[PATCH] drm: Retry i2c transfer of EDID block after failure

[Alex Deucher]

On Tue, Mar 15, 2011 at 7:04 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> Usually EDID retrieval is fine. However, sometimes, especially when the
> machine is loaded, it fails, but succeeds after a few retries.
>
> Based on a patch by Michael Buesch.
>
> Reported-by: Michael Buesch <mb at bu3sch.de>
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>

Looks good.

Reviewed-by: Alex Deucher <alexdeucher at gmail.com>

> ---
>
> I was going to suggest tuning the i2c_adapter.retries of the affected
> components, but that only covers EGAIN and not EREMOTEIO/ETIMEDOUT that
> afflicts us. With the alteration to only retry the transfer of the
> affected block, this looks to be a reasonable idea and complements the
> logic to retry corrupted transfers.
> -Chris
>
> ---
>  drivers/gpu/drm/drm_edid.c |   44 ++++++++++++++++++++++++++------------------
>  1 files changed, 26 insertions(+), 18 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
> index a245d17..ddc3da9 100644
> --- a/drivers/gpu/drm/drm_edid.c
> +++ b/drivers/gpu/drm/drm_edid.c
> @@ -230,24 +230,32 @@ drm_do_probe_ddc_edid(struct i2c_adapter *adapter, unsigned char *buf,
>                      int block, int len)
>  {
>        unsigned char start = block * EDID_LENGTH;
> -       struct i2c_msg msgs[] = {
> -               {
> -                       .addr   = DDC_ADDR,
> -                       .flags  = 0,
> -                       .len    = 1,
> -                       .buf    = &start,
> -               }, {
> -                       .addr   = DDC_ADDR,
> -                       .flags  = I2C_M_RD,
> -                       .len    = len,
> -                       .buf    = buf,
> -               }
> -       };
> +       int ret, retries = 5;
>
> -       if (i2c_transfer(adapter, msgs, 2) == 2)
> -               return 0;
> +       /* The core i2c driver will automatically retry the transfer if the
> +        * adapter reports EAGAIN. However, we find that bit-banging transfers
> +        * are susceptible to errors under a heavily loaded machine and
> +        * generate spurious NAKs and timeouts. Retrying the transfer
> +        * of the individual block a few times seems to overcome this.
> +        */
> +       do {
> +               struct i2c_msg msgs[] = {
> +                       {
> +                               .addr   = DDC_ADDR,
> +                               .flags  = 0,
> +                               .len    = 1,
> +                               .buf    = &start,
> +                       }, {
> +                               .addr   = DDC_ADDR,
> +                               .flags  = I2C_M_RD,
> +                               .len    = len,
> +                               .buf    = buf,
> +                       }
> +               };
> +               ret = i2c_transfer(adapter, msgs, 2);
> +       } while (ret != 2 && --retries);
>
> -       return -1;
> +       return ret == 2 ? 0 : -1;
>  }
>
>  static u8 *
> --
> 1.7.2.3
>
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel
>