[PATCH] swsusp: add read-speed instrumentation
[safe/jmp/linux-2.6] / kernel / power / swap.c
1 /*
2  * linux/kernel/power/swap.c
3  *
4  * This file provides functions for reading the suspend image from
5  * and writing it to a swap partition.
6  *
7  * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz>
8  * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9  *
10  * This file is released under the GPLv2.
11  *
12  */
13
14 #include <linux/module.h>
15 #include <linux/smp_lock.h>
16 #include <linux/file.h>
17 #include <linux/utsname.h>
18 #include <linux/version.h>
19 #include <linux/delay.h>
20 #include <linux/bitops.h>
21 #include <linux/genhd.h>
22 #include <linux/device.h>
23 #include <linux/buffer_head.h>
24 #include <linux/bio.h>
25 #include <linux/swap.h>
26 #include <linux/swapops.h>
27 #include <linux/pm.h>
28
29 #include "power.h"
30
31 extern char resume_file[];
32
33 #define SWSUSP_SIG      "S1SUSPEND"
34
35 static struct swsusp_header {
36         char reserved[PAGE_SIZE - 20 - sizeof(swp_entry_t)];
37         swp_entry_t image;
38         char    orig_sig[10];
39         char    sig[10];
40 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
41
42 /*
43  * Saving part...
44  */
45
46 static unsigned short root_swap = 0xffff;
47
48 static int mark_swapfiles(swp_entry_t start)
49 {
50         int error;
51
52         rw_swap_page_sync(READ, swp_entry(root_swap, 0),
53                           virt_to_page((unsigned long)&swsusp_header), NULL);
54         if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
55             !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
56                 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
57                 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
58                 swsusp_header.image = start;
59                 error = rw_swap_page_sync(WRITE, swp_entry(root_swap, 0),
60                                 virt_to_page((unsigned long)&swsusp_header),
61                                 NULL);
62         } else {
63                 pr_debug("swsusp: Partition is not swap space.\n");
64                 error = -ENODEV;
65         }
66         return error;
67 }
68
69 /**
70  *      swsusp_swap_check - check if the resume device is a swap device
71  *      and get its index (if so)
72  */
73
74 static int swsusp_swap_check(void) /* This is called before saving image */
75 {
76         int res = swap_type_of(swsusp_resume_device);
77
78         if (res >= 0) {
79                 root_swap = res;
80                 return 0;
81         }
82         return res;
83 }
84
85 /**
86  *      write_page - Write one page to given swap location.
87  *      @buf:           Address we're writing.
88  *      @offset:        Offset of the swap page we're writing to.
89  *      @bio_chain:     Link the next write BIO here
90  */
91
92 static int write_page(void *buf, unsigned long offset, struct bio **bio_chain)
93 {
94         swp_entry_t entry;
95         int error = -ENOSPC;
96
97         if (offset) {
98                 struct page *page = virt_to_page(buf);
99
100                 if (bio_chain) {
101                         /*
102                          * Whether or not we successfully allocated a copy page,
103                          * we take a ref on the page here.  It gets undone in
104                          * wait_on_bio_chain().
105                          */
106                         struct page *page_copy;
107                         page_copy = alloc_page(GFP_ATOMIC);
108                         if (page_copy == NULL) {
109                                 WARN_ON_ONCE(1);
110                                 bio_chain = NULL;       /* Go synchronous */
111                                 get_page(page);
112                         } else {
113                                 memcpy(page_address(page_copy),
114                                         page_address(page), PAGE_SIZE);
115                                 page = page_copy;
116                         }
117                 }
118                 entry = swp_entry(root_swap, offset);
119                 error = rw_swap_page_sync(WRITE, entry, page, bio_chain);
120         }
121         return error;
122 }
123
124 /*
125  *      The swap map is a data structure used for keeping track of each page
126  *      written to a swap partition.  It consists of many swap_map_page
127  *      structures that contain each an array of MAP_PAGE_SIZE swap entries.
128  *      These structures are stored on the swap and linked together with the
129  *      help of the .next_swap member.
130  *
131  *      The swap map is created during suspend.  The swap map pages are
132  *      allocated and populated one at a time, so we only need one memory
133  *      page to set up the entire structure.
134  *
135  *      During resume we also only need to use one swap_map_page structure
136  *      at a time.
137  */
138
139 #define MAP_PAGE_ENTRIES        (PAGE_SIZE / sizeof(long) - 1)
140
141 struct swap_map_page {
142         unsigned long           entries[MAP_PAGE_ENTRIES];
143         unsigned long           next_swap;
144 };
145
146 /**
147  *      The swap_map_handle structure is used for handling swap in
148  *      a file-alike way
149  */
150
151 struct swap_map_handle {
152         struct swap_map_page *cur;
153         unsigned long cur_swap;
154         struct bitmap_page *bitmap;
155         unsigned int k;
156 };
157
158 static void release_swap_writer(struct swap_map_handle *handle)
159 {
160         if (handle->cur)
161                 free_page((unsigned long)handle->cur);
162         handle->cur = NULL;
163         if (handle->bitmap)
164                 free_bitmap(handle->bitmap);
165         handle->bitmap = NULL;
166 }
167
168 static void show_speed(struct timeval *start, struct timeval *stop,
169                         unsigned nr_pages, char *msg)
170 {
171         s64 elapsed_centisecs64;
172         int centisecs;
173         int k;
174         int kps;
175
176         elapsed_centisecs64 = timeval_to_ns(stop) - timeval_to_ns(start);
177         do_div(elapsed_centisecs64, NSEC_PER_SEC / 100);
178         centisecs = elapsed_centisecs64;
179         if (centisecs == 0)
180                 centisecs = 1;  /* avoid div-by-zero */
181         k = nr_pages * (PAGE_SIZE / 1024);
182         kps = (k * 100) / centisecs;
183         printk("%s %d kbytes in %d.%02d seconds (%d.%02d MB/s)\n", msg, k,
184                         centisecs / 100, centisecs % 100,
185                         kps / 1000, (kps % 1000) / 10);
186 }
187
188 static int get_swap_writer(struct swap_map_handle *handle)
189 {
190         handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
191         if (!handle->cur)
192                 return -ENOMEM;
193         handle->bitmap = alloc_bitmap(count_swap_pages(root_swap, 0));
194         if (!handle->bitmap) {
195                 release_swap_writer(handle);
196                 return -ENOMEM;
197         }
198         handle->cur_swap = alloc_swap_page(root_swap, handle->bitmap);
199         if (!handle->cur_swap) {
200                 release_swap_writer(handle);
201                 return -ENOSPC;
202         }
203         handle->k = 0;
204         return 0;
205 }
206
207 static int wait_on_bio_chain(struct bio **bio_chain)
208 {
209         struct bio *bio;
210         struct bio *next_bio;
211         int ret = 0;
212
213         if (bio_chain == NULL)
214                 return 0;
215
216         bio = *bio_chain;
217         while (bio) {
218                 struct page *page;
219
220                 next_bio = bio->bi_private;
221                 page = bio->bi_io_vec[0].bv_page;
222                 wait_on_page_locked(page);
223                 if (!PageUptodate(page) || PageError(page))
224                         ret = -EIO;
225                 put_page(page);
226                 bio_put(bio);
227                 bio = next_bio;
228         }
229         *bio_chain = NULL;
230         return ret;
231 }
232
233 static int swap_write_page(struct swap_map_handle *handle, void *buf,
234                                 struct bio **bio_chain)
235 {
236         int error = 0;
237         unsigned long offset;
238
239         if (!handle->cur)
240                 return -EINVAL;
241         offset = alloc_swap_page(root_swap, handle->bitmap);
242         error = write_page(buf, offset, bio_chain);
243         if (error)
244                 return error;
245         handle->cur->entries[handle->k++] = offset;
246         if (handle->k >= MAP_PAGE_ENTRIES) {
247                 error = wait_on_bio_chain(bio_chain);
248                 if (error)
249                         goto out;
250                 offset = alloc_swap_page(root_swap, handle->bitmap);
251                 if (!offset)
252                         return -ENOSPC;
253                 handle->cur->next_swap = offset;
254                 error = write_page(handle->cur, handle->cur_swap, NULL);
255                 if (error)
256                         goto out;
257                 memset(handle->cur, 0, PAGE_SIZE);
258                 handle->cur_swap = offset;
259                 handle->k = 0;
260         }
261 out:
262         return error;
263 }
264
265 static int flush_swap_writer(struct swap_map_handle *handle)
266 {
267         if (handle->cur && handle->cur_swap)
268                 return write_page(handle->cur, handle->cur_swap, NULL);
269         else
270                 return -EINVAL;
271 }
272
273 /**
274  *      save_image - save the suspend image data
275  */
276
277 static int save_image(struct swap_map_handle *handle,
278                       struct snapshot_handle *snapshot,
279                       unsigned int nr_to_write)
280 {
281         unsigned int m;
282         int ret;
283         int error = 0;
284         int nr_pages;
285         int err2;
286         struct bio *bio;
287         struct timeval start;
288         struct timeval stop;
289
290         printk("Saving image data pages (%u pages) ...     ", nr_to_write);
291         m = nr_to_write / 100;
292         if (!m)
293                 m = 1;
294         nr_pages = 0;
295         bio = NULL;
296         do_gettimeofday(&start);
297         do {
298                 ret = snapshot_read_next(snapshot, PAGE_SIZE);
299                 if (ret > 0) {
300                         error = swap_write_page(handle, data_of(*snapshot),
301                                                 &bio);
302                         if (error)
303                                 break;
304                         if (!(nr_pages % m))
305                                 printk("\b\b\b\b%3d%%", nr_pages / m);
306                         nr_pages++;
307                 }
308         } while (ret > 0);
309         err2 = wait_on_bio_chain(&bio);
310         do_gettimeofday(&stop);
311         if (!error)
312                 error = err2;
313         if (!error)
314                 printk("\b\b\b\bdone\n");
315         show_speed(&start, &stop, nr_to_write, "Wrote");
316         return error;
317 }
318
319 /**
320  *      enough_swap - Make sure we have enough swap to save the image.
321  *
322  *      Returns TRUE or FALSE after checking the total amount of swap
323  *      space avaiable from the resume partition.
324  */
325
326 static int enough_swap(unsigned int nr_pages)
327 {
328         unsigned int free_swap = count_swap_pages(root_swap, 1);
329
330         pr_debug("swsusp: free swap pages: %u\n", free_swap);
331         return free_swap > (nr_pages + PAGES_FOR_IO +
332                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
333 }
334
335 /**
336  *      swsusp_write - Write entire image and metadata.
337  *
338  *      It is important _NOT_ to umount filesystems at this point. We want
339  *      them synced (in case something goes wrong) but we DO not want to mark
340  *      filesystem clean: it is not. (And it does not matter, if we resume
341  *      correctly, we'll mark system clean, anyway.)
342  */
343
344 int swsusp_write(void)
345 {
346         struct swap_map_handle handle;
347         struct snapshot_handle snapshot;
348         struct swsusp_info *header;
349         int error;
350
351         if ((error = swsusp_swap_check())) {
352                 printk(KERN_ERR "swsusp: Cannot find swap device, try swapon -a.\n");
353                 return error;
354         }
355         memset(&snapshot, 0, sizeof(struct snapshot_handle));
356         error = snapshot_read_next(&snapshot, PAGE_SIZE);
357         if (error < PAGE_SIZE)
358                 return error < 0 ? error : -EFAULT;
359         header = (struct swsusp_info *)data_of(snapshot);
360         if (!enough_swap(header->pages)) {
361                 printk(KERN_ERR "swsusp: Not enough free swap\n");
362                 return -ENOSPC;
363         }
364         error = get_swap_writer(&handle);
365         if (!error) {
366                 unsigned long start = handle.cur_swap;
367                 error = swap_write_page(&handle, header, NULL);
368                 if (!error)
369                         error = save_image(&handle, &snapshot,
370                                         header->pages - 1);
371                 if (!error) {
372                         flush_swap_writer(&handle);
373                         printk("S");
374                         error = mark_swapfiles(swp_entry(root_swap, start));
375                         printk("|\n");
376                 }
377         }
378         if (error)
379                 free_all_swap_pages(root_swap, handle.bitmap);
380         release_swap_writer(&handle);
381         return error;
382 }
383
384 /*
385  *      Using bio to read from swap.
386  *      This code requires a bit more work than just using buffer heads
387  *      but, it is the recommended way for 2.5/2.6.
388  *      The following are to signal the beginning and end of I/O. Bios
389  *      finish asynchronously, while we want them to happen synchronously.
390  *      A simple atomic_t, and a wait loop take care of this problem.
391  */
392
393 static atomic_t io_done = ATOMIC_INIT(0);
394
395 static int end_io(struct bio *bio, unsigned int num, int err)
396 {
397         if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
398                 printk(KERN_ERR "I/O error reading swsusp image.\n");
399                 return -EIO;
400         }
401         atomic_set(&io_done, 0);
402         return 0;
403 }
404
405 static struct block_device *resume_bdev;
406
407 /**
408  *      submit - submit BIO request.
409  *      @rw:    READ or WRITE.
410  *      @off    physical offset of page.
411  *      @page:  page we're reading or writing.
412  *
413  *      Straight from the textbook - allocate and initialize the bio.
414  *      If we're writing, make sure the page is marked as dirty.
415  *      Then submit it and wait.
416  */
417
418 static int submit(int rw, pgoff_t page_off, void *page)
419 {
420         int error = 0;
421         struct bio *bio;
422
423         bio = bio_alloc(GFP_ATOMIC, 1);
424         if (!bio)
425                 return -ENOMEM;
426         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
427         bio->bi_bdev = resume_bdev;
428         bio->bi_end_io = end_io;
429
430         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
431                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
432                 error = -EFAULT;
433                 goto Done;
434         }
435
436         atomic_set(&io_done, 1);
437         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
438         while (atomic_read(&io_done))
439                 yield();
440         if (rw == READ)
441                 bio_set_pages_dirty(bio);
442  Done:
443         bio_put(bio);
444         return error;
445 }
446
447 static int bio_read_page(pgoff_t page_off, void *page)
448 {
449         return submit(READ, page_off, page);
450 }
451
452 static int bio_write_page(pgoff_t page_off, void *page)
453 {
454         return submit(WRITE, page_off, page);
455 }
456
457 /**
458  *      The following functions allow us to read data using a swap map
459  *      in a file-alike way
460  */
461
462 static void release_swap_reader(struct swap_map_handle *handle)
463 {
464         if (handle->cur)
465                 free_page((unsigned long)handle->cur);
466         handle->cur = NULL;
467 }
468
469 static int get_swap_reader(struct swap_map_handle *handle,
470                                       swp_entry_t start)
471 {
472         int error;
473
474         if (!swp_offset(start))
475                 return -EINVAL;
476         handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_ATOMIC);
477         if (!handle->cur)
478                 return -ENOMEM;
479         error = bio_read_page(swp_offset(start), handle->cur);
480         if (error) {
481                 release_swap_reader(handle);
482                 return error;
483         }
484         handle->k = 0;
485         return 0;
486 }
487
488 static int swap_read_page(struct swap_map_handle *handle, void *buf)
489 {
490         unsigned long offset;
491         int error;
492
493         if (!handle->cur)
494                 return -EINVAL;
495         offset = handle->cur->entries[handle->k];
496         if (!offset)
497                 return -EFAULT;
498         error = bio_read_page(offset, buf);
499         if (error)
500                 return error;
501         if (++handle->k >= MAP_PAGE_ENTRIES) {
502                 handle->k = 0;
503                 offset = handle->cur->next_swap;
504                 if (!offset)
505                         release_swap_reader(handle);
506                 else
507                         error = bio_read_page(offset, handle->cur);
508         }
509         return error;
510 }
511
512 /**
513  *      load_image - load the image using the swap map handle
514  *      @handle and the snapshot handle @snapshot
515  *      (assume there are @nr_pages pages to load)
516  */
517
518 static int load_image(struct swap_map_handle *handle,
519                       struct snapshot_handle *snapshot,
520                       unsigned int nr_pages)
521 {
522         unsigned int m;
523         int ret;
524         int error = 0;
525         struct timeval start;
526         struct timeval stop;
527
528         printk("Loading image data pages (%u pages) ...     ", nr_pages);
529         m = nr_pages / 100;
530         if (!m)
531                 m = 1;
532         nr_pages = 0;
533         do_gettimeofday(&start);
534         do {
535                 ret = snapshot_write_next(snapshot, PAGE_SIZE);
536                 if (ret > 0) {
537                         error = swap_read_page(handle, data_of(*snapshot));
538                         if (error)
539                                 break;
540                         if (!(nr_pages % m))
541                                 printk("\b\b\b\b%3d%%", nr_pages / m);
542                         nr_pages++;
543                 }
544         } while (ret > 0);
545         do_gettimeofday(&stop);
546         if (!error) {
547                 printk("\b\b\b\bdone\n");
548                 if (!snapshot_image_loaded(snapshot))
549                         error = -ENODATA;
550         }
551         show_speed(&start, &stop, nr_pages, "Read");
552         return error;
553 }
554
555 int swsusp_read(void)
556 {
557         int error;
558         struct swap_map_handle handle;
559         struct snapshot_handle snapshot;
560         struct swsusp_info *header;
561
562         if (IS_ERR(resume_bdev)) {
563                 pr_debug("swsusp: block device not initialised\n");
564                 return PTR_ERR(resume_bdev);
565         }
566
567         memset(&snapshot, 0, sizeof(struct snapshot_handle));
568         error = snapshot_write_next(&snapshot, PAGE_SIZE);
569         if (error < PAGE_SIZE)
570                 return error < 0 ? error : -EFAULT;
571         header = (struct swsusp_info *)data_of(snapshot);
572         error = get_swap_reader(&handle, swsusp_header.image);
573         if (!error)
574                 error = swap_read_page(&handle, header);
575         if (!error)
576                 error = load_image(&handle, &snapshot, header->pages - 1);
577         release_swap_reader(&handle);
578
579         blkdev_put(resume_bdev);
580
581         if (!error)
582                 pr_debug("swsusp: Reading resume file was successful\n");
583         else
584                 pr_debug("swsusp: Error %d resuming\n", error);
585         return error;
586 }
587
588 /**
589  *      swsusp_check - Check for swsusp signature in the resume device
590  */
591
592 int swsusp_check(void)
593 {
594         int error;
595
596         resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
597         if (!IS_ERR(resume_bdev)) {
598                 set_blocksize(resume_bdev, PAGE_SIZE);
599                 memset(&swsusp_header, 0, sizeof(swsusp_header));
600                 if ((error = bio_read_page(0, &swsusp_header)))
601                         return error;
602                 if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
603                         memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
604                         /* Reset swap signature now */
605                         error = bio_write_page(0, &swsusp_header);
606                 } else {
607                         return -EINVAL;
608                 }
609                 if (error)
610                         blkdev_put(resume_bdev);
611                 else
612                         pr_debug("swsusp: Signature found, resuming\n");
613         } else {
614                 error = PTR_ERR(resume_bdev);
615         }
616
617         if (error)
618                 pr_debug("swsusp: Error %d check for resume file\n", error);
619
620         return error;
621 }
622
623 /**
624  *      swsusp_close - close swap device.
625  */
626
627 void swsusp_close(void)
628 {
629         if (IS_ERR(resume_bdev)) {
630                 pr_debug("swsusp: block device not initialised\n");
631                 return;
632         }
633
634         blkdev_put(resume_bdev);
635 }