ia64/xen-unstable

view xen/common/page_alloc.c @ 10929:7cde0d938ef4

[IA64] convert more privop_stat to perfc

Convert most privop stats to perfc.

Signed-off-by: Tristan Gingold <tristan.gingold@bull.net>
author awilliam@xenbuild.aw
date Fri Aug 04 09:02:43 2006 -0600 (2006-08-04)
parents 53f552ad4042
children 21f8c507da29
line source
1 /******************************************************************************
2 * page_alloc.c
3 *
4 * Simple buddy heap allocator for Xen.
5 *
6 * Copyright (c) 2002-2004 K A Fraser
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 */
23 #include <xen/config.h>
24 #include <xen/init.h>
25 #include <xen/types.h>
26 #include <xen/lib.h>
27 #include <xen/sched.h>
28 #include <xen/spinlock.h>
29 #include <xen/mm.h>
30 #include <xen/irq.h>
31 #include <xen/softirq.h>
32 #include <xen/shadow.h>
33 #include <xen/domain_page.h>
34 #include <xen/keyhandler.h>
35 #include <xen/perfc.h>
36 #include <asm/page.h>
38 /*
39 * Comma-separated list of hexadecimal page numbers containing bad bytes.
40 * e.g. 'badpage=0x3f45,0x8a321'.
41 */
42 static char opt_badpage[100] = "";
43 string_param("badpage", opt_badpage);
45 /*
46 * Amount of memory to reserve in a low-memory (<4GB) pool for specific
47 * allocation requests. Ordinary requests will not fall back to the
48 * lowmem emergency pool.
49 */
50 static unsigned long lowmem_emergency_pool_pages;
51 static void parse_lowmem_emergency_pool(char *s)
52 {
53 unsigned long long bytes;
54 bytes = parse_size_and_unit(s);
55 lowmem_emergency_pool_pages = bytes >> PAGE_SHIFT;
56 }
57 custom_param("lowmem_emergency_pool", parse_lowmem_emergency_pool);
59 #define round_pgdown(_p) ((_p)&PAGE_MASK)
60 #define round_pgup(_p) (((_p)+(PAGE_SIZE-1))&PAGE_MASK)
62 static DEFINE_SPINLOCK(page_scrub_lock);
63 LIST_HEAD(page_scrub_list);
64 static unsigned long scrub_pages;
66 /*********************
67 * ALLOCATION BITMAP
68 * One bit per page of memory. Bit set => page is allocated.
69 */
71 static unsigned long *alloc_bitmap;
72 #define PAGES_PER_MAPWORD (sizeof(unsigned long) * 8)
74 #define allocated_in_map(_pn) \
75 ( !! (alloc_bitmap[(_pn)/PAGES_PER_MAPWORD] & \
76 (1UL<<((_pn)&(PAGES_PER_MAPWORD-1)))) )
78 /*
79 * Hint regarding bitwise arithmetic in map_{alloc,free}:
80 * -(1<<n) sets all bits >= n.
81 * (1<<n)-1 sets all bits < n.
82 * Variable names in map_{alloc,free}:
83 * *_idx == Index into `alloc_bitmap' array.
84 * *_off == Bit offset within an element of the `alloc_bitmap' array.
85 */
87 static void map_alloc(unsigned long first_page, unsigned long nr_pages)
88 {
89 unsigned long start_off, end_off, curr_idx, end_idx;
91 #ifndef NDEBUG
92 unsigned long i;
93 /* Check that the block isn't already allocated. */
94 for ( i = 0; i < nr_pages; i++ )
95 ASSERT(!allocated_in_map(first_page + i));
96 #endif
98 curr_idx = first_page / PAGES_PER_MAPWORD;
99 start_off = first_page & (PAGES_PER_MAPWORD-1);
100 end_idx = (first_page + nr_pages) / PAGES_PER_MAPWORD;
101 end_off = (first_page + nr_pages) & (PAGES_PER_MAPWORD-1);
103 if ( curr_idx == end_idx )
104 {
105 alloc_bitmap[curr_idx] |= ((1UL<<end_off)-1) & -(1UL<<start_off);
106 }
107 else
108 {
109 alloc_bitmap[curr_idx] |= -(1UL<<start_off);
110 while ( ++curr_idx < end_idx ) alloc_bitmap[curr_idx] = ~0UL;
111 alloc_bitmap[curr_idx] |= (1UL<<end_off)-1;
112 }
113 }
116 static void map_free(unsigned long first_page, unsigned long nr_pages)
117 {
118 unsigned long start_off, end_off, curr_idx, end_idx;
120 #ifndef NDEBUG
121 unsigned long i;
122 /* Check that the block isn't already freed. */
123 for ( i = 0; i < nr_pages; i++ )
124 ASSERT(allocated_in_map(first_page + i));
125 #endif
127 curr_idx = first_page / PAGES_PER_MAPWORD;
128 start_off = first_page & (PAGES_PER_MAPWORD-1);
129 end_idx = (first_page + nr_pages) / PAGES_PER_MAPWORD;
130 end_off = (first_page + nr_pages) & (PAGES_PER_MAPWORD-1);
132 if ( curr_idx == end_idx )
133 {
134 alloc_bitmap[curr_idx] &= -(1UL<<end_off) | ((1UL<<start_off)-1);
135 }
136 else
137 {
138 alloc_bitmap[curr_idx] &= (1UL<<start_off)-1;
139 while ( ++curr_idx != end_idx ) alloc_bitmap[curr_idx] = 0;
140 alloc_bitmap[curr_idx] &= -(1UL<<end_off);
141 }
142 }
146 /*************************
147 * BOOT-TIME ALLOCATOR
148 */
150 /* Initialise allocator to handle up to @max_page pages. */
151 paddr_t init_boot_allocator(paddr_t bitmap_start)
152 {
153 unsigned long bitmap_size;
155 bitmap_start = round_pgup(bitmap_start);
157 /*
158 * Allocate space for the allocation bitmap. Include an extra longword
159 * of padding for possible overrun in map_alloc and map_free.
160 */
161 bitmap_size = max_page / 8;
162 bitmap_size += sizeof(unsigned long);
163 bitmap_size = round_pgup(bitmap_size);
164 alloc_bitmap = (unsigned long *)maddr_to_virt(bitmap_start);
166 /* All allocated by default. */
167 memset(alloc_bitmap, ~0, bitmap_size);
169 return bitmap_start + bitmap_size;
170 }
172 void init_boot_pages(paddr_t ps, paddr_t pe)
173 {
174 unsigned long bad_spfn, bad_epfn, i;
175 char *p;
177 ps = round_pgup(ps);
178 pe = round_pgdown(pe);
179 if ( pe <= ps )
180 return;
182 map_free(ps >> PAGE_SHIFT, (pe - ps) >> PAGE_SHIFT);
184 /* Check new pages against the bad-page list. */
185 p = opt_badpage;
186 while ( *p != '\0' )
187 {
188 bad_spfn = simple_strtoul(p, &p, 0);
189 bad_epfn = bad_spfn;
191 if ( *p == '-' )
192 {
193 p++;
194 bad_epfn = simple_strtoul(p, &p, 0);
195 if ( bad_epfn < bad_spfn )
196 bad_epfn = bad_spfn;
197 }
199 if ( *p == ',' )
200 p++;
201 else if ( *p != '\0' )
202 break;
204 if ( bad_epfn == bad_spfn )
205 printk("Marking page %lx as bad\n", bad_spfn);
206 else
207 printk("Marking pages %lx through %lx as bad\n",
208 bad_spfn, bad_epfn);
210 for ( i = bad_spfn; i <= bad_epfn; i++ )
211 if ( (i < max_page) && !allocated_in_map(i) )
212 map_alloc(i, 1);
213 }
214 }
216 unsigned long alloc_boot_pages(unsigned long nr_pfns, unsigned long pfn_align)
217 {
218 unsigned long pg, i;
220 for ( pg = 0; (pg + nr_pfns) < max_page; pg += pfn_align )
221 {
222 for ( i = 0; i < nr_pfns; i++ )
223 if ( allocated_in_map(pg + i) )
224 break;
226 if ( i == nr_pfns )
227 {
228 map_alloc(pg, nr_pfns);
229 return pg;
230 }
231 }
233 return 0;
234 }
238 /*************************
239 * BINARY BUDDY ALLOCATOR
240 */
242 #define MEMZONE_XEN 0
243 #define MEMZONE_DOM 1
244 #define MEMZONE_DMADOM 2
245 #define NR_ZONES 3
247 #define pfn_dom_zone_type(_pfn) \
248 (((_pfn) <= MAX_DMADOM_PFN) ? MEMZONE_DMADOM : MEMZONE_DOM)
250 static struct list_head heap[NR_ZONES][MAX_ORDER+1];
252 static unsigned long avail[NR_ZONES];
254 static DEFINE_SPINLOCK(heap_lock);
256 void end_boot_allocator(void)
257 {
258 unsigned long i, j;
259 int curr_free = 0, next_free = 0;
261 memset(avail, 0, sizeof(avail));
263 for ( i = 0; i < NR_ZONES; i++ )
264 for ( j = 0; j <= MAX_ORDER; j++ )
265 INIT_LIST_HEAD(&heap[i][j]);
267 /* Pages that are free now go to the domain sub-allocator. */
268 for ( i = 0; i < max_page; i++ )
269 {
270 curr_free = next_free;
271 next_free = !allocated_in_map(i+1);
272 if ( next_free )
273 map_alloc(i+1, 1); /* prevent merging in free_heap_pages() */
274 if ( curr_free )
275 free_heap_pages(pfn_dom_zone_type(i), mfn_to_page(i), 0);
276 }
277 }
279 /* Hand the specified arbitrary page range to the specified heap zone. */
280 void init_heap_pages(
281 unsigned int zone, struct page_info *pg, unsigned long nr_pages)
282 {
283 unsigned long i;
285 ASSERT(zone < NR_ZONES);
287 for ( i = 0; i < nr_pages; i++ )
288 free_heap_pages(zone, pg+i, 0);
289 }
292 /* Allocate 2^@order contiguous pages. */
293 struct page_info *alloc_heap_pages(unsigned int zone, unsigned int order)
294 {
295 int i;
296 struct page_info *pg;
298 ASSERT(zone < NR_ZONES);
300 if ( unlikely(order > MAX_ORDER) )
301 return NULL;
303 spin_lock(&heap_lock);
305 /* Find smallest order which can satisfy the request. */
306 for ( i = order; i <= MAX_ORDER; i++ )
307 if ( !list_empty(&heap[zone][i]) )
308 goto found;
310 /* No suitable memory blocks. Fail the request. */
311 spin_unlock(&heap_lock);
312 return NULL;
314 found:
315 pg = list_entry(heap[zone][i].next, struct page_info, list);
316 list_del(&pg->list);
318 /* We may have to halve the chunk a number of times. */
319 while ( i != order )
320 {
321 PFN_ORDER(pg) = --i;
322 list_add_tail(&pg->list, &heap[zone][i]);
323 pg += 1 << i;
324 }
326 map_alloc(page_to_mfn(pg), 1 << order);
327 avail[zone] -= 1 << order;
329 spin_unlock(&heap_lock);
331 return pg;
332 }
335 /* Free 2^@order set of pages. */
336 void free_heap_pages(
337 unsigned int zone, struct page_info *pg, unsigned int order)
338 {
339 unsigned long mask;
341 ASSERT(zone < NR_ZONES);
342 ASSERT(order <= MAX_ORDER);
344 spin_lock(&heap_lock);
346 map_free(page_to_mfn(pg), 1 << order);
347 avail[zone] += 1 << order;
349 /* Merge chunks as far as possible. */
350 while ( order < MAX_ORDER )
351 {
352 mask = 1 << order;
354 if ( (page_to_mfn(pg) & mask) )
355 {
356 /* Merge with predecessor block? */
357 if ( allocated_in_map(page_to_mfn(pg)-mask) ||
358 (PFN_ORDER(pg-mask) != order) )
359 break;
360 list_del(&(pg-mask)->list);
361 pg -= mask;
362 }
363 else
364 {
365 /* Merge with successor block? */
366 if ( allocated_in_map(page_to_mfn(pg)+mask) ||
367 (PFN_ORDER(pg+mask) != order) )
368 break;
369 list_del(&(pg+mask)->list);
370 }
372 order++;
373 }
375 PFN_ORDER(pg) = order;
376 list_add_tail(&pg->list, &heap[zone][order]);
378 spin_unlock(&heap_lock);
379 }
382 /*
383 * Scrub all unallocated pages in all heap zones. This function is more
384 * convoluted than appears necessary because we do not want to continuously
385 * hold the lock or disable interrupts while scrubbing very large memory areas.
386 */
387 void scrub_heap_pages(void)
388 {
389 void *p;
390 unsigned long pfn;
392 printk("Scrubbing Free RAM: ");
394 for ( pfn = 0; pfn < max_page; pfn++ )
395 {
396 /* Every 100MB, print a progress dot. */
397 if ( (pfn % ((100*1024*1024)/PAGE_SIZE)) == 0 )
398 printk(".");
400 process_pending_timers();
402 /* Quick lock-free check. */
403 if ( allocated_in_map(pfn) )
404 continue;
406 spin_lock_irq(&heap_lock);
408 /* Re-check page status with lock held. */
409 if ( !allocated_in_map(pfn) )
410 {
411 if ( IS_XEN_HEAP_FRAME(mfn_to_page(pfn)) )
412 {
413 p = page_to_virt(mfn_to_page(pfn));
414 memguard_unguard_range(p, PAGE_SIZE);
415 clear_page(p);
416 memguard_guard_range(p, PAGE_SIZE);
417 }
418 else
419 {
420 p = map_domain_page(pfn);
421 clear_page(p);
422 unmap_domain_page(p);
423 }
424 }
426 spin_unlock_irq(&heap_lock);
427 }
429 printk("done.\n");
430 }
434 /*************************
435 * XEN-HEAP SUB-ALLOCATOR
436 */
438 void init_xenheap_pages(paddr_t ps, paddr_t pe)
439 {
440 unsigned long flags;
442 ps = round_pgup(ps);
443 pe = round_pgdown(pe);
444 if ( pe <= ps )
445 return;
447 memguard_guard_range(maddr_to_virt(ps), pe - ps);
449 /*
450 * Yuk! Ensure there is a one-page buffer between Xen and Dom zones, to
451 * prevent merging of power-of-two blocks across the zone boundary.
452 */
453 if ( !IS_XEN_HEAP_FRAME(maddr_to_page(pe)) )
454 pe -= PAGE_SIZE;
456 local_irq_save(flags);
457 init_heap_pages(MEMZONE_XEN, maddr_to_page(ps), (pe - ps) >> PAGE_SHIFT);
458 local_irq_restore(flags);
459 }
462 void *alloc_xenheap_pages(unsigned int order)
463 {
464 unsigned long flags;
465 struct page_info *pg;
466 int i;
468 local_irq_save(flags);
469 pg = alloc_heap_pages(MEMZONE_XEN, order);
470 local_irq_restore(flags);
472 if ( unlikely(pg == NULL) )
473 goto no_memory;
475 memguard_unguard_range(page_to_virt(pg), 1 << (order + PAGE_SHIFT));
477 for ( i = 0; i < (1 << order); i++ )
478 {
479 pg[i].count_info = 0;
480 pg[i].u.inuse._domain = 0;
481 pg[i].u.inuse.type_info = 0;
482 }
484 return page_to_virt(pg);
486 no_memory:
487 printk("Cannot handle page request order %d!\n", order);
488 return NULL;
489 }
492 void free_xenheap_pages(void *v, unsigned int order)
493 {
494 unsigned long flags;
496 if ( v == NULL )
497 return;
499 memguard_guard_range(v, 1 << (order + PAGE_SHIFT));
501 local_irq_save(flags);
502 free_heap_pages(MEMZONE_XEN, virt_to_page(v), order);
503 local_irq_restore(flags);
504 }
508 /*************************
509 * DOMAIN-HEAP SUB-ALLOCATOR
510 */
512 void init_domheap_pages(paddr_t ps, paddr_t pe)
513 {
514 unsigned long s_tot, e_tot, s_dma, e_dma, s_nrm, e_nrm;
516 ASSERT(!in_irq());
518 s_tot = round_pgup(ps) >> PAGE_SHIFT;
519 e_tot = round_pgdown(pe) >> PAGE_SHIFT;
521 s_dma = min(s_tot, MAX_DMADOM_PFN + 1);
522 e_dma = min(e_tot, MAX_DMADOM_PFN + 1);
523 if ( s_dma < e_dma )
524 init_heap_pages(MEMZONE_DMADOM, mfn_to_page(s_dma), e_dma - s_dma);
526 s_nrm = max(s_tot, MAX_DMADOM_PFN + 1);
527 e_nrm = max(e_tot, MAX_DMADOM_PFN + 1);
528 if ( s_nrm < e_nrm )
529 init_heap_pages(MEMZONE_DOM, mfn_to_page(s_nrm), e_nrm - s_nrm);
530 }
533 int assign_pages(
534 struct domain *d,
535 struct page_info *pg,
536 unsigned int order,
537 unsigned int memflags)
538 {
539 unsigned long i;
541 spin_lock(&d->page_alloc_lock);
543 if ( unlikely(test_bit(_DOMF_dying, &d->domain_flags)) )
544 {
545 DPRINTK("Cannot assign page to domain%d -- dying.\n", d->domain_id);
546 goto fail;
547 }
549 if ( !(memflags & MEMF_no_refcount) )
550 {
551 if ( unlikely((d->tot_pages + (1 << order)) > d->max_pages) )
552 {
553 DPRINTK("Over-allocation for domain %u: %u > %u\n",
554 d->domain_id, d->tot_pages + (1 << order), d->max_pages);
555 goto fail;
556 }
558 if ( unlikely(d->tot_pages == 0) )
559 get_knownalive_domain(d);
561 d->tot_pages += 1 << order;
562 }
564 for ( i = 0; i < (1 << order); i++ )
565 {
566 ASSERT(page_get_owner(&pg[i]) == NULL);
567 ASSERT((pg[i].count_info & ~(PGC_allocated | 1)) == 0);
568 page_set_owner(&pg[i], d);
569 wmb(); /* Domain pointer must be visible before updating refcnt. */
570 pg[i].count_info = PGC_allocated | 1;
571 list_add_tail(&pg[i].list, &d->page_list);
572 }
574 spin_unlock(&d->page_alloc_lock);
575 return 0;
577 fail:
578 spin_unlock(&d->page_alloc_lock);
579 return -1;
580 }
583 struct page_info *alloc_domheap_pages(
584 struct domain *d, unsigned int order, unsigned int memflags)
585 {
586 struct page_info *pg = NULL;
587 cpumask_t mask;
588 unsigned long i;
590 ASSERT(!in_irq());
592 if ( !(memflags & MEMF_dma) )
593 {
594 pg = alloc_heap_pages(MEMZONE_DOM, order);
595 /* Failure? Then check if we can fall back to the DMA pool. */
596 if ( unlikely(pg == NULL) &&
597 ((order > MAX_ORDER) ||
598 (avail[MEMZONE_DMADOM] <
599 (lowmem_emergency_pool_pages + (1UL << order)))) )
600 return NULL;
601 }
603 if ( pg == NULL )
604 if ( (pg = alloc_heap_pages(MEMZONE_DMADOM, order)) == NULL )
605 return NULL;
607 mask = pg->u.free.cpumask;
608 tlbflush_filter(mask, pg->tlbflush_timestamp);
610 pg->count_info = 0;
611 pg->u.inuse._domain = 0;
612 pg->u.inuse.type_info = 0;
614 for ( i = 1; i < (1 << order); i++ )
615 {
616 /* Add in any extra CPUs that need flushing because of this page. */
617 cpumask_t extra_cpus_mask;
618 cpus_andnot(extra_cpus_mask, pg[i].u.free.cpumask, mask);
619 tlbflush_filter(extra_cpus_mask, pg[i].tlbflush_timestamp);
620 cpus_or(mask, mask, extra_cpus_mask);
622 pg[i].count_info = 0;
623 pg[i].u.inuse._domain = 0;
624 pg[i].u.inuse.type_info = 0;
625 page_set_owner(&pg[i], NULL);
626 }
628 if ( unlikely(!cpus_empty(mask)) )
629 {
630 perfc_incrc(need_flush_tlb_flush);
631 flush_tlb_mask(mask);
632 }
634 if ( (d != NULL) && assign_pages(d, pg, order, memflags) )
635 {
636 free_heap_pages(pfn_dom_zone_type(page_to_mfn(pg)), pg, order);
637 return NULL;
638 }
640 return pg;
641 }
644 void free_domheap_pages(struct page_info *pg, unsigned int order)
645 {
646 int i, drop_dom_ref;
647 struct domain *d = page_get_owner(pg);
649 ASSERT(!in_irq());
651 if ( unlikely(IS_XEN_HEAP_FRAME(pg)) )
652 {
653 /* NB. May recursively lock from relinquish_memory(). */
654 spin_lock_recursive(&d->page_alloc_lock);
656 for ( i = 0; i < (1 << order); i++ )
657 list_del(&pg[i].list);
659 d->xenheap_pages -= 1 << order;
660 drop_dom_ref = (d->xenheap_pages == 0);
662 spin_unlock_recursive(&d->page_alloc_lock);
663 }
664 else if ( likely(d != NULL) )
665 {
666 /* NB. May recursively lock from relinquish_memory(). */
667 spin_lock_recursive(&d->page_alloc_lock);
669 for ( i = 0; i < (1 << order); i++ )
670 {
671 shadow_drop_references(d, &pg[i]);
672 ASSERT((pg[i].u.inuse.type_info & PGT_count_mask) == 0);
673 pg[i].tlbflush_timestamp = tlbflush_current_time();
674 pg[i].u.free.cpumask = d->domain_dirty_cpumask;
675 list_del(&pg[i].list);
676 }
678 d->tot_pages -= 1 << order;
679 drop_dom_ref = (d->tot_pages == 0);
681 spin_unlock_recursive(&d->page_alloc_lock);
683 if ( likely(!test_bit(_DOMF_dying, &d->domain_flags)) )
684 {
685 free_heap_pages(pfn_dom_zone_type(page_to_mfn(pg)), pg, order);
686 }
687 else
688 {
689 /*
690 * Normally we expect a domain to clear pages before freeing them,
691 * if it cares about the secrecy of their contents. However, after
692 * a domain has died we assume responsibility for erasure.
693 */
694 for ( i = 0; i < (1 << order); i++ )
695 {
696 spin_lock(&page_scrub_lock);
697 list_add(&pg[i].list, &page_scrub_list);
698 scrub_pages++;
699 spin_unlock(&page_scrub_lock);
700 }
701 }
702 }
703 else
704 {
705 /* Freeing anonymous domain-heap pages. */
706 for ( i = 0; i < (1 << order); i++ )
707 pg[i].u.free.cpumask = CPU_MASK_NONE;
708 free_heap_pages(pfn_dom_zone_type(page_to_mfn(pg)), pg, order);
709 drop_dom_ref = 0;
710 }
712 if ( drop_dom_ref )
713 put_domain(d);
714 }
717 unsigned long avail_domheap_pages(void)
718 {
719 unsigned long avail_nrm, avail_dma;
721 avail_nrm = avail[MEMZONE_DOM];
723 avail_dma = avail[MEMZONE_DMADOM];
724 if ( avail_dma > lowmem_emergency_pool_pages )
725 avail_dma -= lowmem_emergency_pool_pages;
726 else
727 avail_dma = 0;
729 return avail_nrm + avail_dma;
730 }
733 static void pagealloc_keyhandler(unsigned char key)
734 {
735 printk("Physical memory information:\n");
736 printk(" Xen heap: %lukB free\n"
737 " DMA heap: %lukB free\n"
738 " Dom heap: %lukB free\n",
739 avail[MEMZONE_XEN]<<(PAGE_SHIFT-10),
740 avail[MEMZONE_DMADOM]<<(PAGE_SHIFT-10),
741 avail[MEMZONE_DOM]<<(PAGE_SHIFT-10));
742 }
745 static __init int pagealloc_keyhandler_init(void)
746 {
747 register_keyhandler('m', pagealloc_keyhandler, "memory info");
748 return 0;
749 }
750 __initcall(pagealloc_keyhandler_init);
754 /*************************
755 * PAGE SCRUBBING
756 */
758 static void page_scrub_softirq(void)
759 {
760 struct list_head *ent;
761 struct page_info *pg;
762 void *p;
763 int i;
764 s_time_t start = NOW();
766 /* Aim to do 1ms of work (ten percent of a 10ms jiffy). */
767 do {
768 spin_lock(&page_scrub_lock);
770 if ( unlikely((ent = page_scrub_list.next) == &page_scrub_list) )
771 {
772 spin_unlock(&page_scrub_lock);
773 return;
774 }
776 /* Peel up to 16 pages from the list. */
777 for ( i = 0; i < 16; i++ )
778 {
779 if ( ent->next == &page_scrub_list )
780 break;
781 ent = ent->next;
782 }
784 /* Remove peeled pages from the list. */
785 ent->next->prev = &page_scrub_list;
786 page_scrub_list.next = ent->next;
787 scrub_pages -= (i+1);
789 spin_unlock(&page_scrub_lock);
791 /* Working backwards, scrub each page in turn. */
792 while ( ent != &page_scrub_list )
793 {
794 pg = list_entry(ent, struct page_info, list);
795 ent = ent->prev;
796 p = map_domain_page(page_to_mfn(pg));
797 clear_page(p);
798 unmap_domain_page(p);
799 free_heap_pages(pfn_dom_zone_type(page_to_mfn(pg)), pg, 0);
800 }
801 } while ( (NOW() - start) < MILLISECS(1) );
802 }
804 unsigned long avail_scrub_pages(void)
805 {
806 return scrub_pages;
807 }
809 static __init int page_scrub_init(void)
810 {
811 open_softirq(PAGE_SCRUB_SOFTIRQ, page_scrub_softirq);
812 return 0;
813 }
814 __initcall(page_scrub_init);
816 /*
817 * Local variables:
818 * mode: C
819 * c-set-style: "BSD"
820 * c-basic-offset: 4
821 * tab-width: 4
822 * indent-tabs-mode: nil
823 * End:
824 */