Skip to contentSkip to conversation

drm/i915/display: guard intel_alpm_get_error() against non-ALPM sinks

PATCH Superseded Stephen J. Fuhry 4 participants 8 messages Jun 11 – Jun 12

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915/display: guard intel_alpm_get_error() against non-ALPM sinks

intel_alpm_get_error() unconditionally reads DP_RECEIVER_ALPM_STATUS via AUX from any DP connection that receives a short-pulse interrupt. However, alpm_dpcd is only populated by intel_edp_init_dpcd(), so it is zero for all external (non-eDP) DP and DP-MST connections. On such connections the AUX read fails, producing a spurious

[drm] *ERROR* Error reading ALPM status

and setting sink_alpm_error = true, which is then visible in the ALPM debug output even on hardware that never enabled ALPM.

intel_alpm_disable() already has the same guard (DISPLAY_VER check plus !intel_dp->alpm_dpcd); extend the same logic to intel_alpm_get_error() so it returns early without attempting the AUX transaction when ALPM is not supported by the sink.

Reproducer: ThinkPad T14 Gen1 (Alder Lake-P / i915 DISPLAY_VER 12) docked via Lenovo ThinkPad USB-C Dock Gen2 with external DP-MST monitors. On every DPMS wake the error fires once on the dock's upstream DP port.

Fixes 2063174c22da ("drm/i915/lobf: Check for sink error and disable LOBF")
---
 drivers/gpu/drm/i915/display/intel_alpm.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/gpu/drm/i915/display/intel_alpm.c +3 -0
index a7350ce8e716..4bb0c00741c4 100644
@@ -611,6 +611,9 @@ bool intel_alpm_get_error(struct intel_dp *intel_dp)
 	u8 val;
 	int r;
 
+	if (!intel_dp->alpm_dpcd)
+		return false;
+
 	r = drm_dp_dpcd_readb(aux, DP_RECEIVER_ALPM_STATUS, &val);
 	if (r != 1) {
 		drm_err(display->drm, "Error reading ALPM status\n");

--
2.39.5

Reviewed-by on v1 1/1
#1 lore raw
Quoted 29 lines · Stephen J. Fuhry

intel_alpm_get_error() unconditionally reads DP_RECEIVER_ALPM_STATUS via
AUX from any DP connection that receives a short-pulse interrupt. However,
alpm_dpcd is only populated by intel_edp_init_dpcd(), so it is zero for
all external (non-eDP) DP and DP-MST connections.  On such connections the
AUX read fails, producing a spurious

  [drm] *ERROR* Error reading ALPM status

and setting sink_alpm_error = true, which is then visible in the ALPM debug output even on hardware that never enabled ALPM.

intel_alpm_disable() already has the same guard (DISPLAY_VER check plus
!intel_dp->alpm_dpcd); extend the same logic to intel_alpm_get_error() so
it returns early without attempting the AUX transaction when ALPM is not
supported by the sink.

Reproducer: ThinkPad T14 Gen1 (Alder Lake-P / i915 DISPLAY_VER 12) docked
via Lenovo ThinkPad USB-C Dock Gen2 with external DP-MST monitors. On
every DPMS wake the error fires once on the dock's upstream DP port.

Quoted 3 lines
Fixes 2063174c22da ("drm/i915/lobf: Check for sink error and disable

LOBF")

As intel_alpm.c is not supporting non-eDP ATM for now I think your fix proposal is ok:

Quoted 23 lines

---
 drivers/gpu/drm/i915/display/intel_alpm.c | 3 +++

 1 file changed, 3 insertions(+)
drivers/gpu/drm/i915/display/intel_alpm.c +3 -0
index a7350ce8e716..4bb0c00741c4 100644
@@ -611,6 +611,9 @@ bool intel_alpm_get_error(struct intel_dp
 *intel_dp)
  	u8 val;
  	int r;
 
+	if (!intel_dp->alpm_dpcd)
+		return false;
+
  	r = drm_dp_dpcd_readb(aux, DP_RECEIVER_ALPM_STATUS, &val);
  	if (r != 1) {

  drm_err(display->drm, "Error reading ALPM status\n");
--
2.39.5

#2 lore raw
In reply to Stephen J. Fuhry
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] intel_alpm_get_error() clears the hardware error state on read, causing a secondary caller within the same short pulse interrupt to miss the error. --

commit 7e6bdc8ef5d560933201e9367240c8dcc3b248fe
Author: Stephen J. Fuhry <[email protected]>

drm/i915/display: guard intel_alpm_get_error() against non-ALPM sinks

intel_alpm_get_error() unconditionally reads DP_RECEIVER_ALPM_STATUS via AUX from any DP connection that receives a short-pulse interrupt. However, alpm_dpcd is only populated by intel_edp_init_dpcd(), so it is zero for all external (non-eDP) DP and DP-MST connections. This patch adds a check to return early when ALPM is not supported.

Quoted 14 lines
drivers/gpu/drm/i915/display/intel_alpm.c +3 -0
index c6963ea420cc9..07c1a96ddc6e2 100644
@@ -629,6 +629,9 @@ bool intel_alpm_get_error(struct intel_dp *intel_dp)
 	u8 val;
 	int r;
 
+	if (!intel_dp->alpm_dpcd)
+		return false;
+
 	r = drm_dp_dpcd_readb(aux, DP_RECEIVER_ALPM_STATUS, &val);
 	if (r != 1) {
 		drm_err(display->drm, "Error reading ALPM status\n");

drivers/gpu/drm/i915/display/intel_alpm.c

Quoted 8 lines
drivers/gpu/drm/i915/display/intel_alpm.c +0 -0
 		return true;
 	}
 
 	if (val & DP_ALPM_LOCK_TIMEOUT_ERROR) {
 		drm_dbg_kms(display->drm, "ALPM lock timeout error\n");
 
 		/* Clearing error */
 		drm_dp_dpcd_writeb(aux, DP_RECEIVER_ALPM_STATUS, val);

[Severity: High]
This is a pre-existing issue, but does clearing the hardware error state here cause subsequent checks to miss the error?

During an HPD short pulse, intel_dp_short_pulse() first calls intel_psr_short_pulse(), which calls psr_alpm_check(). This invokes intel_alpm_get_error() which reads and clears the DP_ALPM_LOCK_TIMEOUT_ERROR bit:

drivers/gpu/drm/i915/display/intel_psr.c:psr_alpm_check() {
    ...
    if (intel_alpm_get_error(intel_dp)) {
        intel_psr_disable_locked(intel_dp);
        psr->sink_not_reliable = true;
    }
    ...
}

Later in the same sequence, intel_dp_short_pulse() directly calls intel_alpm_get_error() a second time:

drivers/gpu/drm/i915/display/intel_dp.c:intel_dp_short_pulse() {
    ...
    intel_psr_short_pulse(intel_dp);

    if (intel_alpm_get_error(intel_dp)) {
        intel_alpm_disable(intel_dp);
        intel_dp->alpm.sink_alpm_error = true;
    }
    ...
}

Because the DPCD error bit was already cleared by the first call, the second call will return false.

While psr_alpm_check() disables PSR via intel_psr_disable_locked(), that function only disables ALPM/LOBF for Panel Replay, and psr_alpm_check() only runs when Panel Replay is NOT enabled.

For standard PSR-enabled eDP panels, does the second check failing mean we skip calling intel_alpm_disable(intel_dp), leaving LOBF improperly enabled despite the ALPM timeout?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#3 lore raw

PATCH 1/1 drm/i915/display: fix ALPM error handling in psr_alpm_check()

In reply to Stephen J. Fuhry

intel_dp_short_pulse() calls intel_psr_short_pulse() before the direct intel_alpm_get_error() check. For PSR2-enabled panels, intel_psr_short_pulse() calls psr_alpm_check(), which reads and clears DP_ALPM_LOCK_TIMEOUT_ERROR via intel_alpm_get_error(). The subsequent direct call in intel_dp_short_pulse() then reads zero and skips calling intel_alpm_disable() and setting sink_alpm_error, leaving LOBF potentially enabled despite the ALPM error.

Fix this by having psr_alpm_check() also call intel_alpm_disable() and set sink_alpm_error when it handles an ALPM error. This mirrors what the direct call in intel_dp_short_pulse() would have done, ensuring correct cleanup regardless of which code path consumes the error bit first.

intel_alpm_disable() is already called under psr->lock from intel_psr_disable_locked() for the Panel Replay path, so the lock ordering is established and safe.

Fixes 2063174c22da ("drm/i915/lobf: Check for sink error and disable LOBF")
---
 drivers/gpu/drm/i915/display/intel_psr.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/display/intel_psr.c +2 -1
index 9382ad1e01d8..0701c8b94aa2 100644
@@ -3869,6 +3869,8 @@ static void psr_alpm_check(struct intel_dp *intel_dp)
 	if (intel_alpm_get_error(intel_dp)) {
 		intel_psr_disable_locked(intel_dp);
 		psr->sink_not_reliable = true;
+		intel_alpm_disable(intel_dp);
+		intel_dp->alpm.sink_alpm_error = true;
 	}
 }
--

2.39.5

#4 lore raw

Re: [PATCH] drm/i915/display: fix ALPM error handling in psr_alpm_check()

Quoted 20 lines · Stephen J. Fuhry

intel_dp_short_pulse() calls intel_psr_short_pulse() before the direct
intel_alpm_get_error() check.  For PSR2-enabled panels, intel_psr_short_pulse()
calls psr_alpm_check(), which reads and clears DP_ALPM_LOCK_TIMEOUT_ERROR via
intel_alpm_get_error().  The subsequent direct call in intel_dp_short_pulse()
then reads zero and skips calling intel_alpm_disable() and setting sink_alpm_error, leaving LOBF potentially enabled despite the ALPM error.

Fix this by having psr_alpm_check() also call intel_alpm_disable() and set
sink_alpm_error when it handles an ALPM error.  This mirrors what the direct
call in intel_dp_short_pulse() would have done, ensuring correct cleanup
regardless of which code path consumes the error bit first.

Quoted 4 lines

intel_alpm_disable() is already called under psr->lock from intel_psr_disable_locked() for the Panel Replay path, so the lock ordering
is established and safe.

This one we don't need. LOBF can't be enabled while PSR2 is enabled. If ALPM error gets triggered when PSR2 is enabled -> PSR2 is disabled. After that LOBF might be used. If there is a problem with LOBF then ALPM error is supposed to get triggered again and then LOBF is disabled.

BR,
Jouni Högander

Quoted 25 lines
Fixes 2063174c22da ("drm/i915/lobf: Check for sink error and disable

LOBF")

---
 drivers/gpu/drm/i915/display/intel_psr.c | 2 ++

 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/display/intel_psr.c +2 -1
index 9382ad1e01d8..0701c8b94aa2 100644
@@ -3869,6 +3869,8 @@ static void psr_alpm_check(struct intel_dp
 *intel_dp)
  	if (intel_alpm_get_error(intel_dp)) {
  		intel_psr_disable_locked(intel_dp);
  		psr->sink_not_reliable = true;
+		intel_alpm_disable(intel_dp);
+		intel_dp->alpm.sink_alpm_error = true;
  	}
  }
--

2.39.5

#5 lore raw

Re: [PATCH] drm/i915/display: fix ALPM error handling in psr_alpm_check()

In reply to Stephen J. Fuhry

Thanks for the explanation Jouni. I missed the has_psr guard in intel_alpm_enable_lobf() -- LOBF can't be active while PSR2 is running, so the double-read consequence is benign.

Dropping this patch.

#6 lore raw

✗ LGCI.VerificationFailed: failure for drm/i915/display: guard intel_alpm_get_error() against non-ALPM sinks

In reply to Stephen J. Fuhry
== Series Details ==

== Series Details ==

Series: drm/i915/display: guard intel_alpm_get_error() against non-ALPM sinks
URL : https://patchwork.freedesktop.org/series/168433/
State : failure

== Summary ==

Series author address '[email protected]' is not on the allowlist, which prevents CI from being automatically triggered. If you want CI to run for this series, ask Patchwork project owners to click 'retest' on the series in Patchwork. Exception occurred during validation, bailing out!
Build URL: http://intel-gfx-ci-public.igk.intel.com:8080/job/xe_pw_trigger/1186757/ (on master)

#7 lore raw

✗ LGCI.VerificationFailed: failure for drm/i915/display: guard intel_alpm_get_error() against non-ALPM sinks (rev2)

In reply to Stephen J. Fuhry
== Series Details ==

== Series Details ==

Series: drm/i915/display: guard intel_alpm_get_error() against non-ALPM sinks (rev2)
URL : https://patchwork.freedesktop.org/series/168435/
State : failure

== Summary ==

Series author address '[email protected]' is not on the allowlist, which prevents CI from being automatically triggered. If you want CI to run for this series, ask Patchwork project owners to click 'retest' on the series in Patchwork. Exception occurred during validation, bailing out!
Build URL: http://gfx-ci.igk.intel.com:8080/job/CI_PW_kernel/179293/ (on built-in)