- 22 Mar, 2006 40 commits
-
-
Chen, Kenneth W authored
follow_hugetlb_page() walks a range of user virtual address and then fills in list of struct page * into an array that is passed from the argument list. It also gets a reference count via get_page(). For compound page, get_page() actually traverse back to head page via page_private() macro and then adds a reference count to the head page. Since we are doing a virt to pte look up, kernel already has a struct page pointer into the head page. So instead of traverse into the small unit page struct and then follow a link back to the head page, optimize that with incrementing the reference count directly on the head page. The benefit is that we don't take a cache miss on accessing page struct for the corresponding user address and more importantly, not to pollute the cache with a "not very useful" round trip of pointer chasing. This adds a moderate performance gain on an I/O intensive database transaction workload. Signed-off-by:
Ken Chen <kenneth.w.chen@intel.com> Cc: David Gibson <david@gibson.dropbear.id.au> Signed-off-by:
Andrew Morton <akpm@osdl.org> Signed-off-by:
Linus Torvalds <torvalds@osdl.org>
-
Chen, Kenneth W authored
Implementation of hugetlbfs_counter() is functionally equivalent to atomic_inc_return(). Use the simpler atomic form. Signed-off-by:
-
David Gibson authored
Quite a long time back, prepare_hugepage_range() replaced is_aligned_hugepage_range() as the callback from mm/mmap.c to arch code to verify if an address range is suitable for a hugepage mapping. is_aligned_hugepage_range() stuck around, but only to implement prepare_hugepage_range() on archs which didn't implement their own. Most archs (everything except ia64 and powerpc) used the same implementation of is_aligned_hugepage_range(). On powerpc, which implements its own prepare_hugepage_range(), the custom version was never used. In addition, "is_aligned_hugepage_range()" was a bad name, because it suggests it returns true iff the given range is a good hugepage range, whereas in fact it returns 0-or-error (so the sense is reversed). This patch cleans up by abolishing is_aligned_hugepage_range(). Instead prepare_hugepage_range() is defined directly. Most archs use the default version, which simply checks the given region is aligned to the size of a hugepage. ia64 and powerpc define custom versions. The ia64 one simply checks that the range is in the correct address space region in addition to being suitably aligned. The powerpc version (just as previously) checks for suitable addresses, and if necessary performs low-level MMU frobbing to set up new areas for use by hugepages. No libhugetlbfs testsuite regressions on ppc64 (POWER5 LPAR). Signed-off-by:
David Gibson <david@gibson.dropbear.id.au> Signed-off-by:
Zhang Yanmin <yanmin.zhang@intel.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@samba.org> Cc: William Lee Irwin III <wli@holomorphy.com> Signed-off-by:
-
David Gibson authored
The optional hugepage callback, hugetlb_free_pgd_range() is presently implemented non-trivially only on ia64 (but I plan to add one for powerpc shortly). It has its own prototype for the function in asm-ia64/pgtable.h. However, since the function is called from generic code, it make sense for its prototype to be in the generic hugetlb.h header file, as the protypes other arch callbacks already are (prepare_hugepage_range(), set_huge_pte_at(), etc.). This patch makes it so. Signed-off-by:
David Gibson <dwg@au1.ibm.com> Cc: William Lee Irwin III <wli@holomorphy.com> Signed-off-by:
-
David Gibson authored
Turns out the hugepage logic in free_pgtables() was doubly broken. The loop coalescing multiple normal page VMAs into one call to free_pgd_range() had an off by one error, which could mean it would coalesce one hugepage VMA into the same bundle (checking 'vma' not 'next' in the loop). I transferred this bug into the new is_vm_hugetlb_page() based version. Here's the fix. This one didn't bite on powerpc previously for the same reason the is_hugepage_only_range() problem didn't: powerpc's hugetlb_free_pgd_range() is identical to free_pgd_range(). It didn't bite on ia64 because the hugepage region is distant enough from any other region that the separated PMD_SIZE distance test would always prevent coalescing the two together. No libhugetlbfs testsuite regressions (ppc64, POWER5). Signed-off-by:
-
David Gibson authored
free_pgtables() has special logic to call hugetlb_free_pgd_range() instead of the normal free_pgd_range() on hugepage VMAs. However, the test it uses to do so is incorrect: it calls is_hugepage_only_range on a hugepage sized range at the start of the vma. is_hugepage_only_range() will return true if the given range has any intersection with a hugepage address region, and in this case the given region need not be hugepage aligned. So, for example, this test can return true if called on, say, a 4k VMA immediately preceding a (nicely aligned) hugepage VMA. At present we get away with this because the powerpc version of hugetlb_free_pgd_range() is just a call to free_pgd_range(). On ia64 (the only other arch with a non-trivial is_hugepage_only_range()) we get away with it for a different reason; the hugepage area is not contiguous with the rest of the user address space, and VMAs are not permitted in between, so the test can't return a false positive there. Nonetheless this should be fixed. We do that in the patch below by replacing the is_hugepage_only_range() test with an explicit test of the VMA using is_vm_hugetlb_page(). This in turn changes behaviour for platforms where is_hugepage_only_range() returns false always (everything except powerpc and ia64). We address this by ensuring that hugetlb_free_pgd_range() is defined to be identical to free_pgd_range() (instead of a no-op) on everything except ia64. Even so, it will prevent some otherwise possible coalescing of calls down to free_pgd_range(). Since this only happens for hugepage VMAs, removing this small optimization seems unlikely to cause any trouble. This patch causes no regressions on the libhugetlbfs testsuite - ppc64 POWER5 (8-way), ppc64 G5 (2-way) and i386 Pentium M (UP). Signed-off-by:
Hugh Dickins <hugh@veritas.com> Signed-off-by:
-
David Gibson authored
Originally, mm/hugetlb.c just handled the hugepage physical allocation path and its {alloc,free}_huge_page() functions were used from the arch specific hugepage code. These days those functions are only used with mm/hugetlb.c itself. Therefore, this patch makes them static and removes their prototypes from hugetlb.h. This requires a small rearrangement of code in mm/hugetlb.c to avoid a forward declaration. This patch causes no regressions on the libhugetlbfs testsuite (ppc64, POWER5). Signed-off-by: -
David Gibson authored
These days, hugepages are demand-allocated at first fault time. There's a somewhat dubious (and racy) heuristic when making a new mmap() to check if there are enough available hugepages to fully satisfy that mapping. A particularly obvious case where the heuristic breaks down is where a process maps its hugepages not as a single chunk, but as a bunch of individually mmap()ed (or shmat()ed) blocks without touching and instantiating the pages in between allocations. In this case the size of each block is compared against the total number of available hugepages. It's thus easy for the process to become overcommitted, because each block mapping will succeed, although the total number of hugepages required by all blocks exceeds the number available. In particular, this defeats such a program which will detect a mapping failure and adjust its hugepage usage downward accordingly. The patch below addresses this problem, by strictly reserving a number of physical hugepages for hugepage inodes which have been mapped, but not instatiated. MAP_SHARED mappings are thus "safe" - they will fail on mmap(), not later with an OOM SIGKILL. MAP_PRIVATE mappings can still trigger an OOM. (Actually SHARED mappings can technically still OOM, but only if the sysadmin explicitly reduces the hugepage pool between mapping and instantiation) This patch appears to address the problem at hand - it allows DB2 to start correctly, for instance, which previously suffered the failure described above. This patch causes no regressions on the libhugetblfs testsuite, and makes a test (designed to catch this problem) pass which previously failed (ppc64, POWER5). Signed-off-by:
-
David Gibson authored
Currently, no lock or mutex is held between allocating a hugepage and inserting it into the pagetables / page cache. When we do go to insert the page into pagetables or page cache, we recheck and may free the newly allocated hugepage. However, since the number of hugepages in the system is strictly limited, and it's usualy to want to use all of them, this can still lead to spurious allocation failures. For example, suppose two processes are both mapping (MAP_SHARED) the same hugepage file, large enough to consume the entire available hugepage pool. If they race instantiating the last page in the mapping, they will both attempt to allocate the last available hugepage. One will fail, of course, returning OOM from the fault and thus causing the process to be killed, despite the fact that the entire mapping can, in fact, be instantiated. The patch fixes this race by the simple method of adding a (sleeping) mutex to serialize the hugepage fault path between allocation and insertion into pagetables and/or page cache. It would be possible to avoid the serialization by catching the allocation failures, waiting on some condition, then rechecking to see if someone else has instantiated the page for us. Given the likely frequency of hugepage instantiations, it seems very doubtful it's worth the extra complexity. This patch causes no regression on the libhugetlbfs testsuite, and one test, which can trigger this race now passes where it previously failed. Actually, the test still sometimes fails, though less often and only as a shmat() failure, rather processes getting OOM killed by the VM. The dodgy heuristic tests in fs/hugetlbfs/inode.c for whether there's enough hugepage space aren't protected by the new mutex, and would be ugly to do so, so there's still a race there. Another patch to replace those tests with something saner for this reason as well as others coming... Signed-off-by:
-
David Gibson authored
Move the loops used in mm/hugetlb.c to clear and copy hugepages to their own functions for clarity. As we do so, we add some checks of need_resched - we are, after all copying megabytes of memory here. We also add might_sleep() accordingly. We generally dropped locks around the clear and copy, already but not everyone has PREEMPT enabled, so we should still be checking explicitly. For this to work, we need to remove the clear_huge_page() from alloc_huge_page(), which is called with the page_table_lock held in the COW path. We move the clear_huge_page() to just after the alloc_huge_page() in the hugepage no-page path. In the COW path, the new page is about to be copied over, so clearing it was just a waste of time anyway. So as a side effect we also fix the fact that we held the page_table_lock for far too long in this path by calling alloc_huge_page() under it. It causes no regressions on the libhugetlbfs testsuite (ppc64, POWER5). Signed-off-by:
-
Zhang, Yanmin authored
2.6.16-rc3 uses hugetlb on-demand paging, but it doesn_t support hugetlb mprotect. From: David Gibson <david@gibson.dropbear.id.au> Remove a test from the mprotect() path which checks that the mprotect()ed range on a hugepage VMA is hugepage aligned (yes, really, the sense of is_aligned_hugepage_range() is the opposite of what you'd guess :-/). In fact, we don't need this test. If the given addresses match the beginning/end of a hugepage VMA they must already be suitably aligned. If they don't, then mprotect_fixup() will attempt to split the VMA. The very first test in split_vma() will check for a badly aligned address on a hugepage VMA and return -EINVAL if necessary. From: "Chen, Kenneth W" <kenneth.w.chen@intel.com> On i386 and x86-64, pte flag _PAGE_PSE collides with _PAGE_PROTNONE. The identify of hugetlb pte is lost when changing page protection via mprotect. A page fault occurs later will trigger a bug check in huge_pte_alloc(). The fix is to always make new pte a hugetlb pte and also to clean up legacy code where _PAGE_PRESENT is forced on in the pre-faulting day. Signed-off-by:
Nishanth Aravamudan <nacc@us.ibm.com> Cc: Andi Kleen <ak@muc.de> Signed-off-by:
-
Steven Pratt authored
The current current get_init_ra_size is not optimal across different IO sizes and max_readahead values. Here is a quick summary of sizes computed under current design and under the attached patch. All of these assume 1st IO at offset 0, or 1st detected sequential IO. 32k max, 4k request old new ----------------- 8k 8k 16k 16k 32k 32k 128k max, 4k request old new ----------------- 32k 16k 64k 32k 128k 64k 128k 128k 128k max, 32k request old new ----------------- 32k 64k <----- 64k 128k 128k 128k 512k max, 4k request old new ----------------- 4k 32k <---- 16k 64k 64k 128k 128k 256k 512k 512k Cc: Oleg Nesterov <oleg@tv-sign.ru> Cc: Steven Pratt <slpratt@austin.ibm.com> Cc: Ram Pai <linuxram@us.ibm.com> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by:
-
Oleg Nesterov authored
If get_next_ra_size() does not grow fast enough, ->prev_page can overrun the ahead window. This means the caller will read the pages from ->ahead_start + ->ahead_size to ->prev_page synchronously. Signed-off-by:
Oleg Nesterov <oleg@tv-sign.ru> Cc: Steven Pratt <slpratt@austin.ibm.com> Cc: Ram Pai <linuxram@us.ibm.com> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by:
-
Hugh Dickins authored
shmem.c was named and shamed in Jesper's "Building 100 kernels" warnings: shmem_parse_mpol is only used when CONFIG_TMPFS parses mount options; and only called from that one site, so mark it inline like its non-NUMA stub. Signed-off-by:
-
Christoph Lameter authored
As suggested by Marcelo: 1. The optimization introduced recently for not calling page_referenced() during zone reclaim makes two additional checks in shrink_list unnecessary. 2. The if (unlikely(sc->may_swap)) in refill_inactive_zone is optimized for the zone_reclaim case. However, most peoples system only does swap. Undo that. Signed-off-by:
Christoph Lameter <clameter@sgi.com> Cc: Marcelo Tosatti <marcelo.tosatti@cyclades.com> Signed-off-by:
-
Michael Buesch authored
Remove the inlining of the new vs old mmap system call common code. This reduces the size of the resulting vmlinux for defconfig as follows: mb@pc1:~/develop/git/linux-2.6$ size vmlinux.mmap* text data bss dec hex filename 3303749 521524 186564 4011837 3d373d vmlinux.mmapinline 3303557 521524 186564 4011645 3d367d vmlinux.mmapnoinline The new sys_mmap2() has also one function call overhead removed, now. (probably it was already optimized to a jmp before, but anyway...) Signed-off-by:
-
Nick Piggin authored
Optimise page_count compound page test and make it consistent with similar functions. Signed-off-by:
Nick Piggin <npiggin@suse.de> Signed-off-by:
-
Nick Piggin authored
Put a few more checks under CONFIG_DEBUG_VM Signed-off-by:
-
Andrew Morton authored
prep_zero_page() uses KM_USER0 and hence may not be used from IRQ context, at least for highmem pages. Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <christoph@lameter.com> Signed-off-by:
-
Nick Piggin authored
Move the prep_ stuff into prep_new_page. Signed-off-by:
-
Nick Piggin authored
set_page_count usage outside mm/ is limited to setting the refcount to 1. Remove set_page_count from outside mm/, and replace those users with init_page_count() and set_page_refcounted(). This allows more debug checking, and tighter control on how code is allowed to play around with page->_count. Signed-off-by:
-
Nick Piggin authored
A couple of places set_page_count(page, 1) that don't need to. Signed-off-by:
-
Nick Piggin authored
Now that compound page handling is properly fixed in the VM, move nommu over to using compound pages rather than rolling their own refcounting. nommu vm page refcounting is broken anyway, but there is no need to have divergent code in the core VM now, nor when it gets fixed. Signed-off-by:
-
Nick Piggin authored
Remove __put_page from outside the core mm/. It is dangerous because it does not handle compound pages nicely, and misses 1->0 transitions. If a user later appears that really needs the extra speed we can reevaluate. Signed-off-by:
-
Nick Piggin authored
Remove page_count and __put_page from x86-64 pageattr Signed-off-by:
Andi Kleen <ak@suse.de> Signed-off-by:
-
Nick Piggin authored
Use page->lru.next to implement the singly linked list of pages rather than the struct deferred_page which needs to be allocated and freed for each page. Signed-off-by:
-
Nick Piggin authored
Stop using __put_page and page_count in i386 pageattr.c Signed-off-by:
-
Nick Piggin authored
sg increments the refcount of constituent pages in its higher order memory allocations when they are about to be mapped by userspace. This is done so the subsequent get_page/put_page when doing the mapping and unmapping does not free the page. Move over to the preferred way, that is, using compound pages instead. This fixes a whole class of possible obscure bugs where a get_user_pages on a constituent page may outlast the user mappings or even the driver. Signed-off-by:
-
Hugh Dickins authored
Now that it's madvisable, remove two pieces of VM_DONTCOPY bogosity: 1. There was and is no logical reason why VM_DONTCOPY should be in the list of flags which forbid vma merging (and those drivers which set it are also setting VM_IO, which itself forbids the merge). 2. It's hard to understand the purpose of the VM_HUGETLB, VM_DONTCOPY block in vm_stat_account: but never mind, it's under CONFIG_HUGETLB, which (unlike CONFIG_HUGETLB_PAGE or CONFIG_HUGETLBFS) has never been defined. Signed-off-by:
-
Wu Fengguang authored
In shrink_inactive_list(), nr_scan is not accounted when nr_taken is 0. But 0 pages taken does not mean 0 pages scanned. Move the goto statement below the accounting code to fix it. Signed-off-by:
Wu Fengguang <wfg@mail.ustc.edu.cn> Signed-off-by:
-
Wu Fengguang authored
In isolate_lru_pages(), *scanned reports one more scan because the scan counter is increased one more time on exit of the while-loop. Change the while-loop to for-loop to fix it. Signed-off-by:
-
Christoph Lameter authored
Add some comments to explain how zone reclaim works. And it fixes the following issues: - PF_SWAPWRITE needs to be set for RECLAIM_SWAP to be able to write out pages to swap. Currently RECLAIM_SWAP may not do that. - remove setting nr_reclaimed pages after slab reclaim since the slab shrinking code does not use that and the nr_reclaimed pages is just right for the intended follow up action. Signed-off-by:
-
Andrew Morton authored
We have: try_to_free_pages ->shrink_caches(struct zone **zones, ..) ->shrink_zone(struct zone *, ...) ->shrink_cache(struct zone *, ...) ->shrink_list(struct list_head *, ...) ->refill_inactive_list((struct zone *, ...) which is fairly irrational. Rename things so that we have try_to_free_pages ->shrink_zones(struct zone **zones, ..) ->shrink_zone(struct zone *, ...) ->shrink_inactive_list(struct zone *, ...) ->shrink_page_list(struct list_head *, ...) ->shrink_active_list(struct zone *, ...) Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <christoph@lameter.com> Signed-off-by:
-
Andrew Morton authored
Change all the vmscan functions to retunr the number-of-reclaimed pages and remove scan_conrtol.nr_reclaimed. Saves ten-odd bytes of text and makes things clearer and more consistent. The patch also changes the behaviour of zone_reclaim() when it falls back to slab shrinking. Christoph says "Setting this to one means that we will rescan and shrink the slab for each allocation if we are out of zone memory and RECLAIM_SLAB is set. Plus if we do an order 0 allocation we do not go off node as intended. "We better set this to zero. This means the allocation will go offnode despite us having potentially freed lots of memory on the zone. Future allocations can then again be done from this zone." Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <christoph@lameter.com> Signed-off-by:
-
Andrew Morton authored
Turn basically everything in vmscan.c into `unsigned long'. This is to avoid the possibility that some piece of code in there might decide to operate upon more than 4G (or even 2G) of pages in one hit. This might be silly, but we'll need it one day. Cc: Christoph Lameter <clameter@sgi.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Signed-off-by:
Rafael J. Wysocki <rjw@sisk.pl> Signed-off-by:
-
Andrew Morton authored
Initialise as much of scan_control as possible at the declaration site. This tidies things up a bit and assures us that all unmentioned fields are zeroed out. Signed-off-by:
-
Christoph Lameter authored
Make nr_to_scan and priority a parameter instead of putting it into scan control. This allows various small optimizations and IMHO makes the code easier to read. Signed-off-by:
-
Andrew Morton authored
Slab duplicates on_each_cpu(). Signed-off-by:
-
Andrew Morton authored
When on_each_cpu() runs the callback on other CPUs, it runs with local interrupts disabled. So we should run the function with local interrupts disabled on this CPU, too. And do the same for UP, so the callback is run in the same environment on both UP and SMP. (strictly it should do preempt_disable() too, but I think local_irq_disable is sufficiently equivalent). Also uninlines on_each_cpu(). softirq.c was the most appropriate file I could find, but it doesn't seem to justify creating a new file. Oh, and fix up that comment over (under?) x86's smp_call_function(). It drives me nuts. Signed-off-by:
-
Christoph Lameter authored
SLAB_NO_REAP is documented as an option that will cause this slab not to be reaped under memory pressure. However, that is not what happens. The only thing that SLAB_NO_REAP controls at the moment is the reclaim of the unused slab elements that were allocated in batch in cache_reap(). Cache_reap() is run every few seconds independently of memory pressure. Could we remove the whole thing? Its only used by three slabs anyways and I cannot find a reason for having this option. There is an additional problem with SLAB_NO_REAP. If set then the recovery of objects from alien caches is switched off. Objects not freed on the same node where they were initially allocated will only be reused if a certain amount of objects accumulates from one alien node (not very likely) or if the cache is explicitly shrunk. (Strangely __cache_shrink does not check for SLAB_NO_REAP) Getting rid of SLAB_NO_REAP fixes the problems with alien cache freeing. Signed-off-by:
-
