Ville Syrjälä writes:
> On Thu, Nov 22, 2012 at 05:22:55AM -0500, Egbert Eich wrote:
> > There are displays which announce EDID extension blocks in the
> > Extension Flag of the EDID base block although they are not EDDC
> > capable (ie. take a segment address at I2C slave address 0x30).
> > We test this by looking for an EDID header which is only possible
> > in the base block.
> > If the segment address is not taken into account, this block will
> > be identical to the base block in which case we stop reading further
> > EEDID blocks, correct the extension flag and just return the base
> > block.
> >
> > v2: Split up EDID fixup code into separate commit.
> >
> > Signed-off-by: Egbert Eich <eich@xxxxxxxx>
> > ---
> > drivers/gpu/drm/drm_edid.c | 13 +++++++++++++
> > 1 files changed, 13 insertions(+), 0 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
> > index a952cfe..5a0e331 100644
> > --- a/drivers/gpu/drm/drm_edid.c
> > +++ b/drivers/gpu/drm/drm_edid.c
> > @@ -364,6 +364,19 @@ drm_do_get_edid(struct drm_connector *connector, struct i2c_adapter *adapter)
> > }
> > if (drm_edid_block_valid(block + (valid_extensions + 1) * EDID_LENGTH, j, print_bad_edid)) {
> > valid_extensions++;
> > + /* Test if base block announced extension blocks although
> > + * display is not EDDC capable.
> > + */
> > + if (j == 2) {
> > + int k;
> > + for (k = 0; k < sizeof(edid_header); k++)
> > + if (block[(EDID_LENGTH * 2) + k] != edid_header[k])
> > + break;
> > + if (k == sizeof(edid_header)) {
> > + valid_extensions = 0;
> > + goto done_fix_extension_count;
> > + }
>
> memcmp()? Also couldn't we just memcmp() the whole block against the base
> block, instead of just the header part?
I don't see an advantage of comparing the entire block with the base block:
the signature should already be unique. However I don't insist ;)
Regarding memcmp() you are definitely right, I will change the code.
>
> Also the comment is somehow misleading. It talks about the base block
> even though we're looking at the extension block.
Reason for this patch:
I had a bug report for a monitor announcing extension blocks in the extension
block flag of the base block (over 200!) although it wasn't EDDC capable.
For some reason it got past the ACK check when the segment number was written
to address 0x30 and happily transferred the base block for any odd numbered
block and some garbage for even ones.
The only reliable way we found to catch this condition early was to check if
block 2 had the header of a base block which will happen when the display
cannot deal with the segment number.
This was what I tried to summarize in very few words - maybe i should reword
it a bit.
Thanks!
Egbert.
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/dri-devel
