Re: [PATCH 1/3] drm/i915/vrr: Use vrr.in_range to determine if we need the AS SDP

[Jani Nikula <jani.nikula@linux.intel.com>]

On Sat, 26 Sep 2026, "xizheTang2005@163.com" <xizheTang2005@163.com> wrote:
> I tested this series against #9252 (applied by semantics on Fedora
> 7.2.7-200.fc44 xe; the distro tree did not take a clean git-am).

The problem with that is, we don't really know what you've actually
tested. You've adapted the patches on top of a distro kernel, with a
number of stable backports, so we don't know if the result is meaningful
or not.

It would be intestesting to see drm-tip before/after this series.


BR,
Jani.



>
> Machine: Lenovo ThinkBook 14 G8+ IPH (21VG), Panther Lake
> 8086:b080, eDP.
>
> The boot / resume full modeset still shows vertical corruption.
> With Adaptive Sync = Always, a lid s2idle resume is sometimes
> clean and sometimes leaves a persistent streak (same policy,
> repeated resumes). With Never or Automatic the streak appears
> and then recovers on its own. Userspace fastsets on an
> already-lit panel do not show it.
>
> intel_dp_needs_as_sdp() now returns crtc_state->vrr.in_range.
> That flag is set from intel_vrr_is_in_range():
>
>   intel_vrr_is_capable(connector) &&
>   vrefresh >= monitor_range.min_vfreq &&
>   vrefresh <= monitor_range.max_vfreq
>
> It does not look at uapi.vrr_enabled / crtc_state->vrr.enable.
> This panel's EDID range descriptor is 30-120 Hz, so both 60 Hz
> and 120 Hz stay in_range while userspace has VRR off. The VRR
> TG is still programmed (fixed_rr, flipline set).
>
> drm.debug=0xe intel_crtc_state_dump:
>
> 1) Adaptive Sync Never, userspace modeset 3072x1920@60:
>
>    vrr: no, fixed rr: yes, vmin: 2016, vmax: 2016, flipline: 2016,
>         pipeline full: 0, guardband: 91 vsync start: 90, vsync end: 84
>    DP SDP: Adaptive-Sync, revision 2, length 9
>    operation mode: 1 (AVT_FIXED_VTOTAL)
>    requested mode: "3072x1920": 60 390950 ... 2016
>
> 2) Same, Never, 3072x1920@120: identical vrr/flipline/AS SDP lines,
>    requested mode 120 781886.
>
> 3) Never, s2idle resume (lid), 3072x1920@120: same dump, full
>    [modeset] on pipe A. Not an in_range readout mismatch on resume.
>
> 4) Always, s2idle resume (lid), 3072x1920@120, full [modeset]:
>
>    vrr: yes, fixed rr: no, vmin: 2016, vmax: 8064, flipline: 2016
>    DP SDP: Adaptive-Sync, revision 2, length 9
>    operation mode: 0 (AVT_DYNAMIC_VTOTAL)
>    this captured resume was corrupt and stayed that way;
>    later Always resumes on the same setup were sometimes clean
>
> https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/9252
> https://patchwork.freedesktop.org/series/174831/
>
> Thanks,
> Xizhe Tang
>
>
>
>

-- 
Jani Nikula, Intel