In commit e11f5bd8228f ("drm: Add support for DP 1.4 Compliance edid
corruption test") the function connector_bad_edid() started assuming
that the memory for the EDID passed to it was big enough to hold
`edid[0x7e] + 1` blocks of data (1 extra for the base block). It
completely ignored the fact that the function was passed `num_blocks`
which indicated how much memory had been allocated for the EDID.
Let's fix this by adding a bounds check.
This is important for handling the case where there's an error in the
first block of the EDID. In that case we will call
connector_bad_edid() without having re-allocated memory based on
`edid[0x7e]`.
Fixes: e11f5bd8228f ("drm: Add support for DP 1.4 Compliance edid corruption test")
Reported-by: Ville Syrjälä <[email protected]>
Signed-off-by: Douglas Anderson <[email protected]>
Reviewed-by: Ville Syrjälä <[email protected]>
---
This problem report came up in the context of a patch I sent out [1]
and this is my attempt at a fix. The problem predates my patch,
though. I don't personally know anything about DP compliance testing
and what should be happening here, nor do I apparently have any
hardware that actually reports a bad EDID. Thus this is just compile
tested. I'm hoping that someone here can test this and make sure it
seems OK to them.
Changes in v2:
- Added a comment/changed math to help make it easier to grok.
drivers/gpu/drm/drm_edid.c | 15 ++++++++++++---
1 file changed, 12 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
index 9c9463ec5465..0383d97c306f 100644
--- a/drivers/gpu/drm/drm_edid.c
+++ b/drivers/gpu/drm/drm_edid.c
@@ -1840,11 +1840,20 @@ static void connector_bad_edid(struct drm_connector *connector,
u8 *edid, int num_blocks)
{
int i;
- u8 num_of_ext = edid[0x7e];
+ u8 last_block;
+
+ /*
+ * 0x7e in the EDID is the number of extension blocks. The EDID
+ * is 1 (base block) + num_ext_blocks big. That means we can think
+ * of 0x7e in the EDID of the _index_ of the last block in the
+ * combined chunk of memory.
+ */
+ last_block = edid[0x7e];
/* Calculate real checksum for the last edid extension block data */
- connector->real_edid_checksum =
- drm_edid_block_checksum(edid + num_of_ext * EDID_LENGTH);
+ if (last_block < num_blocks)
+ connector->real_edid_checksum =
+ drm_edid_block_checksum(edid + last_block * EDID_LENGTH);
if (connector->bad_edid_counter++ && !drm_debug_enabled(DRM_UT_KMS))
return;
--
2.33.0.800.g4c38ced690-goog
Hi,
On Tue, Oct 5, 2021 at 7:29 PM Douglas Anderson <[email protected]> wrote:
>
> In commit e11f5bd8228f ("drm: Add support for DP 1.4 Compliance edid
> corruption test") the function connector_bad_edid() started assuming
> that the memory for the EDID passed to it was big enough to hold
> `edid[0x7e] + 1` blocks of data (1 extra for the base block). It
> completely ignored the fact that the function was passed `num_blocks`
> which indicated how much memory had been allocated for the EDID.
>
> Let's fix this by adding a bounds check.
>
> This is important for handling the case where there's an error in the
> first block of the EDID. In that case we will call
> connector_bad_edid() without having re-allocated memory based on
> `edid[0x7e]`.
>
> Fixes: e11f5bd8228f ("drm: Add support for DP 1.4 Compliance edid corruption test")
> Reported-by: Ville Syrjälä <[email protected]>
> Signed-off-by: Douglas Anderson <[email protected]>
> Reviewed-by: Ville Syrjälä <[email protected]>
> ---
> This problem report came up in the context of a patch I sent out [1]
> and this is my attempt at a fix. The problem predates my patch,
> though. I don't personally know anything about DP compliance testing
> and what should be happening here, nor do I apparently have any
> hardware that actually reports a bad EDID. Thus this is just compile
> tested. I'm hoping that someone here can test this and make sure it
> seems OK to them.
>
> Changes in v2:
> - Added a comment/changed math to help make it easier to grok.
>
> drivers/gpu/drm/drm_edid.c | 15 ++++++++++++---
> 1 file changed, 12 insertions(+), 3 deletions(-)
Pushed this to drm-misc-fixes since the commit it fixes is fairly old.
fdc21c35aaa1 drm/edid: In connector_bad_edid() cap num_of_ext by num_blocks read
-Doug
On Wed, Oct 06, 2021 at 03:45:07PM -0700, Doug Anderson wrote:
> Hi,
>
> On Tue, Oct 5, 2021 at 7:29 PM Douglas Anderson <[email protected]> wrote:
> >
> > In commit e11f5bd8228f ("drm: Add support for DP 1.4 Compliance edid
> > corruption test") the function connector_bad_edid() started assuming
> > that the memory for the EDID passed to it was big enough to hold
> > `edid[0x7e] + 1` blocks of data (1 extra for the base block). It
> > completely ignored the fact that the function was passed `num_blocks`
> > which indicated how much memory had been allocated for the EDID.
> >
> > Let's fix this by adding a bounds check.
> >
> > This is important for handling the case where there's an error in the
> > first block of the EDID. In that case we will call
> > connector_bad_edid() without having re-allocated memory based on
> > `edid[0x7e]`.
> >
> > Fixes: e11f5bd8228f ("drm: Add support for DP 1.4 Compliance edid corruption test")
> > Reported-by: Ville Syrj?l? <[email protected]>
> > Signed-off-by: Douglas Anderson <[email protected]>
> > Reviewed-by: Ville Syrj?l? <[email protected]>
> > ---
> > This problem report came up in the context of a patch I sent out [1]
> > and this is my attempt at a fix. The problem predates my patch,
> > though. I don't personally know anything about DP compliance testing
> > and what should be happening here, nor do I apparently have any
> > hardware that actually reports a bad EDID. Thus this is just compile
> > tested. I'm hoping that someone here can test this and make sure it
> > seems OK to them.
> >
> > Changes in v2:
> > - Added a comment/changed math to help make it easier to grok.
> >
> > drivers/gpu/drm/drm_edid.c | 15 ++++++++++++---
> > 1 file changed, 12 insertions(+), 3 deletions(-)
>
> Pushed this to drm-misc-fixes since the commit it fixes is fairly old.
>
> fdc21c35aaa1 drm/edid: In connector_bad_edid() cap num_of_ext by num_blocks read
BTW seems kasan caught this for us [1]. I didn't notice we had a bug
open about it until now. Just Chris Wilson mentioned it to me in passing
quite a while ago, and I totally forgot about it until I saw your other
patch poking around the same code.
[1] https://gitlab.freedesktop.org/drm/intel/-/issues/4106
--
Ville Syrj?l?
Intel