Fix 4.4 merge issues in zero page support, and add the
missing label.
Change-Id: I4bed7add011e0c9b0e148d1b44132ba1873cf607
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
Zcache could be ineffective if the compressed memory pool is full with
compressed inactive file pages and most of them will be never used again.
So we pick up pages from active file list only, those pages would probably
be accessed again. Compress them in memory can reduce the latency
significantly compared with rereading from disk.
When a file page is shrunk from active file list to inactive file list,
PageActive flag is also cleared.
So adding an extra WasActive page flag for zcache to know whether the
file page was shrunk from the active list.
Change-Id: Ida1f4db17075d1f6f825ef7ce2b3bae4eb799e3f
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Patch-mainline: linux-mm @ 2013-08-06 11:36:17
[vinmenon@codeaurora.org: trivial merge conflict fixes, checkpatch fixes,
fix the definitions of was_active page flag so that it does not create
compile time errors with CONFIG_CLEANCACHE disabled. Also remove the
unnecessary use of PG_was_active in PAGE_FLAGS_CHECK_AT_PREP. Since
was_active is a requirement for zcache, make the definitions dependent on
CONFIG_ZCACHE rather than CONFIG_CLEANCACHE.]
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
Deadlock is observed in zcache reclaim paths due to
different locking sequence.
Core#0: Core#1:
|spin_bug() |do_raw_write_lock()
|do_raw_spin_lock() |_raw_write_lock_irqsave()
|_raw_spin_lock_irqsave() |zcache_rbnode_isolate()
|zcache_flush_inode() |zcache_load_delete_zaddr()
|__cleancache_invalidate_inode() |zcache_evict_zpage()
|truncate_inode_pages_range() |zbud_reclaim_page()
|truncate_inode_pages() |zcache_scan()
|truncate_inode_pages_final() |shrink_slab_node()
|ext4_evict_inode() |shrink_slab()
|evict() |try_to_free_pages()
|dispose_list() |__alloc_pages_nodemask()
|prune_icache_sb() |alloc_kmem_pages_node()
|super_cache_scan() |copy_process.part.52()
|shrink_slab_node() |do_fork()
|shrink_slab() |sys_clone()
|kswapd_shrink_zone.constprop |el0_svc()
|balance_pgdat()
|kswapd()
|kthread()
|ret_from_fork()
The deadlock happens because alternate sequence are
followed while taking
zpool->rb_lock (protects zpool rb tree), and
rbnode->ra_lock (protects radix tree maintained by rbtree node)
Fix the sequence of locks being taken to avoid deadlock.
Change-Id: I32db23268f63eb8eb5aee30e4462c190e2e02f48
Signed-off-by: Shiraz Hashim <shashim@codeaurora.org>
There is a chance of zbud handle being used after a free.
Unable to handle kernel paging request at virtual address ffffffc05be72040
PC is at zcache_store_page+0x59c/0x618
LR is at zcache_store_page+0x59c/0x618
[<ffffffc00019c99c>] zcache_store_page+0x59c/0x618
[<ffffffc0001a70c4>] __cleancache_put_page+0x94/0xcc
[<ffffffc00015da4c>] __delete_from_page_cache+0xc0/0x2cc
[<ffffffc00016d230>] __remove_mapping+0xe4/0x128
[<ffffffc00016e750>] shrink_page_list+0x634/0x95c
[<ffffffc00016f32c>] shrink_inactive_list+0x41c/0x67c
[<ffffffc00016fc14>] shrink_lruvec+0x364/0x510
[<ffffffc00016fe10>] shrink_zone+0x50/0x12c
[<ffffffc000170278>] try_to_free_pages+0x38c/0x56c
[<ffffffc000164e4c>] __alloc_pages_nodemask+0x5e0/0x994
[<ffffffc000165214>] __get_free_pages+0x14/0x60
CRs-Fixed: 968859
Change-Id: I24f6cf8ccbac956d4c3114e70a9f94f5e3bfa1c8
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
zcache store can fail on lowmemory conditions.
Do not throw unnecessary prints on such events, causing
a stall.
Change-Id: I8fba6938f75ef5b6054ee079951b7409279e1c02
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
The following race is possible:
CPU 1 CPU2
zcache_store_page
zbud_alloc
zcache_evict_zpage
zpool = zhandle->zpool;
CRASH
zcache_store_page
zhandle->zpool = zpool
Fix this by properly initializing the zhandle and validating
in zcache_evict_zpage
Change-Id: I02328220b30f415fa1f171236eab3a2e40072fd9
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
When files pages are very low, it is better to clear off
zcache pages, since the freed memory can be used to sustain
an application in foreground. Moreover when file pages
are too low, we don't gain much by holding a few zcache
pages.
Change-Id: I88dd295d24b7de18fb3bc0788e0baeb6bfdb2f6d
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
zero pages need not be compressed and stored unnecessarily.
Rather insert a special handle to the radix tree to identify
a zero page.
Change-Id: Ic92321c4753401a90d69a6e8c61b5119168c9df7
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
zcache allocations happen during reclaim of pages,
and waking up kswapd at this time hinders system
performance. During tests it was seen that without
this patch, there were a lot of kswapd wakeups and
was resulting in bad launches.
Change-Id: Ic0f0240b8fdad6b3fe142b2bdc0366cfd870635e
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
zcache_pool_pages is supposed to store the total
pages in all the pools used by zcache. But at
present zcache_pool_pages is assigned pages of any
particular pool on a load, store or any operation
which modifies the pool size. And this is more
important for external clients which depend on zcache
pool size, like the lowmemorykiller.
Change-Id: Ifdaab8646c40f1fec71dfa5903658fbdc6b3cce5
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
When file pages drops down, at a point zcache pages
can be more than the file pages, and even further when
we have to reclaim the maximum number of file pages
possible to launch an application, we need a way to
reclaim even the zcache pages at least to an extend
which makes it match with the number of file pages.
This can help in better foreground headroom.
Change-Id: I481bfb9961ed5cee47ebeae08eb910bb269b644c
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
zcache store ops cannot sleep since it is called from
an atomic context. Pass the right flags to alloc functions.
Change-Id: I9c7956e895fb7909932315aa3d85845127aa02e1
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
Implemented zbud_ops->evict, so that compressed zpages can be evicted from
zbud memory pool in the case that the compressed pool is full.
zbud already managered the compressed pool based on LRU. The evict was
implemented just by dropping the compressed file page data directly, if
the data is required again then no more disk reading can be saved.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Patch-mainline: linux-mm @ 2013-08-06 11:36:16
[vinmenon@codeaurora.org: trivial merge conflict fixes]
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
Change-Id: Ia66652475e490f0233547511e80abf7587054e65
zcache is a backend for cleancache that takes file pages that are in the
process of being reclaimed and attempts to compress them and store them
in a RAM-based memory pool. This can result in a significant I/O reduction
if system is full with file pages and, in the case where decompressing
from RAM is faster than reading from the disk, can also improve workload
performance.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Patch-mainline: linux-mm @ 2013-08-06 11:36:14
[vinmenon@codeaurora.org: trivial merge conflict fixes, checkpatch fixes]
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
Change-Id: Id05bcc1946eee22b94a83adeb0134a3f6e357ef8
