[PATCH] binder: reduce mmap_sem write-side lock

[PATCH] binder: reduce mmap_sem write-side lock

[Minchan Kim]

binder has used write-side mmap_sem semaphore to release memory
mapped at address space of the process. However, right lock to
release pages is down_read, not down_write because page table lock
already protects the race for parallel freeing.

Please do not use mmap_sem write-side lock which is well known
contented lock.

Cc: Todd Kjos <tkjos@google.com>
Cc: Martijn Coenen <maco@android.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Minchan Kim <minchan@kernel.org>
---
 drivers/android/binder_alloc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/android/binder_alloc.c b/drivers/android/binder_alloc.c
index 022cd80e80cc3..2a62553b20fcc 100644
--- a/drivers/android/binder_alloc.c
+++ b/drivers/android/binder_alloc.c
@@ -964,7 +964,7 @@ enum lru_status binder_alloc_free_page(struct list_head *item,
 		if (!mmget_not_zero(alloc->vma_vm_mm))
 			goto err_mmget;
 		mm = alloc->vma_vm_mm;
-		if (!down_write_trylock(&mm->mmap_sem))
+		if (!down_read_trylock(&mm->mmap_sem))
 			goto err_down_write_mmap_sem_failed;
 	}
 
@@ -980,7 +980,7 @@ enum lru_status binder_alloc_free_page(struct list_head *item,
 
 		trace_binder_unmap_user_end(alloc, index);
 
-		up_write(&mm->mmap_sem);
+		up_read(&mm->mmap_sem);
 		mmput(mm);
 	}
 
-- 
2.21.0.rc0.258.g878e2cd30e-goog

Re: [PATCH] binder: reduce mmap_sem write-side lock

[Greg KH]

On Mon, Feb 18, 2019 at 05:11:45PM +0900, Minchan Kim wrote:
> binder has used write-side mmap_sem semaphore to release memory
> mapped at address space of the process. However, right lock to
> release pages is down_read, not down_write because page table lock
> already protects the race for parallel freeing.
> 
> Please do not use mmap_sem write-side lock which is well known
> contented lock.
> 
> Cc: Todd Kjos <tkjos@google.com>
> Cc: Martijn Coenen <maco@android.com>
> Cc: Arve Hjønnevåg <arve@android.com>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Signed-off-by: Minchan Kim <minchan@kernel.org>
> ---
>  drivers/android/binder_alloc.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)

Is this just needed for the recent binder changes that are in
linux-next, or for older kernels as well?

thanks,

greg k-h

Re: [PATCH] binder: reduce mmap_sem write-side lock

[Minchan Kim]

On Mon, Feb 18, 2019 at 09:32:08AM +0100, Greg KH wrote:
> On Mon, Feb 18, 2019 at 05:11:45PM +0900, Minchan Kim wrote:
> > binder has used write-side mmap_sem semaphore to release memory
> > mapped at address space of the process. However, right lock to
> > release pages is down_read, not down_write because page table lock
> > already protects the race for parallel freeing.
> > 
> > Please do not use mmap_sem write-side lock which is well known
> > contented lock.
> > 
> > Cc: Todd Kjos <tkjos@google.com>
> > Cc: Martijn Coenen <maco@android.com>
> > Cc: Arve Hjønnevåg <arve@android.com>
> > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Signed-off-by: Minchan Kim <minchan@kernel.org>
> > ---
> >  drivers/android/binder_alloc.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> Is this just needed for the recent binder changes that are in
> linux-next, or for older kernels as well?

It has been there for several years but no need to fix older kernels
because down_write of mmap_sem is technically not a bug to stop the
working of binder. Rather than, it's just abuse of the lock so
it's okay to fix only recent kernel.

Thanks.

Re: [PATCH] binder: reduce mmap_sem write-side lock

[Todd Kjos]

On Mon, Feb 18, 2019 at 2:47 AM Minchan Kim <minchan@kernel.org> wrote:
>
> On Mon, Feb 18, 2019 at 09:32:08AM +0100, Greg KH wrote:
> > On Mon, Feb 18, 2019 at 05:11:45PM +0900, Minchan Kim wrote:
> > > binder has used write-side mmap_sem semaphore to release memory
> > > mapped at address space of the process. However, right lock to
> > > release pages is down_read, not down_write because page table lock
> > > already protects the race for parallel freeing.
> > >
> > > Please do not use mmap_sem write-side lock which is well known
> > > contented lock.
> > >
> > > Cc: Todd Kjos <tkjos@google.com>
> > > Cc: Martijn Coenen <maco@android.com>
> > > Cc: Arve Hjønnevåg <arve@android.com>
> > > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > > Signed-off-by: Minchan Kim <minchan@kernel.org>

Acked-by: Todd Kjos <tkjos@google.com>

> > > ---
> > >  drivers/android/binder_alloc.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > Is this just needed for the recent binder changes that are in
> > linux-next, or for older kernels as well?
>
> It has been there for several years but no need to fix older kernels
> because down_write of mmap_sem is technically not a bug to stop the
> working of binder. Rather than, it's just abuse of the lock so
> it's okay to fix only recent kernel.
>
> Thanks.