On Mon, Sep 23, 2019 at 09:16:57AM -0500, Josh Poimboeuf wrote:
> On Mon, Sep 23, 2019 at 03:33:52PM +0200, Peter Zijlstra wrote:
> > On Mon, Sep 23, 2019 at 07:49:01AM -0500, Josh Poimboeuf wrote:
> > > On Mon, Sep 23, 2019 at 11:20:24AM +0200, Peter Zijlstra wrote:
> > > > On Wed, Sep 18, 2019 at 09:04:21PM -0700, Randy Dunlap wrote:
> > > > > On 9/18/19 3:10 PM, Mark Brown wrote:
> > > > > > Hi all,
> > > > > > 
> > > > > > Changes since 20190917:
> > > > > > 
> > > > > 
> > > > > on x86_64:
> > > > > 
> > > > > drivers/gpu/drm/i915/gem/i915_gem_execbuffer.o: warning: objtool: 
> > > > > i915_gem_execbuffer2_ioctl()+0x2fb: call to gen8_canonical_addr() 
> > > > > with UACCESS enabled
> > > > 
> > > > I'm thinking that comes from:
> > > > 
> > > >                                 offset = gen8_canonical_addr(offset & 
> > > > ~UPDATE);
> > > >                                 if (unlikely(__put_user(offset, 
> > > > &urelocs[r-stack].presumed_offset))) {
> > > > 
> > > > however, per commit 6ae865615fc4 (and 2a418cf3f5f1) the compiler really
> > > > should not be sticking gen8_canonical_addr() after __uaccess_begin().
> > > > 
> > > > /me puzzled...
> > > 
> > > I think you're looking at the wrong code.  It has user_access_begin/end
> > > around it:
> > > 
> > >           if (!user_access_begin(user_exec_list, count * 
> > > sizeof(*user_exec_list)))
> > >                   goto end;
> > > 
> > >           for (i = 0; i < args->buffer_count; i++) {
> > >                   if (!(exec2_list[i].offset & UPDATE))
> > >                           continue;
> > > 
> > >                   exec2_list[i].offset =
> > >                           gen8_canonical_addr(exec2_list[i].offset & 
> > > PIN_OFFSET_MASK);
> > >                   unsafe_put_user(exec2_list[i].offset,
> > >                                   &user_exec_list[i].offset,
> > >                                   end_user);
> > >           }
> > > end_user:
> > >           user_access_end();
> > > 
> > 
> > Oh, Duh... Yeah, so the alternative to your solution is to do 2 loops.
> > Not sure which would be better.
> 
> I like your idea better, makes the fix more localized and shrinks the
> uaccess window.  Something like so (not even compile tested):

At which point we should probably also Cc Chris who caused all this
with commit:

  2889caa92321 ("drm/i915: Eliminate lots of iterations over the execobjects 
array")

> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c 
> b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> index b5f6937369ea..406af374f728 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> @@ -2847,6 +2847,11 @@ i915_gem_execbuffer2_ioctl(struct drm_device *dev, 
> void *data,
>                       u64_to_user_ptr(args->buffers_ptr);
>               unsigned int i;
>  
> +             for (i = 0; i < args->buffer_count; i++) {
> +                     exec2_list[i].offset =
> +                             gen8_canonical_addr(exec2_list[i].offset & 
> PIN_OFFSET_MASK);
> +             }
> +
>               /* Copy the new buffer offsets back to the user's exec list. */
>               /*
>                * Note: count * sizeof(*user_exec_list) does not overflow,
> @@ -2862,8 +2867,6 @@ i915_gem_execbuffer2_ioctl(struct drm_device *dev, void 
> *data,
>                       if (!(exec2_list[i].offset & UPDATE))
>                               continue;
>  
> -                     exec2_list[i].offset =
> -                             gen8_canonical_addr(exec2_list[i].offset & 
> PIN_OFFSET_MASK);
>                       unsafe_put_user(exec2_list[i].offset,
>                                       &user_exec_list[i].offset,
>                                       end_user);

Reply via email to