- 24 Feb, 2013 40 commits
-
-
Hugh Dickins authored
Added slightly more detail to the Documentation of merge_across_nodes, a few comments in areas indicated by review, and renamed get_ksm_page()'s argument from "locked" to "lock_it". No functional change. Signed-off-by:
Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Petr Holasek <pholasek@redhat.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Izik Eidus <izik.eidus@ravellosystems.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Greg Thelen authored
Fix several mempolicy leaks in the tmpfs mount logic. These leaks are slow - on the order of one object leaked per mount attempt. Leak 1 (umount doesn't free mpol allocated in mount): while true; do mount -t tmpfs -o mpol=interleave,size=100M nodev /mnt umount /mnt done Leak 2 (errors parsing remount options will leak mpol): mount -t tmpfs -o size=100M nodev /mnt while true; do mount -o remount,mpol=interleave,size=x /mnt 2> /dev/null done umount /mnt Leak 3 (multiple mpol per mount leak mpol): while true; do mount -t tmpfs -o mpol=interleave,mpol=interleave,size=100M nodev /mnt umount /mnt done This patch fixes all of the above. I could have broken the patch into three pieces but is seemed easier to review as one. [akpm@linux-foundation.org: fix handling of mpol_parse_str() errors, per Hugh] Signed-off-by:Greg Thelen <gthelen@google.com> Acked-by:
-
Greg Thelen authored
The tmpfs remount logic preserves filesystem mempolicy if the mpol=M option is not specified in the remount request. A new policy can be specified if mpol=M is given. Before this patch remounting an mpol bound tmpfs without specifying mpol= mount option in the remount request would set the filesystem's mempolicy object to a freed mempolicy object. To reproduce the problem boot a DEBUG_PAGEALLOC kernel and run: # mkdir /tmp/x # mount -t tmpfs -o size=100M,mpol=interleave nodev /tmp/x # grep /tmp/x /proc/mounts nodev /tmp/x tmpfs rw,relatime,size=102400k,mpol=interleave:0-3 0 0 # mount -o remount,size=200M nodev /tmp/x # grep /tmp/x /proc/mounts nodev /tmp/x tmpfs rw,relatime,size=204800k,mpol=??? 0 0 # note ? garbage in mpol=... output above # dd if=/dev/zero of=/tmp/x/f count=1 # panic here Panic: BUG: unable to handle kernel NULL pointer dereference at (null) IP: [< (null)>] (null) [...] Oops: 0010 [#1] SMP DEBUG_PAGEALLOC Call Trace: mpol_shared_policy_init+0xa5/0x160 shmem_get_inode+0x209/0x270 shmem_mknod+0x3e/0xf0 shmem_create+0x18/0x20 vfs_create+0xb5/0x130 do_last+0x9a1/0xea0 path_openat+0xb3/0x4d0 do_filp_open+0x42/0xa0 do_sys_open+0xfe/0x1e0 compat_sys_open+0x1b/0x20 cstar_dispatch+0x7/0x1f Non-debug kernels will not crash immediately because referencing the dangling mpol will not cause a fault. Instead the filesystem will reference a freed mempolicy object, which will cause unpredictable behavior. The problem boils down to a dropped mpol reference below if shmem_parse_options() does not allocate a new mpol: config = *sbinfo shmem_parse_options(data, &config, true) mpol_put(sbinfo->mpol) sbinfo->mpol = config.mpol /* BUG: saves unreferenced mpol */ This patch avoids the crash by not releasing the mempolicy if shmem_parse_options() doesn't create a new mpol. How far back does this issue go? I see it in both 2.6.36 and 3.3. I did not look back further. Signed-off-by: -
Mel Gorman authored
Rob van der Heij reported the following (paraphrased) on private mail. The scenario is that I want to avoid backups to fill up the page cache and purge stuff that is more likely to be used again (this is with s390x Linux on z/VM, so I don't give it as much memory that we don't care anymore). So I have something with LD_PRELOAD that intercepts the close() call (from tar, in this case) and issues a posix_fadvise() just before closing the file. This mostly works, except for small files (less than 14 pages) that remains in page cache after the face. Unfortunately Rob has not had a chance to test this exact patch but the test program below should be reproducing the problem he described. The issue is the per-cpu pagevecs for LRU additions. If the pages are added by one CPU but fadvise() is called on another then the pages remain resident as the invalidate_mapping_pages() only drains the local pagevecs via its call to pagevec_release(). The user-visible effect is that a program that uses fadvise() properly is not obeyed. A possible fix for this is to put the necessary smarts into invalidate_mapping_pages() to globally drain the LRU pagevecs if a pagevec page could not be discarded. The downside with this is that an inode cache shrink would send a global IPI and memory pressure potentially causing global IPI storms is very undesirable. Instead, this patch adds a check during fadvise(POSIX_FADV_DONTNEED) to check if invalidate_mapping_pages() discarded all the requested pages. If a subset of pages are discarded it drains the LRU pagevecs and tries again. If the second attempt fails, it assumes it is due to the pages being mapped, locked or dirty and does not care. With this patch, an application using fadvise() correctly will be obeyed but there is a downside that a malicious application can force the kernel to send global IPIs and increase overhead. If accepted, I would like this to be considered as a -stable candidate. It's not an urgent issue but it's a system call that is not working as advertised which is weak. The following test program demonstrates the problem. It should never report that pages are still resident but will without this patch. It assumes that CPU 0 and 1 exist. int main() { int fd; int pagesize = getpagesize(); ssize_t written = 0, expected; char *buf; unsigned char *vec; int resident, i; cpu_set_t set; /* Prepare a buffer for writing */ expected = FILESIZE_PAGES * pagesize; buf = malloc(expected + 1); if (buf == NULL) { printf("ENOMEM\n"); exit(EXIT_FAILURE); } buf[expected] = 0; memset(buf, 'a', expected); /* Prepare the mincore vec */ vec = malloc(FILESIZE_PAGES); if (vec == NULL) { printf("ENOMEM\n"); exit(EXIT_FAILURE); } /* Bind ourselves to CPU 0 */ CPU_ZERO(&set); CPU_SET(0, &set); if (sched_setaffinity(getpid(), sizeof(set), &set) == -1) { perror("sched_setaffinity"); exit(EXIT_FAILURE); } /* open file, unlink and write buffer */ fd = open("fadvise-test-file", O_CREAT|O_EXCL|O_RDWR); if (fd == -1) { perror("open"); exit(EXIT_FAILURE); } unlink("fadvise-test-file"); while (written < expected) { ssize_t this_write; this_write = write(fd, buf + written, expected - written); if (this_write == -1) { perror("write"); exit(EXIT_FAILURE); } written += this_write; } free(buf); /* * Force ourselves to another CPU. If fadvise only flushes the local * CPUs pagevecs then the fadvise will fail to discard all file pages */ CPU_ZERO(&set); CPU_SET(1, &set); if (sched_setaffinity(getpid(), sizeof(set), &set) == -1) { perror("sched_setaffinity"); exit(EXIT_FAILURE); } /* sync and fadvise to discard the page cache */ fsync(fd); if (posix_fadvise(fd, 0, expected, POSIX_FADV_DONTNEED) == -1) { perror("posix_fadvise"); exit(EXIT_FAILURE); } /* map the file and use mincore to see which parts of it are resident */ buf = mmap(NULL, expected, PROT_READ, MAP_SHARED, fd, 0); if (buf == NULL) { perror("mmap"); exit(EXIT_FAILURE); } if (mincore(buf, expected, vec) == -1) { perror("mincore"); exit(EXIT_FAILURE); } /* Check residency */ for (i = 0, resident = 0; i < FILESIZE_PAGES; i++) { if (vec[i]) resident++; } if (resident != 0) { printf("Nr unexpected pages resident: %d\n", resident); exit(EXIT_FAILURE); } munmap(buf, expected); close(fd); free(vec); exit(EXIT_SUCCESS); } Signed-off-by:Mel Gorman <mgorman@suse.de> Reported-by:
Rob van der Heij <rvdheij@gmail.com> Tested-by:
-
Cliff Wickman authored
We at SGI have a need to address some very high physical address ranges with our GRU (global reference unit), sometimes across partitioned machine boundaries and sometimes with larger addresses than the cpu supports. We do this with the aid of our own 'extended vma' module which mimics the vma. When something (either unmap or exit) frees an 'extended vma' we use the mmu notifiers to clean them up. We had been able to mimic the functions __mmu_notifier_invalidate_range_start() and __mmu_notifier_invalidate_range_end() by locking the per-mm lock and walking the per-mm notifier list. But with the change to a global srcu lock (static in mmu_notifier.c) we can no longer do that. Our module has no access to that lock. So we request that these two functions be exported. Signed-off-by:
Cliff Wickman <cpw@sgi.com> Acked-by:
Robin Holt <holt@sgi.com> Signed-off-by:
-
Michel Lespinasse authored
This change adds a follow_page_mask function which is equivalent to follow_page, but with an extra page_mask argument. follow_page_mask sets *page_mask to HPAGE_PMD_NR - 1 when it encounters a THP page, and to 0 in other cases. __get_user_pages() makes use of this in order to accelerate populating THP ranges - that is, when both the pages and vmas arrays are NULL, we don't need to iterate HPAGE_PMD_NR times to cover a single THP page (and we also avoid taking mm->page_table_lock that many times). Signed-off-by:
Michel Lespinasse <walken@google.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Hugh Dickins <hughd@google.com> Signed-off-by:
-
Michel Lespinasse authored
Use long type for page counts in mm_populate() so as to avoid integer overflow when running the following test code: int main(void) { void *p = mmap(NULL, 0x100000000000, PROT_READ, MAP_PRIVATE | MAP_ANON, -1, 0); printf("p: %p\n", p); mlockall(MCL_CURRENT); printf("done\n"); return 0; } Signed-off-by: -
Zhang Yanfei authored
nr_free_zone_pages(), nr_free_buffer_pages() and nr_free_pagecache_pages() are horribly badly named, so accurately document them with code comments in case of the misuse of them. [akpm@linux-foundation.org: tweak comments] Reviewed-by:
Randy Dunlap <rdunlap@infradead.org> Signed-off-by:
Zhang Yanfei <zhangyanfei@cn.fujitsu.com> Signed-off-by:
-
Naoya Horiguchi authored
error_states[] has two separate states "unevictable LRU page" and "mlocked LRU page", and the former one has the higher priority now. But because of that the latter one is rarely chosen because pages with PageMlocked highly likely have PG_unevictable set. On the other hand, PG_unevictable without PageMlocked is common for ramfs or SHM_LOCKed shared memory, so reversing the priority of these two states helps us clearly distinguish them. Signed-off-by:
Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Chen Gong <gong.chen@linux.intel.com> Cc: Tony Luck <tony.luck@intel.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by:
-
Naoya Horiguchi authored
memory_failure() can't handle memory errors on mlocked pages correctly, because page_action() judges such errors as ones on "unknown pages" instead of ones on "unevictable LRU page" or "mlocked LRU page". In order to determine page_state page_action() checks page flags at the timing of the judgement, but such page flags are not the same with those just after memory_failure() is called, because memory_failure() does unmapping of the error pages before doing page_action(). This unmapping changes the page state, especially page_remove_rmap() (called from try_to_unmap_one()) clears PG_mlocked, so page_action() can't catch mlocked pages after that. With this patch, we store the page flag of the error page before doing unmap, and (only) if the first check with page flags at the time decided the error page is unknown, we do the second check with the stored page flag. This implementation doesn't change error handling for the page types for which the first check can determine the page state correctly. [akpm@linux-foundation.org: tweak comments] Signed-off-by:
-
Hugh Dickins authored
Whilst I run the risk of a flogging for disloyalty to the Lord of Sealand, I do have CONFIG_MEMCG=y CONFIG_MEMCG_KMEM not set, and grow tired of the "mm/memcontrol.c:4972:12: warning: `memcg_propagate_kmem' defined but not used [-Wunused-function]" seen in 3.8-rc: move the #ifdef outwards. Signed-off-by:
Michal Hocko <mhocko@suse.cz> Cc: Glauber Costa <glommer@parallels.com> Signed-off-by:
-
Zhang Yanfei authored
This member of struct virtio_chan is calculated from nr_free_buffer_pages so change its type to unsigned long in case of overflow. Signed-off-by:
-
Zhang Yanfei authored
This variable is calculated from nr_free_pagecache_pages so change its type to unsigned long. Signed-off-by:
-
Zhang Yanfei authored
The three variables are calculated from nr_free_buffer_pages so change their types to unsigned long in case of overflow. Signed-off-by:
-
Zhang Yanfei authored
max_buffer_heads is calculated from nr_free_buffer_pages(), so change its type to unsigned long in case of overflow. Signed-off-by:
-
Zhang Yanfei authored
Now the function nr_free_buffer_pages returns unsigned long, so use %ld to print its return value. Signed-off-by:
-
Zhang Yanfei authored
Currently, the amount of RAM that functions nr_free_*_pages return is held in unsigned int. But in machines with big memory (exceeding 16TB), the amount may be incorrect because of overflow, so fix it. Signed-off-by:
-
Michal Hocko authored
We should encourage all memcg controller initialization independent on a specific mem_cgroup to be done here rather than exploit css_alloc callback and assume that nothing happens before root cgroup is created. Signed-off-by:
Johannes Weiner <hannes@cmpxchg.org> Cc: Tejun Heo <htejun@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Signed-off-by:
-
Michal Hocko authored
memcg_stock are currently initialized during the root cgroup allocation which is OK but it pointlessly pollutes memcg allocation code with something that can be called when the memcg subsystem is initialized by mem_cgroup_init along with other controller specific parts. This patch wraps the current memcg_stock initialization code into a helper calls it from the controller subsystem initialization code. Signed-off-by:
-
Michal Hocko authored
Per-node-zone soft limit tree is currently initialized when the root cgroup is created which is OK but it pointlessly pollutes memcg allocation code with something that can be called when the memcg subsystem is initialized by mem_cgroup_init along with other controller specific parts. While we are at it let's make mem_cgroup_soft_limit_tree_init void because it doesn't make much sense to report memory failure because if we fail to allocate memory that early during the boot then we are screwed anyway (this saves some code). Signed-off-by:
-
Minchan Kim authored
Recently, Luigi reported there are lots of free swap space when OOM happens. It's easily reproduced on zram-over-swap, where many instance of memory hogs are running and laptop_mode is enabled. He said there was no problem when he disabled laptop_mode. The problem when I investigate problem is following as. Assumption for easy explanation: There are no page cache page in system because they all are already reclaimed. 1. try_to_free_pages disable may_writepage when laptop_mode is enabled. 2. shrink_inactive_list isolates victim pages from inactive anon lru list. 3. shrink_page_list adds them to swapcache via add_to_swap but it doesn't pageout because sc->may_writepage is 0 so the page is rotated back into inactive anon lru list. The add_to_swap made the page Dirty by SetPageDirty. 4. 3 couldn't reclaim any pages so do_try_to_free_pages increase priority and retry reclaim with higher priority. 5. shrink_inactlive_list try to isolate victim pages from inactive anon lru list but got failed because it try to isolate pages with ISOLATE_CLEAN mode but inactive anon lru list is full of dirty pages by 3 so it just returns without any reclaim progress. 6. do_try_to_free_pages doesn't set may_writepage due to zero total_scanned. Because sc->nr_scanned is increased by shrink_page_list but we don't call shrink_page_list in 5 due to short of isolated pages. Above loop is continued until OOM happens. The problem didn't happen before [1] was merged because old logic's isolatation in shrink_inactive_list was successful and tried to call shrink_page_list to pageout them but it still ends up failed to page out by may_writepage. But important point is that sc->nr_scanned was increased although we couldn't swap out them so do_try_to_free_pages could set may_writepages. Since commit f80c0673 ("mm: zone_reclaim: make isolate_lru_page() filter-aware") was introduced, it's not a good idea any more to depends on only the number of scanned pages for setting may_writepage. So this patch adds new trigger point of setting may_writepage as below DEF_PRIOIRTY - 2 which is used to show the significant memory pressure in VM so it's good fit for our purpose which would be better to lose power saving or clickety rather than OOM killing. Signed-off-by:
Minchan Kim <minchan@kernel.org> Reported-by:
Luigi Semenzato <semenzato@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by:
-
David Rientjes authored
Make a sweep through mm/ and convert code that uses -1 directly to using the more appropriate NUMA_NO_NODE. Signed-off-by:
David Rientjes <rientjes@google.com> Reviewed-by:
Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com> Signed-off-by:
-
Robin Holt authored
There is a race condition between mmu_notifier_unregister() and __mmu_notifier_release(). Assume two tasks, one calling mmu_notifier_unregister() as a result of a filp_close() ->flush() callout (task A), and the other calling mmu_notifier_release() from an mmput() (task B). A B t1 srcu_read_lock() t2 if (!hlist_unhashed()) t3 srcu_read_unlock() t4 srcu_read_lock() t5 hlist_del_init_rcu() t6 synchronize_srcu() t7 srcu_read_unlock() t8 hlist_del_rcu() <--- NULL pointer deref. Additionally, the list traversal in __mmu_notifier_release() is not protected by the by the mmu_notifier_mm->hlist_lock which can result in callouts to the ->release() notifier from both mmu_notifier_unregister() and __mmu_notifier_release(). -stable suggestions: The stable trees prior to 3.7.y need commits 21a92735 and 70400303 cherry-picked in that order prior to cherry-picking this commit. The 3.7.y tree already has those two commits. Signed-off-by: -
Cody P Schafer authored
Replace open coded pgdat_end_pfn() with helper function. Signed-off-by:
Cody P Schafer <cody@linux.vnet.ibm.com> Cc: David Hansen <dave@linux.vnet.ibm.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mel@csn.ul.ie> Signed-off-by:
-
Cody P Schafer authored
Remove open coding of ensure_zone_is_initialzied(). Signed-off-by:
-
Cody P Schafer authored
ensure_zone_is_initialized() checks if a zone is in a empty & not initialized state (typically occuring after it is created in memory hotplugging), and, if so, calls init_currently_empty_zone() to initialize the zone. Signed-off-by:
-
Cody P Schafer authored
Add a debug message which prints when a page is found outside of the boundaries of the zone it should belong to. Format is: "page $pfn outside zone [ $start_pfn - $end_pfn ]" [akpm@linux-foundation.org: s/pr_debug/pr_err/] Signed-off-by:
-
Cody P Schafer authored
Add pgdat_end_pfn() and pgdat_is_empty() helpers which match the similar zone_*() functions. Change node_end_pfn() to be a wrapper of pgdat_end_pfn(). Signed-off-by:
-
Cody P Schafer authored
Freeing pages to uninitialized zones is not handled by __free_one_page(), and should never happen when the code is correct. Ran into this while writing some code that dynamically onlines extra zones. Signed-off-by:
-
Cody P Schafer authored
Factoring out these 2 checks makes it more clear what we are actually checking for. Signed-off-by:
-
Cody P Schafer authored
Add 2 helpers (zone_end_pfn() and zone_spans_pfn()) to reduce code duplication. This also switches to using them in compaction (where an additional variable needed to be renamed), page_alloc, vmstat, memory_hotplug, and kmemleak. Note that in compaction.c I avoid calling zone_end_pfn() repeatedly because I expect at some point the sycronization issues with start_pfn & spanned_pages will need fixing, either by actually using the seqlock or clever memory barrier usage. Signed-off-by:
-
Cody P Schafer authored
Instead of directly utilizing a combination of config options to determine this, add a macro to specifically address it. Signed-off-by:
-
Johannes Weiner authored
The fact that mlock calls get_user_pages, and get_user_pages might call mlock when expanding a stack looks like a potential recursion. However, mlock makes sure the requested range is already contained within a vma, so no stack expansion will actually happen from mlock. Should this ever change: the stack expansion mlocks only the newly expanded range and so will not result in recursive expansion. Signed-off-by:
Al Viro <viro@ZenIV.linux.org.uk> Cc: Hugh Dickins <hughd@google.com> Acked-by:
-
Johannes Weiner authored
An inactive file list is considered low when its active counterpart is bigger, regardless of whether it is a global zone LRU list or a memcg zone LRU list. The only difference is in how the LRU size is assessed. get_lru_size() does the right thing for both global and memcg reclaim situations. Get rid of inactive_file_is_low_global() and mem_cgroup_inactive_file_is_low() by using get_lru_size() and compare the numbers in common code. Signed-off-by:
-
Johannes Weiner authored
In shmem_find_get_pages_and_swap(), use the faster radix tree iterator construct from commit 78c1d784 ("radix-tree: introduce bit-optimized iterator"). Signed-off-by:
-
Hugh Dickins authored
Complaints are rare, but lockdep still does not understand the way ksm_memory_callback(MEM_GOING_OFFLINE) takes ksm_thread_mutex, and holds it until the ksm_memory_callback(MEM_OFFLINE): that appears to be a problem because notifier callbacks are made under down_read of blocking_notifier_head->rwsem (so first the mutex is taken while holding the rwsem, then later the rwsem is taken while still holding the mutex); but is not in fact a problem because mem_hotplug_mutex is held throughout the dance. There was an attempt to fix this with mutex_lock_nested(); but if that happened to fool lockdep two years ago, apparently it does so no longer. I had hoped to eradicate this issue in extending KSM page migration not to need the ksm_thread_mutex. But then realized that although the page migration itself is safe, we do still need to lock out ksmd and other users of get_ksm_page() while offlining memory - at some point between MEM_GOING_OFFLINE and MEM_OFFLINE, the struct pages themselves may vanish, and get_ksm_page()'s accesses to them become a violation. So, give up on holding ksm_thread_mutex itself from MEM_GOING_OFFLINE to MEM_OFFLINE, and add a KSM_RUN_OFFLINE flag, and wait_while_offlining() checks, to achieve the same lockout without being caught by lockdep. This is less elegant for KSM, but it's more important to keep lockdep useful to other users - and I apologize for how long it took to fix. Signed-off-by:
Gerald Schaefer <gerald.schaefer@de.ibm.com> Tested-by:
-
Hugh Dickins authored
No functional change, but the only purpose of the offlining argument to migrate_pages() etc, was to ensure that __unmap_and_move() could migrate a KSM page for memory hotremove (which took ksm_thread_mutex) but not for other callers. Now all cases are safe, remove the arg. Signed-off-by:
-
Hugh Dickins authored
Migration of KSM pages is now safe: remove the PageKsm restrictions from mempolicy.c and migrate.c. But keep PageKsm out of __unmap_and_move()'s anon_vma contortions, which are irrelevant to KSM: it looks as if that code was preventing hotremove migration of KSM pages, unless they happened to be in swapcache. There is some question as to whether enforcing a NUMA mempolicy migration ought to migrate KSM pages, mapped into entirely unrelated processes; but moving page_mapcount > 1 is only permitted with MPOL_MF_MOVE_ALL anyway, and it seems reasonable to assume that you wouldn't set MADV_MERGEABLE on any area where this is a worry. Signed-off-by:
-
Hugh Dickins authored
The new KSM NUMA merge_across_nodes knob introduces a problem, when it's set to non-default 0: if a KSM page is migrated to a different NUMA node, how do we migrate its stable node to the right tree? And what if that collides with an existing stable node? ksm_migrate_page() can do no more than it's already doing, updating stable_node->kpfn: the stable tree itself cannot be manipulated without holding ksm_thread_mutex. So accept that a stable tree may temporarily indicate a page belonging to the wrong NUMA node, leave updating until the next pass of ksmd, just be careful not to merge other pages on to a misplaced page. Note nid of holding tree in stable_node, and recognize that it will not always match nid of kpfn. A misplaced KSM page is discovered, either when ksm_do_scan() next comes around to one of its rmap_items (we now have to go to cmp_and_merge_page even on pages in a stable tree), or when stable_tree_search() arrives at a matching node for another page, and this node page is found misplaced. In each case, move the misplaced stable_node to a list of migrate_nodes (and use the address of migrate_nodes as magic by which to identify them): we don't need them in a tree. If stable_tree_search() finds no match for a page, but it's currently exiled to this list, then slot its stable_node right there into the tree, bringing all of its mappings with it; otherwise they get migrated one by one to the original page of the colliding node. stable_tree_search() is now modelled more like stable_tree_insert(), in order to handle these insertions of migrated nodes. remove_node_from_stable_tree(), remove_all_stable_nodes() and ksm_check_stable_tree() have to handle the migrate_nodes list as well as the stable tree itself. Less obviously, we do need to prune the list of stale entries from time to time (scan_get_next_rmap_item() does it once each full scan): whereas stale nodes in the stable tree get naturally pruned as searches try to brush past them, these migrate_nodes may get forgotten and accumulate. Signed-off-by:
-
Hugh Dickins authored
KSM page migration is already supported in the case of memory hotremove, which takes the ksm_thread_mutex across all its migrations to keep life simple. But the new KSM NUMA merge_across_nodes knob introduces a problem, when it's set to non-default 0: if a KSM page is migrated to a different NUMA node, how do we migrate its stable node to the right tree? And what if that collides with an existing stable node? So far there's no provision for that, and this patch does not attempt to deal with it either. But how will I test a solution, when I don't know how to hotremove memory? The best answer is to enable KSM page migration in all cases now, and test more common cases. With THP and compaction added since KSM came in, page migration is now mainstream, and it's a shame that a KSM page can frustrate freeing a page block. Without worrying about merge_across_nodes 0 for now, this patch gets KSM page migration working reliably for default merge_across_nodes 1 (but leave the patch enabling it until near the end of the series). It's much simpler than I'd originally imagined, and does not require an additional tier of locking: page migration relies on the page lock, KSM page reclaim relies on the page lock, the page lock is enough for KSM page migration too. Almost all the care has to be in get_ksm_page(): that's the function which worries about when a stable node is stale and should be freed, now it also has to worry about the KSM page being migrated. The only new overhead is an additional put/get/lock/unlock_page when stable_tree_search() arrives at a matching node: to make sure migration respects the raised page count, and so does not migrate the page while we're busy with it here. That's probably avoidable, either by changing internal interfaces from using kpage to stable_node, or by moving the ksm_migrate_page() callsite into a page_freeze_refs() section (even if not swapcache); but this works well, I've no urge to pull it apart now. (Descents of the stable tree may pass through nodes whose KSM pages are under migration: being unlocked, the raised page count does not prevent that, nor need it: it's safe to memcmp against either old or new page.) You might worry about mremap, and whether page migration's rmap_walk to remove migration entries will find all the KSM locations where it inserted earlier: that should already be handled, by the satisfyingly heavy hammer of move_vma()'s call to ksm_madvise(,,,MADV_UNMERGEABLE,). Signed-off-by:
-
