Now that we're murder-synchronous, this code path will never be
called (and if it does, it doesn't tell us anything useful other
than we killed a task that was already being killed by somebody
else but hadn't gotten its' signal yet)
Signed-off-by: San Mehat <san@google.com>
As it turns out, the CONFIG_PROFILING interfaces leak a
task struct if the notifier chain returns NOTIFY_OK.. doh.
This patch reworks lowmemkiller to use the new generic task
free notifier chain.
Signed-off-by: San Mehat <san@google.com>
binder_deferred_release was not unmapping the page from the buffer
before freeing it, causing memory corruption. This only happened
when page(s) had not been freed by binder_update_page_range, which
properly unmaps the pages.
This only happens on architectures with VIPT aliasing.
To reproduce, create a program which opens, mmaps, munmaps, then closes
the binder very quickly. This should leave a page allocated when the
binder is released. When binder_deferrred_release is called on the
close, the page will remain mapped to the address in the linear
proc->buffer. Later, we may map the same physical page to a different
virtual address that has different coloring, and this may cause
aliasing to occur.
PAGE_POISONING will greatly increase your chances of noticing any
problems.
Change-Id: I6941bf212881b8bf846bdfda43d3609c7ae4892e
Signed-off-by: Christopher Lais <chris+android@zenthought.org>
This patch optimizes lowmemkiller to not do any work when it has an outstanding
kill-request. This greatly reduces the pressure on the task_list lock
(improving interactivity), as well as improving the vmscan performance
when under heavy memory pressure (by up to 20x in tests).
Note: For this enhancement to work, you need CONFIG_PROFILING
Signed-off-by: San Mehat <san@google.com>
Under certain circumstances, a process can take awhile to
handle a sig-kill (especially if it's in a scheduler group with
a very low share ratio). When this occurs, lowmemkiller returns
to vmscan indicating the process memory has been freed - even
though the process is still waiting to die. Since the memory
hasn't actually freed, lowmemkiller is called again shortly after,
and picks the same process to die; regardless of the fact that
it has already been 'scheduled' to die and the memory has already
been reported to vmscan as having been freed.
Solution is to check fatal_signal_pending() on the selected
task, and if it's already pending destruction return; indicating
to vmscan that no resources were freed on this pass.
Signed-off-by: San Mehat <san@google.com>
Some drivers flush the global workqueue when closed. This would deadlock if
the last reference to the file was released from the binder.
Change-Id: Ifdabc0b383fecb20836d1bbb9786c632402a14e1
Signed-off-by: Arve Hjønnevåg <arve@android.com>
The timed output device never previously checked the return value of sscanf,
resulting in an uninitialized int being passed to enable() if input value
was invalid.
Signed-off-by: Mike Lockwood <lockwood@android.com>
These drivers are no longer being developed and the original authors
seem to have abandonded them and hence, do not want them in the mainline
kernel tree.
So sad :(
Cc: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
It's causing lots of build errors, so just mark it as broken. It is
scheduled to be removed in 2.6.33 anyway.
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This was done using a semantic patch (http://coccinelle.lip6.fr/) that
checks that the declaration is not inside a function definition, that the
defined variable is not exported using EXPORTED_SYMBOL, etc, and that the
defined variable does not occur in any other file. If these conditions
hold, static is added before the declaration.
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Move module_params to near the end of the source file so that
their references are already known/defined. Fixes build errors:
drivers/staging/android/lowmemorykiller.c: In function '__check_cost':
drivers/staging/android/lowmemorykiller.c:60: error: 'lowmem_shrinker' undeclared (first use in this function)
drivers/staging/android/lowmemorykiller.c: At top level:
drivers/staging/android/lowmemorykiller.c:60: error: 'lowmem_shrinker' undeclared here (not in a function)
drivers/staging/android/lowmemorykiller.c:60: warning: type defaults to 'int' in declaration of 'type name'
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Move the lowmemorykiller.txt into the actual source file which is really the
correct place for it, and delete lowmemorykiller.txt.
Signed-off-by: Daniel Walker <dwalker@fifo99.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This file has a lot of long line problems due to the massive indent issues
that exist. This just cleans up some of the really really long ones.
Signed-off-by: Daniel Walker <dwalker@fifo99.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
An initial cleanup of all the binder_stat statements. The binder
command and return stats still need some assistance tho.
Signed-off-by: Daniel Walker <dwalker@fifo99.com>
Cc: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Replaced a manual hlist_head declaration with a macro based one.
Also reorganized the globals to be grouped better.
Signed-off-by: Daniel Walker <dwalker@fifo99.com>
Cc: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Declare the binder_deferred_state enum, and use the new enum
for one of the binder_defer_work function arguments. This
should keep the argument within the confines of the enum
instead of the whole int range.
Signed-off-by: Daniel Walker <dwalker@fifo99.com>
Cc: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
I removed the binder_transaction_buffer_release predefine, and put
the actual function in place of it.
Signed-off-by: Daniel Walker <dwalker@fifo99.com>
Cc: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
I moved the continual,
if (binder_debug_mask & mask)
printk()
into a single macro so it's all in one place. It could be refined further
from there.
Signed-off-by: Daniel Walker <dwalker@fifo99.com>
Cc: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
I'm about to merge "oom: move oom_adj value from task_struct to
mm_struct", and this fixup is needed to repair linux-next's
drivers/staging/android/lowmemorykiller.c.
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Wrap all ANDROID config items with a #if to keep from asking if you want
specific Android drivers even if you say N to CONFIG_ANDROID
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This cleans up the majority of the checkpatch warnings in the android
binder driver. All that is left now is a bunch of too-long-line stuff.
Cc: San Mehat <san@android.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
get_mm_rss() atomically dereferences the actual without checking for a
NULL pointer, which is possible since task_lock() is not held.
Cc: San Mehat <san@android.com>
Cc: Arve Hjønnevåg <arve@android.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Clean up the code in lowmem_shrink() for the Android low memory killer so
that it follows the kernel coding style.
It's unnecessary to check for p->oomkilladj >= min_adj if the selected
task's oomkilladj score is stored since get_mm_rss() will always be
greater than zero.
Cc: San Mehat <san@android.com>
Cc: Arve Hjønnevåg <arve@android.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
From: Arve Hjønnevåg <arve@android.com>
This allows processes to be killed when the kernel evict cache pages in
an attempt to get more contiguous free memory.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Cc: David Rientjes <rientjes@google.com>
Cc: San Mehat <san@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Use NR_ACTIVE plus NR_INACTIVE as a size estimate for our fake cache
instead the sum of rss. Neither method is accurate.
Also skip the process scan, if the amount of memory available is above
the largest threshold set.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Cc: David Rientjes <rientjes@google.com>
Cc: San Mehat <san@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
If a transaction that contains a file descriptor fails on a later object,
the new file descriptor needs to be closed. If this is a binder file
descriptor we would deadlock in flush. If there were no other references to
the file at this point release would also be called.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
If a thread is part of a transaction stack, it is only allowed to make
another call if it was the target of the top transaction on the stack.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This prevents breaking fget_light if a single threaded application
allows incoming file descriptors (in replies or on nodes).
Should also prevent inserting a file in the wrong files_struct if the
receving process execs in the middle of a transaction (between
task_get_unused_fd_flags and task_fd_install).
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Check that datasize is not smaller than one flat_binder_object.
Check that offsets are aligned.
Check that offsets_size is aligned.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Only allow a binder file pointer to be mmapped once. The buffer management
code cannot deal with more then one area.
Also remove leftover mutex_unlock if mmap fails.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Many platforms do not support mappings without VM_EXEC.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
