]> git.baikalelectronics.ru Git - kernel.git/commit
drm/panfrost: Simplify lock_region calculation
authorAlyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tue, 24 Aug 2021 17:30:25 +0000 (13:30 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 22 Sep 2021 10:26:38 +0000 (12:26 +0200)
commitf6baa98b58400dd0a837af116ff375e91d231376
tree073bfaea9f415b3fbf9fda33a3796dc9212f611a
parent08c66c9694bd60e9c55b55fbf7899bb03585e311
drm/panfrost: Simplify lock_region calculation

commit b5fab345654c603c07525100d744498f28786929 upstream.

In lock_region, simplify the calculation of the region_width parameter.
This field is the size, but encoded as ceil(log2(size)) - 1.
ceil(log2(size)) may be computed directly as fls(size - 1). However, we
want to use the 64-bit versions as the amount to lock can exceed
32-bits.

This avoids undefined (and completely wrong) behaviour when locking all
memory (size ~0). In this case, the old code would "round up" ~0 to the
nearest page, overflowing to 0. Since fls(0) == 0, this would calculate
a region width of 10 + 0 = 10. But then the code would shift by
(region_width - 11) = -1. As shifting by a negative number is undefined,
UBSAN flags the bug. Of course, even if it were defined the behaviour is
wrong, instead of locking all memory almost none would get locked.

The new form of the calculation corrects this special case and avoids
the undefined behaviour.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reported-and-tested-by: Chris Morgan <macromorgan@hotmail.com>
Fixes: 7e59668a6fb5 ("drm/panfrost: Add initial panfrost driver")
Cc: <stable@vger.kernel.org>
Reviewed-by: Steven Price <steven.price@arm.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210824173028.7528-2-alyssa.rosenzweig@collabora.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/gpu/drm/panfrost/panfrost_mmu.c