Recently, I got many reports about perfermance degradation in embedded system(Android mobile phone, webOS TV and so on) and failed to fork easily. The problem was fragmentation caused by zram and GPU driver pages. Their pages cannot be migrated so compaction cannot work well, either so reclaimer ends up shrinking all of working set pages. It made system very slow and even to fail to fork easily. Other pain point is that they cannot work with CMA. Most of CMA memory space could be idle(ie, it could be used for movable pages unless driver is using) but if driver(i.e., zram) cannot migrate his page, that memory space could be wasted. In our product which has big CMA memory, it reclaims zones too exccessively although there are lots of free space in CMA so system was very slow easily. To solve these problem, this patch try to add facility to migrate non-lru pages via introducing new friend functions of migratepage in address_space_operation and new page flags. (isolate_page, putback_page) (PG_movable, PG_isolated) For details, please read description in "mm/compaction: support non-lru movable page migration". Originally, Gioh Kim tried to support this feature but he moved so I took over the work. But I took many code from his work and changed a little bit. Thanks, Gioh! And I should mention Konstantin Khlebnikov. He really heped Gioh at that time so he should deserve to have many credit, too. Thanks, Konstantin! This patchset consists of five parts. 1. clean up migration mm: use put_page to free page instead of putback_lru_page 2. add non-lru page migration feature mm/compaction: support non-lru movable page migration mm: add non-lru movable page support document 3. rework KVM memory-ballooning mm/balloon: use general movable page feature into balloon 4. zsmalloc clean-up for preparing page migration zsmalloc: keep max_object in size_class zsmalloc: squeeze inuse into page->mapping zsmalloc: remove page_mapcount_reset zsmalloc: squeeze freelist into page->mapping zsmalloc: move struct zs_meta from mapping to freelist zsmalloc: factor page chain functionality out zsmalloc: separate free_zspage from putback_zspage zsmalloc: zs_compact refactoring 5. add zsmalloc page migration zsmalloc: migrate head page of zspage zsmalloc: use single linked list for page chain zsmalloc: migrate tail pages in zspage zram: use __GFP_MOVABLE for memory allocation * From v2 * rebase on mmotm-2016-03-29-15-54-16 * check PageMovable before lock_page - Joonsoo * check PageMovable before PageIsolated checking - Joonsoo * add more description about rule * From v1 * rebase on v4.5-mmotm-2016-03-17-15-04 * reordering patches to merge clean-up patches first * add Acked-by/Reviewed-by from Vlastimil and Sergey * use each own mount model instead of reusing anon_inode_fs - Al Viro * small changes - YiPing, Gioh Minchan Kim (16): mm: use put_page to free page instead of putback_lru_page mm/compaction: support non-lru movable page migration mm: add non-lru movable page support document mm/balloon: use general movable page feature into balloon zsmalloc: keep max_object in size_class zsmalloc: squeeze inuse into page->mapping zsmalloc: remove page_mapcount_reset zsmalloc: squeeze freelist into page->mapping zsmalloc: move struct zs_meta from mapping to freelist zsmalloc: factor page chain functionality out zsmalloc: separate free_zspage from putback_zspage zsmalloc: zs_compact refactoring zsmalloc: migrate head page of zspage zsmalloc: use single linked list for page chain zsmalloc: migrate tail pages in zspage zram: use __GFP_MOVABLE for memory allocation Documentation/filesystems/Locking | 4 + Documentation/filesystems/vfs.txt | 16 +- Documentation/vm/page_migration | 69 +- drivers/block/zram/zram_drv.c | 3 +- drivers/virtio/virtio_balloon.c | 53 +- fs/proc/page.c | 3 + include/linux/balloon_compaction.h | 49 +- include/linux/fs.h | 2 + include/linux/migrate.h | 2 + include/linux/page-flags.h | 47 +- include/uapi/linux/kernel-page-flags.h | 1 + include/uapi/linux/magic.h | 2 + mm/balloon_compaction.c | 101 +-- mm/compaction.c | 15 +- mm/migrate.c | 238 ++++-- mm/vmscan.c | 2 +- mm/zsmalloc.c | 1253 ++++++++++++++++++++++++-------- 17 files changed, 1368 insertions(+), 492 deletions(-) -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 01/16] mm: use put_page to free page instead of putback_lru_page
Procedure of page migration is as follows: First of all, it should isolate a page from LRU and try to migrate the page. If it is successful, it releases the page for freeing. Otherwise, it should put the page back to LRU list. For LRU pages, we have used putback_lru_page for both freeing and putback to LRU list. It's okay because put_page is aware of LRU list so if it releases last refcount of the page, it removes the page from LRU list. However, It makes unnecessary operations (e.g., lru_cache_add, pagevec and flags operations. It would be not significant but no worth to do) and harder to support new non-lru page migration because put_page isn't aware of non-lru page's data structure. To solve the problem, we can add new hook in put_page with PageMovable flags check but it can increase overhead in hot path and needs new locking scheme to stabilize the flag check with put_page. So, this patch cleans it up to divide two semantic(ie, put and putback). If migration is successful, use put_page instead of putback_lru_page and use putback_lru_page only on failure. That makes code more readable and doesn't add overhead in put_page. Comment from Vlastimil "Yeah, and compaction (perhaps also other migration users) has to drain the lru pvec... Getting rid of this stuff is worth even by itself." Cc: Mel Gorman <mgorman at suse.de> Cc: Hugh Dickins <hughd at google.com> Cc: Naoya Horiguchi <n-horiguchi at ah.jp.nec.com> Acked-by: Vlastimil Babka <vbabka at suse.cz> Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/migrate.c | 50 +++++++++++++++++++++++++++++++------------------- 1 file changed, 31 insertions(+), 19 deletions(-) diff --git a/mm/migrate.c b/mm/migrate.c index 6c822a7b27e0..53529c805752 100644 --- a/mm/migrate.c +++ b/mm/migrate.c @@ -913,6 +913,14 @@ static int __unmap_and_move(struct page *page, struct page *newpage, put_anon_vma(anon_vma); unlock_page(page); out: + /* If migration is successful, move newpage to right list */ + if (rc == MIGRATEPAGE_SUCCESS) { + if (unlikely(__is_movable_balloon_page(newpage))) + put_page(newpage); + else + putback_lru_page(newpage); + } + return rc; } @@ -946,6 +954,12 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page, if (page_count(page) == 1) { /* page was freed from under us. So we are done. */ + ClearPageActive(page); + ClearPageUnevictable(page); + if (put_new_page) + put_new_page(newpage, private); + else + put_page(newpage); goto out; } @@ -958,10 +972,8 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page, } rc = __unmap_and_move(page, newpage, force, mode); - if (rc == MIGRATEPAGE_SUCCESS) { - put_new_page = NULL; + if (rc == MIGRATEPAGE_SUCCESS) set_page_owner_migrate_reason(newpage, reason); - } out: if (rc != -EAGAIN) { @@ -974,28 +986,28 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page, list_del(&page->lru); dec_zone_page_state(page, NR_ISOLATED_ANON + page_is_file_cache(page)); - /* Soft-offlined page shouldn't go through lru cache list */ + } + + /* + * If migration is successful, drop the reference grabbed during + * isolation. Otherwise, restore the page to LRU list unless we + * want to retry. + */ + if (rc == MIGRATEPAGE_SUCCESS) { + put_page(page); if (reason == MR_MEMORY_FAILURE) { - put_page(page); if (!test_set_page_hwpoison(page)) num_poisoned_pages_inc(); - } else + } + } else { + if (rc != -EAGAIN) putback_lru_page(page); + if (put_new_page) + put_new_page(newpage, private); + else + put_page(newpage); } - /* - * If migration was not successful and there's a freeing callback, use - * it. Otherwise, putback_lru_page() will drop the reference grabbed - * during isolation. - */ - if (put_new_page) - put_new_page(newpage, private); - else if (unlikely(__is_movable_balloon_page(newpage))) { - /* drop our reference, page already in the balloon */ - put_page(newpage); - } else - putback_lru_page(newpage); - if (result) { if (rc) *result = rc; -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 02/16] mm/compaction: support non-lru movable page migration
We have allowed migration for only LRU pages until now and it was enough to make high-order pages. But recently, embedded system(e.g., webOS, android) uses lots of non-movable pages(e.g., zram, GPU memory) so we have seen several reports about troubles of small high-order allocation. For fixing the problem, there were several efforts (e,g,. enhance compaction algorithm, SLUB fallback to 0-order page, reserved memory, vmalloc and so on) but if there are lots of non-movable pages in system, their solutions are void in the long run. So, this patch is to support facility to change non-movable pages with movable. For the feature, this patch introduces functions related to migration to address_space_operations as well as some page flags. Basically, this patch supports two page-flags and two functions related to page migration. The flag and page->mapping stability are protected by PG_lock. PG_movable PG_isolated bool (*isolate_page) (struct page *, isolate_mode_t); void (*putback_page) (struct page *); Duty of subsystem want to make their pages as migratable are as follows: 1. It should register address_space to page->mapping then mark the page as PG_movable via __SetPageMovable. 2. It should mark the page as PG_isolated via SetPageIsolated if isolation is sucessful and return true. 3. If migration is successful, it should clear PG_isolated and PG_movable of the page for free preparation then release the reference of the page to free. 4. If migration fails, putback function of subsystem should clear PG_isolated via ClearPageIsolated. 5. If a subsystem want to release isolated page, it should clear PG_isolated but not PG_movable. Instead, VM will do it. Cc: Vlastimil Babka <vbabka at suse.cz> Cc: Mel Gorman <mgorman at suse.de> Cc: Hugh Dickins <hughd at google.com> Cc: dri-devel at lists.freedesktop.org Cc: virtualization at lists.linux-foundation.org Signed-off-by: Gioh Kim <gurugio at hanmail.net> Signed-off-by: Minchan Kim <minchan at kernel.org> --- Documentation/filesystems/Locking | 4 + Documentation/filesystems/vfs.txt | 5 + fs/proc/page.c | 3 + include/linux/fs.h | 2 + include/linux/migrate.h | 2 + include/linux/page-flags.h | 31 ++++++ include/uapi/linux/kernel-page-flags.h | 1 + mm/compaction.c | 14 ++- mm/migrate.c | 174 +++++++++++++++++++++++++++++---- 9 files changed, 217 insertions(+), 19 deletions(-) diff --git a/Documentation/filesystems/Locking b/Documentation/filesystems/Locking index 619af9bfdcb3..0bb79560abb3 100644 --- a/Documentation/filesystems/Locking +++ b/Documentation/filesystems/Locking @@ -195,7 +195,9 @@ unlocks and drops the reference. int (*releasepage) (struct page *, int); void (*freepage)(struct page *); int (*direct_IO)(struct kiocb *, struct iov_iter *iter, loff_t offset); + bool (*isolate_page) (struct page *, isolate_mode_t); int (*migratepage)(struct address_space *, struct page *, struct page *); + void (*putback_page) (struct page *); int (*launder_page)(struct page *); int (*is_partially_uptodate)(struct page *, unsigned long, unsigned long); int (*error_remove_page)(struct address_space *, struct page *); @@ -219,7 +221,9 @@ invalidatepage: yes releasepage: yes freepage: yes direct_IO: +isolate_page: yes migratepage: yes (both) +putback_page: yes launder_page: yes is_partially_uptodate: yes error_remove_page: yes diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt index b02a7d598258..4c1b6c3b4bc8 100644 --- a/Documentation/filesystems/vfs.txt +++ b/Documentation/filesystems/vfs.txt @@ -592,9 +592,14 @@ struct address_space_operations { int (*releasepage) (struct page *, int); void (*freepage)(struct page *); ssize_t (*direct_IO)(struct kiocb *, struct iov_iter *iter, loff_t offset); + /* isolate a page for migration */ + bool (*isolate_page) (struct page *, isolate_mode_t); /* migrate the contents of a page to the specified target */ int (*migratepage) (struct page *, struct page *); + /* put the page back to right list */ + void (*putback_page) (struct page *); int (*launder_page) (struct page *); + int (*is_partially_uptodate) (struct page *, unsigned long, unsigned long); void (*is_dirty_writeback) (struct page *, bool *, bool *); diff --git a/fs/proc/page.c b/fs/proc/page.c index 3ecd445e830d..ce3d08a4ad8d 100644 --- a/fs/proc/page.c +++ b/fs/proc/page.c @@ -157,6 +157,9 @@ u64 stable_page_flags(struct page *page) if (page_is_idle(page)) u |= 1 << KPF_IDLE; + if (PageMovable(page)) + u |= 1 << KPF_MOVABLE; + u |= kpf_copy_bit(k, KPF_LOCKED, PG_locked); u |= kpf_copy_bit(k, KPF_SLAB, PG_slab); diff --git a/include/linux/fs.h b/include/linux/fs.h index da9e67d937e5..36f2d610e7a8 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -401,6 +401,8 @@ struct address_space_operations { */ int (*migratepage) (struct address_space *, struct page *, struct page *, enum migrate_mode); + bool (*isolate_page)(struct page *, isolate_mode_t); + void (*putback_page)(struct page *); int (*launder_page) (struct page *); int (*is_partially_uptodate) (struct page *, unsigned long, unsigned long); diff --git a/include/linux/migrate.h b/include/linux/migrate.h index 9b50325e4ddf..404fbfefeb33 100644 --- a/include/linux/migrate.h +++ b/include/linux/migrate.h @@ -37,6 +37,8 @@ extern int migrate_page(struct address_space *, struct page *, struct page *, enum migrate_mode); extern int migrate_pages(struct list_head *l, new_page_t new, free_page_t free, unsigned long private, enum migrate_mode mode, int reason); +extern bool isolate_movable_page(struct page *page, isolate_mode_t mode); +extern void putback_movable_page(struct page *page); extern int migrate_prep(void); extern int migrate_prep_local(void); diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h index f4ed4f1b0c77..77ebf8fdbc6e 100644 --- a/include/linux/page-flags.h +++ b/include/linux/page-flags.h @@ -129,6 +129,10 @@ enum pageflags { /* Compound pages. Stored in first tail page's flags */ PG_double_map = PG_private_2, + + /* non-lru movable pages */ + PG_movable = PG_reclaim, + PG_isolated = PG_owner_priv_1, }; #ifndef __GENERATING_BOUNDS_H @@ -614,6 +618,33 @@ static inline void __ClearPageBalloon(struct page *page) atomic_set(&page->_mapcount, -1); } +#define PAGE_MOVABLE_MAPCOUNT_VALUE (-255) + +static inline int PageMovable(struct page *page) +{ + return ((test_bit(PG_movable, &(page)->flags) && + atomic_read(&page->_mapcount) == PAGE_MOVABLE_MAPCOUNT_VALUE) + || PageBalloon(page)); +} + +/* Caller should hold a PG_lock */ +static inline void __SetPageMovable(struct page *page, + struct address_space *mapping) +{ + page->mapping = mapping; + __set_bit(PG_movable, &page->flags); + atomic_set(&page->_mapcount, PAGE_MOVABLE_MAPCOUNT_VALUE); +} + +static inline void __ClearPageMovable(struct page *page) +{ + atomic_set(&page->_mapcount, -1); + __clear_bit(PG_movable, &(page)->flags); + page->mapping = NULL; +} + +PAGEFLAG(Isolated, isolated, PF_ANY); + /* * If network-based swap is enabled, sl*b must keep track of whether pages * were allocated from pfmemalloc reserves. diff --git a/include/uapi/linux/kernel-page-flags.h b/include/uapi/linux/kernel-page-flags.h index 5da5f8751ce7..a184fd2434fa 100644 --- a/include/uapi/linux/kernel-page-flags.h +++ b/include/uapi/linux/kernel-page-flags.h @@ -34,6 +34,7 @@ #define KPF_BALLOON 23 #define KPF_ZERO_PAGE 24 #define KPF_IDLE 25 +#define KPF_MOVABLE 26 #endif /* _UAPILINUX_KERNEL_PAGE_FLAGS_H */ diff --git a/mm/compaction.c b/mm/compaction.c index ccf97b02b85f..7557aedddaee 100644 --- a/mm/compaction.c +++ b/mm/compaction.c @@ -703,7 +703,7 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, /* * Check may be lockless but that's ok as we recheck later. - * It's possible to migrate LRU pages and balloon pages + * It's possible to migrate LRU and movable kernel pages. * Skip any other type of page */ is_lru = PageLRU(page); @@ -714,6 +714,18 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, goto isolate_success; } } + + if (unlikely(PageMovable(page)) && + !PageIsolated(page)) { + if (locked) { + spin_unlock_irqrestore(&zone->lru_lock, + flags); + locked = false; + } + + if (isolate_movable_page(page, isolate_mode)) + goto isolate_success; + } } /* diff --git a/mm/migrate.c b/mm/migrate.c index 53529c805752..b56bf2b3fe8c 100644 --- a/mm/migrate.c +++ b/mm/migrate.c @@ -73,6 +73,85 @@ int migrate_prep_local(void) return 0; } +bool isolate_movable_page(struct page *page, isolate_mode_t mode) +{ + bool ret = false; + + /* + * Avoid burning cycles with pages that are yet under __free_pages(), + * or just got freed under us. + * + * In case we 'win' a race for a movable page being freed under us and + * raise its refcount preventing __free_pages() from doing its job + * the put_page() at the end of this block will take care of + * release this page, thus avoiding a nasty leakage. + */ + if (unlikely(!get_page_unless_zero(page))) + goto out; + + /* + * Check PG_movable before holding a PG_lock because page's owner + * assumes anybody doesn't touch PG_lock of newly allocated page. + */ + if (unlikely(!PageMovable(page))) + goto out_putpage; + /* + * As movable pages are not isolated from LRU lists, concurrent + * compaction threads can race against page migration functions + * as well as race against the releasing a page. + * + * In order to avoid having an already isolated movable page + * being (wrongly) re-isolated while it is under migration, + * or to avoid attempting to isolate pages being released, + * lets be sure we have the page lock + * before proceeding with the movable page isolation steps. + */ + if (unlikely(!trylock_page(page))) + goto out_putpage; + + if (!PageMovable(page) || PageIsolated(page)) + goto out_no_isolated; + + ret = page->mapping->a_ops->isolate_page(page, mode); + if (!ret) + goto out_no_isolated; + + WARN_ON_ONCE(!PageIsolated(page)); + unlock_page(page); + return ret; + +out_no_isolated: + unlock_page(page); +out_putpage: + put_page(page); +out: + return ret; +} + +/* It should be called on page which is PG_movable */ +void putback_movable_page(struct page *page) +{ + /* + * 'lock_page()' stabilizes the page and prevents races against + * concurrent isolation threads attempting to re-isolate it. + */ + VM_BUG_ON_PAGE(!PageMovable(page), page); + + lock_page(page); + if (PageIsolated(page)) { + struct address_space *mapping; + + mapping = page_mapping(page); + mapping->a_ops->putback_page(page); + WARN_ON_ONCE(PageIsolated(page)); + } else { + __ClearPageMovable(page); + } + unlock_page(page); + /* drop the extra ref count taken for movable page isolation */ + put_page(page); +} + /* * Put previously isolated pages back onto the appropriate lists * from where they were once taken off for compaction/migration. @@ -94,10 +173,18 @@ void putback_movable_pages(struct list_head *l) list_del(&page->lru); dec_zone_page_state(page, NR_ISOLATED_ANON + page_is_file_cache(page)); - if (unlikely(isolated_balloon_page(page))) + if (unlikely(isolated_balloon_page(page))) { balloon_page_putback(page); - else + } else if (unlikely(PageMovable(page))) { + if (PageIsolated(page)) { + putback_movable_page(page); + } else { + __ClearPageMovable(page); + put_page(page); + } + } else { putback_lru_page(page); + } } } @@ -592,7 +679,7 @@ void migrate_page_copy(struct page *newpage, struct page *page) ***********************************************************/ /* - * Common logic to directly migrate a single page suitable for + * Common logic to directly migrate a single LRU page suitable for * pages that do not use PagePrivate/PagePrivate2. * * Pages are locked upon entry and exit. @@ -755,24 +842,54 @@ static int move_to_new_page(struct page *newpage, struct page *page, enum migrate_mode mode) { struct address_space *mapping; - int rc; + int rc = -EAGAIN; + bool lru_movable = true; VM_BUG_ON_PAGE(!PageLocked(page), page); VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); mapping = page_mapping(page); - if (!mapping) - rc = migrate_page(mapping, newpage, page, mode); - else if (mapping->a_ops->migratepage) - /* - * Most pages have a mapping and most filesystems provide a - * migratepage callback. Anonymous pages are part of swap - * space which also has its own migratepage callback. This - * is the most common path for page migration. - */ - rc = mapping->a_ops->migratepage(mapping, newpage, page, mode); - else - rc = fallback_migrate_page(mapping, newpage, page, mode); + /* + * In case of non-lru page, it could be released after + * isolation step. In that case, we shouldn't try + * fallback migration which was designed for LRU pages. + * + * The rule for such case is that subsystem should clear + * PG_isolated but remains PG_movable so VM should catch + * it and clear PG_movable for it. + */ + if (unlikely(PageMovable(page))) { + lru_movable = false; + VM_BUG_ON_PAGE(!mapping, page); + if (!PageIsolated(page)) { + rc = MIGRATEPAGE_SUCCESS; + __ClearPageMovable(page); + goto out; + } + } + + if (likely(lru_movable)) { + if (!mapping) + rc = migrate_page(mapping, newpage, page, mode); + else if (mapping->a_ops->migratepage) + /* + * Most pages have a mapping and most filesystems + * provide a migratepage callback. Anonymous pages + * are part of swap space which also has its own + * migratepage callback. This is the most common path + * for page migration. + */ + rc = mapping->a_ops->migratepage(mapping, newpage, + page, mode); + else + rc = fallback_migrate_page(mapping, newpage, + page, mode); + } else { + rc = mapping->a_ops->migratepage(mapping, newpage, + page, mode); + WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS && + PageIsolated(page)); + } /* * When successful, old pagecache page->mapping must be cleared before @@ -782,6 +899,7 @@ static int move_to_new_page(struct page *newpage, struct page *page, if (!PageAnon(page)) page->mapping = NULL; } +out: return rc; } @@ -960,6 +1078,8 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page, put_new_page(newpage, private); else put_page(newpage); + if (PageMovable(page)) + __ClearPageMovable(page); goto out; } @@ -1000,8 +1120,26 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page, num_poisoned_pages_inc(); } } else { - if (rc != -EAGAIN) - putback_lru_page(page); + if (rc != -EAGAIN) { + /* + * subsystem couldn't remove PG_movable since page is + * isolated so PageMovable check is not racy in here. + * But PageIsolated check can be racy but it's okay + * because putback_movable_page checks it under PG_lock + * again. + */ + if (unlikely(PageMovable(page))) { + if (PageIsolated(page)) + putback_movable_page(page); + else { + __ClearPageMovable(page); + put_page(page); + } + } else { + putback_lru_page(page); + } + } + if (put_new_page) put_new_page(newpage, private); else -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 03/16] mm: add non-lru movable page support document
This patch describes what a subsystem should do for non-lru movable page supporting. Cc: Jonathan Corbet <corbet at lwn.net> Signed-off-by: Minchan Kim <minchan at kernel.org> --- Documentation/filesystems/vfs.txt | 11 ++++++- Documentation/vm/page_migration | 69 ++++++++++++++++++++++++++++++++++++++- 2 files changed, 78 insertions(+), 2 deletions(-) diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt index 4c1b6c3b4bc8..d63142f8ed7b 100644 --- a/Documentation/filesystems/vfs.txt +++ b/Documentation/filesystems/vfs.txt @@ -752,12 +752,21 @@ struct address_space_operations { and transfer data directly between the storage and the application's address space. + isolate_page: Called by the VM when isolating a movable non-lru page. + If page is successfully isolated, we should mark the page as + PG_isolated via __SetPageIsolated. + migrate_page: This is used to compact the physical memory usage. If the VM wants to relocate a page (maybe off a memory card that is signalling imminent failure) it will pass a new page and an old page to this function. migrate_page should transfer any private data across and update any references - that it has to the page. + that it has to the page. If migrated page is non-lru page, + we should clear PG_isolated and PG_movable via __ClearPageIsolated + and __ClearPageMovable. + + putback_page: Called by the VM when isolated page's migration fails. + We should clear PG_isolated marked in isolated_page function. launder_page: Called before freeing a page - it writes back the dirty page. To prevent redirtying the page, it is kept locked during the whole diff --git a/Documentation/vm/page_migration b/Documentation/vm/page_migration index fea5c0864170..c4e7551a414e 100644 --- a/Documentation/vm/page_migration +++ b/Documentation/vm/page_migration @@ -142,5 +142,72 @@ is increased so that the page cannot be freed while page migration occurs. 20. The new page is moved to the LRU and can be scanned by the swapper etc again. -Christoph Lameter, May 8, 2006. +C. Non-LRU Page migration +------------------------- + +Although original migration aimed for reducing the latency of memory access +for NUMA, compaction who want to create high-order page is also main customer. + +Ppage migration's disadvantage is that it was designed to migrate only +*LRU* pages. However, there are potential non-lru movable pages which can be +migrated in system, for example, zsmalloc, virtio-balloon pages. +For virtio-balloon pages, some parts of migration code path was hooked up +and added virtio-balloon specific functions to intercept logi. +It's too specific to one subsystem so other subsystem who want to make +their pages movable should add own specific hooks in migration path. + +To solve such problem, VM supports non-LRU page migration which provides +generic functions for non-LRU movable pages without needing subsystem +specific hook in mm/{migrate|compact}.c. + +If a subsystem want to make own pages movable, it should mark pages as +PG_movable via __SetPageMovable. __SetPageMovable needs address_space for +argument for register functions which will be called by VM. + +Three functions in address_space_operation related to non-lru movable page: + + bool (*isolate_page) (struct page *, isolate_mode_t); + int (*migratepage) (struct address_space *, + struct page *, struct page *, enum migrate_mode); + void (*putback_page)(struct page *); + +1. Isolation + +What VM expected on isolate_page of subsystem is to set PG_isolated flags +of the page if it was successful. With that, concurrent isolation among +CPUs skips the isolated page by other CPU earlier. VM calls isolate_page +under PG_lock of page. If a subsystem cannot isolate the page, it should +return false. +2. Migration + +After successful isolation, VM calls migratepage. The migratepage's goal is +to move content of the old page to new page and set up struct page fields +of new page. If migration is successful, subsystem should release old page's +refcount to free. Keep in mind that subsystem should clear PG_movable and +PG_isolated before releasing the refcount. If everything are done, user +should return MIGRATEPAGE_SUCCESS. If subsystem cannot migrate the page +at the moment, migratepage can return -EAGAIN. On -EAGAIN, VM will retry page +migration because VM interprets -EAGAIN as "temporal migration failure". + +3. Putback + +If migration was unsuccessful, VM calls putback_page. The subsystem should +insert isolated page to own data structure again if it has. And subsystem +should clear PG_isolated which was marked in isolation step. + +Note about releasing page: + +Subsystem can release pages whenever it want but if it releses the page +which is already isolated, it should clear PG_isolated but doesn't touch +PG_movable under PG_lock. Instead of it, VM will clear PG_movable after +his job done. Otherweise, subsystem should clear both page flags before +releasing the page. + +Note about PG_isolated: + +PG_isolated check on a page is valid only if the page's flag is already +set to PG_movable. + +Christoph Lameter, May 8, 2006. +Minchan Kim, Mar 28, 2016. -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 04/16] mm/balloon: use general movable page feature into balloon
Now, VM has a feature to migrate non-lru movable pages so balloon doesn't need custom migration hooks in migrate.c and compact.c. Instead, this patch implements page->mapping ->{isolate|migrate|putback} functions. With that, we could remove hooks for ballooning in general migration functions and make balloon compaction simple. Cc: virtualization at lists.linux-foundation.org Cc: Rafael Aquini <aquini at redhat.com> Cc: Konstantin Khlebnikov <koct9i at gmail.com> Signed-off-by: Gioh Kim <gurugio at hanmail.net> Signed-off-by: Minchan Kim <minchan at kernel.org> --- drivers/virtio/virtio_balloon.c | 53 ++++++++++++++++--- include/linux/balloon_compaction.h | 49 ++++-------------- include/linux/page-flags.h | 56 +++++++++++--------- include/uapi/linux/magic.h | 1 + mm/balloon_compaction.c | 101 ++++++++----------------------------- mm/compaction.c | 7 --- mm/migrate.c | 22 ++------ mm/vmscan.c | 2 +- 8 files changed, 119 insertions(+), 172 deletions(-) diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c index 7b6d74f0c72f..0c16192d2684 100644 --- a/drivers/virtio/virtio_balloon.c +++ b/drivers/virtio/virtio_balloon.c @@ -30,6 +30,7 @@ #include <linux/oom.h> #include <linux/wait.h> #include <linux/mm.h> +#include <linux/mount.h> /* * Balloon device works in 4K page units. So each page is pointed to by @@ -45,6 +46,10 @@ static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES; module_param(oom_pages, int, S_IRUSR | S_IWUSR); MODULE_PARM_DESC(oom_pages, "pages to free on OOM"); +#ifdef CONFIG_BALLOON_COMPACTION +static struct vfsmount *balloon_mnt; +#endif + struct virtio_balloon { struct virtio_device *vdev; struct virtqueue *inflate_vq, *deflate_vq, *stats_vq; @@ -482,10 +487,29 @@ static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info, mutex_unlock(&vb->balloon_lock); + ClearPageIsolated(page); put_page(page); /* balloon reference */ return MIGRATEPAGE_SUCCESS; } + +static struct dentry *balloon_mount(struct file_system_type *fs_type, + int flags, const char *dev_name, void *data) +{ + static const struct dentry_operations ops = { + .d_dname = simple_dname, + }; + + return mount_pseudo(fs_type, "balloon-kvm:", NULL, &ops, + BALLOON_KVM_MAGIC); +} + +static struct file_system_type balloon_fs = { + .name = "balloon-kvm", + .mount = balloon_mount, + .kill_sb = kill_anon_super, +}; + #endif /* CONFIG_BALLOON_COMPACTION */ static int virtballoon_probe(struct virtio_device *vdev) @@ -515,10 +539,6 @@ static int virtballoon_probe(struct virtio_device *vdev) vb->vdev = vdev; balloon_devinfo_init(&vb->vb_dev_info); -#ifdef CONFIG_BALLOON_COMPACTION - vb->vb_dev_info.migratepage = virtballoon_migratepage; -#endif - err = init_vqs(vb); if (err) goto out_free_vb; @@ -527,13 +547,32 @@ static int virtballoon_probe(struct virtio_device *vdev) vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY; err = register_oom_notifier(&vb->nb); if (err < 0) - goto out_oom_notify; + goto out_del_vqs; + +#ifdef CONFIG_BALLOON_COMPACTION + balloon_mnt = kern_mount(&balloon_fs); + if (IS_ERR(balloon_mnt)) { + err = PTR_ERR(balloon_mnt); + unregister_oom_notifier(&vb->nb); + goto out_del_vqs; + } + vb->vb_dev_info.migratepage = virtballoon_migratepage; + vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb); + if (IS_ERR(vb->vb_dev_info.inode)) { + err = PTR_ERR(vb->vb_dev_info.inode); + kern_unmount(balloon_mnt); + unregister_oom_notifier(&vb->nb); + vb->vb_dev_info.inode = NULL; + goto out_del_vqs; + } + vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops; +#endif virtio_device_ready(vdev); return 0; -out_oom_notify: +out_del_vqs: vdev->config->del_vqs(vdev); out_free_vb: kfree(vb); @@ -567,6 +606,8 @@ static void virtballoon_remove(struct virtio_device *vdev) cancel_work_sync(&vb->update_balloon_stats_work); remove_common(vb); + if (vb->vb_dev_info.inode) + iput(vb->vb_dev_info.inode); kfree(vb); } diff --git a/include/linux/balloon_compaction.h b/include/linux/balloon_compaction.h index 9b0a15d06a4f..4c693bf3abdf 100644 --- a/include/linux/balloon_compaction.h +++ b/include/linux/balloon_compaction.h @@ -48,6 +48,7 @@ #include <linux/migrate.h> #include <linux/gfp.h> #include <linux/err.h> +#include <linux/fs.h> /* * Balloon device information descriptor. @@ -62,6 +63,7 @@ struct balloon_dev_info { struct list_head pages; /* Pages enqueued & handled to Host */ int (*migratepage)(struct balloon_dev_info *, struct page *newpage, struct page *page, enum migrate_mode mode); + struct inode *inode; }; extern struct page *balloon_page_enqueue(struct balloon_dev_info *b_dev_info); @@ -73,45 +75,19 @@ static inline void balloon_devinfo_init(struct balloon_dev_info *balloon) spin_lock_init(&balloon->pages_lock); INIT_LIST_HEAD(&balloon->pages); balloon->migratepage = NULL; + balloon->inode = NULL; } #ifdef CONFIG_BALLOON_COMPACTION -extern bool balloon_page_isolate(struct page *page); +extern const struct address_space_operations balloon_aops; +extern bool balloon_page_isolate(struct page *page, + isolate_mode_t mode); extern void balloon_page_putback(struct page *page); -extern int balloon_page_migrate(struct page *newpage, +extern int balloon_page_migrate(struct address_space *mapping, + struct page *newpage, struct page *page, enum migrate_mode mode); /* - * __is_movable_balloon_page - helper to perform @page PageBalloon tests - */ -static inline bool __is_movable_balloon_page(struct page *page) -{ - return PageBalloon(page); -} - -/* - * balloon_page_movable - test PageBalloon to identify balloon pages - * and PagePrivate to check that the page is not - * isolated and can be moved by compaction/migration. - * - * As we might return false positives in the case of a balloon page being just - * released under us, this need to be re-tested later, under the page lock. - */ -static inline bool balloon_page_movable(struct page *page) -{ - return PageBalloon(page) && PagePrivate(page); -} - -/* - * isolated_balloon_page - identify an isolated balloon page on private - * compaction/migration page lists. - */ -static inline bool isolated_balloon_page(struct page *page) -{ - return PageBalloon(page); -} - -/* * balloon_page_insert - insert a page into the balloon's page list and make * the page->private assignment accordingly. * @balloon : pointer to balloon device @@ -123,8 +99,7 @@ static inline bool isolated_balloon_page(struct page *page) static inline void balloon_page_insert(struct balloon_dev_info *balloon, struct page *page) { - __SetPageBalloon(page); - SetPagePrivate(page); + __SetPageBalloon(page, balloon->inode->i_mapping); set_page_private(page, (unsigned long)balloon); list_add(&page->lru, &balloon->pages); } @@ -141,10 +116,8 @@ static inline void balloon_page_delete(struct page *page) { __ClearPageBalloon(page); set_page_private(page, 0); - if (PagePrivate(page)) { - ClearPagePrivate(page); + if (!PageIsolated(page)) list_del(&page->lru); - } } /* @@ -166,7 +139,7 @@ static inline gfp_t balloon_mapping_gfp_mask(void) static inline void balloon_page_insert(struct balloon_dev_info *balloon, struct page *page) { - __SetPageBalloon(page); + __SetPageBalloon(page, NULL); list_add(&page->lru, &balloon->pages); } diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h index 77ebf8fdbc6e..603c47752126 100644 --- a/include/linux/page-flags.h +++ b/include/linux/page-flags.h @@ -599,32 +599,13 @@ static inline void __ClearPageBuddy(struct page *page) extern bool is_free_buddy_page(struct page *page); -#define PAGE_BALLOON_MAPCOUNT_VALUE (-256) - -static inline int PageBalloon(struct page *page) -{ - return atomic_read(&page->_mapcount) == PAGE_BALLOON_MAPCOUNT_VALUE; -} - -static inline void __SetPageBalloon(struct page *page) -{ - VM_BUG_ON_PAGE(atomic_read(&page->_mapcount) != -1, page); - atomic_set(&page->_mapcount, PAGE_BALLOON_MAPCOUNT_VALUE); -} - -static inline void __ClearPageBalloon(struct page *page) -{ - VM_BUG_ON_PAGE(!PageBalloon(page), page); - atomic_set(&page->_mapcount, -1); -} - -#define PAGE_MOVABLE_MAPCOUNT_VALUE (-255) +#define PAGE_MOVABLE_MAPCOUNT_VALUE (-256) +#define PAGE_BALLOON_MAPCOUNT_VALUE PAGE_MOVABLE_MAPCOUNT_VALUE static inline int PageMovable(struct page *page) { - return ((test_bit(PG_movable, &(page)->flags) && - atomic_read(&page->_mapcount) == PAGE_MOVABLE_MAPCOUNT_VALUE) - || PageBalloon(page)); + return (test_bit(PG_movable, &(page)->flags) && + atomic_read(&page->_mapcount) == PAGE_MOVABLE_MAPCOUNT_VALUE); } /* Caller should hold a PG_lock */ @@ -645,6 +626,35 @@ static inline void __ClearPageMovable(struct page *page) PAGEFLAG(Isolated, isolated, PF_ANY); +static inline int PageBalloon(struct page *page) +{ + return atomic_read(&page->_mapcount) == PAGE_BALLOON_MAPCOUNT_VALUE + && PagePrivate2(page); +} + +static inline void __SetPageBalloon(struct page *page, + struct address_space *mapping) +{ + VM_BUG_ON_PAGE(atomic_read(&page->_mapcount) != -1, page); +#ifdef CONFIG_BALLOON_COMPACTION + __SetPageMovable(page, mapping); +#else + atomic_set(&page->_mapcount, PAGE_BALLOON_MAPCOUNT_VALUE); +#endif + SetPagePrivate2(page); +} + +static inline void __ClearPageBalloon(struct page *page) +{ + VM_BUG_ON_PAGE(!PageBalloon(page), page); +#ifdef CONFIG_BALLOON_COMPACTION + __ClearPageMovable(page); +#else + atomic_set(&page->_mapcount, -1); +#endif + ClearPagePrivate2(page); +} + /* * If network-based swap is enabled, sl*b must keep track of whether pages * were allocated from pfmemalloc reserves. diff --git a/include/uapi/linux/magic.h b/include/uapi/linux/magic.h index 0de181ad73d5..e1fbe72c39c0 100644 --- a/include/uapi/linux/magic.h +++ b/include/uapi/linux/magic.h @@ -78,5 +78,6 @@ #define BTRFS_TEST_MAGIC 0x73727279 #define NSFS_MAGIC 0x6e736673 #define BPF_FS_MAGIC 0xcafe4a11 +#define BALLOON_KVM_MAGIC 0x13661366 #endif /* __LINUX_MAGIC_H__ */ diff --git a/mm/balloon_compaction.c b/mm/balloon_compaction.c index 57b3e9bd6bc5..1fbc7fb387bb 100644 --- a/mm/balloon_compaction.c +++ b/mm/balloon_compaction.c @@ -70,7 +70,7 @@ struct page *balloon_page_dequeue(struct balloon_dev_info *b_dev_info) */ if (trylock_page(page)) { #ifdef CONFIG_BALLOON_COMPACTION - if (!PagePrivate(page)) { + if (PageIsolated(page)) { /* raced with isolation */ unlock_page(page); continue; @@ -106,110 +106,53 @@ EXPORT_SYMBOL_GPL(balloon_page_dequeue); #ifdef CONFIG_BALLOON_COMPACTION -static inline void __isolate_balloon_page(struct page *page) +/* __isolate_lru_page() counterpart for a ballooned page */ +bool balloon_page_isolate(struct page *page, isolate_mode_t mode) { struct balloon_dev_info *b_dev_info = balloon_page_device(page); unsigned long flags; spin_lock_irqsave(&b_dev_info->pages_lock, flags); - ClearPagePrivate(page); list_del(&page->lru); b_dev_info->isolated_pages++; spin_unlock_irqrestore(&b_dev_info->pages_lock, flags); + SetPageIsolated(page); + + return true; } -static inline void __putback_balloon_page(struct page *page) +/* putback_lru_page() counterpart for a ballooned page */ +void balloon_page_putback(struct page *page) { struct balloon_dev_info *b_dev_info = balloon_page_device(page); unsigned long flags; + ClearPageIsolated(page); spin_lock_irqsave(&b_dev_info->pages_lock, flags); - SetPagePrivate(page); list_add(&page->lru, &b_dev_info->pages); b_dev_info->isolated_pages--; spin_unlock_irqrestore(&b_dev_info->pages_lock, flags); } -/* __isolate_lru_page() counterpart for a ballooned page */ -bool balloon_page_isolate(struct page *page) -{ - /* - * Avoid burning cycles with pages that are yet under __free_pages(), - * or just got freed under us. - * - * In case we 'win' a race for a balloon page being freed under us and - * raise its refcount preventing __free_pages() from doing its job - * the put_page() at the end of this block will take care of - * release this page, thus avoiding a nasty leakage. - */ - if (likely(get_page_unless_zero(page))) { - /* - * As balloon pages are not isolated from LRU lists, concurrent - * compaction threads can race against page migration functions - * as well as race against the balloon driver releasing a page. - * - * In order to avoid having an already isolated balloon page - * being (wrongly) re-isolated while it is under migration, - * or to avoid attempting to isolate pages being released by - * the balloon driver, lets be sure we have the page lock - * before proceeding with the balloon page isolation steps. - */ - if (likely(trylock_page(page))) { - /* - * A ballooned page, by default, has PagePrivate set. - * Prevent concurrent compaction threads from isolating - * an already isolated balloon page by clearing it. - */ - if (balloon_page_movable(page)) { - __isolate_balloon_page(page); - unlock_page(page); - return true; - } - unlock_page(page); - } - put_page(page); - } - return false; -} - -/* putback_lru_page() counterpart for a ballooned page */ -void balloon_page_putback(struct page *page) -{ - /* - * 'lock_page()' stabilizes the page and prevents races against - * concurrent isolation threads attempting to re-isolate it. - */ - lock_page(page); - - if (__is_movable_balloon_page(page)) { - __putback_balloon_page(page); - /* drop the extra ref count taken for page isolation */ - put_page(page); - } else { - WARN_ON(1); - dump_page(page, "not movable balloon page"); - } - unlock_page(page); -} - /* move_to_new_page() counterpart for a ballooned page */ -int balloon_page_migrate(struct page *newpage, - struct page *page, enum migrate_mode mode) +int balloon_page_migrate(struct address_space *mapping, + struct page *newpage, struct page *page, + enum migrate_mode mode) { struct balloon_dev_info *balloon = balloon_page_device(page); - int rc = -EAGAIN; VM_BUG_ON_PAGE(!PageLocked(page), page); VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); + VM_BUG_ON_PAGE(!PageMovable(page), page); + VM_BUG_ON_PAGE(!PageIsolated(page), page); - if (WARN_ON(!__is_movable_balloon_page(page))) { - dump_page(page, "not movable balloon page"); - return rc; - } - - if (balloon && balloon->migratepage) - rc = balloon->migratepage(balloon, newpage, page, mode); - - return rc; + return balloon->migratepage(balloon, newpage, page, mode); } + +const struct address_space_operations balloon_aops = { + .migratepage = balloon_page_migrate, + .isolate_page = balloon_page_isolate, + .putback_page = balloon_page_putback, +}; +EXPORT_SYMBOL_GPL(balloon_aops); #endif /* CONFIG_BALLOON_COMPACTION */ diff --git a/mm/compaction.c b/mm/compaction.c index 7557aedddaee..e336c620fd7b 100644 --- a/mm/compaction.c +++ b/mm/compaction.c @@ -708,13 +708,6 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, */ is_lru = PageLRU(page); if (!is_lru) { - if (unlikely(balloon_page_movable(page))) { - if (balloon_page_isolate(page)) { - /* Successfully isolated */ - goto isolate_success; - } - } - if (unlikely(PageMovable(page)) && !PageIsolated(page)) { if (locked) { diff --git a/mm/migrate.c b/mm/migrate.c index b56bf2b3fe8c..028814625eea 100644 --- a/mm/migrate.c +++ b/mm/migrate.c @@ -157,8 +157,8 @@ void putback_movable_page(struct page *page) * from where they were once taken off for compaction/migration. * * This function shall be used whenever the isolated pageset has been - * built from lru, balloon, hugetlbfs page. See isolate_migratepages_range() - * and isolate_huge_page(). + * built from lru, movable, hugetlbfs page. + * See isolate_migratepages_range() and isolate_huge_page(). */ void putback_movable_pages(struct list_head *l) { @@ -173,9 +173,7 @@ void putback_movable_pages(struct list_head *l) list_del(&page->lru); dec_zone_page_state(page, NR_ISOLATED_ANON + page_is_file_cache(page)); - if (unlikely(isolated_balloon_page(page))) { - balloon_page_putback(page); - } else if (unlikely(PageMovable(page))) { + if (unlikely(PageMovable(page))) { if (PageIsolated(page)) { putback_movable_page(page); } else { @@ -977,18 +975,6 @@ static int __unmap_and_move(struct page *page, struct page *newpage, if (unlikely(!trylock_page(newpage))) goto out_unlock; - if (unlikely(isolated_balloon_page(page))) { - /* - * A ballooned page does not need any special attention from - * physical to virtual reverse mapping procedures. - * Skip any attempt to unmap PTEs or to remap swap cache, - * in order to avoid burning cycles at rmap level, and perform - * the page migration right away (proteced by page lock). - */ - rc = balloon_page_migrate(newpage, page, mode); - goto out_unlock_both; - } - /* * Corner case handling: * 1. When a new swap-cache page is read into, it is added to the LRU @@ -1033,7 +1019,7 @@ static int __unmap_and_move(struct page *page, struct page *newpage, out: /* If migration is successful, move newpage to right list */ if (rc == MIGRATEPAGE_SUCCESS) { - if (unlikely(__is_movable_balloon_page(newpage))) + if (unlikely(PageMovable(newpage))) put_page(newpage); else putback_lru_page(newpage); diff --git a/mm/vmscan.c b/mm/vmscan.c index d82196244340..c7696a2e11c7 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -1254,7 +1254,7 @@ unsigned long reclaim_clean_pages_from_list(struct zone *zone, list_for_each_entry_safe(page, next, page_list, lru) { if (page_is_file_cache(page) && !PageDirty(page) && - !isolated_balloon_page(page)) { + !PageIsolated(page)) { ClearPageActive(page); list_move(&page->lru, &clean_pages); } -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 05/16] zsmalloc: keep max_object in size_class
Every zspage in a size_class has same number of max objects so we could move it to a size_class. Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 32 +++++++++++++++----------------- 1 file changed, 15 insertions(+), 17 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index a0890e9003e2..8649d0243e6c 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -32,8 +32,6 @@ * page->freelist: points to the first free object in zspage. * Free objects are linked together using in-place * metadata. - * page->objects: maximum number of objects we can store in this - * zspage (class->zspage_order * PAGE_SIZE / class->size) * page->lru: links together first pages of various zspages. * Basically forming list of zspages in a fullness group. * page->mapping: class index and fullness group of the zspage @@ -211,6 +209,7 @@ struct size_class { * of ZS_ALIGN. */ int size; + int objs_per_zspage; unsigned int index; struct zs_size_stat stats; @@ -627,21 +626,22 @@ static inline void zs_pool_stat_destroy(struct zs_pool *pool) * the pool (not yet implemented). This function returns fullness * status of the given page. */ -static enum fullness_group get_fullness_group(struct page *first_page) +static enum fullness_group get_fullness_group(struct size_class *class, + struct page *first_page) { - int inuse, max_objects; + int inuse, objs_per_zspage; enum fullness_group fg; VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); inuse = first_page->inuse; - max_objects = first_page->objects; + objs_per_zspage = class->objs_per_zspage; if (inuse == 0) fg = ZS_EMPTY; - else if (inuse == max_objects) + else if (inuse == objs_per_zspage) fg = ZS_FULL; - else if (inuse <= 3 * max_objects / fullness_threshold_frac) + else if (inuse <= 3 * objs_per_zspage / fullness_threshold_frac) fg = ZS_ALMOST_EMPTY; else fg = ZS_ALMOST_FULL; @@ -728,7 +728,7 @@ static enum fullness_group fix_fullness_group(struct size_class *class, enum fullness_group currfg, newfg; get_zspage_mapping(first_page, &class_idx, &currfg); - newfg = get_fullness_group(first_page); + newfg = get_fullness_group(class, first_page); if (newfg == currfg) goto out; @@ -1008,9 +1008,6 @@ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) init_zspage(class, first_page); first_page->freelist = location_to_obj(first_page, 0); - /* Maximum number of objects we can store in this zspage */ - first_page->objects = class->pages_per_zspage * PAGE_SIZE / class->size; - error = 0; /* Success */ cleanup: @@ -1238,11 +1235,11 @@ static bool can_merge(struct size_class *prev, int size, int pages_per_zspage) return true; } -static bool zspage_full(struct page *first_page) +static bool zspage_full(struct size_class *class, struct page *first_page) { VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - return first_page->inuse == first_page->objects; + return first_page->inuse == class->objs_per_zspage; } unsigned long zs_get_total_pages(struct zs_pool *pool) @@ -1628,7 +1625,7 @@ static int migrate_zspage(struct zs_pool *pool, struct size_class *class, } /* Stop if there is no more space */ - if (zspage_full(d_page)) { + if (zspage_full(class, d_page)) { unpin_tag(handle); ret = -ENOMEM; break; @@ -1687,7 +1684,7 @@ static enum fullness_group putback_zspage(struct zs_pool *pool, { enum fullness_group fullness; - fullness = get_fullness_group(first_page); + fullness = get_fullness_group(class, first_page); insert_zspage(class, fullness, first_page); set_zspage_mapping(first_page, class->index, fullness); @@ -1936,8 +1933,9 @@ struct zs_pool *zs_create_pool(const char *name, gfp_t flags) class->size = size; class->index = i; class->pages_per_zspage = pages_per_zspage; - if (pages_per_zspage == 1 && - get_maxobj_per_zspage(size, pages_per_zspage) == 1) + class->objs_per_zspage = class->pages_per_zspage * + PAGE_SIZE / class->size; + if (pages_per_zspage == 1 && class->objs_per_zspage == 1) class->huge = true; spin_lock_init(&class->lock); pool->size_class[i] = class; -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 06/16] zsmalloc: squeeze inuse into page->mapping
Currently, we store class:fullness into page->mapping. The number of class we can support is 255 and fullness is 4 so (8 + 2 = 10bit) is enough to represent them. Meanwhile, the bits we need to store in-use objects in zspage is that 11bit is enough. For example, If we assume that 64K PAGE_SIZE, class_size 32 which is worst case, class->pages_per_zspage become 1 so the number of objects in zspage is 2048 so 11bit is enough. The next class is 32 + 256(i.e., ZS_SIZE_CLASS_DELTA). With worst case that ZS_MAX_PAGES_PER_ZSPAGE, 64K * 4 / (32 + 256) = 910 so 11bit is still enough. So, we could squeeze inuse object count to page->mapping. Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 103 ++++++++++++++++++++++++++++++++++++++++------------------ 1 file changed, 71 insertions(+), 32 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index 8649d0243e6c..4dd72a803568 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -34,8 +34,7 @@ * metadata. * page->lru: links together first pages of various zspages. * Basically forming list of zspages in a fullness group. - * page->mapping: class index and fullness group of the zspage - * page->inuse: the number of objects that are used in this zspage + * page->mapping: override by struct zs_meta * * Usage of struct page flags: * PG_private: identifies the first component page @@ -132,6 +131,13 @@ /* each chunk includes extra space to keep handle */ #define ZS_MAX_ALLOC_SIZE PAGE_SIZE +#define CLASS_BITS 8 +#define CLASS_MASK ((1 << CLASS_BITS) - 1) +#define FULLNESS_BITS 2 +#define FULLNESS_MASK ((1 << FULLNESS_BITS) - 1) +#define INUSE_BITS 11 +#define INUSE_MASK ((1 << INUSE_BITS) - 1) + /* * On systems with 4K page size, this gives 255 size classes! There is a * trader-off here: @@ -145,7 +151,7 @@ * ZS_MIN_ALLOC_SIZE and ZS_SIZE_CLASS_DELTA must be multiple of ZS_ALIGN * (reason above) */ -#define ZS_SIZE_CLASS_DELTA (PAGE_SIZE >> 8) +#define ZS_SIZE_CLASS_DELTA (PAGE_SIZE >> CLASS_BITS) /* * We do not maintain any list for completely empty or full pages @@ -155,7 +161,7 @@ enum fullness_group { ZS_ALMOST_EMPTY, _ZS_NR_FULLNESS_GROUPS, - ZS_EMPTY, + ZS_EMPTY = _ZS_NR_FULLNESS_GROUPS, ZS_FULL }; @@ -263,14 +269,11 @@ struct zs_pool { #endif }; -/* - * A zspage's class index and fullness group - * are encoded in its (first)page->mapping - */ -#define CLASS_IDX_BITS 28 -#define FULLNESS_BITS 4 -#define CLASS_IDX_MASK ((1 << CLASS_IDX_BITS) - 1) -#define FULLNESS_MASK ((1 << FULLNESS_BITS) - 1) +struct zs_meta { + unsigned long class:CLASS_BITS; + unsigned long fullness:FULLNESS_BITS; + unsigned long inuse:INUSE_BITS; +}; struct mapping_area { #ifdef CONFIG_PGTABLE_MAPPING @@ -412,28 +415,61 @@ static int is_last_page(struct page *page) return PagePrivate2(page); } +static int get_zspage_inuse(struct page *first_page) +{ + struct zs_meta *m; + + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + + m = (struct zs_meta *)&first_page->mapping; + + return m->inuse; +} + +static void set_zspage_inuse(struct page *first_page, int val) +{ + struct zs_meta *m; + + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + + m = (struct zs_meta *)&first_page->mapping; + m->inuse = val; +} + +static void mod_zspage_inuse(struct page *first_page, int val) +{ + struct zs_meta *m; + + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + + m = (struct zs_meta *)&first_page->mapping; + m->inuse += val; +} + static void get_zspage_mapping(struct page *first_page, unsigned int *class_idx, enum fullness_group *fullness) { - unsigned long m; + struct zs_meta *m; + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = (unsigned long)first_page->mapping; - *fullness = m & FULLNESS_MASK; - *class_idx = (m >> FULLNESS_BITS) & CLASS_IDX_MASK; + m = (struct zs_meta *)&first_page->mapping; + *fullness = m->fullness; + *class_idx = m->class; } static void set_zspage_mapping(struct page *first_page, unsigned int class_idx, enum fullness_group fullness) { - unsigned long m; + struct zs_meta *m; + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = ((class_idx & CLASS_IDX_MASK) << FULLNESS_BITS) | - (fullness & FULLNESS_MASK); - first_page->mapping = (struct address_space *)m; + m = (struct zs_meta *)&first_page->mapping; + m->fullness = fullness; + m->class = class_idx; } /* @@ -632,9 +668,7 @@ static enum fullness_group get_fullness_group(struct size_class *class, int inuse, objs_per_zspage; enum fullness_group fg; - VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - - inuse = first_page->inuse; + inuse = get_zspage_inuse(first_page); objs_per_zspage = class->objs_per_zspage; if (inuse == 0) @@ -677,10 +711,10 @@ static void insert_zspage(struct size_class *class, /* * We want to see more ZS_FULL pages and less almost - * empty/full. Put pages with higher ->inuse first. + * empty/full. Put pages with higher inuse first. */ list_add_tail(&first_page->lru, &(*head)->lru); - if (first_page->inuse >= (*head)->inuse) + if (get_zspage_inuse(first_page) >= get_zspage_inuse(*head)) *head = first_page; } @@ -896,7 +930,7 @@ static void free_zspage(struct page *first_page) struct page *nextp, *tmp, *head_extra; VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - VM_BUG_ON_PAGE(first_page->inuse, first_page); + VM_BUG_ON_PAGE(get_zspage_inuse(first_page), first_page); head_extra = (struct page *)page_private(first_page); @@ -992,7 +1026,7 @@ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) SetPagePrivate(page); set_page_private(page, 0); first_page = page; - first_page->inuse = 0; + set_zspage_inuse(page, 0); } if (i == 1) set_page_private(first_page, (unsigned long)page); @@ -1237,9 +1271,7 @@ static bool can_merge(struct size_class *prev, int size, int pages_per_zspage) static bool zspage_full(struct size_class *class, struct page *first_page) { - VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - - return first_page->inuse == class->objs_per_zspage; + return get_zspage_inuse(first_page) == class->objs_per_zspage; } unsigned long zs_get_total_pages(struct zs_pool *pool) @@ -1372,7 +1404,7 @@ static unsigned long obj_malloc(struct size_class *class, /* record handle in first_page->private */ set_page_private(first_page, handle); kunmap_atomic(vaddr); - first_page->inuse++; + mod_zspage_inuse(first_page, 1); zs_stat_inc(class, OBJ_USED, 1); return obj; @@ -1457,7 +1489,7 @@ static void obj_free(struct size_class *class, unsigned long obj) set_page_private(first_page, 0); kunmap_atomic(vaddr); first_page->freelist = (void *)obj; - first_page->inuse--; + mod_zspage_inuse(first_page, -1); zs_stat_dec(class, OBJ_USED, 1); } @@ -2002,6 +2034,13 @@ static int __init zs_init(void) if (ret) goto notifier_fail; + /* + * A zspage's class index, fullness group, inuse object count are + * encoded in its (first)page->mapping so sizeof(struct zs_meta) + * should be less than sizeof(page->mapping(i.e., unsigned long)). + */ + BUILD_BUG_ON(sizeof(struct zs_meta) > sizeof(unsigned long)); + init_zs_size_classes(); #ifdef CONFIG_ZPOOL -- 1.9.1
We don't use page->_mapcount any more so no need to reset. Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 1 - 1 file changed, 1 deletion(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index 4dd72a803568..0f6cce9b9119 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -922,7 +922,6 @@ static void reset_page(struct page *page) set_page_private(page, 0); page->mapping = NULL; page->freelist = NULL; - page_mapcount_reset(page); } static void free_zspage(struct page *first_page) -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 08/16] zsmalloc: squeeze freelist into page->mapping
Zsmalloc stores first free object's position into first_page->freelist in each zspage. If we change it with object index from first_page instead of location, we could squeeze it into page->mapping because the number of bit we need to store offset is at most 11bit. Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 158 +++++++++++++++++++++++++++++++++++----------------------- 1 file changed, 96 insertions(+), 62 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index 0f6cce9b9119..807998462539 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -18,9 +18,7 @@ * Usage of struct page fields: * page->private: points to the first component (0-order) page * page->index (union with page->freelist): offset of the first object - * starting in this page. For the first page, this is - * always 0, so we use this field (aka freelist) to point - * to the first free object in zspage. + * starting in this page. * page->lru: links together all component pages (except the first page) * of a zspage * @@ -29,9 +27,6 @@ * page->private: refers to the component page after the first page * If the page is first_page for huge object, it stores handle. * Look at size_class->huge. - * page->freelist: points to the first free object in zspage. - * Free objects are linked together using in-place - * metadata. * page->lru: links together first pages of various zspages. * Basically forming list of zspages in a fullness group. * page->mapping: override by struct zs_meta @@ -131,6 +126,7 @@ /* each chunk includes extra space to keep handle */ #define ZS_MAX_ALLOC_SIZE PAGE_SIZE +#define FREEOBJ_BITS 11 #define CLASS_BITS 8 #define CLASS_MASK ((1 << CLASS_BITS) - 1) #define FULLNESS_BITS 2 @@ -228,17 +224,17 @@ struct size_class { /* * Placed within free objects to form a singly linked list. - * For every zspage, first_page->freelist gives head of this list. + * For every zspage, first_page->freeobj gives head of this list. * * This must be power of 2 and less than or equal to ZS_ALIGN */ struct link_free { union { /* - * Position of next free chunk (encodes <PFN, obj_idx>) + * free object list * It's valid for non-allocated object */ - void *next; + unsigned long next; /* * Handle of allocated object. */ @@ -270,6 +266,7 @@ struct zs_pool { }; struct zs_meta { + unsigned long freeobj:FREEOBJ_BITS; unsigned long class:CLASS_BITS; unsigned long fullness:FULLNESS_BITS; unsigned long inuse:INUSE_BITS; @@ -446,6 +443,26 @@ static void mod_zspage_inuse(struct page *first_page, int val) m->inuse += val; } +static void set_freeobj(struct page *first_page, int idx) +{ + struct zs_meta *m; + + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + + m = (struct zs_meta *)&first_page->mapping; + m->freeobj = idx; +} + +static unsigned long get_freeobj(struct page *first_page) +{ + struct zs_meta *m; + + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + + m = (struct zs_meta *)&first_page->mapping; + return m->freeobj; +} + static void get_zspage_mapping(struct page *first_page, unsigned int *class_idx, enum fullness_group *fullness) @@ -837,30 +854,33 @@ static struct page *get_next_page(struct page *page) return next; } -/* - * Encode <page, obj_idx> as a single handle value. - * We use the least bit of handle for tagging. - */ -static void *location_to_obj(struct page *page, unsigned long obj_idx) +static void objidx_to_page_and_offset(struct size_class *class, + struct page *first_page, + unsigned long obj_idx, + struct page **obj_page, + unsigned long *offset_in_page) { - unsigned long obj; + int i; + unsigned long offset; + struct page *cursor; + int nr_page; - if (!page) { - VM_BUG_ON(obj_idx); - return NULL; - } + offset = obj_idx * class->size; + cursor = first_page; + nr_page = offset >> PAGE_SHIFT; - obj = page_to_pfn(page) << OBJ_INDEX_BITS; - obj |= ((obj_idx) & OBJ_INDEX_MASK); - obj <<= OBJ_TAG_BITS; + *offset_in_page = offset & ~PAGE_MASK; + + for (i = 0; i < nr_page; i++) + cursor = get_next_page(cursor); - return (void *)obj; + *obj_page = cursor; } -/* - * Decode <page, obj_idx> pair from the given object handle. We adjust the - * decoded obj_idx back to its original value since it was adjusted in - * location_to_obj(). +/** + * obj_to_location - get (<page>, <obj_idx>) from encoded object value + * @page: page object resides in zspage + * @obj_idx: object index */ static void obj_to_location(unsigned long obj, struct page **page, unsigned long *obj_idx) @@ -870,6 +890,23 @@ static void obj_to_location(unsigned long obj, struct page **page, *obj_idx = (obj & OBJ_INDEX_MASK); } +/** + * location_to_obj - get obj value encoded from (<page>, <obj_idx>) + * @page: page object resides in zspage + * @obj_idx: object index + */ +static unsigned long location_to_obj(struct page *page, + unsigned long obj_idx) +{ + unsigned long obj; + + obj = page_to_pfn(page) << OBJ_INDEX_BITS; + obj |= obj_idx & OBJ_INDEX_MASK; + obj <<= OBJ_TAG_BITS; + + return obj; +} + static unsigned long handle_to_obj(unsigned long handle) { return *(unsigned long *)handle; @@ -885,17 +922,6 @@ static unsigned long obj_to_head(struct size_class *class, struct page *page, return *(unsigned long *)obj; } -static unsigned long obj_idx_to_offset(struct page *page, - unsigned long obj_idx, int class_size) -{ - unsigned long off = 0; - - if (!is_first_page(page)) - off = page->index; - - return off + obj_idx * class_size; -} - static inline int trypin_tag(unsigned long handle) { unsigned long *ptr = (unsigned long *)handle; @@ -952,6 +978,7 @@ static void free_zspage(struct page *first_page) /* Initialize a newly allocated zspage */ static void init_zspage(struct size_class *class, struct page *first_page) { + int freeobj = 1; unsigned long off = 0; struct page *page = first_page; @@ -960,14 +987,11 @@ static void init_zspage(struct size_class *class, struct page *first_page) while (page) { struct page *next_page; struct link_free *link; - unsigned int i = 1; void *vaddr; /* * page->index stores offset of first object starting - * in the page. For the first page, this is always 0, - * so we use first_page->index (aka ->freelist) to store - * head of corresponding zspage's freelist. + * in the page. */ if (page != first_page) page->index = off; @@ -976,7 +1000,7 @@ static void init_zspage(struct size_class *class, struct page *first_page) link = (struct link_free *)vaddr + off / sizeof(*link); while ((off += class->size) < PAGE_SIZE) { - link->next = location_to_obj(page, i++); + link->next = freeobj++ << OBJ_ALLOCATED_TAG; link += class->size / sizeof(*link); } @@ -986,11 +1010,21 @@ static void init_zspage(struct size_class *class, struct page *first_page) * page (if present) */ next_page = get_next_page(page); - link->next = location_to_obj(next_page, 0); + if (next_page) { + link->next = freeobj++ << OBJ_ALLOCATED_TAG; + } else { + /* + * Reset OBJ_ALLOCATED_TAG bit to last link for + * migration to know it is allocated object or not. + */ + link->next = -1 << OBJ_ALLOCATED_TAG; + } kunmap_atomic(vaddr); page = next_page; off %= PAGE_SIZE; } + + set_freeobj(first_page, 0); } /* @@ -1040,7 +1074,6 @@ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) init_zspage(class, first_page); - first_page->freelist = location_to_obj(first_page, 0); error = 0; /* Success */ cleanup: @@ -1320,7 +1353,7 @@ void *zs_map_object(struct zs_pool *pool, unsigned long handle, obj_to_location(obj, &page, &obj_idx); get_zspage_mapping(get_first_page(page), &class_idx, &fg); class = pool->size_class[class_idx]; - off = obj_idx_to_offset(page, obj_idx, class->size); + off = (class->size * obj_idx) & ~PAGE_MASK; area = &get_cpu_var(zs_map_area); area->vm_mm = mm; @@ -1359,7 +1392,7 @@ void zs_unmap_object(struct zs_pool *pool, unsigned long handle) obj_to_location(obj, &page, &obj_idx); get_zspage_mapping(get_first_page(page), &class_idx, &fg); class = pool->size_class[class_idx]; - off = obj_idx_to_offset(page, obj_idx, class->size); + off = (class->size * obj_idx) & ~PAGE_MASK; area = this_cpu_ptr(&zs_map_area); if (off + class->size <= PAGE_SIZE) @@ -1385,17 +1418,17 @@ static unsigned long obj_malloc(struct size_class *class, struct link_free *link; struct page *m_page; - unsigned long m_objidx, m_offset; + unsigned long m_offset; void *vaddr; handle |= OBJ_ALLOCATED_TAG; - obj = (unsigned long)first_page->freelist; - obj_to_location(obj, &m_page, &m_objidx); - m_offset = obj_idx_to_offset(m_page, m_objidx, class->size); + obj = get_freeobj(first_page); + objidx_to_page_and_offset(class, first_page, obj, + &m_page, &m_offset); vaddr = kmap_atomic(m_page); link = (struct link_free *)vaddr + m_offset / sizeof(*link); - first_page->freelist = link->next; + set_freeobj(first_page, link->next >> OBJ_ALLOCATED_TAG); if (!class->huge) /* record handle in the header of allocated chunk */ link->handle = handle; @@ -1406,6 +1439,8 @@ static unsigned long obj_malloc(struct size_class *class, mod_zspage_inuse(first_page, 1); zs_stat_inc(class, OBJ_USED, 1); + obj = location_to_obj(m_page, obj); + return obj; } @@ -1475,19 +1510,17 @@ static void obj_free(struct size_class *class, unsigned long obj) obj &= ~OBJ_ALLOCATED_TAG; obj_to_location(obj, &f_page, &f_objidx); + f_offset = (class->size * f_objidx) & ~PAGE_MASK; first_page = get_first_page(f_page); - - f_offset = obj_idx_to_offset(f_page, f_objidx, class->size); - vaddr = kmap_atomic(f_page); /* Insert this object in containing zspage's freelist */ link = (struct link_free *)(vaddr + f_offset); - link->next = first_page->freelist; + link->next = get_freeobj(first_page) << OBJ_ALLOCATED_TAG; if (class->huge) set_page_private(first_page, 0); kunmap_atomic(vaddr); - first_page->freelist = (void *)obj; + set_freeobj(first_page, f_objidx); mod_zspage_inuse(first_page, -1); zs_stat_dec(class, OBJ_USED, 1); } @@ -1543,8 +1576,8 @@ static void zs_object_copy(struct size_class *class, unsigned long dst, obj_to_location(src, &s_page, &s_objidx); obj_to_location(dst, &d_page, &d_objidx); - s_off = obj_idx_to_offset(s_page, s_objidx, class->size); - d_off = obj_idx_to_offset(d_page, d_objidx, class->size); + s_off = (class->size * s_objidx) & ~PAGE_MASK; + d_off = (class->size * d_objidx) & ~PAGE_MASK; if (s_off + class->size > PAGE_SIZE) s_size = PAGE_SIZE - s_off; @@ -2034,9 +2067,10 @@ static int __init zs_init(void) goto notifier_fail; /* - * A zspage's class index, fullness group, inuse object count are - * encoded in its (first)page->mapping so sizeof(struct zs_meta) - * should be less than sizeof(page->mapping(i.e., unsigned long)). + * A zspage's a free object index, class index, fullness group, + * inuse object count are encoded in its (first)page->mapping + * so sizeof(struct zs_meta) should be less than + * sizeof(page->mapping(i.e., unsigned long)). */ BUILD_BUG_ON(sizeof(struct zs_meta) > sizeof(unsigned long)); -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 09/16] zsmalloc: move struct zs_meta from mapping to freelist
For supporting migration from VM, we need to have address_space on every page so zsmalloc shouldn't use page->mapping. So, this patch moves zs_meta from mapping to freelist. Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index 807998462539..d4d33a819832 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -29,7 +29,7 @@ * Look at size_class->huge. * page->lru: links together first pages of various zspages. * Basically forming list of zspages in a fullness group. - * page->mapping: override by struct zs_meta + * page->freelist: override by struct zs_meta * * Usage of struct page flags: * PG_private: identifies the first component page @@ -418,7 +418,7 @@ static int get_zspage_inuse(struct page *first_page) VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = (struct zs_meta *)&first_page->mapping; + m = (struct zs_meta *)&first_page->freelist; return m->inuse; } @@ -429,7 +429,7 @@ static void set_zspage_inuse(struct page *first_page, int val) VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = (struct zs_meta *)&first_page->mapping; + m = (struct zs_meta *)&first_page->freelist; m->inuse = val; } @@ -439,7 +439,7 @@ static void mod_zspage_inuse(struct page *first_page, int val) VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = (struct zs_meta *)&first_page->mapping; + m = (struct zs_meta *)&first_page->freelist; m->inuse += val; } @@ -449,7 +449,7 @@ static void set_freeobj(struct page *first_page, int idx) VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = (struct zs_meta *)&first_page->mapping; + m = (struct zs_meta *)&first_page->freelist; m->freeobj = idx; } @@ -459,7 +459,7 @@ static unsigned long get_freeobj(struct page *first_page) VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = (struct zs_meta *)&first_page->mapping; + m = (struct zs_meta *)&first_page->freelist; return m->freeobj; } @@ -471,7 +471,7 @@ static void get_zspage_mapping(struct page *first_page, VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = (struct zs_meta *)&first_page->mapping; + m = (struct zs_meta *)&first_page->freelist; *fullness = m->fullness; *class_idx = m->class; } @@ -484,7 +484,7 @@ static void set_zspage_mapping(struct page *first_page, VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); - m = (struct zs_meta *)&first_page->mapping; + m = (struct zs_meta *)&first_page->freelist; m->fullness = fullness; m->class = class_idx; } @@ -946,7 +946,6 @@ static void reset_page(struct page *page) clear_bit(PG_private, &page->flags); clear_bit(PG_private_2, &page->flags); set_page_private(page, 0); - page->mapping = NULL; page->freelist = NULL; } @@ -1056,6 +1055,7 @@ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) INIT_LIST_HEAD(&page->lru); if (i == 0) { /* first page */ + page->freelist = NULL; SetPagePrivate(page); set_page_private(page, 0); first_page = page; @@ -2068,9 +2068,9 @@ static int __init zs_init(void) /* * A zspage's a free object index, class index, fullness group, - * inuse object count are encoded in its (first)page->mapping + * inuse object count are encoded in its (first)page->freelist * so sizeof(struct zs_meta) should be less than - * sizeof(page->mapping(i.e., unsigned long)). + * sizeof(page->freelist(i.e., void *)). */ BUILD_BUG_ON(sizeof(struct zs_meta) > sizeof(unsigned long)); -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 10/16] zsmalloc: factor page chain functionality out
For migration, we need to create sub-page chain of zspage dynamically so this patch factors it out from alloc_zspage. As a minor refactoring, it makes OBJ_ALLOCATED_TAG assign more clear in obj_malloc(it could be another patch but it's trivial so I want to put together in this patch). Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 80 ++++++++++++++++++++++++++++++++++------------------------- 1 file changed, 46 insertions(+), 34 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index d4d33a819832..14bcc741eead 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -981,7 +981,9 @@ static void init_zspage(struct size_class *class, struct page *first_page) unsigned long off = 0; struct page *page = first_page; - VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + first_page->freelist = NULL; + INIT_LIST_HEAD(&first_page->lru); + set_zspage_inuse(first_page, 0); while (page) { struct page *next_page; @@ -1026,13 +1028,44 @@ static void init_zspage(struct size_class *class, struct page *first_page) set_freeobj(first_page, 0); } +static void create_page_chain(struct page *pages[], int nr_pages) +{ + int i; + struct page *page; + struct page *prev_page = NULL; + struct page *first_page = NULL; + + for (i = 0; i < nr_pages; i++) { + page = pages[i]; + + INIT_LIST_HEAD(&page->lru); + if (i == 0) { + SetPagePrivate(page); + set_page_private(page, 0); + first_page = page; + } + + if (i == 1) + set_page_private(first_page, (unsigned long)page); + if (i >= 1) + set_page_private(page, (unsigned long)first_page); + if (i >= 2) + list_add(&page->lru, &prev_page->lru); + if (i == nr_pages - 1) + SetPagePrivate2(page); + + prev_page = page; + } +} + /* * Allocate a zspage for the given size class */ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) { - int i, error; - struct page *first_page = NULL, *uninitialized_var(prev_page); + int i; + struct page *first_page = NULL; + struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE]; /* * Allocate individual pages and link them together as: @@ -1045,43 +1078,23 @@ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) * (i.e. no other sub-page has this flag set) and PG_private_2 to * identify the last page. */ - error = -ENOMEM; for (i = 0; i < class->pages_per_zspage; i++) { struct page *page; page = alloc_page(flags); - if (!page) - goto cleanup; - - INIT_LIST_HEAD(&page->lru); - if (i == 0) { /* first page */ - page->freelist = NULL; - SetPagePrivate(page); - set_page_private(page, 0); - first_page = page; - set_zspage_inuse(page, 0); + if (!page) { + while (--i >= 0) + __free_page(pages[i]); + return NULL; } - if (i == 1) - set_page_private(first_page, (unsigned long)page); - if (i >= 1) - set_page_private(page, (unsigned long)first_page); - if (i >= 2) - list_add(&page->lru, &prev_page->lru); - if (i == class->pages_per_zspage - 1) /* last page */ - SetPagePrivate2(page); - prev_page = page; + + pages[i] = page; } + create_page_chain(pages, class->pages_per_zspage); + first_page = pages[0]; init_zspage(class, first_page); - error = 0; /* Success */ - -cleanup: - if (unlikely(error) && first_page) { - free_zspage(first_page); - first_page = NULL; - } - return first_page; } @@ -1421,7 +1434,6 @@ static unsigned long obj_malloc(struct size_class *class, unsigned long m_offset; void *vaddr; - handle |= OBJ_ALLOCATED_TAG; obj = get_freeobj(first_page); objidx_to_page_and_offset(class, first_page, obj, &m_page, &m_offset); @@ -1431,10 +1443,10 @@ static unsigned long obj_malloc(struct size_class *class, set_freeobj(first_page, link->next >> OBJ_ALLOCATED_TAG); if (!class->huge) /* record handle in the header of allocated chunk */ - link->handle = handle; + link->handle = handle | OBJ_ALLOCATED_TAG; else /* record handle in first_page->private */ - set_page_private(first_page, handle); + set_page_private(first_page, handle | OBJ_ALLOCATED_TAG); kunmap_atomic(vaddr); mod_zspage_inuse(first_page, 1); zs_stat_inc(class, OBJ_USED, 1); -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 11/16] zsmalloc: separate free_zspage from putback_zspage
Currently, putback_zspage does free zspage under class->lock if fullness become ZS_EMPTY but it makes trouble to implement locking scheme for new zspage migration. So, this patch is to separate free_zspage from putback_zspage and free zspage out of class->lock which is preparation for zspage migration. Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 46 +++++++++++++++++++++++----------------------- 1 file changed, 23 insertions(+), 23 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index 14bcc741eead..b11dcd718502 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -949,7 +949,8 @@ static void reset_page(struct page *page) page->freelist = NULL; } -static void free_zspage(struct page *first_page) +static void free_zspage(struct zs_pool *pool, struct size_class *class, + struct page *first_page) { struct page *nextp, *tmp, *head_extra; @@ -972,6 +973,11 @@ static void free_zspage(struct page *first_page) } reset_page(head_extra); __free_page(head_extra); + + zs_stat_dec(class, OBJ_ALLOCATED, get_maxobj_per_zspage( + class->size, class->pages_per_zspage)); + atomic_long_sub(class->pages_per_zspage, + &pool->pages_allocated); } /* Initialize a newly allocated zspage */ @@ -1559,13 +1565,8 @@ void zs_free(struct zs_pool *pool, unsigned long handle) spin_lock(&class->lock); obj_free(class, obj); fullness = fix_fullness_group(class, first_page); - if (fullness == ZS_EMPTY) { - zs_stat_dec(class, OBJ_ALLOCATED, get_maxobj_per_zspage( - class->size, class->pages_per_zspage)); - atomic_long_sub(class->pages_per_zspage, - &pool->pages_allocated); - free_zspage(first_page); - } + if (fullness == ZS_EMPTY) + free_zspage(pool, class, first_page); spin_unlock(&class->lock); unpin_tag(handle); @@ -1752,7 +1753,7 @@ static struct page *isolate_target_page(struct size_class *class) * @class: destination class * @first_page: target page * - * Return @fist_page's fullness_group + * Return @first_page's updated fullness_group */ static enum fullness_group putback_zspage(struct zs_pool *pool, struct size_class *class, @@ -1764,15 +1765,6 @@ static enum fullness_group putback_zspage(struct zs_pool *pool, insert_zspage(class, fullness, first_page); set_zspage_mapping(first_page, class->index, fullness); - if (fullness == ZS_EMPTY) { - zs_stat_dec(class, OBJ_ALLOCATED, get_maxobj_per_zspage( - class->size, class->pages_per_zspage)); - atomic_long_sub(class->pages_per_zspage, - &pool->pages_allocated); - - free_zspage(first_page); - } - return fullness; } @@ -1835,23 +1827,31 @@ static void __zs_compact(struct zs_pool *pool, struct size_class *class) if (!migrate_zspage(pool, class, &cc)) break; - putback_zspage(pool, class, dst_page); + VM_BUG_ON_PAGE(putback_zspage(pool, class, + dst_page) == ZS_EMPTY, dst_page); } /* Stop if we couldn't find slot */ if (dst_page == NULL) break; - putback_zspage(pool, class, dst_page); - if (putback_zspage(pool, class, src_page) == ZS_EMPTY) + VM_BUG_ON_PAGE(putback_zspage(pool, class, + dst_page) == ZS_EMPTY, dst_page); + if (putback_zspage(pool, class, src_page) == ZS_EMPTY) { pool->stats.pages_compacted += class->pages_per_zspage; - spin_unlock(&class->lock); + spin_unlock(&class->lock); + free_zspage(pool, class, src_page); + } else { + spin_unlock(&class->lock); + } + cond_resched(); spin_lock(&class->lock); } if (src_page) - putback_zspage(pool, class, src_page); + VM_BUG_ON_PAGE(putback_zspage(pool, class, + src_page) == ZS_EMPTY, src_page); spin_unlock(&class->lock); } -- 1.9.1
Currently, we rely on class->lock to prevent zspage destruction. It was okay until now because the critical section is short but with run-time migration, it could be long so class->lock is not a good apporach any more. So, this patch introduces [un]freeze_zspage functions which freeze allocated objects in the zspage with pinning tag so user cannot free using object. With those functions, this patch redesign compaction. Those functions will be used for implementing zspage runtime migrations, too. Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 393 ++++++++++++++++++++++++++++++++++++++-------------------- 1 file changed, 257 insertions(+), 136 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index b11dcd718502..ac8ca7b10720 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -922,6 +922,13 @@ static unsigned long obj_to_head(struct size_class *class, struct page *page, return *(unsigned long *)obj; } +static inline int testpin_tag(unsigned long handle) +{ + unsigned long *ptr = (unsigned long *)handle; + + return test_bit(HANDLE_PIN_BIT, ptr); +} + static inline int trypin_tag(unsigned long handle) { unsigned long *ptr = (unsigned long *)handle; @@ -949,8 +956,7 @@ static void reset_page(struct page *page) page->freelist = NULL; } -static void free_zspage(struct zs_pool *pool, struct size_class *class, - struct page *first_page) +static void free_zspage(struct zs_pool *pool, struct page *first_page) { struct page *nextp, *tmp, *head_extra; @@ -973,11 +979,6 @@ static void free_zspage(struct zs_pool *pool, struct size_class *class, } reset_page(head_extra); __free_page(head_extra); - - zs_stat_dec(class, OBJ_ALLOCATED, get_maxobj_per_zspage( - class->size, class->pages_per_zspage)); - atomic_long_sub(class->pages_per_zspage, - &pool->pages_allocated); } /* Initialize a newly allocated zspage */ @@ -1325,6 +1326,11 @@ static bool zspage_full(struct size_class *class, struct page *first_page) return get_zspage_inuse(first_page) == class->objs_per_zspage; } +static bool zspage_empty(struct size_class *class, struct page *first_page) +{ + return get_zspage_inuse(first_page) == 0; +} + unsigned long zs_get_total_pages(struct zs_pool *pool) { return atomic_long_read(&pool->pages_allocated); @@ -1455,7 +1461,6 @@ static unsigned long obj_malloc(struct size_class *class, set_page_private(first_page, handle | OBJ_ALLOCATED_TAG); kunmap_atomic(vaddr); mod_zspage_inuse(first_page, 1); - zs_stat_inc(class, OBJ_USED, 1); obj = location_to_obj(m_page, obj); @@ -1510,6 +1515,7 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size) } obj = obj_malloc(class, first_page, handle); + zs_stat_inc(class, OBJ_USED, 1); /* Now move the zspage to another fullness group, if required */ fix_fullness_group(class, first_page); record_obj(handle, obj); @@ -1540,7 +1546,6 @@ static void obj_free(struct size_class *class, unsigned long obj) kunmap_atomic(vaddr); set_freeobj(first_page, f_objidx); mod_zspage_inuse(first_page, -1); - zs_stat_dec(class, OBJ_USED, 1); } void zs_free(struct zs_pool *pool, unsigned long handle) @@ -1564,10 +1569,19 @@ void zs_free(struct zs_pool *pool, unsigned long handle) spin_lock(&class->lock); obj_free(class, obj); + zs_stat_dec(class, OBJ_USED, 1); fullness = fix_fullness_group(class, first_page); - if (fullness == ZS_EMPTY) - free_zspage(pool, class, first_page); + if (fullness == ZS_EMPTY) { + zs_stat_dec(class, OBJ_ALLOCATED, get_maxobj_per_zspage( + class->size, class->pages_per_zspage)); + spin_unlock(&class->lock); + atomic_long_sub(class->pages_per_zspage, + &pool->pages_allocated); + free_zspage(pool, first_page); + goto out; + } spin_unlock(&class->lock); +out: unpin_tag(handle); free_handle(pool, handle); @@ -1637,127 +1651,66 @@ static void zs_object_copy(struct size_class *class, unsigned long dst, kunmap_atomic(s_addr); } -/* - * Find alloced object in zspage from index object and - * return handle. - */ -static unsigned long find_alloced_obj(struct size_class *class, - struct page *page, int index) +static unsigned long handle_from_obj(struct size_class *class, + struct page *first_page, int obj_idx) { - unsigned long head; - int offset = 0; - unsigned long handle = 0; - void *addr = kmap_atomic(page); - - if (!is_first_page(page)) - offset = page->index; - offset += class->size * index; - - while (offset < PAGE_SIZE) { - head = obj_to_head(class, page, addr + offset); - if (head & OBJ_ALLOCATED_TAG) { - handle = head & ~OBJ_ALLOCATED_TAG; - if (trypin_tag(handle)) - break; - handle = 0; - } + struct page *page; + unsigned long offset_in_page; + void *addr; + unsigned long head, handle = 0; - offset += class->size; - index++; - } + objidx_to_page_and_offset(class, first_page, obj_idx, + &page, &offset_in_page); + addr = kmap_atomic(page); + head = obj_to_head(class, page, addr + offset_in_page); + if (head & OBJ_ALLOCATED_TAG) + handle = head & ~OBJ_ALLOCATED_TAG; kunmap_atomic(addr); + return handle; } -struct zs_compact_control { - /* Source page for migration which could be a subpage of zspage. */ - struct page *s_page; - /* Destination page for migration which should be a first page - * of zspage. */ - struct page *d_page; - /* Starting object index within @s_page which used for live object - * in the subpage. */ - int index; -}; - -static int migrate_zspage(struct zs_pool *pool, struct size_class *class, - struct zs_compact_control *cc) +static int migrate_zspage(struct size_class *class, struct page *dst_page, + struct page *src_page) { - unsigned long used_obj, free_obj; unsigned long handle; - struct page *s_page = cc->s_page; - struct page *d_page = cc->d_page; - unsigned long index = cc->index; - int ret = 0; + unsigned long old_obj, new_obj; + int i; + int nr_migrated = 0; - while (1) { - handle = find_alloced_obj(class, s_page, index); - if (!handle) { - s_page = get_next_page(s_page); - if (!s_page) - break; - index = 0; + for (i = 0; i < class->objs_per_zspage; i++) { + handle = handle_from_obj(class, src_page, i); + if (!handle) continue; - } - - /* Stop if there is no more space */ - if (zspage_full(class, d_page)) { - unpin_tag(handle); - ret = -ENOMEM; + if (zspage_full(class, dst_page)) break; - } - - used_obj = handle_to_obj(handle); - free_obj = obj_malloc(class, d_page, handle); - zs_object_copy(class, free_obj, used_obj); - index++; + old_obj = handle_to_obj(handle); + new_obj = obj_malloc(class, dst_page, handle); + zs_object_copy(class, new_obj, old_obj); + nr_migrated++; /* * record_obj updates handle's value to free_obj and it will * invalidate lock bit(ie, HANDLE_PIN_BIT) of handle, which * breaks synchronization using pin_tag(e,g, zs_free) so * let's keep the lock bit. */ - free_obj |= BIT(HANDLE_PIN_BIT); - record_obj(handle, free_obj); - unpin_tag(handle); - obj_free(class, used_obj); + new_obj |= BIT(HANDLE_PIN_BIT); + record_obj(handle, new_obj); + obj_free(class, old_obj); } - - /* Remember last position in this iteration */ - cc->s_page = s_page; - cc->index = index; - - return ret; -} - -static struct page *isolate_target_page(struct size_class *class) -{ - int i; - struct page *page; - - for (i = 0; i < _ZS_NR_FULLNESS_GROUPS; i++) { - page = class->fullness_list[i]; - if (page) { - remove_zspage(class, i, page); - break; - } - } - - return page; + return nr_migrated; } /* * putback_zspage - add @first_page into right class's fullness list - * @pool: target pool * @class: destination class * @first_page: target page * * Return @first_page's updated fullness_group */ -static enum fullness_group putback_zspage(struct zs_pool *pool, - struct size_class *class, - struct page *first_page) +static enum fullness_group putback_zspage(struct size_class *class, + struct page *first_page) { enum fullness_group fullness; @@ -1768,17 +1721,155 @@ static enum fullness_group putback_zspage(struct zs_pool *pool, return fullness; } +/* + * freeze_zspage - freeze all objects in a zspage + * @class: size class of the page + * @first_page: first page of zspage + * + * Freeze all allocated objects in a zspage so objects couldn't be + * freed until unfreeze objects. It should be called under class->lock. + * + * RETURNS: + * the number of pinned objects + */ +static int freeze_zspage(struct size_class *class, struct page *first_page) +{ + unsigned long obj_idx; + struct page *obj_page; + unsigned long offset; + void *addr; + int nr_freeze = 0; + + for (obj_idx = 0; obj_idx < class->objs_per_zspage; obj_idx++) { + unsigned long head; + + objidx_to_page_and_offset(class, first_page, obj_idx, + &obj_page, &offset); + addr = kmap_atomic(obj_page); + head = obj_to_head(class, obj_page, addr + offset); + if (head & OBJ_ALLOCATED_TAG) { + unsigned long handle = head & ~OBJ_ALLOCATED_TAG; + + if (!trypin_tag(handle)) { + kunmap_atomic(addr); + break; + } + nr_freeze++; + } + kunmap_atomic(addr); + } + + return nr_freeze; +} + +/* + * unfreeze_page - unfreeze objects freezed by freeze_zspage in a zspage + * @class: size class of the page + * @first_page: freezed zspage to unfreeze + * @nr_obj: the number of objects to unfreeze + * + * unfreeze objects in a zspage. + */ +static void unfreeze_zspage(struct size_class *class, struct page *first_page, + int nr_obj) +{ + unsigned long obj_idx; + struct page *obj_page; + unsigned long offset; + void *addr; + int nr_unfreeze = 0; + + for (obj_idx = 0; obj_idx < class->objs_per_zspage && + nr_unfreeze < nr_obj; obj_idx++) { + unsigned long head; + + objidx_to_page_and_offset(class, first_page, obj_idx, + &obj_page, &offset); + addr = kmap_atomic(obj_page); + head = obj_to_head(class, obj_page, addr + offset); + if (head & OBJ_ALLOCATED_TAG) { + unsigned long handle = head & ~OBJ_ALLOCATED_TAG; + + VM_BUG_ON(!testpin_tag(handle)); + unpin_tag(handle); + nr_unfreeze++; + } + kunmap_atomic(addr); + } +} + +/* + * isolate_source_page - isolate a zspage for migration source + * @class: size class of zspage for isolation + * + * Returns a zspage which are isolated from list so anyone can + * allocate a object from that page. As well, freeze all objects + * allocated in the zspage so anyone cannot access that objects + * (e.g., zs_map_object, zs_free). + */ static struct page *isolate_source_page(struct size_class *class) { int i; struct page *page = NULL; for (i = ZS_ALMOST_EMPTY; i >= ZS_ALMOST_FULL; i--) { + int inuse, freezed; + page = class->fullness_list[i]; if (!page) continue; remove_zspage(class, i, page); + + inuse = get_zspage_inuse(page); + freezed = freeze_zspage(class, page); + + if (inuse != freezed) { + unfreeze_zspage(class, page, freezed); + putback_zspage(class, page); + page = NULL; + continue; + } + + break; + } + + return page; +} + +/* + * isolate_target_page - isolate a zspage for migration target + * @class: size class of zspage for isolation + * + * Returns a zspage which are isolated from list so anyone can + * allocate a object from that page. As well, freeze all objects + * allocated in the zspage so anyone cannot access that objects + * (e.g., zs_map_object, zs_free). + */ +static struct page *isolate_target_page(struct size_class *class) +{ + int i; + struct page *page; + + for (i = 0; i < _ZS_NR_FULLNESS_GROUPS; i++) { + int inuse, freezed; + + page = class->fullness_list[i]; + if (!page) + continue; + + remove_zspage(class, i, page); + + inuse = get_zspage_inuse(page); + freezed = freeze_zspage(class, page); + + if (inuse != freezed) { + unfreeze_zspage(class, page, freezed); + putback_zspage(class, page); + page = NULL; + continue; + } + break; } @@ -1793,9 +1884,11 @@ static struct page *isolate_source_page(struct size_class *class) static unsigned long zs_can_compact(struct size_class *class) { unsigned long obj_wasted; + unsigned long obj_allocated, obj_used; - obj_wasted = zs_stat_get(class, OBJ_ALLOCATED) - - zs_stat_get(class, OBJ_USED); + obj_allocated = zs_stat_get(class, OBJ_ALLOCATED); + obj_used = zs_stat_get(class, OBJ_USED); + obj_wasted = obj_allocated - obj_used; obj_wasted /= get_maxobj_per_zspage(class->size, class->pages_per_zspage); @@ -1805,53 +1898,81 @@ static unsigned long zs_can_compact(struct size_class *class) static void __zs_compact(struct zs_pool *pool, struct size_class *class) { - struct zs_compact_control cc; - struct page *src_page; + struct page *src_page = NULL; struct page *dst_page = NULL; - spin_lock(&class->lock); - while ((src_page = isolate_source_page(class))) { + while (1) { + int nr_migrated; - if (!zs_can_compact(class)) + spin_lock(&class->lock); + if (!zs_can_compact(class)) { + spin_unlock(&class->lock); break; + } - cc.index = 0; - cc.s_page = src_page; + /* + * Isolate source page and freeze all objects in a zspage + * to prevent zspage destroying. + */ + if (!src_page) { + src_page = isolate_source_page(class); + if (!src_page) { + spin_unlock(&class->lock); + break; + } + } - while ((dst_page = isolate_target_page(class))) { - cc.d_page = dst_page; - /* - * If there is no more space in dst_page, resched - * and see if anyone had allocated another zspage. - */ - if (!migrate_zspage(pool, class, &cc)) + /* Isolate target page and freeze all objects in the zspage */ + if (!dst_page) { + dst_page = isolate_target_page(class); + if (!dst_page) { + spin_unlock(&class->lock); break; + } + } + spin_unlock(&class->lock); + + nr_migrated = migrate_zspage(class, dst_page, src_page); - VM_BUG_ON_PAGE(putback_zspage(pool, class, - dst_page) == ZS_EMPTY, dst_page); + if (zspage_full(class, dst_page)) { + spin_lock(&class->lock); + putback_zspage(class, dst_page); + unfreeze_zspage(class, dst_page, + class->objs_per_zspage); + spin_unlock(&class->lock); + dst_page = NULL; } - /* Stop if we couldn't find slot */ - if (dst_page == NULL) - break; + if (zspage_empty(class, src_page)) { + free_zspage(pool, src_page); + spin_lock(&class->lock); + zs_stat_dec(class, OBJ_ALLOCATED, + get_maxobj_per_zspage( + class->size, class->pages_per_zspage)); + atomic_long_sub(class->pages_per_zspage, + &pool->pages_allocated); - VM_BUG_ON_PAGE(putback_zspage(pool, class, - dst_page) == ZS_EMPTY, dst_page); - if (putback_zspage(pool, class, src_page) == ZS_EMPTY) { pool->stats.pages_compacted += class->pages_per_zspage; spin_unlock(&class->lock); - free_zspage(pool, class, src_page); - } else { - spin_unlock(&class->lock); + src_page = NULL; } + } - cond_resched(); - spin_lock(&class->lock); + if (!src_page && !dst_page) + return; + + spin_lock(&class->lock); + if (src_page) { + putback_zspage(class, src_page); + unfreeze_zspage(class, src_page, + class->objs_per_zspage); } - if (src_page) - VM_BUG_ON_PAGE(putback_zspage(pool, class, - src_page) == ZS_EMPTY, src_page); + if (dst_page) { + putback_zspage(class, dst_page); + unfreeze_zspage(class, dst_page, + class->objs_per_zspage); + } spin_unlock(&class->lock); } -- 1.9.1
This patch introduces run-time migration feature for zspage. To begin with, it supports only head page migration for easy review(later patches will support tail page migration). For migration, it supports three functions * zs_page_isolate It isolates a zspage which includes a subpage VM want to migrate from class so anyone cannot allocate new object from the zspage. IOW, allocation freeze * zs_page_migrate First of all, it freezes zspage to prevent zspage destrunction so anyone cannot free object. Then, It copies content from oldpage to newpage and create new page-chain with new page. If it was successful, drop the refcount of old page to free and putback new zspage to right data structure of zsmalloc. Lastly, unfreeze zspages so we allows object allocation/free from now on. * zs_page_putback It returns isolated zspage to right fullness_group list if it fails to migrate a page. NOTE: A hurdle to support migration is that destroying zspage while migration is going on. Once a zspage is isolated, anyone cannot allocate object from the zspage but can deallocate object freely so a zspage could be destroyed until all of objects in zspage are freezed to prevent deallocation. The problem is large window betwwen zs_page_isolate and freeze_zspage in zs_page_migrate so the zspage could be destroyed. A easy approach to solve the problem is that object freezing in zs_page_isolate but it has a drawback that any object cannot be deallocated until migration fails after isolation. However, There is large time gab between isolation and migration so any object freeing in other CPU should spin by pin_tag which would cause big latency. So, this patch introduces lock_zspage which holds PG_lock of all pages in a zspage right before freeing the zspage. VM migration locks the page, too right before calling ->migratepage so such race doesn't exist any more. Signed-off-by: Minchan Kim <minchan at kernel.org> --- include/uapi/linux/magic.h | 1 + mm/zsmalloc.c | 332 +++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 318 insertions(+), 15 deletions(-) diff --git a/include/uapi/linux/magic.h b/include/uapi/linux/magic.h index e1fbe72c39c0..93b1affe4801 100644 --- a/include/uapi/linux/magic.h +++ b/include/uapi/linux/magic.h @@ -79,5 +79,6 @@ #define NSFS_MAGIC 0x6e736673 #define BPF_FS_MAGIC 0xcafe4a11 #define BALLOON_KVM_MAGIC 0x13661366 +#define ZSMALLOC_MAGIC 0x58295829 #endif /* __LINUX_MAGIC_H__ */ diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index ac8ca7b10720..f6c9138c3be0 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -56,6 +56,8 @@ #include <linux/debugfs.h> #include <linux/zsmalloc.h> #include <linux/zpool.h> +#include <linux/mount.h> +#include <linux/migrate.h> /* * This must be power of 2 and greater than of equal to sizeof(link_free). @@ -182,6 +184,8 @@ struct zs_size_stat { static struct dentry *zs_stat_root; #endif +static struct vfsmount *zsmalloc_mnt; + /* * number of size_classes */ @@ -263,6 +267,7 @@ struct zs_pool { #ifdef CONFIG_ZSMALLOC_STAT struct dentry *stat_dentry; #endif + struct inode *inode; }; struct zs_meta { @@ -412,6 +417,29 @@ static int is_last_page(struct page *page) return PagePrivate2(page); } +/* + * Indicate that whether zspage is isolated for page migration. + * Protected by size_class lock + */ +static void SetZsPageIsolate(struct page *first_page) +{ + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + SetPageUptodate(first_page); +} + +static int ZsPageIsolate(struct page *first_page) +{ + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + + return PageUptodate(first_page); +} + +static void ClearZsPageIsolate(struct page *first_page) +{ + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + ClearPageUptodate(first_page); +} + static int get_zspage_inuse(struct page *first_page) { struct zs_meta *m; @@ -783,8 +811,11 @@ static enum fullness_group fix_fullness_group(struct size_class *class, if (newfg == currfg) goto out; - remove_zspage(class, currfg, first_page); - insert_zspage(class, newfg, first_page); + /* Later, putback will insert page to right list */ + if (!ZsPageIsolate(first_page)) { + remove_zspage(class, currfg, first_page); + insert_zspage(class, newfg, first_page); + } set_zspage_mapping(first_page, class_idx, newfg); out: @@ -950,12 +981,31 @@ static void unpin_tag(unsigned long handle) static void reset_page(struct page *page) { + if (!PageIsolated(page)) + __ClearPageMovable(page); + ClearPageIsolated(page); clear_bit(PG_private, &page->flags); clear_bit(PG_private_2, &page->flags); set_page_private(page, 0); page->freelist = NULL; } +/** + * lock_zspage - lock all pages in the zspage + * @first_page: head page of the zspage + * + * To prevent destroy during migration, zspage freeing should + * hold locks of all pages in a zspage + */ +void lock_zspage(struct page *first_page) +{ + struct page *cursor = first_page; + + do { + while (!trylock_page(cursor)); + } while ((cursor = get_next_page(cursor)) != NULL); +} + static void free_zspage(struct zs_pool *pool, struct page *first_page) { struct page *nextp, *tmp, *head_extra; @@ -963,26 +1013,31 @@ static void free_zspage(struct zs_pool *pool, struct page *first_page) VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); VM_BUG_ON_PAGE(get_zspage_inuse(first_page), first_page); + lock_zspage(first_page); head_extra = (struct page *)page_private(first_page); - reset_page(first_page); - __free_page(first_page); - /* zspage with only 1 system page */ if (!head_extra) - return; + goto out; list_for_each_entry_safe(nextp, tmp, &head_extra->lru, lru) { list_del(&nextp->lru); reset_page(nextp); - __free_page(nextp); + unlock_page(nextp); + put_page(nextp); } reset_page(head_extra); - __free_page(head_extra); + unlock_page(head_extra); + put_page(head_extra); +out: + reset_page(first_page); + unlock_page(first_page); + put_page(first_page); } /* Initialize a newly allocated zspage */ -static void init_zspage(struct size_class *class, struct page *first_page) +static void init_zspage(struct size_class *class, struct page *first_page, + struct address_space *mapping) { int freeobj = 1; unsigned long off = 0; @@ -991,6 +1046,9 @@ static void init_zspage(struct size_class *class, struct page *first_page) first_page->freelist = NULL; INIT_LIST_HEAD(&first_page->lru); set_zspage_inuse(first_page, 0); + BUG_ON(!trylock_page(first_page)); + __SetPageMovable(first_page, mapping); + unlock_page(first_page); while (page) { struct page *next_page; @@ -1065,10 +1123,45 @@ static void create_page_chain(struct page *pages[], int nr_pages) } } +static void replace_sub_page(struct size_class *class, struct page *first_page, + struct page *newpage, struct page *oldpage) +{ + struct page *page; + struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE] = {NULL,}; + int idx = 0; + + page = first_page; + do { + if (page == oldpage) + pages[idx] = newpage; + else + pages[idx] = page; + idx++; + } while ((page = get_next_page(page)) != NULL); + + create_page_chain(pages, class->pages_per_zspage); + + if (is_first_page(oldpage)) { + enum fullness_group fg; + int class_idx; + + SetZsPageIsolate(newpage); + get_zspage_mapping(oldpage, &class_idx, &fg); + set_zspage_mapping(newpage, class_idx, fg); + set_freeobj(newpage, get_freeobj(oldpage)); + set_zspage_inuse(newpage, get_zspage_inuse(oldpage)); + if (class->huge) + set_page_private(newpage, page_private(oldpage)); + } + + __SetPageMovable(newpage, oldpage->mapping); +} + /* * Allocate a zspage for the given size class */ -static struct page *alloc_zspage(struct size_class *class, gfp_t flags) +static struct page *alloc_zspage(struct zs_pool *pool, + struct size_class *class) { int i; struct page *first_page = NULL; @@ -1088,7 +1181,7 @@ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) for (i = 0; i < class->pages_per_zspage; i++) { struct page *page; - page = alloc_page(flags); + page = alloc_page(pool->flags); if (!page) { while (--i >= 0) __free_page(pages[i]); @@ -1100,7 +1193,7 @@ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) create_page_chain(pages, class->pages_per_zspage); first_page = pages[0]; - init_zspage(class, first_page); + init_zspage(class, first_page, pool->inode->i_mapping); return first_page; } @@ -1499,7 +1592,7 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size) if (!first_page) { spin_unlock(&class->lock); - first_page = alloc_zspage(class, pool->flags); + first_page = alloc_zspage(pool, class); if (unlikely(!first_page)) { free_handle(pool, handle); return 0; @@ -1559,6 +1652,7 @@ void zs_free(struct zs_pool *pool, unsigned long handle) if (unlikely(!handle)) return; + /* Once handle is pinned, page|object migration cannot work */ pin_tag(handle); obj = handle_to_obj(handle); obj_to_location(obj, &f_page, &f_objidx); @@ -1714,6 +1808,9 @@ static enum fullness_group putback_zspage(struct size_class *class, { enum fullness_group fullness; + VM_BUG_ON_PAGE(!list_empty(&first_page->lru), first_page); + VM_BUG_ON_PAGE(ZsPageIsolate(first_page), first_page); + fullness = get_fullness_group(class, first_page); insert_zspage(class, fullness, first_page); set_zspage_mapping(first_page, class->index, fullness); @@ -2059,6 +2156,173 @@ static int zs_register_shrinker(struct zs_pool *pool) return register_shrinker(&pool->shrinker); } +bool zs_page_isolate(struct page *page, isolate_mode_t mode) +{ + struct zs_pool *pool; + struct size_class *class; + int class_idx; + enum fullness_group fullness; + struct page *first_page; + + /* + * The page is locked so it couldn't be destroyed. + * For detail, look at lock_zspage in free_zspage. + */ + VM_BUG_ON_PAGE(!PageLocked(page), page); + VM_BUG_ON_PAGE(PageIsolated(page), page); + /* + * In this implementation, it allows only first page migration. + */ + VM_BUG_ON_PAGE(!is_first_page(page), page); + first_page = page; + + /* + * Without class lock, fullness is meaningless while constant + * class_idx is okay. We will get it under class lock at below, + * again. + */ + get_zspage_mapping(first_page, &class_idx, &fullness); + pool = page->mapping->private_data; + class = pool->size_class[class_idx]; + + if (!spin_trylock(&class->lock)) + return false; + + get_zspage_mapping(first_page, &class_idx, &fullness); + remove_zspage(class, fullness, first_page); + SetZsPageIsolate(first_page); + SetPageIsolated(page); + spin_unlock(&class->lock); + + return true; +} + +int zs_page_migrate(struct address_space *mapping, struct page *newpage, + struct page *page, enum migrate_mode mode) +{ + struct zs_pool *pool; + struct size_class *class; + int class_idx; + enum fullness_group fullness; + struct page *first_page; + void *s_addr, *d_addr, *addr; + int ret = -EBUSY; + int offset = 0; + int freezed = 0; + + VM_BUG_ON_PAGE(!PageMovable(page), page); + VM_BUG_ON_PAGE(!PageIsolated(page), page); + + first_page = page; + get_zspage_mapping(first_page, &class_idx, &fullness); + pool = page->mapping->private_data; + class = pool->size_class[class_idx]; + + /* + * Get stable fullness under class->lock + */ + if (!spin_trylock(&class->lock)) + return ret; + + get_zspage_mapping(first_page, &class_idx, &fullness); + if (get_zspage_inuse(first_page) == 0) + goto out_class_unlock; + + freezed = freeze_zspage(class, first_page); + if (freezed != get_zspage_inuse(first_page)) + goto out_unfreeze; + + /* copy contents from page to newpage */ + s_addr = kmap_atomic(page); + d_addr = kmap_atomic(newpage); + memcpy(d_addr, s_addr, PAGE_SIZE); + kunmap_atomic(d_addr); + kunmap_atomic(s_addr); + + if (!is_first_page(page)) + offset = page->index; + + addr = kmap_atomic(page); + do { + unsigned long handle; + unsigned long head; + unsigned long new_obj, old_obj; + unsigned long obj_idx; + struct page *dummy; + + head = obj_to_head(class, page, addr + offset); + if (head & OBJ_ALLOCATED_TAG) { + handle = head & ~OBJ_ALLOCATED_TAG; + if (!testpin_tag(handle)) + BUG(); + + old_obj = handle_to_obj(handle); + obj_to_location(old_obj, &dummy, &obj_idx); + new_obj = location_to_obj(newpage, obj_idx); + new_obj |= BIT(HANDLE_PIN_BIT); + record_obj(handle, new_obj); + } + offset += class->size; + } while (offset < PAGE_SIZE); + kunmap_atomic(addr); + + replace_sub_page(class, first_page, newpage, page); + first_page = newpage; + get_page(newpage); + VM_BUG_ON_PAGE(get_fullness_group(class, first_page) =+ ZS_EMPTY, first_page); + ClearZsPageIsolate(first_page); + putback_zspage(class, first_page); + + /* Migration complete. Free old page */ + ClearPageIsolated(page); + reset_page(page); + put_page(page); + ret = MIGRATEPAGE_SUCCESS; + +out_unfreeze: + unfreeze_zspage(class, first_page, freezed); +out_class_unlock: + spin_unlock(&class->lock); + + return ret; +} + +void zs_page_putback(struct page *page) +{ + struct zs_pool *pool; + struct size_class *class; + int class_idx; + enum fullness_group fullness; + struct page *first_page; + + VM_BUG_ON_PAGE(!PageMovable(page), page); + VM_BUG_ON_PAGE(!PageIsolated(page), page); + + first_page = page; + get_zspage_mapping(first_page, &class_idx, &fullness); + pool = page->mapping->private_data; + class = pool->size_class[class_idx]; + + /* + * If there is race betwwen zs_free and here, free_zspage + * in zs_free will wait the page lock of @page without + * destroying of zspage. + */ + INIT_LIST_HEAD(&first_page->lru); + spin_lock(&class->lock); + ClearPageIsolated(page); + ClearZsPageIsolate(first_page); + putback_zspage(class, first_page); + spin_unlock(&class->lock); +} + +const struct address_space_operations zsmalloc_aops = { + .isolate_page = zs_page_isolate, + .migratepage = zs_page_migrate, + .putback_page = zs_page_putback, +}; + /** * zs_create_pool - Creates an allocation pool to work from. * @flags: allocation flags used to allocate pool metadata @@ -2145,6 +2409,15 @@ struct zs_pool *zs_create_pool(const char *name, gfp_t flags) if (zs_pool_stat_create(pool, name)) goto err; + pool->inode = alloc_anon_inode(zsmalloc_mnt->mnt_sb); + if (IS_ERR(pool->inode)) { + pool->inode = NULL; + goto err; + } + + pool->inode->i_mapping->a_ops = &zsmalloc_aops; + pool->inode->i_mapping->private_data = pool; + /* * Not critical, we still can use the pool * and user can trigger compaction manually. @@ -2164,6 +2437,8 @@ void zs_destroy_pool(struct zs_pool *pool) int i; zs_unregister_shrinker(pool); + if (pool->inode) + iput(pool->inode); zs_pool_stat_destroy(pool); for (i = 0; i < zs_size_classes; i++) { @@ -2192,10 +2467,33 @@ void zs_destroy_pool(struct zs_pool *pool) } EXPORT_SYMBOL_GPL(zs_destroy_pool); +static struct dentry *zs_mount(struct file_system_type *fs_type, + int flags, const char *dev_name, void *data) +{ + static const struct dentry_operations ops = { + .d_dname = simple_dname, + }; + + return mount_pseudo(fs_type, "zsmalloc:", NULL, &ops, ZSMALLOC_MAGIC); +} + +static struct file_system_type zsmalloc_fs = { + .name = "zsmalloc", + .mount = zs_mount, + .kill_sb = kill_anon_super, +}; + static int __init zs_init(void) { - int ret = zs_register_cpu_notifier(); + int ret; + + zsmalloc_mnt = kern_mount(&zsmalloc_fs); + if (IS_ERR(zsmalloc_mnt)) { + ret = PTR_ERR(zsmalloc_mnt); + goto out; + } + ret = zs_register_cpu_notifier(); if (ret) goto notifier_fail; @@ -2218,6 +2516,7 @@ static int __init zs_init(void) pr_err("zs stat initialization failed\n"); goto stat_fail; } + return 0; stat_fail: @@ -2226,7 +2525,8 @@ static int __init zs_init(void) #endif notifier_fail: zs_unregister_cpu_notifier(); - + kern_unmount(zsmalloc_mnt); +out: return ret; } @@ -2237,6 +2537,8 @@ static void __exit zs_exit(void) #endif zs_unregister_cpu_notifier(); + kern_unmount(zsmalloc_mnt); + zs_stat_exit(); } -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 14/16] zsmalloc: use single linked list for page chain
For tail page migration, we shouldn't use page->lru which was used for page chaining because VM will use it for own purpose so that we need another field for chaining. For chaining, singly linked list is enough and page->index of tail page to point first object offset in the page could be replaced in run-time calculation. So, this patch change page->lru list for chaining with singly linked list via page->freelist squeeze and introduces get_first_obj_ofs to get first object offset in a page. With that, it could maintain page chaining without using page->lru. Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 119 ++++++++++++++++++++++++++++++++++++++-------------------- 1 file changed, 78 insertions(+), 41 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index f6c9138c3be0..a41cf3ef2077 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -17,10 +17,7 @@ * * Usage of struct page fields: * page->private: points to the first component (0-order) page - * page->index (union with page->freelist): offset of the first object - * starting in this page. - * page->lru: links together all component pages (except the first page) - * of a zspage + * page->index (union with page->freelist): override by struct zs_meta * * For _first_ page only: * @@ -271,10 +268,19 @@ struct zs_pool { }; struct zs_meta { - unsigned long freeobj:FREEOBJ_BITS; - unsigned long class:CLASS_BITS; - unsigned long fullness:FULLNESS_BITS; - unsigned long inuse:INUSE_BITS; + union { + /* first page */ + struct { + unsigned long freeobj:FREEOBJ_BITS; + unsigned long class:CLASS_BITS; + unsigned long fullness:FULLNESS_BITS; + unsigned long inuse:INUSE_BITS; + }; + /* tail pages */ + struct { + struct page *next; + }; + }; }; struct mapping_area { @@ -491,6 +497,34 @@ static unsigned long get_freeobj(struct page *first_page) return m->freeobj; } +static void set_next_page(struct page *page, struct page *next) +{ + struct zs_meta *m; + + VM_BUG_ON_PAGE(is_first_page(page), page); + + m = (struct zs_meta *)&page->index; + m->next = next; +} + +static struct page *get_next_page(struct page *page) +{ + struct page *next; + + if (is_last_page(page)) + next = NULL; + else if (is_first_page(page)) + next = (struct page *)page_private(page); + else { + struct zs_meta *m = (struct zs_meta *)&page->index; + + VM_BUG_ON(!m->next); + next = m->next; + } + + return next; +} + static void get_zspage_mapping(struct page *first_page, unsigned int *class_idx, enum fullness_group *fullness) @@ -871,18 +905,30 @@ static struct page *get_first_page(struct page *page) return (struct page *)page_private(page); } -static struct page *get_next_page(struct page *page) +int get_first_obj_ofs(struct size_class *class, struct page *first_page, + struct page *page) { - struct page *next; + int pos, bound; + int page_idx = 0; + int ofs = 0; + struct page *cursor = first_page; - if (is_last_page(page)) - next = NULL; - else if (is_first_page(page)) - next = (struct page *)page_private(page); - else - next = list_entry(page->lru.next, struct page, lru); + if (first_page == page) + goto out; - return next; + while (page != cursor) { + page_idx++; + cursor = get_next_page(cursor); + } + + bound = PAGE_SIZE * page_idx; + pos = (((class->objs_per_zspage * class->size) * + page_idx / class->pages_per_zspage) / class->size + ) * class->size; + + ofs = (pos + class->size) % PAGE_SIZE; +out: + return ofs; } static void objidx_to_page_and_offset(struct size_class *class, @@ -1008,27 +1054,25 @@ void lock_zspage(struct page *first_page) static void free_zspage(struct zs_pool *pool, struct page *first_page) { - struct page *nextp, *tmp, *head_extra; + struct page *nextp, *tmp; VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); VM_BUG_ON_PAGE(get_zspage_inuse(first_page), first_page); lock_zspage(first_page); - head_extra = (struct page *)page_private(first_page); + nextp = (struct page *)page_private(first_page); /* zspage with only 1 system page */ - if (!head_extra) + if (!nextp) goto out; - list_for_each_entry_safe(nextp, tmp, &head_extra->lru, lru) { - list_del(&nextp->lru); - reset_page(nextp); - unlock_page(nextp); - put_page(nextp); - } - reset_page(head_extra); - unlock_page(head_extra); - put_page(head_extra); + do { + tmp = nextp; + nextp = get_next_page(nextp); + reset_page(tmp); + unlock_page(tmp); + put_page(tmp); + } while (nextp); out: reset_page(first_page); unlock_page(first_page); @@ -1055,13 +1099,6 @@ static void init_zspage(struct size_class *class, struct page *first_page, struct link_free *link; void *vaddr; - /* - * page->index stores offset of first object starting - * in the page. - */ - if (page != first_page) - page->index = off; - vaddr = kmap_atomic(page); link = (struct link_free *)vaddr + off / sizeof(*link); @@ -1103,7 +1140,6 @@ static void create_page_chain(struct page *pages[], int nr_pages) for (i = 0; i < nr_pages; i++) { page = pages[i]; - INIT_LIST_HEAD(&page->lru); if (i == 0) { SetPagePrivate(page); set_page_private(page, 0); @@ -1112,10 +1148,12 @@ static void create_page_chain(struct page *pages[], int nr_pages) if (i == 1) set_page_private(first_page, (unsigned long)page); - if (i >= 1) + if (i >= 1) { + set_next_page(page, NULL); set_page_private(page, (unsigned long)first_page); + } if (i >= 2) - list_add(&page->lru, &prev_page->lru); + set_next_page(prev_page, page); if (i == nr_pages - 1) SetPagePrivate2(page); @@ -2239,8 +2277,7 @@ int zs_page_migrate(struct address_space *mapping, struct page *newpage, kunmap_atomic(d_addr); kunmap_atomic(s_addr); - if (!is_first_page(page)) - offset = page->index; + offset = get_first_obj_ofs(class, first_page, page); addr = kmap_atomic(page); do { -- 1.9.1
This patch enables tail page migration of zspage. In this point, I tested zsmalloc regression with micro-benchmark which does zs_malloc/map/unmap/zs_free for all size class in every CPU(my system is 12) during 20 sec. It shows 1% regression which is really small when we consider the benefit of this feature and realworkload overhead(i.e., most overhead comes from compression). Signed-off-by: Minchan Kim <minchan at kernel.org> --- mm/zsmalloc.c | 129 +++++++++++++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 114 insertions(+), 15 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index a41cf3ef2077..e24f4a160892 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -551,6 +551,19 @@ static void set_zspage_mapping(struct page *first_page, m->class = class_idx; } +static bool check_isolated_page(struct page *first_page) +{ + struct page *cursor; + + for (cursor = first_page; cursor != NULL; cursor + get_next_page(cursor)) { + if (PageIsolated(cursor)) + return true; + } + + return false; +} + /* * zsmalloc divides the pool into various size classes where each * class maintains a list of zspages where each zspage is divided @@ -1052,6 +1065,44 @@ void lock_zspage(struct page *first_page) } while ((cursor = get_next_page(cursor)) != NULL); } +int trylock_zspage(struct page *first_page, struct page *locked_page) +{ + struct page *cursor, *fail; + + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); + + for (cursor = first_page; cursor != NULL; cursor + get_next_page(cursor)) { + if (cursor != locked_page) { + if (!trylock_page(cursor)) { + fail = cursor; + goto unlock; + } + } + } + + return 1; +unlock: + for (cursor = first_page; cursor != fail; cursor + get_next_page(cursor)) { + if (cursor != locked_page) + unlock_page(cursor); + } + + return 0; +} + +void unlock_zspage(struct page *first_page, struct page *locked_page) +{ + struct page *cursor = first_page; + + for (; cursor != NULL; cursor = get_next_page(cursor)) { + VM_BUG_ON_PAGE(!PageLocked(cursor), cursor); + if (cursor != locked_page) + unlock_page(cursor); + } +} + static void free_zspage(struct zs_pool *pool, struct page *first_page) { struct page *nextp, *tmp; @@ -1090,15 +1141,16 @@ static void init_zspage(struct size_class *class, struct page *first_page, first_page->freelist = NULL; INIT_LIST_HEAD(&first_page->lru); set_zspage_inuse(first_page, 0); - BUG_ON(!trylock_page(first_page)); - __SetPageMovable(first_page, mapping); - unlock_page(first_page); while (page) { struct page *next_page; struct link_free *link; void *vaddr; + BUG_ON(!trylock_page(page)); + __SetPageMovable(page, mapping); + unlock_page(page); + vaddr = kmap_atomic(page); link = (struct link_free *)vaddr + off / sizeof(*link); @@ -1848,6 +1900,7 @@ static enum fullness_group putback_zspage(struct size_class *class, VM_BUG_ON_PAGE(!list_empty(&first_page->lru), first_page); VM_BUG_ON_PAGE(ZsPageIsolate(first_page), first_page); + VM_BUG_ON_PAGE(check_isolated_page(first_page), first_page); fullness = get_fullness_group(class, first_page); insert_zspage(class, fullness, first_page); @@ -1954,6 +2007,12 @@ static struct page *isolate_source_page(struct size_class *class) if (!page) continue; + /* To prevent race between object and page migration */ + if (!trylock_zspage(page, NULL)) { + page = NULL; + continue; + } + remove_zspage(class, i, page); inuse = get_zspage_inuse(page); @@ -1962,6 +2021,7 @@ static struct page *isolate_source_page(struct size_class *class) if (inuse != freezed) { unfreeze_zspage(class, page, freezed); putback_zspage(class, page); + unlock_zspage(page, NULL); page = NULL; continue; } @@ -1993,6 +2053,12 @@ static struct page *isolate_target_page(struct size_class *class) if (!page) continue; + /* To prevent race between object and page migration */ + if (!trylock_zspage(page, NULL)) { + page = NULL; + continue; + } + remove_zspage(class, i, page); inuse = get_zspage_inuse(page); @@ -2001,6 +2067,7 @@ static struct page *isolate_target_page(struct size_class *class) if (inuse != freezed) { unfreeze_zspage(class, page, freezed); putback_zspage(class, page); + unlock_zspage(page, NULL); page = NULL; continue; } @@ -2074,11 +2141,13 @@ static void __zs_compact(struct zs_pool *pool, struct size_class *class) putback_zspage(class, dst_page); unfreeze_zspage(class, dst_page, class->objs_per_zspage); + unlock_zspage(dst_page, NULL); spin_unlock(&class->lock); dst_page = NULL; } if (zspage_empty(class, src_page)) { + unlock_zspage(src_page, NULL); free_zspage(pool, src_page); spin_lock(&class->lock); zs_stat_dec(class, OBJ_ALLOCATED, @@ -2101,12 +2170,14 @@ static void __zs_compact(struct zs_pool *pool, struct size_class *class) putback_zspage(class, src_page); unfreeze_zspage(class, src_page, class->objs_per_zspage); + unlock_zspage(src_page, NULL); } if (dst_page) { putback_zspage(class, dst_page); unfreeze_zspage(class, dst_page, class->objs_per_zspage); + unlock_zspage(dst_page, NULL); } spin_unlock(&class->lock); @@ -2209,10 +2280,11 @@ bool zs_page_isolate(struct page *page, isolate_mode_t mode) VM_BUG_ON_PAGE(!PageLocked(page), page); VM_BUG_ON_PAGE(PageIsolated(page), page); /* - * In this implementation, it allows only first page migration. + * first_page will not be destroyed by PG_lock of @page but it could + * be migrated out. For prohibiting it, zs_page_migrate calls + * trylock_zspage so it closes the race. */ - VM_BUG_ON_PAGE(!is_first_page(page), page); - first_page = page; + first_page = get_first_page(page); /* * Without class lock, fullness is meaningless while constant @@ -2226,9 +2298,18 @@ bool zs_page_isolate(struct page *page, isolate_mode_t mode) if (!spin_trylock(&class->lock)) return false; + if (check_isolated_page(first_page)) + goto skip_isolate; + + /* + * If this is first time isolation for zspage, isolate zspage from + * size_class to prevent further allocations from the zspage. + */ get_zspage_mapping(first_page, &class_idx, &fullness); remove_zspage(class, fullness, first_page); SetZsPageIsolate(first_page); + +skip_isolate: SetPageIsolated(page); spin_unlock(&class->lock); @@ -2251,7 +2332,7 @@ int zs_page_migrate(struct address_space *mapping, struct page *newpage, VM_BUG_ON_PAGE(!PageMovable(page), page); VM_BUG_ON_PAGE(!PageIsolated(page), page); - first_page = page; + first_page = get_first_page(page); get_zspage_mapping(first_page, &class_idx, &fullness); pool = page->mapping->private_data; class = pool->size_class[class_idx]; @@ -2266,6 +2347,13 @@ int zs_page_migrate(struct address_space *mapping, struct page *newpage, if (get_zspage_inuse(first_page) == 0) goto out_class_unlock; + /* + * It prevents first_page migration during tail page opeartion for + * get_first_page's stability. + */ + if (!trylock_zspage(first_page, page)) + goto out_class_unlock; + freezed = freeze_zspage(class, first_page); if (freezed != get_zspage_inuse(first_page)) goto out_unfreeze; @@ -2304,21 +2392,26 @@ int zs_page_migrate(struct address_space *mapping, struct page *newpage, kunmap_atomic(addr); replace_sub_page(class, first_page, newpage, page); - first_page = newpage; + first_page = get_first_page(newpage); get_page(newpage); VM_BUG_ON_PAGE(get_fullness_group(class, first_page) = ZS_EMPTY, first_page); - ClearZsPageIsolate(first_page); - putback_zspage(class, first_page); + if (!check_isolated_page(first_page)) { + INIT_LIST_HEAD(&first_page->lru); + ClearZsPageIsolate(first_page); + putback_zspage(class, first_page); + } + /* Migration complete. Free old page */ ClearPageIsolated(page); reset_page(page); put_page(page); ret = MIGRATEPAGE_SUCCESS; - + page = newpage; out_unfreeze: unfreeze_zspage(class, first_page, freezed); + unlock_zspage(first_page, page); out_class_unlock: spin_unlock(&class->lock); @@ -2336,7 +2429,7 @@ void zs_page_putback(struct page *page) VM_BUG_ON_PAGE(!PageMovable(page), page); VM_BUG_ON_PAGE(!PageIsolated(page), page); - first_page = page; + first_page = get_first_page(page); get_zspage_mapping(first_page, &class_idx, &fullness); pool = page->mapping->private_data; class = pool->size_class[class_idx]; @@ -2346,11 +2439,17 @@ void zs_page_putback(struct page *page) * in zs_free will wait the page lock of @page without * destroying of zspage. */ - INIT_LIST_HEAD(&first_page->lru); spin_lock(&class->lock); ClearPageIsolated(page); - ClearZsPageIsolate(first_page); - putback_zspage(class, first_page); + /* + * putback zspage to right list if this is last isolated page + * putback in the zspage. + */ + if (!check_isolated_page(first_page)) { + INIT_LIST_HEAD(&first_page->lru); + ClearZsPageIsolate(first_page); + putback_zspage(class, first_page); + } spin_unlock(&class->lock); } -- 1.9.1
Minchan Kim
2016-Mar-30 07:12 UTC
[PATCH v3 16/16] zram: use __GFP_MOVABLE for memory allocation
Zsmalloc is ready for page migration so zram can use __GFP_MOVABLE from now on. I did test to see how it helps to make higher order pages. Test scenario is as follows. KVM guest, 1G memory, ext4 formated zram block device, for i in `seq 1 8`; do dd if=/dev/vda1 of=mnt/test$i.txt bs=128M count=1 & done wait `pidof dd` for i in `seq 1 2 8`; do rm -rf mnt/test$i.txt done fstrim -v mnt echo "init" cat /proc/buddyinfo echo "compaction" echo 1 > /proc/sys/vm/compact_memory cat /proc/buddyinfo old: init Node 0, zone DMA 208 120 51 41 11 0 0 0 0 0 0 Node 0, zone DMA32 16380 13777 9184 3805 789 54 3 0 0 0 0 compaction Node 0, zone DMA 132 82 40 39 16 2 1 0 0 0 0 Node 0, zone DMA32 5219 5526 4969 3455 1831 677 139 15 0 0 0 new: init Node 0, zone DMA 379 115 97 19 2 0 0 0 0 0 0 Node 0, zone DMA32 18891 16774 10862 3947 637 21 0 0 0 0 0 compaction 1 Node 0, zone DMA 214 66 87 29 10 3 0 0 0 0 0 Node 0, zone DMA32 1612 3139 3154 2469 1745 990 384 94 7 0 0 As you can see, compaction made so many high-order pages. Yay! Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky at gmail.com> Signed-off-by: Minchan Kim <minchan at kernel.org> --- drivers/block/zram/zram_drv.c | 3 ++- mm/zsmalloc.c | 2 +- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c index 370c2f76016d..10f6ff1cf6a0 100644 --- a/drivers/block/zram/zram_drv.c +++ b/drivers/block/zram/zram_drv.c @@ -514,7 +514,8 @@ static struct zram_meta *zram_meta_alloc(char *pool_name, u64 disksize) goto out_error; } - meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM); + meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO|__GFP_HIGHMEM + |__GFP_MOVABLE); if (!meta->mem_pool) { pr_err("Error creating memory pool\n"); goto out_error; diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index e24f4a160892..4b1ccb68f2ee 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -308,7 +308,7 @@ static void destroy_handle_cache(struct zs_pool *pool) static unsigned long alloc_handle(struct zs_pool *pool) { return (unsigned long)kmem_cache_alloc(pool->handle_cachep, - pool->flags & ~__GFP_HIGHMEM); + pool->flags & ~(__GFP_HIGHMEM|__GFP_MOVABLE)); } static void free_handle(struct zs_pool *pool, unsigned long handle) -- 1.9.1
On Wed, 30 Mar 2016 16:11:59 +0900 Minchan Kim <minchan at kernel.org> wrote:> Recently, I got many reports about perfermance degradation > in embedded system(Android mobile phone, webOS TV and so on) > and failed to fork easily. > > The problem was fragmentation caused by zram and GPU driver > pages. Their pages cannot be migrated so compaction cannot > work well, either so reclaimer ends up shrinking all of working > set pages. It made system very slow and even to fail to fork > easily. > > Other pain point is that they cannot work with CMA. > Most of CMA memory space could be idle(ie, it could be used > for movable pages unless driver is using) but if driver(i.e., > zram) cannot migrate his page, that memory space could be > wasted. In our product which has big CMA memory, it reclaims > zones too exccessively although there are lots of free space > in CMA so system was very slow easily. > > To solve these problem, this patch try to add facility to > migrate non-lru pages via introducing new friend functions > of migratepage in address_space_operation and new page flags. > > (isolate_page, putback_page) > (PG_movable, PG_isolated) > > For details, please read description in > "mm/compaction: support non-lru movable page migration".OK, I grabbed all these. I wonder about testing coverage during the -next period. How many people are likely to exercise these code paths in a serious way before it all hits mainline?
On (03/30/16 16:11), Andrew Morton wrote: [..]> > For details, please read description in > > "mm/compaction: support non-lru movable page migration". > > OK, I grabbed all these. > > I wonder about testing coverage during the -next period. How many > people are likely to exercise these code paths in a serious way before > it all hits mainline?I'm hammering the zsmalloc part. -ss
On Wed, Mar 30, 2016 at 04:11:41PM -0700, Andrew Morton wrote:> On Wed, 30 Mar 2016 16:11:59 +0900 Minchan Kim <minchan at kernel.org> wrote: > > > Recently, I got many reports about perfermance degradation > > in embedded system(Android mobile phone, webOS TV and so on) > > and failed to fork easily. > > > > The problem was fragmentation caused by zram and GPU driver > > pages. Their pages cannot be migrated so compaction cannot > > work well, either so reclaimer ends up shrinking all of working > > set pages. It made system very slow and even to fail to fork > > easily. > > > > Other pain point is that they cannot work with CMA. > > Most of CMA memory space could be idle(ie, it could be used > > for movable pages unless driver is using) but if driver(i.e., > > zram) cannot migrate his page, that memory space could be > > wasted. In our product which has big CMA memory, it reclaims > > zones too exccessively although there are lots of free space > > in CMA so system was very slow easily. > > > > To solve these problem, this patch try to add facility to > > migrate non-lru pages via introducing new friend functions > > of migratepage in address_space_operation and new page flags. > > > > (isolate_page, putback_page) > > (PG_movable, PG_isolated) > > > > For details, please read description in > > "mm/compaction: support non-lru movable page migration". > > OK, I grabbed all these. > > I wonder about testing coverage during the -next period. How many > people are likely to exercise these code paths in a serious way before > it all hits mainline?I asked this patchset to production team in my company for stress testing. They alaways catch zram/zsmalloc bugs I have missed so I hope they help me well, too. About ballooning part, I hope Rafael Aquini get a time to review and test it. Other than that, IOW, linux-next will have a enough time to test common migration part modification, I guess. :) Thanks.
Vlastimil Babka
2016-Apr-01 12:58 UTC
[PATCH v3 01/16] mm: use put_page to free page instead of putback_lru_page
On 03/30/2016 09:12 AM, Minchan Kim wrote:> Procedure of page migration is as follows: > > First of all, it should isolate a page from LRU and try to > migrate the page. If it is successful, it releases the page > for freeing. Otherwise, it should put the page back to LRU > list. > > For LRU pages, we have used putback_lru_page for both freeing > and putback to LRU list. It's okay because put_page is aware of > LRU list so if it releases last refcount of the page, it removes > the page from LRU list. However, It makes unnecessary operations > (e.g., lru_cache_add, pagevec and flags operations. It would be > not significant but no worth to do) and harder to support new > non-lru page migration because put_page isn't aware of non-lru > page's data structure. > > To solve the problem, we can add new hook in put_page with > PageMovable flags check but it can increase overhead in > hot path and needs new locking scheme to stabilize the flag check > with put_page. > > So, this patch cleans it up to divide two semantic(ie, put and putback). > If migration is successful, use put_page instead of putback_lru_page and > use putback_lru_page only on failure. That makes code more readable > and doesn't add overhead in put_page. > > Comment from Vlastimil > "Yeah, and compaction (perhaps also other migration users) has to drain > the lru pvec... Getting rid of this stuff is worth even by itself." > > Cc: Mel Gorman <mgorman at suse.de> > Cc: Hugh Dickins <hughd at google.com> > Cc: Naoya Horiguchi <n-horiguchi at ah.jp.nec.com> > Acked-by: Vlastimil Babka <vbabka at suse.cz> > Signed-off-by: Minchan Kim <minchan at kernel.org>[...]> @@ -974,28 +986,28 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page, > list_del(&page->lru); > dec_zone_page_state(page, NR_ISOLATED_ANON + > page_is_file_cache(page)); > - /* Soft-offlined page shouldn't go through lru cache list */ > + } > + > + /* > + * If migration is successful, drop the reference grabbed during > + * isolation. Otherwise, restore the page to LRU list unless we > + * want to retry. > + */ > + if (rc == MIGRATEPAGE_SUCCESS) { > + put_page(page); > if (reason == MR_MEMORY_FAILURE) { > - put_page(page); > if (!test_set_page_hwpoison(page)) > num_poisoned_pages_inc(); > - } else > + }Hmm, I didn't notice it previously, or it's due to rebasing, but it seems that you restricted the memory failure handling (i.e. setting hwpoison) to MIGRATE_SUCCESS, while previously it was done for all non-EAGAIN results. I think that goes against the intention of hwpoison, which is IIRC to catch and kill the poor process that still uses the page? Also (but not your fault) the put_page() preceding test_set_page_hwpoison(page)) IMHO deserves a comment saying which pin we are releasing and which one we still have (hopefully? if I read description of da1b13ccfbebe right) otherwise it looks like doing something with a page that we just potentially freed.> + } else { > + if (rc != -EAGAIN) > putback_lru_page(page); > + if (put_new_page) > + put_new_page(newpage, private); > + else > + put_page(newpage); > } > > - /* > - * If migration was not successful and there's a freeing callback, use > - * it. Otherwise, putback_lru_page() will drop the reference grabbed > - * during isolation. > - */ > - if (put_new_page) > - put_new_page(newpage, private); > - else if (unlikely(__is_movable_balloon_page(newpage))) { > - /* drop our reference, page already in the balloon */ > - put_page(newpage); > - } else > - putback_lru_page(newpage); > - > if (result) { > if (rc) > *result = rc; >
Vlastimil Babka
2016-Apr-01 14:38 UTC
[PATCH v3 03/16] mm: add non-lru movable page support document
On 03/30/2016 09:12 AM, Minchan Kim wrote:> This patch describes what a subsystem should do for non-lru movable > page supporting.Intentionally reading this first without studying the code to better catch things that would seem obvious otherwise.> Cc: Jonathan Corbet <corbet at lwn.net> > Signed-off-by: Minchan Kim <minchan at kernel.org> > --- > Documentation/filesystems/vfs.txt | 11 ++++++- > Documentation/vm/page_migration | 69 ++++++++++++++++++++++++++++++++++++++- > 2 files changed, 78 insertions(+), 2 deletions(-) > > diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt > index 4c1b6c3b4bc8..d63142f8ed7b 100644 > --- a/Documentation/filesystems/vfs.txt > +++ b/Documentation/filesystems/vfs.txt > @@ -752,12 +752,21 @@ struct address_space_operations { > and transfer data directly between the storage and the > application's address space. > > + isolate_page: Called by the VM when isolating a movable non-lru page. > + If page is successfully isolated, we should mark the page as > + PG_isolated via __SetPageIsolated.Patch 02 changelog suggests SetPageIsolated, so this is confusing. I guess the main point is that there might be parallel attempts and only one is allowed to succeed, right? Whether it's done by atomic ops or otherwise doesn't matter to e.g. compaction.> migrate_page: This is used to compact the physical memory usage. > If the VM wants to relocate a page (maybe off a memory card > that is signalling imminent failure) it will pass a new page > and an old page to this function. migrate_page should > transfer any private data across and update any references > - that it has to the page. > + that it has to the page. If migrated page is non-lru page, > + we should clear PG_isolated and PG_movable via __ClearPageIsolated > + and __ClearPageMovable.Similar concern as __SetPageIsolated.> + > + putback_page: Called by the VM when isolated page's migration fails. > + We should clear PG_isolated marked in isolated_page function.Note this kind of wording is less confusing and could be used above wrt my concerns.> > launder_page: Called before freeing a page - it writes back the dirty page. To > prevent redirtying the page, it is kept locked during the whole > diff --git a/Documentation/vm/page_migration b/Documentation/vm/page_migration > index fea5c0864170..c4e7551a414e 100644 > --- a/Documentation/vm/page_migration > +++ b/Documentation/vm/page_migration > @@ -142,5 +142,72 @@ is increased so that the page cannot be freed while page migration occurs. > 20. The new page is moved to the LRU and can be scanned by the swapper > etc again. > > -Christoph Lameter, May 8, 2006. > +C. Non-LRU Page migration > +------------------------- > + > +Although original migration aimed for reducing the latency of memory access > +for NUMA, compaction who want to create high-order page is also main customer. > + > +Ppage migration's disadvantage is that it was designed to migrate only > +*LRU* pages. However, there are potential non-lru movable pages which can be > +migrated in system, for example, zsmalloc, virtio-balloon pages. > +For virtio-balloon pages, some parts of migration code path was hooked up > +and added virtio-balloon specific functions to intercept logi.logi -> logic?> +It's too specific to one subsystem so other subsystem who want to make > +their pages movable should add own specific hooks in migration path.s/should/would have to/ I guess?> +To solve such problem, VM supports non-LRU page migration which provides > +generic functions for non-LRU movable pages without needing subsystem > +specific hook in mm/{migrate|compact}.c. > + > +If a subsystem want to make own pages movable, it should mark pages as > +PG_movable via __SetPageMovable. __SetPageMovable needs address_space for > +argument for register functions which will be called by VM. > + > +Three functions in address_space_operation related to non-lru movable page: > + > + bool (*isolate_page) (struct page *, isolate_mode_t); > + int (*migratepage) (struct address_space *, > + struct page *, struct page *, enum migrate_mode); > + void (*putback_page)(struct page *); > + > +1. Isolation > + > +What VM expected on isolate_page of subsystem is to set PG_isolated flags > +of the page if it was successful. With that, concurrent isolation among > +CPUs skips the isolated page by other CPU earlier. VM calls isolate_page > +under PG_lock of page. If a subsystem cannot isolate the page, it should > +return false.Ah, I see, so it's designed with page lock to handle the concurrent isolations etc. In http://marc.info/?l=linux-mm&m=143816716511904&w=2 Mel has warned about doing this in general under page_lock and suggested that each user handles concurrent calls to isolate_page() internally. Might be more generic that way, even if all current implementers will actually use the page lock. Also it's worth reading that mail in full and incorporating here, as there are more concerns related to concurrency that should be documented, e.g. with pages that can be mapped to userspace. Not a case with zram and balloon pages I guess, but one of Gioh's original use cases was a driver which IIRC could map pages. So the design and documentation should keep that in mind.> +2. Migration > + > +After successful isolation, VM calls migratepage. The migratepage's goal is > +to move content of the old page to new page and set up struct page fields > +of new page. If migration is successful, subsystem should release old page's > +refcount to free. Keep in mind that subsystem should clear PG_movable and > +PG_isolated before releasing the refcount. If everything are done, user > +should return MIGRATEPAGE_SUCCESS. If subsystem cannot migrate the page > +at the moment, migratepage can return -EAGAIN. On -EAGAIN, VM will retry page > +migration because VM interprets -EAGAIN as "temporal migration failure". > + > +3. Putback > + > +If migration was unsuccessful, VM calls putback_page. The subsystem should > +insert isolated page to own data structure again if it has. And subsystem > +should clear PG_isolated which was marked in isolation step. > + > +Note about releasing page: > + > +Subsystem can release pages whenever it want but if it releses the page > +which is already isolated, it should clear PG_isolated but doesn't touch > +PG_movable under PG_lock. Instead of it, VM will clear PG_movable after > +his job done. Otherweise, subsystem should clear both page flags before > +releasing the page.I don't understand this right now. But maybe I will get it after reading the patches and suggest some improved wording here.> + > +Note about PG_isolated: > + > +PG_isolated check on a page is valid only if the page's flag is already > +set to PG_movable.But it's not possible to check both atomically, so I guess it implies checking under page lock? If that's true, should be explicit. Thanks!> +Christoph Lameter, May 8, 2006. > +Minchan Kim, Mar 28, 2016. >
Vlastimil Babka
2016-Apr-01 21:29 UTC
[PATCH v3 02/16] mm/compaction: support non-lru movable page migration
Might have been better as a separate migration patch and then a compaction patch. It's prefixed mm/compaction, but most changed are in mm/migrate.c On 03/30/2016 09:12 AM, Minchan Kim wrote:> We have allowed migration for only LRU pages until now and it was > enough to make high-order pages. But recently, embedded system(e.g., > webOS, android) uses lots of non-movable pages(e.g., zram, GPU memory) > so we have seen several reports about troubles of small high-order > allocation. For fixing the problem, there were several efforts > (e,g,. enhance compaction algorithm, SLUB fallback to 0-order page, > reserved memory, vmalloc and so on) but if there are lots of > non-movable pages in system, their solutions are void in the long run. > > So, this patch is to support facility to change non-movable pages > with movable. For the feature, this patch introduces functions related > to migration to address_space_operations as well as some page flags. > > Basically, this patch supports two page-flags and two functions related > to page migration. The flag and page->mapping stability are protected > by PG_lock. > > PG_movable > PG_isolated > > bool (*isolate_page) (struct page *, isolate_mode_t); > void (*putback_page) (struct page *); > > Duty of subsystem want to make their pages as migratable are > as follows: > > 1. It should register address_space to page->mapping then mark > the page as PG_movable via __SetPageMovable. > > 2. It should mark the page as PG_isolated via SetPageIsolated > if isolation is sucessful and return true.Ah another thing to document (especially in the comments/Doc) is that the subsystem must not expect anything to survive in page.lru (or fields that union it) after having isolated successfully.> 3. If migration is successful, it should clear PG_isolated and > PG_movable of the page for free preparation then release the > reference of the page to free. > > 4. If migration fails, putback function of subsystem should > clear PG_isolated via ClearPageIsolated. > > 5. If a subsystem want to release isolated page, it should > clear PG_isolated but not PG_movable. Instead, VM will do it.Under lock? Or just with ClearPageIsolated?> Cc: Vlastimil Babka <vbabka at suse.cz> > Cc: Mel Gorman <mgorman at suse.de> > Cc: Hugh Dickins <hughd at google.com> > Cc: dri-devel at lists.freedesktop.org > Cc: virtualization at lists.linux-foundation.org > Signed-off-by: Gioh Kim <gurugio at hanmail.net> > Signed-off-by: Minchan Kim <minchan at kernel.org> > --- > Documentation/filesystems/Locking | 4 + > Documentation/filesystems/vfs.txt | 5 + > fs/proc/page.c | 3 + > include/linux/fs.h | 2 + > include/linux/migrate.h | 2 + > include/linux/page-flags.h | 31 ++++++ > include/uapi/linux/kernel-page-flags.h | 1 + > mm/compaction.c | 14 ++- > mm/migrate.c | 174 +++++++++++++++++++++++++++++---- > 9 files changed, 217 insertions(+), 19 deletions(-) > > diff --git a/Documentation/filesystems/Locking b/Documentation/filesystems/Locking > index 619af9bfdcb3..0bb79560abb3 100644 > --- a/Documentation/filesystems/Locking > +++ b/Documentation/filesystems/Locking > @@ -195,7 +195,9 @@ unlocks and drops the reference. > int (*releasepage) (struct page *, int); > void (*freepage)(struct page *); > int (*direct_IO)(struct kiocb *, struct iov_iter *iter, loff_t offset); > + bool (*isolate_page) (struct page *, isolate_mode_t); > int (*migratepage)(struct address_space *, struct page *, struct page *); > + void (*putback_page) (struct page *); > int (*launder_page)(struct page *); > int (*is_partially_uptodate)(struct page *, unsigned long, unsigned long); > int (*error_remove_page)(struct address_space *, struct page *); > @@ -219,7 +221,9 @@ invalidatepage: yes > releasepage: yes > freepage: yes > direct_IO: > +isolate_page: yes > migratepage: yes (both) > +putback_page: yes > launder_page: yes > is_partially_uptodate: yes > error_remove_page: yes > diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt > index b02a7d598258..4c1b6c3b4bc8 100644 > --- a/Documentation/filesystems/vfs.txt > +++ b/Documentation/filesystems/vfs.txt > @@ -592,9 +592,14 @@ struct address_space_operations { > int (*releasepage) (struct page *, int); > void (*freepage)(struct page *); > ssize_t (*direct_IO)(struct kiocb *, struct iov_iter *iter, loff_t offset); > + /* isolate a page for migration */ > + bool (*isolate_page) (struct page *, isolate_mode_t); > /* migrate the contents of a page to the specified target */ > int (*migratepage) (struct page *, struct page *); > + /* put the page back to right list */... "after a failed migration" ?> + void (*putback_page) (struct page *); > int (*launder_page) (struct page *); > + > int (*is_partially_uptodate) (struct page *, unsigned long, > unsigned long); > void (*is_dirty_writeback) (struct page *, bool *, bool *); > diff --git a/fs/proc/page.c b/fs/proc/page.c > index 3ecd445e830d..ce3d08a4ad8d 100644 > --- a/fs/proc/page.c > +++ b/fs/proc/page.c > @@ -157,6 +157,9 @@ u64 stable_page_flags(struct page *page) > if (page_is_idle(page)) > u |= 1 << KPF_IDLE; > > + if (PageMovable(page)) > + u |= 1 << KPF_MOVABLE; > + > u |= kpf_copy_bit(k, KPF_LOCKED, PG_locked); > > u |= kpf_copy_bit(k, KPF_SLAB, PG_slab); > diff --git a/include/linux/fs.h b/include/linux/fs.h > index da9e67d937e5..36f2d610e7a8 100644 > --- a/include/linux/fs.h > +++ b/include/linux/fs.h > @@ -401,6 +401,8 @@ struct address_space_operations { > */ > int (*migratepage) (struct address_space *, > struct page *, struct page *, enum migrate_mode); > + bool (*isolate_page)(struct page *, isolate_mode_t); > + void (*putback_page)(struct page *); > int (*launder_page) (struct page *); > int (*is_partially_uptodate) (struct page *, unsigned long, > unsigned long); > diff --git a/include/linux/migrate.h b/include/linux/migrate.h > index 9b50325e4ddf..404fbfefeb33 100644 > --- a/include/linux/migrate.h > +++ b/include/linux/migrate.h > @@ -37,6 +37,8 @@ extern int migrate_page(struct address_space *, > struct page *, struct page *, enum migrate_mode); > extern int migrate_pages(struct list_head *l, new_page_t new, free_page_t free, > unsigned long private, enum migrate_mode mode, int reason); > +extern bool isolate_movable_page(struct page *page, isolate_mode_t mode); > +extern void putback_movable_page(struct page *page); > > extern int migrate_prep(void); > extern int migrate_prep_local(void); > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h > index f4ed4f1b0c77..77ebf8fdbc6e 100644 > --- a/include/linux/page-flags.h > +++ b/include/linux/page-flags.h > @@ -129,6 +129,10 @@ enum pageflags { > > /* Compound pages. Stored in first tail page's flags */ > PG_double_map = PG_private_2, > + > + /* non-lru movable pages */ > + PG_movable = PG_reclaim, > + PG_isolated = PG_owner_priv_1,Documentation should probably state that these fields alias and subsystem supporting the movable pages shouldn't use them elsewhere. Also I'm a bit uncomfortable how isolate_movable_page() blindly expects that page->mapping->a_ops->isolate_page exists for PageMovable() pages. What if it's a false positive on a PG_reclaim page? Can we rely on PG_reclaim always (and without races) implying PageLRU() so that we don't even attempt isolate_movable_page()?> }; > > #ifndef __GENERATING_BOUNDS_H > @@ -614,6 +618,33 @@ static inline void __ClearPageBalloon(struct page *page) > atomic_set(&page->_mapcount, -1); > } > > +#define PAGE_MOVABLE_MAPCOUNT_VALUE (-255)IIRC this was what Gioh's previous attempts used instead of PG_movable? Is it still needed? Doesn't it prevent a driver providing movable *and* mapped pages? If it's to distinguish the PG_reclaim alias that I mention above, it seems like an overkill to me. Why would be need both special mapcount value and a flag? Checking that page->mapping->a_ops->isolate_page exists before calling it should be enough to resolve the ambiguity?> + > +static inline int PageMovable(struct page *page) > +{ > + return ((test_bit(PG_movable, &(page)->flags) && > + atomic_read(&page->_mapcount) == PAGE_MOVABLE_MAPCOUNT_VALUE) > + || PageBalloon(page)); > +} > + > +/* Caller should hold a PG_lock */ > +static inline void __SetPageMovable(struct page *page, > + struct address_space *mapping) > +{ > + page->mapping = mapping; > + __set_bit(PG_movable, &page->flags); > + atomic_set(&page->_mapcount, PAGE_MOVABLE_MAPCOUNT_VALUE); > +} > + > +static inline void __ClearPageMovable(struct page *page) > +{ > + atomic_set(&page->_mapcount, -1); > + __clear_bit(PG_movable, &(page)->flags); > + page->mapping = NULL; > +} > + > +PAGEFLAG(Isolated, isolated, PF_ANY); > + > /* > * If network-based swap is enabled, sl*b must keep track of whether pages > * were allocated from pfmemalloc reserves. > diff --git a/include/uapi/linux/kernel-page-flags.h b/include/uapi/linux/kernel-page-flags.h > index 5da5f8751ce7..a184fd2434fa 100644 > --- a/include/uapi/linux/kernel-page-flags.h > +++ b/include/uapi/linux/kernel-page-flags.h > @@ -34,6 +34,7 @@ > #define KPF_BALLOON 23 > #define KPF_ZERO_PAGE 24 > #define KPF_IDLE 25 > +#define KPF_MOVABLE 26 > > > #endif /* _UAPILINUX_KERNEL_PAGE_FLAGS_H */ > diff --git a/mm/compaction.c b/mm/compaction.c > index ccf97b02b85f..7557aedddaee 100644 > --- a/mm/compaction.c > +++ b/mm/compaction.c > @@ -703,7 +703,7 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, > > /* > * Check may be lockless but that's ok as we recheck later. > - * It's possible to migrate LRU pages and balloon pages > + * It's possible to migrate LRU and movable kernel pages. > * Skip any other type of page > */ > is_lru = PageLRU(page); > @@ -714,6 +714,18 @@ isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, > goto isolate_success; > } > } > + > + if (unlikely(PageMovable(page)) && > + !PageIsolated(page)) { > + if (locked) { > + spin_unlock_irqrestore(&zone->lru_lock, > + flags); > + locked = false; > + } > + > + if (isolate_movable_page(page, isolate_mode)) > + goto isolate_success; > + } > } > > /* > diff --git a/mm/migrate.c b/mm/migrate.c > index 53529c805752..b56bf2b3fe8c 100644 > --- a/mm/migrate.c > +++ b/mm/migrate.c > @@ -73,6 +73,85 @@ int migrate_prep_local(void) > return 0; > } > > +bool isolate_movable_page(struct page *page, isolate_mode_t mode) > +{ > + bool ret = false;Maintaining "ret" seems useless here. All the "goto out*" statements are executed only when ret is false, and ret == true is returned by a different return.> + > + /* > + * Avoid burning cycles with pages that are yet under __free_pages(), > + * or just got freed under us. > + * > + * In case we 'win' a race for a movable page being freed under us and > + * raise its refcount preventing __free_pages() from doing its job > + * the put_page() at the end of this block will take care of > + * release this page, thus avoiding a nasty leakage. > + */ > + if (unlikely(!get_page_unless_zero(page))) > + goto out; > + > + /* > + * Check PG_movable before holding a PG_lock because page's owner > + * assumes anybody doesn't touch PG_lock of newly allocated page. > + */ > + if (unlikely(!PageMovable(page))) > + goto out_putpage; > + /* > + * As movable pages are not isolated from LRU lists, concurrent > + * compaction threads can race against page migration functions > + * as well as race against the releasing a page. > + * > + * In order to avoid having an already isolated movable page > + * being (wrongly) re-isolated while it is under migration, > + * or to avoid attempting to isolate pages being released, > + * lets be sure we have the page lock > + * before proceeding with the movable page isolation steps. > + */ > + if (unlikely(!trylock_page(page))) > + goto out_putpage; > + > + if (!PageMovable(page) || PageIsolated(page)) > + goto out_no_isolated; > + > + ret = page->mapping->a_ops->isolate_page(page, mode); > + if (!ret) > + goto out_no_isolated; > + > + WARN_ON_ONCE(!PageIsolated(page)); > + unlock_page(page); > + return ret; > + > +out_no_isolated: > + unlock_page(page); > +out_putpage: > + put_page(page); > +out: > + return ret; > +} > + > +/* It should be called on page which is PG_movable */ > +void putback_movable_page(struct page *page) > +{ > + /* > + * 'lock_page()' stabilizes the page and prevents races against > + * concurrent isolation threads attempting to re-isolate it. > + */ > + VM_BUG_ON_PAGE(!PageMovable(page), page); > + > + lock_page(page); > + if (PageIsolated(page)) { > + struct address_space *mapping; > + > + mapping = page_mapping(page); > + mapping->a_ops->putback_page(page); > + WARN_ON_ONCE(PageIsolated(page)); > + } else { > + __ClearPageMovable(page); > + } > + unlock_page(page); > + /* drop the extra ref count taken for movable page isolation */ > + put_page(page); > +} > + > /* > * Put previously isolated pages back onto the appropriate lists > * from where they were once taken off for compaction/migration. > @@ -94,10 +173,18 @@ void putback_movable_pages(struct list_head *l) > list_del(&page->lru); > dec_zone_page_state(page, NR_ISOLATED_ANON + > page_is_file_cache(page)); > - if (unlikely(isolated_balloon_page(page))) > + if (unlikely(isolated_balloon_page(page))) { > balloon_page_putback(page); > - else > + } else if (unlikely(PageMovable(page))) { > + if (PageIsolated(page)) { > + putback_movable_page(page); > + } else { > + __ClearPageMovable(page);We don't do lock_page() here, so what prevents parallel compaction isolating the same page?> + put_page(page); > + } > + } else { > putback_lru_page(page); > + } > } > } > > @@ -592,7 +679,7 @@ void migrate_page_copy(struct page *newpage, struct page *page) > ***********************************************************/ > > /* > - * Common logic to directly migrate a single page suitable for > + * Common logic to directly migrate a single LRU page suitable for > * pages that do not use PagePrivate/PagePrivate2. > * > * Pages are locked upon entry and exit. > @@ -755,24 +842,54 @@ static int move_to_new_page(struct page *newpage, struct page *page, > enum migrate_mode mode) > { > struct address_space *mapping; > - int rc; > + int rc = -EAGAIN; > + bool lru_movable = true; > > VM_BUG_ON_PAGE(!PageLocked(page), page); > VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); > > mapping = page_mapping(page); > - if (!mapping) > - rc = migrate_page(mapping, newpage, page, mode); > - else if (mapping->a_ops->migratepage) > - /* > - * Most pages have a mapping and most filesystems provide a > - * migratepage callback. Anonymous pages are part of swap > - * space which also has its own migratepage callback. This > - * is the most common path for page migration. > - */ > - rc = mapping->a_ops->migratepage(mapping, newpage, page, mode); > - else > - rc = fallback_migrate_page(mapping, newpage, page, mode); > + /* > + * In case of non-lru page, it could be released after > + * isolation step. In that case, we shouldn't try > + * fallback migration which was designed for LRU pages. > + * > + * The rule for such case is that subsystem should clear > + * PG_isolated but remains PG_movable so VM should catch > + * it and clear PG_movable for it. > + */ > + if (unlikely(PageMovable(page))) {Can false positive from PG_reclaim occur here?> + lru_movable = false; > + VM_BUG_ON_PAGE(!mapping, page); > + if (!PageIsolated(page)) { > + rc = MIGRATEPAGE_SUCCESS; > + __ClearPageMovable(page); > + goto out; > + } > + } > + > + if (likely(lru_movable)) { > + if (!mapping) > + rc = migrate_page(mapping, newpage, page, mode); > + else if (mapping->a_ops->migratepage) > + /* > + * Most pages have a mapping and most filesystems > + * provide a migratepage callback. Anonymous pages > + * are part of swap space which also has its own > + * migratepage callback. This is the most common path > + * for page migration. > + */ > + rc = mapping->a_ops->migratepage(mapping, newpage, > + page, mode); > + else > + rc = fallback_migrate_page(mapping, newpage, > + page, mode); > + } else { > + rc = mapping->a_ops->migratepage(mapping, newpage, > + page, mode); > + WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS && > + PageIsolated(page)); > + } > > /* > * When successful, old pagecache page->mapping must be cleared before > @@ -782,6 +899,7 @@ static int move_to_new_page(struct page *newpage, struct page *page, > if (!PageAnon(page)) > page->mapping = NULL; > } > +out: > return rc; > } > > @@ -960,6 +1078,8 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page, > put_new_page(newpage, private); > else > put_page(newpage); > + if (PageMovable(page)) > + __ClearPageMovable(page); > goto out; > } > > @@ -1000,8 +1120,26 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page, > num_poisoned_pages_inc(); > } > } else { > - if (rc != -EAGAIN) > - putback_lru_page(page); > + if (rc != -EAGAIN) { > + /* > + * subsystem couldn't remove PG_movable since page is > + * isolated so PageMovable check is not racy in here. > + * But PageIsolated check can be racy but it's okay > + * because putback_movable_page checks it under PG_lock > + * again. > + */ > + if (unlikely(PageMovable(page))) { > + if (PageIsolated(page)) > + putback_movable_page(page); > + else { > + __ClearPageMovable(page);Again, we don't do lock_page() here, so what prevents parallel compaction isolating the same page? Sorry for so many questions, hope they all have good answers and this series is a success :) Thanks for picking it up.> + put_page(page); > + } > + } else { > + putback_lru_page(page); > + } > + } > + > if (put_new_page) > put_new_page(newpage, private); > else >
Balbir Singh
2016-Apr-04 05:53 UTC
[PATCH v3 01/16] mm: use put_page to free page instead of putback_lru_page
On 30/03/16 18:12, Minchan Kim wrote:> Procedure of page migration is as follows: > > First of all, it should isolate a page from LRU and try to > migrate the page. If it is successful, it releases the page > for freeing. Otherwise, it should put the page back to LRU > list. > > For LRU pages, we have used putback_lru_page for both freeing > and putback to LRU list. It's okay because put_page is aware of > LRU list so if it releases last refcount of the page, it removes > the page from LRU list. However, It makes unnecessary operations > (e.g., lru_cache_add, pagevec and flags operations. It would be > not significant but no worth to do) and harder to support new > non-lru page migration because put_page isn't aware of non-lru > page's data structure. > > To solve the problem, we can add new hook in put_page with > PageMovable flags check but it can increase overhead in > hot path and needs new locking scheme to stabilize the flag check > with put_page. > > So, this patch cleans it up to divide two semantic(ie, put and putback). > If migration is successful, use put_page instead of putback_lru_page and > use putback_lru_page only on failure. That makes code more readable > and doesn't add overhead in put_page.So effectively when we return from unmap_and_move() the page is either put_page or putback_lru_page() and the page is gone from under us.
On Wed, Mar 30, 2016 at 04:11:59PM +0900, Minchan Kim wrote:> Recently, I got many reports about perfermance degradation > in embedded system(Android mobile phone, webOS TV and so on) > and failed to fork easily. > > The problem was fragmentation caused by zram and GPU driver > pages. Their pages cannot be migrated so compaction cannot > work well, either so reclaimer ends up shrinking all of working > set pages. It made system very slow and even to fail to fork > easily. > > Other pain point is that they cannot work with CMA. > Most of CMA memory space could be idle(ie, it could be used > for movable pages unless driver is using) but if driver(i.e., > zram) cannot migrate his page, that memory space could be > wasted. In our product which has big CMA memory, it reclaims > zones too exccessively although there are lots of free space > in CMA so system was very slow easily. > > To solve these problem, this patch try to add facility to > migrate non-lru pages via introducing new friend functions > of migratepage in address_space_operation and new page flags. > > (isolate_page, putback_page) > (PG_movable, PG_isolated) > > For details, please read description in > "mm/compaction: support non-lru movable page migration".Thanks, this mirrors what we see with the ARM Mali GPU drivers too. One thing with the current design which worries me is the potential for multiple calls due to many separated pages being migrated. On GPUs (or any other device) which has an IOMMU and L2 cache, which isn't coherent with the CPU, we must do L2 cache flush & invalidation per page. I guess batching pages isn't easily possible? -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 648 bytes Desc: not available URL: <http://lists.linuxfoundation.org/pipermail/virtualization/attachments/20160404/cd3f7748/attachment.sig>
Vlastimil Babka
2016-Apr-05 12:03 UTC
[PATCH v3 04/16] mm/balloon: use general movable page feature into balloon
On 03/30/2016 09:12 AM, Minchan Kim wrote:> Now, VM has a feature to migrate non-lru movable pages so > balloon doesn't need custom migration hooks in migrate.c > and compact.c. Instead, this patch implements page->mapping > ->{isolate|migrate|putback} functions. > > With that, we could remove hooks for ballooning in general > migration functions and make balloon compaction simple. > > Cc: virtualization at lists.linux-foundation.org > Cc: Rafael Aquini <aquini at redhat.com> > Cc: Konstantin Khlebnikov <koct9i at gmail.com> > Signed-off-by: Gioh Kim <gurugio at hanmail.net> > Signed-off-by: Minchan Kim <minchan at kernel.org>I'm not familiar with the inode and pseudofs stuff, so just some things I noticed:> -#define PAGE_MOVABLE_MAPCOUNT_VALUE (-255) > +#define PAGE_MOVABLE_MAPCOUNT_VALUE (-256) > +#define PAGE_BALLOON_MAPCOUNT_VALUE PAGE_MOVABLE_MAPCOUNT_VALUE > > static inline int PageMovable(struct page *page) > { > - return ((test_bit(PG_movable, &(page)->flags) && > - atomic_read(&page->_mapcount) == PAGE_MOVABLE_MAPCOUNT_VALUE) > - || PageBalloon(page)); > + return (test_bit(PG_movable, &(page)->flags) && > + atomic_read(&page->_mapcount) == PAGE_MOVABLE_MAPCOUNT_VALUE); > } > > /* Caller should hold a PG_lock */ > @@ -645,6 +626,35 @@ static inline void __ClearPageMovable(struct page *page) > > PAGEFLAG(Isolated, isolated, PF_ANY); > > +static inline int PageBalloon(struct page *page) > +{ > + return atomic_read(&page->_mapcount) == PAGE_BALLOON_MAPCOUNT_VALUE > + && PagePrivate2(page); > +}Hmm so you are now using PG_private_2 flag here, but it's not documented. Also the only caller of PageBalloon() seems to be stable_page_flags(). Which will now report all movable pages with PG_private_2 as KPF_BALOON. Seems like an overkill and also not reliable. Could it test e.g. page->mapping instead? Or maybe if we manage to get rid of PAGE_MOVABLE_MAPCOUNT_VALUE, we can keep PAGE_BALLOON_MAPCOUNT_VALUE to simply distinguish balloon pages for stable_page_flags().> @@ -1033,7 +1019,7 @@ static int __unmap_and_move(struct page *page, struct page *newpage, > out: > /* If migration is successful, move newpage to right list */ > if (rc == MIGRATEPAGE_SUCCESS) { > - if (unlikely(__is_movable_balloon_page(newpage))) > + if (unlikely(PageMovable(newpage))) > put_page(newpage); > else > putback_lru_page(newpage);Hmm shouldn't the condition have been changed to if (unlikely(__is_movable_balloon_page(newpage)) || PageMovable(newpage) by patch 02/16? And this patch should be just removing the balloon-specific check? Otherwise it seems like between patches 02 and 04, other kinds of PageMovable pages were unnecessarily/wrongly routed through putback_lru_page()?> diff --git a/mm/vmscan.c b/mm/vmscan.c > index d82196244340..c7696a2e11c7 100644 > --- a/mm/vmscan.c > +++ b/mm/vmscan.c > @@ -1254,7 +1254,7 @@ unsigned long reclaim_clean_pages_from_list(struct zone *zone, > > list_for_each_entry_safe(page, next, page_list, lru) { > if (page_is_file_cache(page) && !PageDirty(page) && > - !isolated_balloon_page(page)) { > + !PageIsolated(page)) { > ClearPageActive(page); > list_move(&page->lru, &clean_pages); > }This looks like the same comment as above at first glance. But looking closer, it's even weirder. isolated_balloon_page() was simply PageBalloon() after d6d86c0a7f8dd... weird already. You replace it with check for !PageIsolated() which looks like a more correct check, so ok. Except the potential false positive with PG_owner_priv_1. Thanks.
On Mon, Apr 04, 2016 at 03:17:18PM +0200, John Einar Reitan wrote:> On Wed, Mar 30, 2016 at 04:11:59PM +0900, Minchan Kim wrote: > > Recently, I got many reports about perfermance degradation > > in embedded system(Android mobile phone, webOS TV and so on) > > and failed to fork easily. > > > > The problem was fragmentation caused by zram and GPU driver > > pages. Their pages cannot be migrated so compaction cannot > > work well, either so reclaimer ends up shrinking all of working > > set pages. It made system very slow and even to fail to fork > > easily. > > > > Other pain point is that they cannot work with CMA. > > Most of CMA memory space could be idle(ie, it could be used > > for movable pages unless driver is using) but if driver(i.e., > > zram) cannot migrate his page, that memory space could be > > wasted. In our product which has big CMA memory, it reclaims > > zones too exccessively although there are lots of free space > > in CMA so system was very slow easily. > > > > To solve these problem, this patch try to add facility to > > migrate non-lru pages via introducing new friend functions > > of migratepage in address_space_operation and new page flags. > > > > (isolate_page, putback_page) > > (PG_movable, PG_isolated) > > > > For details, please read description in > > "mm/compaction: support non-lru movable page migration". > > Thanks, this mirrors what we see with the ARM Mali GPU drivers too. > > One thing with the current design which worries me is the potential > for multiple calls due to many separated pages being migrated. > On GPUs (or any other device) which has an IOMMU and L2 cache, which > isn't coherent with the CPU, we must do L2 cache flush & invalidation > per page. I guess batching pages isn't easily possible? >Hmm, I think it seems to cause many code stirring but surely worth to work. So, IMMO, it would be better to add such feature after soft landing of current work. Anyway, I will Cc'ed you in next revision. Thanks.
Sergey Senozhatsky
2016-Apr-17 15:08 UTC
[PATCH v3 06/16] zsmalloc: squeeze inuse into page->mapping
Hello, On (03/30/16 16:12), Minchan Kim wrote: [..]> +static int get_zspage_inuse(struct page *first_page) > +{ > + struct zs_meta *m; > + > + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > + > + m = (struct zs_meta *)&first_page->mapping;..> +static void set_zspage_inuse(struct page *first_page, int val) > +{ > + struct zs_meta *m; > + > + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > + > + m = (struct zs_meta *)&first_page->mapping;..> +static void mod_zspage_inuse(struct page *first_page, int val) > +{ > + struct zs_meta *m; > + > + VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > + > + m = (struct zs_meta *)&first_page->mapping;..> static void get_zspage_mapping(struct page *first_page, > unsigned int *class_idx, > enum fullness_group *fullness) > { > - unsigned long m; > + struct zs_meta *m; > + > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > + m = (struct zs_meta *)&first_page->mapping;..> static void set_zspage_mapping(struct page *first_page, > unsigned int class_idx, > enum fullness_group fullness) > { > + struct zs_meta *m; > + > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > > + m = (struct zs_meta *)&first_page->mapping; > + m->fullness = fullness; > + m->class = class_idx; > }a nitpick: this struct zs_meta *m; VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); m = (struct zs_meta *)&first_page->mapping; seems to be common in several places, may be it makes sense to factor it out and turn into a macro or a static inline helper? other than that, looks good to me Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky at gmail.com> -ss
Sergey Senozhatsky
2016-Apr-17 15:08 UTC
[PATCH v3 05/16] zsmalloc: keep max_object in size_class
Hello, On (03/30/16 16:12), Minchan Kim wrote:> > Every zspage in a size_class has same number of max objects so > we could move it to a size_class. >Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky at gmail.com> -ss
Sergey Senozhatsky
2016-Apr-17 15:11 UTC
[PATCH v3 07/16] zsmalloc: remove page_mapcount_reset
Hello, On (03/30/16 16:12), Minchan Kim wrote:> We don't use page->_mapcount any more so no need to reset. > > Signed-off-by: Minchan Kim <minchan at kernel.org>Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky at gmail.com> -ss> --- > mm/zsmalloc.c | 1 - > 1 file changed, 1 deletion(-) > > diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c > index 4dd72a803568..0f6cce9b9119 100644 > --- a/mm/zsmalloc.c > +++ b/mm/zsmalloc.c > @@ -922,7 +922,6 @@ static void reset_page(struct page *page) > set_page_private(page, 0); > page->mapping = NULL; > page->freelist = NULL; > - page_mapcount_reset(page); > } > > static void free_zspage(struct page *first_page) > -- > 1.9.1 >
Sergey Senozhatsky
2016-Apr-17 15:22 UTC
[PATCH v3 09/16] zsmalloc: move struct zs_meta from mapping to freelist
Hello, On (03/30/16 16:12), Minchan Kim wrote:> For supporting migration from VM, we need to have address_space > on every page so zsmalloc shouldn't use page->mapping. So, > this patch moves zs_meta from mapping to freelist. > > Signed-off-by: Minchan Kim <minchan at kernel.org>a small get_zspage_meta() helper would make this patch shorter :) Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky at gmail.com> -ss> --- > mm/zsmalloc.c | 22 +++++++++++----------- > 1 file changed, 11 insertions(+), 11 deletions(-) > > diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c > index 807998462539..d4d33a819832 100644 > --- a/mm/zsmalloc.c > +++ b/mm/zsmalloc.c > @@ -29,7 +29,7 @@ > * Look at size_class->huge. > * page->lru: links together first pages of various zspages. > * Basically forming list of zspages in a fullness group. > - * page->mapping: override by struct zs_meta > + * page->freelist: override by struct zs_meta > * > * Usage of struct page flags: > * PG_private: identifies the first component page > @@ -418,7 +418,7 @@ static int get_zspage_inuse(struct page *first_page) > > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > > - m = (struct zs_meta *)&first_page->mapping; > + m = (struct zs_meta *)&first_page->freelist; > > return m->inuse; > } > @@ -429,7 +429,7 @@ static void set_zspage_inuse(struct page *first_page, int val) > > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > > - m = (struct zs_meta *)&first_page->mapping; > + m = (struct zs_meta *)&first_page->freelist; > m->inuse = val; > } > > @@ -439,7 +439,7 @@ static void mod_zspage_inuse(struct page *first_page, int val) > > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > > - m = (struct zs_meta *)&first_page->mapping; > + m = (struct zs_meta *)&first_page->freelist; > m->inuse += val; > } > > @@ -449,7 +449,7 @@ static void set_freeobj(struct page *first_page, int idx) > > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > > - m = (struct zs_meta *)&first_page->mapping; > + m = (struct zs_meta *)&first_page->freelist; > m->freeobj = idx; > } > > @@ -459,7 +459,7 @@ static unsigned long get_freeobj(struct page *first_page) > > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > > - m = (struct zs_meta *)&first_page->mapping; > + m = (struct zs_meta *)&first_page->freelist; > return m->freeobj; > } > > @@ -471,7 +471,7 @@ static void get_zspage_mapping(struct page *first_page, > > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > > - m = (struct zs_meta *)&first_page->mapping; > + m = (struct zs_meta *)&first_page->freelist; > *fullness = m->fullness; > *class_idx = m->class; > } > @@ -484,7 +484,7 @@ static void set_zspage_mapping(struct page *first_page, > > VM_BUG_ON_PAGE(!is_first_page(first_page), first_page); > > - m = (struct zs_meta *)&first_page->mapping; > + m = (struct zs_meta *)&first_page->freelist; > m->fullness = fullness; > m->class = class_idx; > } > @@ -946,7 +946,6 @@ static void reset_page(struct page *page) > clear_bit(PG_private, &page->flags); > clear_bit(PG_private_2, &page->flags); > set_page_private(page, 0); > - page->mapping = NULL; > page->freelist = NULL; > } > > @@ -1056,6 +1055,7 @@ static struct page *alloc_zspage(struct size_class *class, gfp_t flags) > > INIT_LIST_HEAD(&page->lru); > if (i == 0) { /* first page */ > + page->freelist = NULL; > SetPagePrivate(page); > set_page_private(page, 0); > first_page = page; > @@ -2068,9 +2068,9 @@ static int __init zs_init(void) > > /* > * A zspage's a free object index, class index, fullness group, > - * inuse object count are encoded in its (first)page->mapping > + * inuse object count are encoded in its (first)page->freelist > * so sizeof(struct zs_meta) should be less than > - * sizeof(page->mapping(i.e., unsigned long)). > + * sizeof(page->freelist(i.e., void *)). > */ > BUILD_BUG_ON(sizeof(struct zs_meta) > sizeof(unsigned long)); > > -- > 1.9.1 >
Sergey Senozhatsky
2016-Apr-17 15:56 UTC
[PATCH v3 08/16] zsmalloc: squeeze freelist into page->mapping
Hello, On (03/30/16 16:12), Minchan Kim wrote: [..]> +static void objidx_to_page_and_offset(struct size_class *class, > + struct page *first_page, > + unsigned long obj_idx, > + struct page **obj_page, > + unsigned long *offset_in_page) > { > - unsigned long obj; > + int i; > + unsigned long offset; > + struct page *cursor; > + int nr_page; > > - if (!page) { > - VM_BUG_ON(obj_idx); > - return NULL; > - } > + offset = obj_idx * class->size;so we already know the `offset' before we call objidx_to_page_and_offset(), thus we can drop `struct size_class *class' and `obj_idx', and pass `long obj_offset' (which is `obj_idx * class->size') instead, right? we also _may be_ can return `cursor' from the function. static struct page *objidx_to_page_and_offset(struct page *first_page, unsigned long obj_offset, unsigned long *offset_in_page); this can save ~20 instructions, which is not so terrible for a hot path like obj_malloc(). what do you think? well, seems that `unsigned long *offset_in_page' can be calculated outside of this function too, it's basically *offset_in_page = (obj_idx * class->size) & ~PAGE_MASK; so we don't need to supply it to this function, nor modify it there. which can save ~40 instructions on my system. does this sound silly? -ss> + cursor = first_page; > + nr_page = offset >> PAGE_SHIFT; > > - obj = page_to_pfn(page) << OBJ_INDEX_BITS; > - obj |= ((obj_idx) & OBJ_INDEX_MASK); > - obj <<= OBJ_TAG_BITS; > + *offset_in_page = offset & ~PAGE_MASK; > + > + for (i = 0; i < nr_page; i++) > + cursor = get_next_page(cursor); > > - return (void *)obj; > + *obj_page = cursor; > }-ss
Sergey Senozhatsky
2016-Apr-18 00:33 UTC
[PATCH v3 10/16] zsmalloc: factor page chain functionality out
Hello, On (03/30/16 16:12), Minchan Kim wrote:> @@ -1421,7 +1434,6 @@ static unsigned long obj_malloc(struct size_class *class, > unsigned long m_offset; > void *vaddr; > > - handle |= OBJ_ALLOCATED_TAG;a nitpick, why did you replace this ALLOCATED_TAG assignment with 2 'handle | OBJ_ALLOCATED_TAG'? -ss> obj = get_freeobj(first_page); > objidx_to_page_and_offset(class, first_page, obj, > &m_page, &m_offset); > @@ -1431,10 +1443,10 @@ static unsigned long obj_malloc(struct size_class *class, > set_freeobj(first_page, link->next >> OBJ_ALLOCATED_TAG); > if (!class->huge) > /* record handle in the header of allocated chunk */ > - link->handle = handle; > + link->handle = handle | OBJ_ALLOCATED_TAG; > else > /* record handle in first_page->private */ > - set_page_private(first_page, handle); > + set_page_private(first_page, handle | OBJ_ALLOCATED_TAG); > kunmap_atomic(vaddr); > mod_zspage_inuse(first_page, 1); > zs_stat_inc(class, OBJ_USED, 1);
Sergey Senozhatsky
2016-Apr-18 01:04 UTC
[PATCH v3 11/16] zsmalloc: separate free_zspage from putback_zspage
Hello Minchan, On (03/30/16 16:12), Minchan Kim wrote: [..]> @@ -1835,23 +1827,31 @@ static void __zs_compact(struct zs_pool *pool, struct size_class *class) > if (!migrate_zspage(pool, class, &cc)) > break; > > - putback_zspage(pool, class, dst_page); > + VM_BUG_ON_PAGE(putback_zspage(pool, class, > + dst_page) == ZS_EMPTY, dst_page);can this VM_BUG_ON_PAGE() condition ever be true?> } > /* Stop if we couldn't find slot */ > if (dst_page == NULL) > break; > - putback_zspage(pool, class, dst_page); > - if (putback_zspage(pool, class, src_page) == ZS_EMPTY) > + VM_BUG_ON_PAGE(putback_zspage(pool, class, > + dst_page) == ZS_EMPTY, dst_page);hm... this VM_BUG_ON_PAGE(dst_page) is sort of confusing. under what circumstances it can be true? a minor nit, it took me some time (need some coffee I guess) to correctly parse this macro wrapper VM_BUG_ON_PAGE(putback_zspage(pool, class, dst_page) == ZS_EMPTY, dst_page); may be do it like: fullness = putback_zspage(pool, class, dst_page); VM_BUG_ON_PAGE(fullness == ZS_EMPTY, dst_page); well, if we want to VM_BUG_ON_PAGE() at all. there haven't been any problems with compaction, is there any specific reason these macros were added?> + if (putback_zspage(pool, class, src_page) == ZS_EMPTY) { > pool->stats.pages_compacted += class->pages_per_zspage; > - spin_unlock(&class->lock); > + spin_unlock(&class->lock); > + free_zspage(pool, class, src_page);do we really need to free_zspage() out of class->lock? wouldn't something like this if (putback_zspage(pool, class, src_page) == ZS_EMPTY) { pool->stats.pages_compacted += class->pages_per_zspage; free_zspage(pool, class, src_page); } spin_unlock(&class->lock); be simpler? besides, free_zspage() now updates class stats out of class lock, not critical but still. -ss> + } else { > + spin_unlock(&class->lock); > + } > + > cond_resched(); > spin_lock(&class->lock); > } > > if (src_page) > - putback_zspage(pool, class, src_page); > + VM_BUG_ON_PAGE(putback_zspage(pool, class, > + src_page) == ZS_EMPTY, src_page); > > spin_unlock(&class->lock); > }
Reasonably Related Threads
- [PATCH v3 08/16] zsmalloc: squeeze freelist into page->mapping
- [PATCH v1 11/19] zsmalloc: squeeze freelist into page->mapping
- [PATCH v1 11/19] zsmalloc: squeeze freelist into page->mapping
- [PATCH v3 08/16] zsmalloc: squeeze freelist into page->mapping
- [PATCH v1 11/19] zsmalloc: squeeze freelist into page->mapping