Re: [PATCH 6/6] drivers/IB,core: reduce scope of mmap_sem

[Jason Gunthorpe <jgg@mellanox.com>]

On Tue, Jan 15, 2019 at 10:13:00AM -0800, Davidlohr Bueso wrote:
> ib_umem_get() uses gup_longterm() and relies on the lock to
> stabilze the vma_list, so we cannot really get rid of mmap_sem
> altogether, but now that the counter is atomic, we can get of
> some complexity that mmap_sem brings with only pinned_vm.
> 
> Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
>  drivers/infiniband/core/umem.c | 41 ++---------------------------------------
>  1 file changed, 2 insertions(+), 39 deletions(-)
> 
> diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
> index bf556215aa7e..baa2412bf6fb 100644
> +++ b/drivers/infiniband/core/umem.c
> @@ -160,15 +160,12 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
>  
>  	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
>  
> -	down_write(&mm->mmap_sem);
> -	new_pinned = atomic_long_read(&mm->pinned_vm) + npages;
> +	new_pinned = atomic_long_add_return(npages, &mm->pinned_vm);
>  	if (new_pinned > lock_limit && !capable(CAP_IPC_LOCK)) {

I thought a patch had been made for this to use check_overflow...

npages is controlled by userspace, so can we protect pinned_vm from
overflow in some way that still allows it to be atomic?

Jason