Re: [PATCH] ANDROID: binder: change down_write to down_read

[Minchan Kim]

On Wed, Mar 28, 2018 at 07:53:54AM +0200, Greg Kroah-Hartman wrote:
> On Wed, Mar 28, 2018 at 11:42:31AM +0900, Minchan Kim wrote:
> > binder_update_page_range needs down_write of mmap_sem because
> > vm_insert_page need to change vma->vm_flags to VM_MIXEDMAP unless
> > it is set. However, when I profile binder working, it seems
> > every binder buffers should be mapped in advance by binder_mmap.
> > It means we could set VM_MIXEDMAP in bider_mmap time which is
> > already hold a mmap_sem as down_write so binder_update_page_range
> > doesn't need to hold a mmap_sem as down_write.
> > 
> > Android suffers from mmap_sem contention so let's reduce mmap_sem
> > down_write.
> > 
> > Cc: Martijn Coenen <maco@xxxxxxxxxxx>
> > Cc: Todd Kjos <tkjos@xxxxxxxxxx>
> > Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> > Signed-off-by: Minchan Kim <minchan@xxxxxxxxxx>
> > ---
> >  drivers/android/binder.c       | 2 +-
> >  drivers/android/binder_alloc.c | 8 +++++---
> >  2 files changed, 6 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/android/binder.c b/drivers/android/binder.c
> > index 764b63a5aade..9a14c6dd60c4 100644
> > --- a/drivers/android/binder.c
> > +++ b/drivers/android/binder.c
> > @@ -4722,7 +4722,7 @@ static int binder_mmap(struct file *filp, struct vm_area_struct *vma)
> >  		failure_string = "bad vm_flags";
> >  		goto err_bad_arg;
> >  	}
> > -	vma->vm_flags = (vma->vm_flags | VM_DONTCOPY) & ~VM_MAYWRITE;
> > +	vma->vm_flags |= (VM_DONTCOPY | VM_MIXEDMAP) & ~VM_MAYWRITE;
> >  	vma->vm_ops = &binder_vm_ops;
> >  	vma->vm_private_data = proc;
> >  
> > diff --git a/drivers/android/binder_alloc.c b/drivers/android/binder_alloc.c
> > index 5a426c877dfb..a184bf12eb15 100644
> > --- a/drivers/android/binder_alloc.c
> > +++ b/drivers/android/binder_alloc.c
> > @@ -219,7 +219,7 @@ static int binder_update_page_range(struct binder_alloc *alloc, int allocate,
> >  		mm = alloc->vma_vm_mm;
> >  
> >  	if (mm) {
> > -		down_write(&mm->mmap_sem);
> > +		down_read(&mm->mmap_sem);
> >  		vma = alloc->vma;
> >  	}
> >  
> > @@ -229,6 +229,8 @@ static int binder_update_page_range(struct binder_alloc *alloc, int allocate,
> >  		goto err_no_vma;
> >  	}
> >  
> > +	WARN_ON_ONCE(vma && !(vma->vm_flags & VM_MIXEDMAP));
> 
> What can we do with this warning?  What about systems that run in 'panic
> on warn' mode?  Is this something we should handle directly instead?

Hmm, it seems we don't need above warning there because vm_insert_page will hit
BUG_ON in that case. I will remove it at respin. Thanks for the review, Greg.

I will wait other binder maintainer's opinion about below question.

> > it seems
> > every binder buffers should be mapped in advance by binder_mmap.