| 1 |
1275 |
phoenix |
/*
|
| 2 |
|
|
* linux/mm/vmscan.c
|
| 3 |
|
|
*
|
| 4 |
|
|
* The pageout daemon, decides which pages to evict (swap out) and
|
| 5 |
|
|
* does the actual work of freeing them.
|
| 6 |
|
|
*
|
| 7 |
|
|
* Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
|
| 8 |
|
|
*
|
| 9 |
|
|
* Swap reorganised 29.12.95, Stephen Tweedie.
|
| 10 |
|
|
* kswapd added: 7.1.96 sct
|
| 11 |
|
|
* Removed kswapd_ctl limits, and swap out as many pages as needed
|
| 12 |
|
|
* to bring the system back to freepages.high: 2.4.97, Rik van Riel.
|
| 13 |
|
|
* Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com).
|
| 14 |
|
|
* Multiqueue VM started 5.8.00, Rik van Riel.
|
| 15 |
|
|
*/
|
| 16 |
|
|
|
| 17 |
|
|
#include <linux/slab.h>
|
| 18 |
|
|
#include <linux/kernel_stat.h>
|
| 19 |
|
|
#include <linux/swap.h>
|
| 20 |
|
|
#include <linux/swapctl.h>
|
| 21 |
|
|
#include <linux/smp_lock.h>
|
| 22 |
|
|
#include <linux/pagemap.h>
|
| 23 |
|
|
#include <linux/init.h>
|
| 24 |
|
|
#include <linux/highmem.h>
|
| 25 |
|
|
#include <linux/file.h>
|
| 26 |
|
|
|
| 27 |
|
|
#include <asm/pgalloc.h>
|
| 28 |
|
|
|
| 29 |
|
|
/*
|
| 30 |
|
|
* "vm_passes" is the number of vm passes before failing the
|
| 31 |
|
|
* memory balancing. Take into account 3 passes are needed
|
| 32 |
|
|
* for a flush/wait/free cycle and that we only scan 1/vm_cache_scan_ratio
|
| 33 |
|
|
* of the inactive list at each pass.
|
| 34 |
|
|
*/
|
| 35 |
|
|
int vm_passes = 60;
|
| 36 |
|
|
|
| 37 |
|
|
/*
|
| 38 |
|
|
* "vm_cache_scan_ratio" is how much of the inactive LRU queue we will scan
|
| 39 |
|
|
* in one go. A value of 6 for vm_cache_scan_ratio implies that we'll
|
| 40 |
|
|
* scan 1/6 of the inactive lists during a normal aging round.
|
| 41 |
|
|
*/
|
| 42 |
|
|
int vm_cache_scan_ratio = 6;
|
| 43 |
|
|
|
| 44 |
|
|
/*
|
| 45 |
|
|
* "vm_mapped_ratio" controls the pageout rate, the smaller, the earlier
|
| 46 |
|
|
* we'll start to pageout.
|
| 47 |
|
|
*/
|
| 48 |
|
|
int vm_mapped_ratio = 100;
|
| 49 |
|
|
|
| 50 |
|
|
/*
|
| 51 |
|
|
* "vm_lru_balance_ratio" controls the balance between active and
|
| 52 |
|
|
* inactive cache. The bigger vm_balance is, the easier the
|
| 53 |
|
|
* active cache will grow, because we'll rotate the active list
|
| 54 |
|
|
* slowly. A value of 2 means we'll go towards a balance of
|
| 55 |
|
|
* 1/3 of the cache being inactive.
|
| 56 |
|
|
*/
|
| 57 |
|
|
int vm_lru_balance_ratio = 2;
|
| 58 |
|
|
|
| 59 |
|
|
/*
|
| 60 |
|
|
* "vm_vfs_scan_ratio" is what proportion of the VFS queues we will scan
|
| 61 |
|
|
* in one go. A value of 6 for vm_vfs_scan_ratio implies that 1/6th of
|
| 62 |
|
|
* the unused-inode, dentry and dquot caches will be freed during a normal
|
| 63 |
|
|
* aging round.
|
| 64 |
|
|
*/
|
| 65 |
|
|
int vm_vfs_scan_ratio = 6;
|
| 66 |
|
|
|
| 67 |
|
|
/*
|
| 68 |
|
|
* The swap-out function returns 1 if it successfully
|
| 69 |
|
|
* scanned all the pages it was asked to (`count').
|
| 70 |
|
|
* It returns zero if it couldn't do anything,
|
| 71 |
|
|
*
|
| 72 |
|
|
* rss may decrease because pages are shared, but this
|
| 73 |
|
|
* doesn't count as having freed a page.
|
| 74 |
|
|
*/
|
| 75 |
|
|
|
| 76 |
|
|
/* mm->page_table_lock is held. mmap_sem is not held */
|
| 77 |
|
|
static inline int try_to_swap_out(struct mm_struct * mm, struct vm_area_struct* vma, unsigned long address, pte_t * page_table, struct page *page, zone_t * classzone)
|
| 78 |
|
|
{
|
| 79 |
|
|
pte_t pte;
|
| 80 |
|
|
swp_entry_t entry;
|
| 81 |
|
|
|
| 82 |
|
|
/* Don't look at this pte if it's been accessed recently. */
|
| 83 |
|
|
if ((vma->vm_flags & VM_LOCKED) || ptep_test_and_clear_young(page_table)) {
|
| 84 |
|
|
mark_page_accessed(page);
|
| 85 |
|
|
return 0;
|
| 86 |
|
|
}
|
| 87 |
|
|
|
| 88 |
|
|
/* Don't bother unmapping pages that are active */
|
| 89 |
|
|
if (PageActive(page))
|
| 90 |
|
|
return 0;
|
| 91 |
|
|
|
| 92 |
|
|
/* Don't bother replenishing zones not under pressure.. */
|
| 93 |
|
|
if (!memclass(page_zone(page), classzone))
|
| 94 |
|
|
return 0;
|
| 95 |
|
|
|
| 96 |
|
|
if (TryLockPage(page))
|
| 97 |
|
|
return 0;
|
| 98 |
|
|
|
| 99 |
|
|
/* From this point on, the odds are that we're going to
|
| 100 |
|
|
* nuke this pte, so read and clear the pte. This hook
|
| 101 |
|
|
* is needed on CPUs which update the accessed and dirty
|
| 102 |
|
|
* bits in hardware.
|
| 103 |
|
|
*/
|
| 104 |
|
|
flush_cache_page(vma, address);
|
| 105 |
|
|
pte = ptep_get_and_clear(page_table);
|
| 106 |
|
|
flush_tlb_page(vma, address);
|
| 107 |
|
|
|
| 108 |
|
|
if (pte_dirty(pte))
|
| 109 |
|
|
set_page_dirty(page);
|
| 110 |
|
|
|
| 111 |
|
|
/*
|
| 112 |
|
|
* Is the page already in the swap cache? If so, then
|
| 113 |
|
|
* we can just drop our reference to it without doing
|
| 114 |
|
|
* any IO - it's already up-to-date on disk.
|
| 115 |
|
|
*/
|
| 116 |
|
|
if (PageSwapCache(page)) {
|
| 117 |
|
|
entry.val = page->index;
|
| 118 |
|
|
swap_duplicate(entry);
|
| 119 |
|
|
set_swap_pte:
|
| 120 |
|
|
set_pte(page_table, swp_entry_to_pte(entry));
|
| 121 |
|
|
drop_pte:
|
| 122 |
|
|
mm->rss--;
|
| 123 |
|
|
UnlockPage(page);
|
| 124 |
|
|
{
|
| 125 |
|
|
int freeable = page_count(page) - !!page->buffers <= 2;
|
| 126 |
|
|
page_cache_release(page);
|
| 127 |
|
|
return freeable;
|
| 128 |
|
|
}
|
| 129 |
|
|
}
|
| 130 |
|
|
|
| 131 |
|
|
/*
|
| 132 |
|
|
* Is it a clean page? Then it must be recoverable
|
| 133 |
|
|
* by just paging it in again, and we can just drop
|
| 134 |
|
|
* it.. or if it's dirty but has backing store,
|
| 135 |
|
|
* just mark the page dirty and drop it.
|
| 136 |
|
|
*
|
| 137 |
|
|
* However, this won't actually free any real
|
| 138 |
|
|
* memory, as the page will just be in the page cache
|
| 139 |
|
|
* somewhere, and as such we should just continue
|
| 140 |
|
|
* our scan.
|
| 141 |
|
|
*
|
| 142 |
|
|
* Basically, this just makes it possible for us to do
|
| 143 |
|
|
* some real work in the future in "refill_inactive()".
|
| 144 |
|
|
*/
|
| 145 |
|
|
if (page->mapping)
|
| 146 |
|
|
goto drop_pte;
|
| 147 |
|
|
if (!PageDirty(page))
|
| 148 |
|
|
goto drop_pte;
|
| 149 |
|
|
|
| 150 |
|
|
/*
|
| 151 |
|
|
* Anonymous buffercache pages can be left behind by
|
| 152 |
|
|
* concurrent truncate and pagefault.
|
| 153 |
|
|
*/
|
| 154 |
|
|
if (page->buffers)
|
| 155 |
|
|
goto preserve;
|
| 156 |
|
|
|
| 157 |
|
|
/*
|
| 158 |
|
|
* This is a dirty, swappable page. First of all,
|
| 159 |
|
|
* get a suitable swap entry for it, and make sure
|
| 160 |
|
|
* we have the swap cache set up to associate the
|
| 161 |
|
|
* page with that swap entry.
|
| 162 |
|
|
*/
|
| 163 |
|
|
for (;;) {
|
| 164 |
|
|
entry = get_swap_page();
|
| 165 |
|
|
if (!entry.val)
|
| 166 |
|
|
break;
|
| 167 |
|
|
/* Add it to the swap cache and mark it dirty
|
| 168 |
|
|
* (adding to the page cache will clear the dirty
|
| 169 |
|
|
* and uptodate bits, so we need to do it again)
|
| 170 |
|
|
*/
|
| 171 |
|
|
if (add_to_swap_cache(page, entry) == 0) {
|
| 172 |
|
|
SetPageUptodate(page);
|
| 173 |
|
|
set_page_dirty(page);
|
| 174 |
|
|
goto set_swap_pte;
|
| 175 |
|
|
}
|
| 176 |
|
|
/* Raced with "speculative" read_swap_cache_async */
|
| 177 |
|
|
swap_free(entry);
|
| 178 |
|
|
}
|
| 179 |
|
|
|
| 180 |
|
|
/* No swap space left */
|
| 181 |
|
|
preserve:
|
| 182 |
|
|
set_pte(page_table, pte);
|
| 183 |
|
|
UnlockPage(page);
|
| 184 |
|
|
return 0;
|
| 185 |
|
|
}
|
| 186 |
|
|
|
| 187 |
|
|
/* mm->page_table_lock is held. mmap_sem is not held */
|
| 188 |
|
|
static inline int swap_out_pmd(struct mm_struct * mm, struct vm_area_struct * vma, pmd_t *dir, unsigned long address, unsigned long end, int count, zone_t * classzone)
|
| 189 |
|
|
{
|
| 190 |
|
|
pte_t * pte;
|
| 191 |
|
|
unsigned long pmd_end;
|
| 192 |
|
|
|
| 193 |
|
|
if (pmd_none(*dir))
|
| 194 |
|
|
return count;
|
| 195 |
|
|
if (pmd_bad(*dir)) {
|
| 196 |
|
|
pmd_ERROR(*dir);
|
| 197 |
|
|
pmd_clear(dir);
|
| 198 |
|
|
return count;
|
| 199 |
|
|
}
|
| 200 |
|
|
|
| 201 |
|
|
pte = pte_offset(dir, address);
|
| 202 |
|
|
|
| 203 |
|
|
pmd_end = (address + PMD_SIZE) & PMD_MASK;
|
| 204 |
|
|
if (end > pmd_end)
|
| 205 |
|
|
end = pmd_end;
|
| 206 |
|
|
|
| 207 |
|
|
do {
|
| 208 |
|
|
if (pte_present(*pte)) {
|
| 209 |
|
|
struct page *page = pte_page(*pte);
|
| 210 |
|
|
|
| 211 |
|
|
if (VALID_PAGE(page) && !PageReserved(page)) {
|
| 212 |
|
|
count -= try_to_swap_out(mm, vma, address, pte, page, classzone);
|
| 213 |
|
|
if (!count) {
|
| 214 |
|
|
address += PAGE_SIZE;
|
| 215 |
|
|
break;
|
| 216 |
|
|
}
|
| 217 |
|
|
}
|
| 218 |
|
|
}
|
| 219 |
|
|
address += PAGE_SIZE;
|
| 220 |
|
|
pte++;
|
| 221 |
|
|
} while (address && (address < end));
|
| 222 |
|
|
mm->swap_address = address;
|
| 223 |
|
|
return count;
|
| 224 |
|
|
}
|
| 225 |
|
|
|
| 226 |
|
|
/* mm->page_table_lock is held. mmap_sem is not held */
|
| 227 |
|
|
static inline int swap_out_pgd(struct mm_struct * mm, struct vm_area_struct * vma, pgd_t *dir, unsigned long address, unsigned long end, int count, zone_t * classzone)
|
| 228 |
|
|
{
|
| 229 |
|
|
pmd_t * pmd;
|
| 230 |
|
|
unsigned long pgd_end;
|
| 231 |
|
|
|
| 232 |
|
|
if (pgd_none(*dir))
|
| 233 |
|
|
return count;
|
| 234 |
|
|
if (pgd_bad(*dir)) {
|
| 235 |
|
|
pgd_ERROR(*dir);
|
| 236 |
|
|
pgd_clear(dir);
|
| 237 |
|
|
return count;
|
| 238 |
|
|
}
|
| 239 |
|
|
|
| 240 |
|
|
pmd = pmd_offset(dir, address);
|
| 241 |
|
|
|
| 242 |
|
|
pgd_end = (address + PGDIR_SIZE) & PGDIR_MASK;
|
| 243 |
|
|
if (pgd_end && (end > pgd_end))
|
| 244 |
|
|
end = pgd_end;
|
| 245 |
|
|
|
| 246 |
|
|
do {
|
| 247 |
|
|
count = swap_out_pmd(mm, vma, pmd, address, end, count, classzone);
|
| 248 |
|
|
if (!count)
|
| 249 |
|
|
break;
|
| 250 |
|
|
address = (address + PMD_SIZE) & PMD_MASK;
|
| 251 |
|
|
pmd++;
|
| 252 |
|
|
} while (address && (address < end));
|
| 253 |
|
|
return count;
|
| 254 |
|
|
}
|
| 255 |
|
|
|
| 256 |
|
|
/* mm->page_table_lock is held. mmap_sem is not held */
|
| 257 |
|
|
static inline int swap_out_vma(struct mm_struct * mm, struct vm_area_struct * vma, unsigned long address, int count, zone_t * classzone)
|
| 258 |
|
|
{
|
| 259 |
|
|
pgd_t *pgdir;
|
| 260 |
|
|
unsigned long end;
|
| 261 |
|
|
|
| 262 |
|
|
/* Don't swap out areas which are reserved */
|
| 263 |
|
|
if (vma->vm_flags & VM_RESERVED)
|
| 264 |
|
|
return count;
|
| 265 |
|
|
|
| 266 |
|
|
pgdir = pgd_offset(mm, address);
|
| 267 |
|
|
|
| 268 |
|
|
end = vma->vm_end;
|
| 269 |
|
|
BUG_ON(address >= end);
|
| 270 |
|
|
do {
|
| 271 |
|
|
count = swap_out_pgd(mm, vma, pgdir, address, end, count, classzone);
|
| 272 |
|
|
if (!count)
|
| 273 |
|
|
break;
|
| 274 |
|
|
address = (address + PGDIR_SIZE) & PGDIR_MASK;
|
| 275 |
|
|
pgdir++;
|
| 276 |
|
|
} while (address && (address < end));
|
| 277 |
|
|
return count;
|
| 278 |
|
|
}
|
| 279 |
|
|
|
| 280 |
|
|
/* Placeholder for swap_out(): may be updated by fork.c:mmput() */
|
| 281 |
|
|
struct mm_struct *swap_mm = &init_mm;
|
| 282 |
|
|
|
| 283 |
|
|
/*
|
| 284 |
|
|
* Returns remaining count of pages to be swapped out by followup call.
|
| 285 |
|
|
*/
|
| 286 |
|
|
static inline int swap_out_mm(struct mm_struct * mm, int count, int * mmcounter, zone_t * classzone)
|
| 287 |
|
|
{
|
| 288 |
|
|
unsigned long address;
|
| 289 |
|
|
struct vm_area_struct* vma;
|
| 290 |
|
|
|
| 291 |
|
|
/*
|
| 292 |
|
|
* Find the proper vm-area after freezing the vma chain
|
| 293 |
|
|
* and ptes.
|
| 294 |
|
|
*/
|
| 295 |
|
|
spin_lock(&mm->page_table_lock);
|
| 296 |
|
|
address = mm->swap_address;
|
| 297 |
|
|
if (address == TASK_SIZE || swap_mm != mm) {
|
| 298 |
|
|
/* We raced: don't count this mm but try again */
|
| 299 |
|
|
++*mmcounter;
|
| 300 |
|
|
goto out_unlock;
|
| 301 |
|
|
}
|
| 302 |
|
|
vma = find_vma(mm, address);
|
| 303 |
|
|
if (vma) {
|
| 304 |
|
|
if (address < vma->vm_start)
|
| 305 |
|
|
address = vma->vm_start;
|
| 306 |
|
|
|
| 307 |
|
|
for (;;) {
|
| 308 |
|
|
count = swap_out_vma(mm, vma, address, count, classzone);
|
| 309 |
|
|
vma = vma->vm_next;
|
| 310 |
|
|
if (!vma)
|
| 311 |
|
|
break;
|
| 312 |
|
|
if (!count)
|
| 313 |
|
|
goto out_unlock;
|
| 314 |
|
|
address = vma->vm_start;
|
| 315 |
|
|
}
|
| 316 |
|
|
}
|
| 317 |
|
|
/* Indicate that we reached the end of address space */
|
| 318 |
|
|
mm->swap_address = TASK_SIZE;
|
| 319 |
|
|
|
| 320 |
|
|
out_unlock:
|
| 321 |
|
|
spin_unlock(&mm->page_table_lock);
|
| 322 |
|
|
return count;
|
| 323 |
|
|
}
|
| 324 |
|
|
|
| 325 |
|
|
static int FASTCALL(swap_out(zone_t * classzone));
|
| 326 |
|
|
static int swap_out(zone_t * classzone)
|
| 327 |
|
|
{
|
| 328 |
|
|
int counter, nr_pages = SWAP_CLUSTER_MAX;
|
| 329 |
|
|
struct mm_struct *mm;
|
| 330 |
|
|
|
| 331 |
|
|
counter = mmlist_nr << 1;
|
| 332 |
|
|
do {
|
| 333 |
|
|
if (unlikely(current->need_resched)) {
|
| 334 |
|
|
__set_current_state(TASK_RUNNING);
|
| 335 |
|
|
schedule();
|
| 336 |
|
|
}
|
| 337 |
|
|
|
| 338 |
|
|
spin_lock(&mmlist_lock);
|
| 339 |
|
|
mm = swap_mm;
|
| 340 |
|
|
while (mm->swap_address == TASK_SIZE || mm == &init_mm) {
|
| 341 |
|
|
mm->swap_address = 0;
|
| 342 |
|
|
mm = list_entry(mm->mmlist.next, struct mm_struct, mmlist);
|
| 343 |
|
|
if (mm == swap_mm)
|
| 344 |
|
|
goto empty;
|
| 345 |
|
|
swap_mm = mm;
|
| 346 |
|
|
}
|
| 347 |
|
|
|
| 348 |
|
|
/* Make sure the mm doesn't disappear when we drop the lock.. */
|
| 349 |
|
|
atomic_inc(&mm->mm_users);
|
| 350 |
|
|
spin_unlock(&mmlist_lock);
|
| 351 |
|
|
|
| 352 |
|
|
nr_pages = swap_out_mm(mm, nr_pages, &counter, classzone);
|
| 353 |
|
|
|
| 354 |
|
|
mmput(mm);
|
| 355 |
|
|
|
| 356 |
|
|
if (!nr_pages)
|
| 357 |
|
|
return 1;
|
| 358 |
|
|
} while (--counter >= 0);
|
| 359 |
|
|
|
| 360 |
|
|
return 0;
|
| 361 |
|
|
|
| 362 |
|
|
empty:
|
| 363 |
|
|
spin_unlock(&mmlist_lock);
|
| 364 |
|
|
return 0;
|
| 365 |
|
|
}
|
| 366 |
|
|
|
| 367 |
|
|
static void FASTCALL(refill_inactive(int nr_pages, zone_t * classzone));
|
| 368 |
|
|
static int FASTCALL(shrink_cache(int nr_pages, zone_t * classzone, unsigned int gfp_mask, int * failed_swapout));
|
| 369 |
|
|
static int shrink_cache(int nr_pages, zone_t * classzone, unsigned int gfp_mask, int * failed_swapout)
|
| 370 |
|
|
{
|
| 371 |
|
|
struct list_head * entry;
|
| 372 |
|
|
int max_scan = (classzone->nr_inactive_pages + classzone->nr_active_pages) / vm_cache_scan_ratio;
|
| 373 |
|
|
int max_mapped = vm_mapped_ratio * nr_pages;
|
| 374 |
|
|
|
| 375 |
|
|
while (max_scan && classzone->nr_inactive_pages && (entry = inactive_list.prev) != &inactive_list) {
|
| 376 |
|
|
struct page * page;
|
| 377 |
|
|
|
| 378 |
|
|
if (unlikely(current->need_resched)) {
|
| 379 |
|
|
spin_unlock(&pagemap_lru_lock);
|
| 380 |
|
|
__set_current_state(TASK_RUNNING);
|
| 381 |
|
|
schedule();
|
| 382 |
|
|
spin_lock(&pagemap_lru_lock);
|
| 383 |
|
|
continue;
|
| 384 |
|
|
}
|
| 385 |
|
|
|
| 386 |
|
|
page = list_entry(entry, struct page, lru);
|
| 387 |
|
|
|
| 388 |
|
|
BUG_ON(!PageLRU(page));
|
| 389 |
|
|
BUG_ON(PageActive(page));
|
| 390 |
|
|
|
| 391 |
|
|
list_del(entry);
|
| 392 |
|
|
list_add(entry, &inactive_list);
|
| 393 |
|
|
|
| 394 |
|
|
/*
|
| 395 |
|
|
* Zero page counts can happen because we unlink the pages
|
| 396 |
|
|
* _after_ decrementing the usage count..
|
| 397 |
|
|
*/
|
| 398 |
|
|
if (unlikely(!page_count(page)))
|
| 399 |
|
|
continue;
|
| 400 |
|
|
|
| 401 |
|
|
if (!memclass(page_zone(page), classzone))
|
| 402 |
|
|
continue;
|
| 403 |
|
|
|
| 404 |
|
|
max_scan--;
|
| 405 |
|
|
|
| 406 |
|
|
/* Racy check to avoid trylocking when not worthwhile */
|
| 407 |
|
|
if (!page->buffers && (page_count(page) != 1 || !page->mapping))
|
| 408 |
|
|
goto page_mapped;
|
| 409 |
|
|
|
| 410 |
|
|
/*
|
| 411 |
|
|
* The page is locked. IO in progress?
|
| 412 |
|
|
* Move it to the back of the list.
|
| 413 |
|
|
*/
|
| 414 |
|
|
if (unlikely(TryLockPage(page))) {
|
| 415 |
|
|
if (PageLaunder(page) && (gfp_mask & __GFP_FS)) {
|
| 416 |
|
|
page_cache_get(page);
|
| 417 |
|
|
spin_unlock(&pagemap_lru_lock);
|
| 418 |
|
|
wait_on_page(page);
|
| 419 |
|
|
page_cache_release(page);
|
| 420 |
|
|
spin_lock(&pagemap_lru_lock);
|
| 421 |
|
|
}
|
| 422 |
|
|
continue;
|
| 423 |
|
|
}
|
| 424 |
|
|
|
| 425 |
|
|
if (PageDirty(page) && is_page_cache_freeable(page) && page->mapping) {
|
| 426 |
|
|
/*
|
| 427 |
|
|
* It is not critical here to write it only if
|
| 428 |
|
|
* the page is unmapped beause any direct writer
|
| 429 |
|
|
* like O_DIRECT would set the PG_dirty bitflag
|
| 430 |
|
|
* on the phisical page after having successfully
|
| 431 |
|
|
* pinned it and after the I/O to the page is finished,
|
| 432 |
|
|
* so the direct writes to the page cannot get lost.
|
| 433 |
|
|
*/
|
| 434 |
|
|
int (*writepage)(struct page *);
|
| 435 |
|
|
|
| 436 |
|
|
writepage = page->mapping->a_ops->writepage;
|
| 437 |
|
|
if ((gfp_mask & __GFP_FS) && writepage) {
|
| 438 |
|
|
ClearPageDirty(page);
|
| 439 |
|
|
SetPageLaunder(page);
|
| 440 |
|
|
page_cache_get(page);
|
| 441 |
|
|
spin_unlock(&pagemap_lru_lock);
|
| 442 |
|
|
|
| 443 |
|
|
writepage(page);
|
| 444 |
|
|
page_cache_release(page);
|
| 445 |
|
|
|
| 446 |
|
|
spin_lock(&pagemap_lru_lock);
|
| 447 |
|
|
continue;
|
| 448 |
|
|
}
|
| 449 |
|
|
}
|
| 450 |
|
|
|
| 451 |
|
|
/*
|
| 452 |
|
|
* If the page has buffers, try to free the buffer mappings
|
| 453 |
|
|
* associated with this page. If we succeed we try to free
|
| 454 |
|
|
* the page as well.
|
| 455 |
|
|
*/
|
| 456 |
|
|
if (page->buffers) {
|
| 457 |
|
|
spin_unlock(&pagemap_lru_lock);
|
| 458 |
|
|
|
| 459 |
|
|
/* avoid to free a locked page */
|
| 460 |
|
|
page_cache_get(page);
|
| 461 |
|
|
|
| 462 |
|
|
if (try_to_release_page(page, gfp_mask)) {
|
| 463 |
|
|
if (!page->mapping) {
|
| 464 |
|
|
/*
|
| 465 |
|
|
* We must not allow an anon page
|
| 466 |
|
|
* with no buffers to be visible on
|
| 467 |
|
|
* the LRU, so we unlock the page after
|
| 468 |
|
|
* taking the lru lock
|
| 469 |
|
|
*/
|
| 470 |
|
|
spin_lock(&pagemap_lru_lock);
|
| 471 |
|
|
UnlockPage(page);
|
| 472 |
|
|
__lru_cache_del(page);
|
| 473 |
|
|
|
| 474 |
|
|
/* effectively free the page here */
|
| 475 |
|
|
page_cache_release(page);
|
| 476 |
|
|
|
| 477 |
|
|
if (--nr_pages)
|
| 478 |
|
|
continue;
|
| 479 |
|
|
break;
|
| 480 |
|
|
} else {
|
| 481 |
|
|
/*
|
| 482 |
|
|
* The page is still in pagecache so undo the stuff
|
| 483 |
|
|
* before the try_to_release_page since we've not
|
| 484 |
|
|
* finished and we can now try the next step.
|
| 485 |
|
|
*/
|
| 486 |
|
|
page_cache_release(page);
|
| 487 |
|
|
|
| 488 |
|
|
spin_lock(&pagemap_lru_lock);
|
| 489 |
|
|
}
|
| 490 |
|
|
} else {
|
| 491 |
|
|
/* failed to drop the buffers so stop here */
|
| 492 |
|
|
UnlockPage(page);
|
| 493 |
|
|
page_cache_release(page);
|
| 494 |
|
|
|
| 495 |
|
|
spin_lock(&pagemap_lru_lock);
|
| 496 |
|
|
continue;
|
| 497 |
|
|
}
|
| 498 |
|
|
}
|
| 499 |
|
|
|
| 500 |
|
|
spin_lock(&pagecache_lock);
|
| 501 |
|
|
|
| 502 |
|
|
/*
|
| 503 |
|
|
* This is the non-racy check for busy page.
|
| 504 |
|
|
* It is critical to check PageDirty _after_ we made sure
|
| 505 |
|
|
* the page is freeable so not in use by anybody.
|
| 506 |
|
|
* At this point we're guaranteed that page->buffers is NULL,
|
| 507 |
|
|
* nobody can refill page->buffers under us because we still
|
| 508 |
|
|
* hold the page lock.
|
| 509 |
|
|
*/
|
| 510 |
|
|
if (!page->mapping || page_count(page) > 1) {
|
| 511 |
|
|
spin_unlock(&pagecache_lock);
|
| 512 |
|
|
UnlockPage(page);
|
| 513 |
|
|
page_mapped:
|
| 514 |
|
|
if (--max_mapped < 0) {
|
| 515 |
|
|
spin_unlock(&pagemap_lru_lock);
|
| 516 |
|
|
|
| 517 |
|
|
nr_pages -= kmem_cache_reap(gfp_mask);
|
| 518 |
|
|
if (nr_pages <= 0)
|
| 519 |
|
|
goto out;
|
| 520 |
|
|
|
| 521 |
|
|
shrink_dcache_memory(vm_vfs_scan_ratio, gfp_mask);
|
| 522 |
|
|
shrink_icache_memory(vm_vfs_scan_ratio, gfp_mask);
|
| 523 |
|
|
#ifdef CONFIG_QUOTA
|
| 524 |
|
|
shrink_dqcache_memory(vm_vfs_scan_ratio, gfp_mask);
|
| 525 |
|
|
#endif
|
| 526 |
|
|
|
| 527 |
|
|
if (!*failed_swapout)
|
| 528 |
|
|
*failed_swapout = !swap_out(classzone);
|
| 529 |
|
|
|
| 530 |
|
|
max_mapped = nr_pages * vm_mapped_ratio;
|
| 531 |
|
|
|
| 532 |
|
|
spin_lock(&pagemap_lru_lock);
|
| 533 |
|
|
refill_inactive(nr_pages, classzone);
|
| 534 |
|
|
}
|
| 535 |
|
|
continue;
|
| 536 |
|
|
|
| 537 |
|
|
}
|
| 538 |
|
|
if (PageDirty(page)) {
|
| 539 |
|
|
spin_unlock(&pagecache_lock);
|
| 540 |
|
|
UnlockPage(page);
|
| 541 |
|
|
continue;
|
| 542 |
|
|
}
|
| 543 |
|
|
|
| 544 |
|
|
__lru_cache_del(page);
|
| 545 |
|
|
|
| 546 |
|
|
/* point of no return */
|
| 547 |
|
|
if (likely(!PageSwapCache(page))) {
|
| 548 |
|
|
__remove_inode_page(page);
|
| 549 |
|
|
spin_unlock(&pagecache_lock);
|
| 550 |
|
|
} else {
|
| 551 |
|
|
swp_entry_t swap;
|
| 552 |
|
|
swap.val = page->index;
|
| 553 |
|
|
__delete_from_swap_cache(page);
|
| 554 |
|
|
spin_unlock(&pagecache_lock);
|
| 555 |
|
|
swap_free(swap);
|
| 556 |
|
|
}
|
| 557 |
|
|
|
| 558 |
|
|
UnlockPage(page);
|
| 559 |
|
|
|
| 560 |
|
|
/* effectively free the page here */
|
| 561 |
|
|
page_cache_release(page);
|
| 562 |
|
|
|
| 563 |
|
|
if (--nr_pages)
|
| 564 |
|
|
continue;
|
| 565 |
|
|
break;
|
| 566 |
|
|
}
|
| 567 |
|
|
spin_unlock(&pagemap_lru_lock);
|
| 568 |
|
|
|
| 569 |
|
|
out:
|
| 570 |
|
|
return nr_pages;
|
| 571 |
|
|
}
|
| 572 |
|
|
|
| 573 |
|
|
/*
|
| 574 |
|
|
* This moves pages from the active list to
|
| 575 |
|
|
* the inactive list.
|
| 576 |
|
|
*
|
| 577 |
|
|
* We move them the other way when we see the
|
| 578 |
|
|
* reference bit on the page.
|
| 579 |
|
|
*/
|
| 580 |
|
|
static void refill_inactive(int nr_pages, zone_t * classzone)
|
| 581 |
|
|
{
|
| 582 |
|
|
struct list_head * entry;
|
| 583 |
|
|
unsigned long ratio;
|
| 584 |
|
|
|
| 585 |
|
|
ratio = (unsigned long) nr_pages * classzone->nr_active_pages / (((unsigned long) classzone->nr_inactive_pages * vm_lru_balance_ratio) + 1);
|
| 586 |
|
|
|
| 587 |
|
|
entry = active_list.prev;
|
| 588 |
|
|
while (ratio && entry != &active_list) {
|
| 589 |
|
|
struct page * page;
|
| 590 |
|
|
|
| 591 |
|
|
page = list_entry(entry, struct page, lru);
|
| 592 |
|
|
entry = entry->prev;
|
| 593 |
|
|
if (PageTestandClearReferenced(page)) {
|
| 594 |
|
|
list_del(&page->lru);
|
| 595 |
|
|
list_add(&page->lru, &active_list);
|
| 596 |
|
|
continue;
|
| 597 |
|
|
}
|
| 598 |
|
|
|
| 599 |
|
|
ratio--;
|
| 600 |
|
|
|
| 601 |
|
|
del_page_from_active_list(page);
|
| 602 |
|
|
add_page_to_inactive_list(page);
|
| 603 |
|
|
SetPageReferenced(page);
|
| 604 |
|
|
}
|
| 605 |
|
|
|
| 606 |
|
|
if (entry != &active_list) {
|
| 607 |
|
|
list_del(&active_list);
|
| 608 |
|
|
list_add(&active_list, entry);
|
| 609 |
|
|
}
|
| 610 |
|
|
}
|
| 611 |
|
|
|
| 612 |
|
|
static int FASTCALL(shrink_caches(zone_t * classzone, unsigned int gfp_mask, int nr_pages, int * failed_swapout));
|
| 613 |
|
|
static int shrink_caches(zone_t * classzone, unsigned int gfp_mask, int nr_pages, int * failed_swapout)
|
| 614 |
|
|
{
|
| 615 |
|
|
nr_pages -= kmem_cache_reap(gfp_mask);
|
| 616 |
|
|
if (nr_pages <= 0)
|
| 617 |
|
|
goto out;
|
| 618 |
|
|
|
| 619 |
|
|
spin_lock(&pagemap_lru_lock);
|
| 620 |
|
|
refill_inactive(nr_pages, classzone);
|
| 621 |
|
|
|
| 622 |
|
|
nr_pages = shrink_cache(nr_pages, classzone, gfp_mask, failed_swapout);
|
| 623 |
|
|
|
| 624 |
|
|
out:
|
| 625 |
|
|
return nr_pages;
|
| 626 |
|
|
}
|
| 627 |
|
|
|
| 628 |
|
|
static int check_classzone_need_balance(zone_t * classzone);
|
| 629 |
|
|
|
| 630 |
|
|
int try_to_free_pages_zone(zone_t *classzone, unsigned int gfp_mask)
|
| 631 |
|
|
{
|
| 632 |
|
|
gfp_mask = pf_gfp_mask(gfp_mask);
|
| 633 |
|
|
|
| 634 |
|
|
for (;;) {
|
| 635 |
|
|
int tries = vm_passes;
|
| 636 |
|
|
int failed_swapout = !(gfp_mask & __GFP_IO);
|
| 637 |
|
|
int nr_pages = SWAP_CLUSTER_MAX;
|
| 638 |
|
|
|
| 639 |
|
|
do {
|
| 640 |
|
|
nr_pages = shrink_caches(classzone, gfp_mask, nr_pages, &failed_swapout);
|
| 641 |
|
|
if (nr_pages <= 0)
|
| 642 |
|
|
return 1;
|
| 643 |
|
|
shrink_dcache_memory(vm_vfs_scan_ratio, gfp_mask);
|
| 644 |
|
|
shrink_icache_memory(vm_vfs_scan_ratio, gfp_mask);
|
| 645 |
|
|
#ifdef CONFIG_QUOTA
|
| 646 |
|
|
shrink_dqcache_memory(vm_vfs_scan_ratio, gfp_mask);
|
| 647 |
|
|
#endif
|
| 648 |
|
|
if (!failed_swapout)
|
| 649 |
|
|
failed_swapout = !swap_out(classzone);
|
| 650 |
|
|
} while (--tries);
|
| 651 |
|
|
|
| 652 |
|
|
#ifdef CONFIG_OOM_KILLER
|
| 653 |
|
|
out_of_memory();
|
| 654 |
|
|
#else
|
| 655 |
|
|
if (likely(current->pid != 1))
|
| 656 |
|
|
break;
|
| 657 |
|
|
if (!check_classzone_need_balance(classzone))
|
| 658 |
|
|
break;
|
| 659 |
|
|
|
| 660 |
|
|
__set_current_state(TASK_RUNNING);
|
| 661 |
|
|
yield();
|
| 662 |
|
|
#endif
|
| 663 |
|
|
}
|
| 664 |
|
|
|
| 665 |
|
|
return 0;
|
| 666 |
|
|
}
|
| 667 |
|
|
|
| 668 |
|
|
int try_to_free_pages(unsigned int gfp_mask)
|
| 669 |
|
|
{
|
| 670 |
|
|
pg_data_t *pgdat;
|
| 671 |
|
|
zonelist_t *zonelist;
|
| 672 |
|
|
unsigned long pf_free_pages;
|
| 673 |
|
|
int error = 0;
|
| 674 |
|
|
|
| 675 |
|
|
pf_free_pages = current->flags & PF_FREE_PAGES;
|
| 676 |
|
|
current->flags &= ~PF_FREE_PAGES;
|
| 677 |
|
|
|
| 678 |
|
|
for_each_pgdat(pgdat) {
|
| 679 |
|
|
zonelist = pgdat->node_zonelists + (gfp_mask & GFP_ZONEMASK);
|
| 680 |
|
|
error |= try_to_free_pages_zone(zonelist->zones[0], gfp_mask);
|
| 681 |
|
|
}
|
| 682 |
|
|
|
| 683 |
|
|
current->flags |= pf_free_pages;
|
| 684 |
|
|
return error;
|
| 685 |
|
|
}
|
| 686 |
|
|
|
| 687 |
|
|
DECLARE_WAIT_QUEUE_HEAD(kswapd_wait);
|
| 688 |
|
|
|
| 689 |
|
|
static int check_classzone_need_balance(zone_t * classzone)
|
| 690 |
|
|
{
|
| 691 |
|
|
zone_t * first_zone;
|
| 692 |
|
|
int class_idx = zone_idx(classzone);
|
| 693 |
|
|
|
| 694 |
|
|
first_zone = classzone->zone_pgdat->node_zones;
|
| 695 |
|
|
while (classzone >= first_zone) {
|
| 696 |
|
|
if (classzone->free_pages > classzone->watermarks[class_idx].high)
|
| 697 |
|
|
return 0;
|
| 698 |
|
|
classzone--;
|
| 699 |
|
|
}
|
| 700 |
|
|
return 1;
|
| 701 |
|
|
}
|
| 702 |
|
|
|
| 703 |
|
|
static int kswapd_balance_pgdat(pg_data_t * pgdat)
|
| 704 |
|
|
{
|
| 705 |
|
|
int need_more_balance = 0, i;
|
| 706 |
|
|
zone_t * zone;
|
| 707 |
|
|
|
| 708 |
|
|
for (i = pgdat->nr_zones-1; i >= 0; i--) {
|
| 709 |
|
|
zone = pgdat->node_zones + i;
|
| 710 |
|
|
if (unlikely(current->need_resched))
|
| 711 |
|
|
schedule();
|
| 712 |
|
|
if (!zone->need_balance || !zone->size)
|
| 713 |
|
|
continue;
|
| 714 |
|
|
if (!try_to_free_pages_zone(zone, GFP_KSWAPD)) {
|
| 715 |
|
|
zone->need_balance = 0;
|
| 716 |
|
|
__set_current_state(TASK_INTERRUPTIBLE);
|
| 717 |
|
|
schedule_timeout(HZ*5);
|
| 718 |
|
|
continue;
|
| 719 |
|
|
}
|
| 720 |
|
|
if (check_classzone_need_balance(zone))
|
| 721 |
|
|
need_more_balance = 1;
|
| 722 |
|
|
else
|
| 723 |
|
|
zone->need_balance = 0;
|
| 724 |
|
|
}
|
| 725 |
|
|
|
| 726 |
|
|
return need_more_balance;
|
| 727 |
|
|
}
|
| 728 |
|
|
|
| 729 |
|
|
static void kswapd_balance(void)
|
| 730 |
|
|
{
|
| 731 |
|
|
int need_more_balance;
|
| 732 |
|
|
pg_data_t * pgdat;
|
| 733 |
|
|
|
| 734 |
|
|
do {
|
| 735 |
|
|
need_more_balance = 0;
|
| 736 |
|
|
|
| 737 |
|
|
for_each_pgdat(pgdat)
|
| 738 |
|
|
need_more_balance |= kswapd_balance_pgdat(pgdat);
|
| 739 |
|
|
} while (need_more_balance);
|
| 740 |
|
|
}
|
| 741 |
|
|
|
| 742 |
|
|
static int kswapd_can_sleep_pgdat(pg_data_t * pgdat)
|
| 743 |
|
|
{
|
| 744 |
|
|
zone_t * zone;
|
| 745 |
|
|
int i;
|
| 746 |
|
|
|
| 747 |
|
|
for (i = pgdat->nr_zones-1; i >= 0; i--) {
|
| 748 |
|
|
zone = pgdat->node_zones + i;
|
| 749 |
|
|
if (!zone->need_balance || !zone->size)
|
| 750 |
|
|
continue;
|
| 751 |
|
|
return 0;
|
| 752 |
|
|
}
|
| 753 |
|
|
|
| 754 |
|
|
return 1;
|
| 755 |
|
|
}
|
| 756 |
|
|
|
| 757 |
|
|
static int kswapd_can_sleep(void)
|
| 758 |
|
|
{
|
| 759 |
|
|
pg_data_t * pgdat;
|
| 760 |
|
|
|
| 761 |
|
|
for_each_pgdat(pgdat) {
|
| 762 |
|
|
if (!kswapd_can_sleep_pgdat(pgdat))
|
| 763 |
|
|
return 0;
|
| 764 |
|
|
}
|
| 765 |
|
|
|
| 766 |
|
|
return 1;
|
| 767 |
|
|
}
|
| 768 |
|
|
|
| 769 |
|
|
/*
|
| 770 |
|
|
* The background pageout daemon, started as a kernel thread
|
| 771 |
|
|
* from the init process.
|
| 772 |
|
|
*
|
| 773 |
|
|
* This basically trickles out pages so that we have _some_
|
| 774 |
|
|
* free memory available even if there is no other activity
|
| 775 |
|
|
* that frees anything up. This is needed for things like routing
|
| 776 |
|
|
* etc, where we otherwise might have all activity going on in
|
| 777 |
|
|
* asynchronous contexts that cannot page things out.
|
| 778 |
|
|
*
|
| 779 |
|
|
* If there are applications that are active memory-allocators
|
| 780 |
|
|
* (most normal use), this basically shouldn't matter.
|
| 781 |
|
|
*/
|
| 782 |
|
|
int kswapd(void *unused)
|
| 783 |
|
|
{
|
| 784 |
|
|
struct task_struct *tsk = current;
|
| 785 |
|
|
DECLARE_WAITQUEUE(wait, tsk);
|
| 786 |
|
|
|
| 787 |
|
|
daemonize();
|
| 788 |
|
|
strcpy(tsk->comm, "kswapd");
|
| 789 |
|
|
sigfillset(&tsk->blocked);
|
| 790 |
|
|
|
| 791 |
|
|
/*
|
| 792 |
|
|
* Tell the memory management that we're a "memory allocator",
|
| 793 |
|
|
* and that if we need more memory we should get access to it
|
| 794 |
|
|
* regardless (see "__alloc_pages()"). "kswapd" should
|
| 795 |
|
|
* never get caught in the normal page freeing logic.
|
| 796 |
|
|
*
|
| 797 |
|
|
* (Kswapd normally doesn't need memory anyway, but sometimes
|
| 798 |
|
|
* you need a small amount of memory in order to be able to
|
| 799 |
|
|
* page out something else, and this flag essentially protects
|
| 800 |
|
|
* us from recursively trying to free more memory as we're
|
| 801 |
|
|
* trying to free the first piece of memory in the first place).
|
| 802 |
|
|
*/
|
| 803 |
|
|
tsk->flags |= PF_MEMALLOC;
|
| 804 |
|
|
|
| 805 |
|
|
/*
|
| 806 |
|
|
* Kswapd main loop.
|
| 807 |
|
|
*/
|
| 808 |
|
|
for (;;) {
|
| 809 |
|
|
__set_current_state(TASK_INTERRUPTIBLE);
|
| 810 |
|
|
add_wait_queue(&kswapd_wait, &wait);
|
| 811 |
|
|
|
| 812 |
|
|
mb();
|
| 813 |
|
|
if (kswapd_can_sleep())
|
| 814 |
|
|
schedule();
|
| 815 |
|
|
|
| 816 |
|
|
__set_current_state(TASK_RUNNING);
|
| 817 |
|
|
remove_wait_queue(&kswapd_wait, &wait);
|
| 818 |
|
|
|
| 819 |
|
|
/*
|
| 820 |
|
|
* If we actually get into a low-memory situation,
|
| 821 |
|
|
* the processes needing more memory will wake us
|
| 822 |
|
|
* up on a more timely basis.
|
| 823 |
|
|
*/
|
| 824 |
|
|
kswapd_balance();
|
| 825 |
|
|
run_task_queue(&tq_disk);
|
| 826 |
|
|
}
|
| 827 |
|
|
}
|
| 828 |
|
|
|
| 829 |
|
|
static int __init kswapd_init(void)
|
| 830 |
|
|
{
|
| 831 |
|
|
printk("Starting kswapd\n");
|
| 832 |
|
|
swap_setup();
|
| 833 |
|
|
kernel_thread(kswapd, NULL, CLONE_FS | CLONE_FILES | CLONE_SIGNAL);
|
| 834 |
|
|
return 0;
|
| 835 |
|
|
}
|
| 836 |
|
|
|
| 837 |
|
|
module_init(kswapd_init)
|
