[PATCH v2] drm/i915/cdclk: Don't force cd2x pipe select on Xe3 in bxt_sanitize_cdclk()

[PATCH v2] drm/i915/cdclk: Don't force cd2x pipe select on Xe3 in bxt_sanitize_cdclk()

[Chang Jiong(常炯)]

Since commit 2ee8dbd880b1 ("drm/i915/cdclk: Fix up CDCLK_FREQ_DECIMAL
without a full PLL re-enable") bxt_sanitize_cdclk() normalizes the
CD2X pipe select field of the CDCLK_CTL readback to PIPE_NONE before
comparing against the expected value.

However bxt_cdclk_ctl() does not include the CD2X pipe select field
at all on DISPLAY_VER() >= 30, so on Xe3 (PTL) the normalized
readback (pipe field forced to the PIPE_NONE encoding, 0b111) can
never match the expected value (pipe field 0). The comparison now
fails on every boot and we always take the full CDCLK PLL
disable+re-enable path with the BIOS framebuffer actively scanning
out - exactly what this code was trying to avoid. On a Panther Lake
laptop (Arc B390, 8086:b080, eDP 3120x2080@120) this causes
"CPU pipe A FIFO underrun" during probe and persistent full-screen
scanout corruption for the first KMS client (the SDDM greeter);
the corruption lasts until the next full modeset.

Only normalize the CD2X pipe select field on platforms where
bxt_cdclk_ctl() actually programs it, mirroring its DISPLAY_VER()
check.

Fixes: 2ee8dbd880b1 ("drm/i915/cdclk: Fix up CDCLK_FREQ_DECIMAL without a full PLL re-enable")
Cc: stable@vger.kernel.org
Signed-off-by: Chang Jiong(常炯) <xspeed1989@gmail.com>
---
v2: No code changes. Resent with From matching Signed-off-by
    (v1 was sent from 304750714@qq.com, which is not on the
    Intel CI allowlist).

 drivers/gpu/drm/i915/display/intel_cdclk.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.c b/drivers/gpu/drm/i915/display/intel_cdclk.c
index 7bc9b956554b..0438f5ac78c0 100644
--- a/drivers/gpu/drm/i915/display/intel_cdclk.c
+++ b/drivers/gpu/drm/i915/display/intel_cdclk.c
@@ -2366,8 +2366,10 @@ static void bxt_sanitize_cdclk(struct intel_display *display)
 	 * dividers both syncing to an active pipe, or asynchronously
 	 * (PIPE_NONE).
 	 */
-	cdctl &= ~bxt_cdclk_cd2x_pipe(display, INVALID_PIPE);
-	cdctl |= bxt_cdclk_cd2x_pipe(display, INVALID_PIPE);
+	if (DISPLAY_VER(display) < 30) {
+		cdctl &= ~bxt_cdclk_cd2x_pipe(display, INVALID_PIPE);
+		cdctl |= bxt_cdclk_cd2x_pipe(display, INVALID_PIPE);
+	}
 
 	if (cdctl != expected) {
 		if (DISPLAY_VER(display) < 20) {
-- 
2.55.0

Re: [PATCH v2] drm/i915/cdclk: Don't force cd2x pipe select on Xe3 in bxt_sanitize_cdclk()

[Zhefu Zhang]

I hit the same bug independently on a Panther Lake laptop
(Xiaomi Book Pro 14 2026, 8086:b080, Xe3_LPD display 30 stepping B0).

Arch linux 7.1.8 stock reproduces the corruption with two
"CPU pipe A FIFO underrun" errors. With the same change applied on 7.1.8,
the panel is clean and the boot log has no FIFO underrun. Also verified
the same change on current drm-tip (b801f94c49ffdb0f65e37baf31d68570356d4983):
normal panel, no FIFO underrun.

Tested-by: Zhefu Zhang <a723356@gmail.com>

Re: [PATCH v2] drm/i915/cdclk: Don't force cd2x pipe select on Xe3 in bxt_sanitize_cdclk()

[Jani Nikula]

On Sun, 16 Aug 2026, Chang Jiong(常炯) <xspeed1989@gmail.com> wrote:
> Since commit 2ee8dbd880b1 ("drm/i915/cdclk: Fix up CDCLK_FREQ_DECIMAL
> without a full PLL re-enable") bxt_sanitize_cdclk() normalizes the
> CD2X pipe select field of the CDCLK_CTL readback to PIPE_NONE before
> comparing against the expected value.
>
> However bxt_cdclk_ctl() does not include the CD2X pipe select field
> at all on DISPLAY_VER() >= 30, so on Xe3 (PTL) the normalized
> readback (pipe field forced to the PIPE_NONE encoding, 0b111) can
> never match the expected value (pipe field 0). The comparison now
> fails on every boot and we always take the full CDCLK PLL
> disable+re-enable path with the BIOS framebuffer actively scanning
> out - exactly what this code was trying to avoid. On a Panther Lake
> laptop (Arc B390, 8086:b080, eDP 3120x2080@120) this causes
> "CPU pipe A FIFO underrun" during probe and persistent full-screen
> scanout corruption for the first KMS client (the SDDM greeter);
> the corruption lasts until the next full modeset.
>
> Only normalize the CD2X pipe select field on platforms where
> bxt_cdclk_ctl() actually programs it, mirroring its DISPLAY_VER()
> check.
>
> Fixes: 2ee8dbd880b1 ("drm/i915/cdclk: Fix up CDCLK_FREQ_DECIMAL without a full PLL re-enable")
> Cc: stable@vger.kernel.org
> Signed-off-by: Chang Jiong(常炯) <xspeed1989@gmail.com>

Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8550
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8569
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8901

The first one has a patch from Ville that is identical to this one.

I'll try to kick CI to get this tested before merging.

> ---
> v2: No code changes. Resent with From matching Signed-off-by
>     (v1 was sent from 304750714@qq.com, which is not on the
>     Intel CI allowlist).
>
>  drivers/gpu/drm/i915/display/intel_cdclk.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.c b/drivers/gpu/drm/i915/display/intel_cdclk.c
> index 7bc9b956554b..0438f5ac78c0 100644
> --- a/drivers/gpu/drm/i915/display/intel_cdclk.c
> +++ b/drivers/gpu/drm/i915/display/intel_cdclk.c
> @@ -2366,8 +2366,10 @@ static void bxt_sanitize_cdclk(struct intel_display *display)
>  	 * dividers both syncing to an active pipe, or asynchronously
>  	 * (PIPE_NONE).
>  	 */
> -	cdctl &= ~bxt_cdclk_cd2x_pipe(display, INVALID_PIPE);
> -	cdctl |= bxt_cdclk_cd2x_pipe(display, INVALID_PIPE);
> +	if (DISPLAY_VER(display) < 30) {
> +		cdctl &= ~bxt_cdclk_cd2x_pipe(display, INVALID_PIPE);
> +		cdctl |= bxt_cdclk_cd2x_pipe(display, INVALID_PIPE);
> +	}
>  
>  	if (cdctl != expected) {
>  		if (DISPLAY_VER(display) < 20) {

-- 
Jani Nikula, Intel