On Thu, Nov 22, 2018 at 2:32 AM Christopher Lameter [off-list ref] wrote:

On Sun, 11 Nov 2018, Nicolas Boichat wrote:

quoted

SLAB_CACHE_DMA32 is only available after explicit kmem_cache_create calls,
no default cache is created for kmalloc. Add a test in check_slab_flags
for this.

This does not define the dma32 kmalloc array. Is that intentional?

Yes that's intentional, AFAICT there is no user, so there is no point
creating the cache.

 (okay, I could find one, but it's probably broken:
git grep GFP_DMA32 | grep k[a-z]*alloc
drivers/media/platform/vivid/vivid-osd.c: dev->video_vbase =
kzalloc(dev->video_buffer_size, GFP_KERNEL | GFP_DMA32);
).

In that
case you need to fail any request for GFP_DMA32 coming in via kmalloc.

Well, we do check for these in check_slab_flags (aka GFP_SLAB_BUG_MASK
before patch 1/3 of this series), so, with or without this patch,
calls with GFP_DMA32 will end up failing in check_slab_flags.