)
615 break;
616 if (filp->f_flags & O_NONBLOCK)
617 return -EAGAIN;
618 if (wait_event_interruptible(dbq->wait,
619 dbq->head != dbq->tail))
620 return -ERESTARTSYS;
621 continue;
622 }
623
624 /*
625 * Even though we have an smp_wmb() in the ISR, the core
626 * might speculatively execute the "dbell = ..." below while
627 * it's evaluating the if-statement above. In that case, the
628 * value put into dbell could be stale if the core accepts the
629 * speculation. To prevent that, we need a read memory barrier
630 * here as well.
631 */
632 smp_rmb();
633
634 /* Copy the data to a temporary local buffer, because
635 * we can't call copy_to_user() from inside a spinlock
636 */
637 dbell = dbq->q[dbq->head];
638 dbq->head = nextp(dbq->head);
639
640 spin_unlock_irqrestore(&dbq->lock, flags);
641
642 if (put_user(dbell, p))
643 return -EFAULT;
644 p++;
645 count += sizeof(uint32_t);
646 len -= sizeof(uint32_t);
647 }
648
649 return count;
650}
651
652/*
653 * Open the driver and prepare for reading doorbells.
654 *
655 * Every time an application opens the driver, we create a doorbell queue
656 * for that file handle. This queue is used for any incoming doorbells.
657 */
658static int fsl_hv_open(struct inode *inode, struct file *filp)
659{
660 struct doorbell_queue *dbq;
661 unsigned long flags;
662 int ret = 0;
663
664 dbq = kzalloc(sizeof(struct doorbell_queue), GFP_KERNEL);
665 if (!dbq) {
666 pr_err("fsl-hv: out of memory\n");
667 return -ENOMEM;
668 }
669
670 spin_lock_init(&dbq->lock);
671 init_waitqueue_head(&dbq->wait);
672
673 spin_lock_irqsave(&db_list_lock, flags);
674 list_add(&dbq->list, &db_list);
675 spin_unlock_irqrestore(&db_list_lock, flags);
676
677 filp->private_data = dbq;
678
679 return ret;
680}
681
682/*
683 * Close the driver
684 */
685static int fsl_hv_close(struct inode *inode, struct file *filp)
686{
687 struct doorbell_queue *dbq = filp->private_data;
688 unsigned long flags;
689
690 int ret = 0;
691
692 spin_lock_irqsave(&db_list_lock, flags);
693 list_del(&dbq->list);
694 spin_unlock_irqrestore(&db_list_lock, flags);
695
696 kfree(dbq);
697
698 return ret;
699}
700
701static const struct file_operations fsl_hv_fops = {
702 .owner = THIS_MODULE,
703 .open = fsl_hv_open,
704 .release = fsl_hv_close,
705 .poll = fsl_hv_poll,
706 .read = fsl_hv_read,
707 .unlocked_ioctl = fsl_hv_ioctl,
708};
709
710static struct miscdevice fsl_hv_misc_dev = {
711 MISC_DYNAMIC_MINOR,
712 "fsl-hv",
713 &fsl_hv_fops
714};
715
716static irqreturn_t fsl_hv_shutdown_isr(int irq, void *data)
717{
718 orderly_poweroff(false);
719
720 return IRQ_HANDLED;
721}
722
723/*
724 * Returns the handle of the parent of the given node
725 *
726 * The handle is the value of the 'hv-handle' property
727 */
728static int get_parent_handle(struct device_node *np)
729{
730 struct device_node *parent;
731 const uint32_t *prop;
732 uint32_t handle;
733 int len;
734
735 parent = of_get_parent(np);
736 if (!parent)
737 /* It's not really possible for this to fail */
738 return -ENODEV;
739
740 /*
741 * The proper name for the handle property is "hv-handle", but some
742 * older versions of the hypervisor used "reg".
743 */
744 prop = of_get_property(parent, "hv-handle", &len);
745 if (!prop)
746 prop = of_get_property(parent, "reg", &len);
747
748 if (!prop || (len != sizeof(uint32_t))) {
749 /* This can happen only if the node is malformed */
750 of_node_put(parent);
751 return -ENODEV;
752 }
753
754 handle = be32_to_cpup(prop);
755 of_node_put(parent);
756
757 return handle;
758}
759
760/*
761 * Register a callback for failover events
762 *
763 * This function is called by device drivers to register their callback
764 * functions for fail-over events.
765 */
766int fsl_hv_failover_register(struct notifier_block *nb)
767{
768 return blocking_notifier_chain_register(&failover_subscribers, nb);
769}
770EXPORT_SYMBOL(fsl_hv_failover_register);
771
772/*
773 * Unregister a callback for failover events
774 */
775int fsl_hv_failover_unregister(struct notifier_block *nb)
776{
777 return blocking_notifier_chain_unregister(&failover_subscribers, nb);
778}
779EXPORT_SYMBOL(fsl_hv_failover_unregister);
780
781/*
782 * Return TRUE if we're running under FSL hypervisor
783 *
784 * This function checks to see if we're running under the Freescale
785 * hypervisor, and returns zero if we're not, or non-zero if we are.
786 *
787 * First, it checks if MSR[GS]==1, which means we're running under some
788 * hypervisor. Then it checks if there is a hypervisor node in the device
789 * tree. Currently, that means there needs to be a node in the root called
790 * "hypervisor" and which has a property named "fsl,hv-version".
791 */
792static int has_fsl_hypervisor(void)
793{
794 struct device_node *node;
795 int ret;
796
797 if (!(mfmsr() & MSR_GS))
798 return 0;
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 */
819static 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
902out_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 */
919static 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
932module_init(fsl_hypervisor_init);
933module_exit(fsl_hypervisor_exit);
934
935MODULE_AUTHOR("Timur Tabi <timur@freescale.com>");
936MODULE_DESCRIPTION("Freescale hypervisor management driver");
937MODULE_LICENSE("GPL v2");