On Thu, Nov 27, 2025 at 11:40:54PM +0200, Ville Syrjälä wrote:
> On Thu, Nov 27, 2025 at 12:44:06PM +0800, Chia-Lin Kao (AceLan) wrote:
> > Some USB-C hubs and adapters have buggy firmware where multi-byte AUX
> > reads from DPCD address 0x00000 consistently timeout, while single-byte
> > reads from the same address work correctly.
> > 
> > Known affected devices that exhibit this issue:
> > - Lenovo USB-C to VGA adapter (VIA VL817 chipset)
> >   idVendor=17ef, idProduct=7217
> > - Dell DA310 USB-C mobile adapter hub
> >   idVendor=413c, idProduct=c010
> > 
> > Analysis of the failure pattern shows:
> > - Single-byte probes to 0xf0000 (LTTPR) succeed
> > - Single-byte probes to 0x00102 (TRAINING_AUX_RD_INTERVAL) succeed
> > - 15-byte reads from 0x00000 (DPCD capabilities) timeout with -ETIMEDOUT
> > - Retrying does not help - the failure is consistent across all attempts
> 
> I thought we changed that to the more sensible 16 bytes.
> Anyone know what happened to that patch?
> 
> Anyways, does 16 bytes work better than 15 bytes?
Change DP_RECEIVER_CAP_SIZE from 15 to 16 and it doesn't help.
I can see it tried to read 16 bytes but it still timed out.

> 
> > 
> > The issue appears to be a firmware bug in the AUX transaction handling
> > that specifically affects multi-byte reads from the base DPCD address.
> > 
> > Add a fallback mechanism that attempts byte-by-byte reading when the
> > normal multi-byte drm_dp_read_dpcd_caps() fails. This workaround only
> > activates for adapters that fail the standard read path, ensuring no
> > impact on correctly functioning hardware.
> > 
> > The byte-by-byte read uses drm_dp_dpcd_readb() to read each of the 15
> > DPCD capability bytes individually, working around the firmware bug
> > while maintaining compatibility with all other adapters.
> > 
> > Tested with:
> > - Lenovo USB-C to VGA adapter (VIA VL817) - now works with fallback
> > - Dell DA310 USB-C hub - now works with fallback
> > - Dell/Analogix Slimport adapter - continues to work with normal path
> > 
> > Signed-off-by: Chia-Lin Kao (AceLan) <[email protected]>
> > ---
> >  .../drm/i915/display/intel_dp_link_training.c | 21 ++++++++++++++++++-
> >  1 file changed, 20 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/display/intel_dp_link_training.c 
> > b/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> > index aad5fe14962f..738a5bb4adb3 100644
> > --- a/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> > +++ b/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> > @@ -213,6 +213,7 @@ static int intel_dp_init_lttpr(struct intel_dp 
> > *intel_dp, const u8 dpcd[DP_RECEI
> >  int intel_dp_read_dprx_caps(struct intel_dp *intel_dp, u8 
> > dpcd[DP_RECEIVER_CAP_SIZE])
> >  {
> >     struct intel_display *display = to_intel_display(intel_dp);
> > +   int ret, i;
> >  
> >     if (intel_dp_is_edp(intel_dp))
> >             return 0;
> > @@ -226,7 +227,25 @@ int intel_dp_read_dprx_caps(struct intel_dp *intel_dp, 
> > u8 dpcd[DP_RECEIVER_CAP_S
> >                                   
> > DP_LT_TUNABLE_PHY_REPEATER_FIELD_DATA_STRUCTURE_REV))
> >                     return -EIO;
> >  
> > -   if (drm_dp_read_dpcd_caps(&intel_dp->aux, dpcd))
> > +   ret = drm_dp_read_dpcd_caps(&intel_dp->aux, dpcd);
> > +   if (ret == 0)
> > +           return 0;
> > +
> > +   /*
> > +    * Workaround for USB-C hubs/adapters with buggy firmware that fail
> > +    * multi-byte AUX reads from DPCD address 0x00000 but work with
> > +    * single-byte reads. Known affected devices:
> > +    * - Lenovo USB-C to VGA adapter (VIA VL817, idVendor=17ef, 
> > idProduct=7217)
> > +    * - Dell DA310 USB-C hub (idVendor=413c, idProduct=c010)
> > +    * Read the DPCD capabilities byte-by-byte as a fallback.
> > +    */
> > +   for (i = 0; i < DP_RECEIVER_CAP_SIZE; i++) {
> > +           ret = drm_dp_dpcd_readb(&intel_dp->aux, DP_DPCD_REV + i, 
> > &dpcd[i]);
> > +           if (ret < 0)
> > +                   return -EIO;
> > +   }
> 
> Doing this in i915 specific code doesn't make sense.
Right, I'll move the workaround to the drm driver, and submit v2.

> 
> > +
> > +   if (dpcd[DP_DPCD_REV] == 0)
> >             return -EIO;
> >  
> >     return 0;
> > -- 
> > 2.43.0
> 
> -- 
> Ville Syrjälä
> Intel

Reply via email to