Skip to contentSkip to conversation

drm/amd/display: Fix flip-done timeouts on mode1 reset

PATCH In review [email protected] 2 participants 5 messages Jul 23 – Jul 24

Subsystems (4)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH 1/1 drm/amd/display: Fix flip-done timeouts on mode1 reset

From: Leo Li <[email protected]>

The vblank on/off callbacks mixed use of amdgpu_irq_get/put() and amdgpu_dm_crtc_set_vupdate_irq() to enable and disable IRQs.

With get/put, base driver will callback into DC to disable IRQs when refcount == 0. With set_vupdate_irq(), DC is called directly to disable IRQs, bypassing base driver's refcount tracking.

During gpu reset, base driver can restore IRQs via amdgpu_irq_gpu_reset_resume_helper() > amdgpu_irq_update(). So if get/put() is not used (i.e. refcount == 0), then vupdate_irq will be disabled.

This is problematic if DRM requests vblank on before amdgpu_irq_update() is called: drm_vblank_on() > set_vupdate_irq() enables vupdate_irq, but the refcount is still 0. gpu_reset_resume_helper() > irq_update() then immediately disables it, thus leading to flip done timeouts.

This is made worse on DCN since VUPDATE_NO_LOCK is the only IRQ enabled. Prior to 8382cd234981, a combination of GRPH_FLIP and VSTARTUP IRQs were used, and they used get/put(). This explains why 8382cd234981 exposed this issue.

Fix by using get/put() instead of set_vupdate_irq(). DCE is unchanged, since it relies on unbalanced enable/disable calls based on VRR status, and hence requires direct set_vupdate_irq(). Plus, it also uses GRPH_FLIP and VLINE IRQs, which are properly tracked by get/put().

Fixes 8382cd234981 ("drm/amd/display: consolidate DCN vblank/flip handling onto vupdate_no_lock")
---
 .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c  | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c +11 -2
index 05d6915f9a6b0..079d4ccc88da6 100644
@@ -287,10 +287,19 @@ static inline int amdgpu_dm_crtc_set_vblank(struct drm_crtc *crtc, bool enable)
 	 * is enabled. On DCE, vupdate is only needed in VRR mode.
 	 */
 	if (amdgpu_ip_version(adev, DCE_HWIP, 0) != 0) {
-		rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, enable);
+		if (enable) {
+			rc = amdgpu_irq_get(adev, &adev->vupdate_irq, irq_type);
+			drm_dbg_vbl(crtc->dev, "Get vupdate_irq ret=%d\n", rc);
+		} else {
+			rc = amdgpu_irq_put(adev, &adev->vupdate_irq, irq_type);
+			drm_dbg_vbl(crtc->dev, "Put vupdate_irq ret=%d\n", rc);
+		}
 	} else if (dc_supports_vrr(dm->dc->ctx->dce_version)) {
 		if (enable) {
-			/* vblank irq on -> Only need vupdate irq in vrr mode */
+			/* vblank irq on -> Only need vupdate irq in vrr mode
+			 * Not ref-counted since we need explicit enable/disable
+			 * for DCE VRR handling
+			 */
 			if (amdgpu_dm_crtc_vrr_active(acrtc_state))
 				rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, true);
 		} else {
Signature · 1 line
2.54.0
#1 lore raw
Quoted 42 lines · On 7/23/26 13:01, [email protected] wrote:

From: Leo Li <[email protected]>

The vblank on/off callbacks mixed use of amdgpu_irq_get/put() and amdgpu_dm_crtc_set_vupdate_irq() to enable and disable IRQs.

With get/put, base driver will callback into DC to disable IRQs when refcount == 0. With set_vupdate_irq(), DC is called directly to disable IRQs, bypassing base driver's refcount tracking.

During gpu reset, base driver can restore IRQs via amdgpu_irq_gpu_reset_resume_helper() > amdgpu_irq_update(). So if get/put() is not used (i.e. refcount == 0), then vupdate_irq will be disabled.

This is problematic if DRM requests vblank on before amdgpu_irq_update() is called: drm_vblank_on() > set_vupdate_irq() enables vupdate_irq, but the refcount is still 0. gpu_reset_resume_helper() > irq_update() then immediately disables it, thus leading to flip done timeouts.

This is made worse on DCN since VUPDATE_NO_LOCK is the only IRQ enabled. Prior to 8382cd234981, a combination of GRPH_FLIP and VSTARTUP IRQs were used, and they used get/put(). This explains why 8382cd234981 exposed this issue.

Fix by using get/put() instead of set_vupdate_irq(). DCE is unchanged, since it relies on unbalanced enable/disable calls based on VRR status, and hence requires direct set_vupdate_irq(). Plus, it also uses GRPH_FLIP and VLINE IRQs, which are properly tracked by get/put().

Fixes 8382cd234981 ("drm/amd/display: consolidate DCN vblank/flip handling onto vupdate_no_lock")
---
  .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c  | 13 +++++++++++--
  1 file changed, 11 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c +0 -0
index 05d6915f9a6b0..079d4ccc88da6 100644
@@ -287,10 +287,19 @@ static inline int amdgpu_dm_crtc_set_vblank(struct drm_crtc *crtc, bool enable)
  	 * is enabled. On DCE, vupdate is only needed in VRR mode.

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c static inline int amdgpu_dm_crtc_set_vblank(struct drm_crtc *crtc, bool enable) · line 296

Quoted 10 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c +7 -1
  	 */
  	if (amdgpu_ip_version(adev, DCE_HWIP, 0) != 0) {
-		rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, enable);
+		if (enable) {
+			rc = amdgpu_irq_get(adev, &adev->vupdate_irq, irq_type);
+			drm_dbg_vbl(crtc->dev, "Get vupdate_irq ret=%d\n", rc);
+		} else {
+			rc = amdgpu_irq_put(adev, &adev->vupdate_irq, irq_type);
+			drm_dbg_vbl(crtc->dev, "Put vupdate_irq ret=%d\n", rc);
+		}

For completeness in error handling, shouldn't you pass up the return code on non-zero? It looks like it could pass up to DRM core then.

Quoted 10 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c +4 -1
  	} else if (dc_supports_vrr(dm->dc->ctx->dce_version)) {
  		if (enable) {
-			/* vblank irq on -> Only need vupdate irq in vrr mode */
+			/* vblank irq on -> Only need vupdate irq in vrr mode
+			 * Not ref-counted since we need explicit enable/disable
+			 * for DCE VRR handling
+			 */
  			if (amdgpu_dm_crtc_vrr_active(acrtc_state))
  				rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, true);
  		} else {
author maintainer
#2 lore raw
Quoted 17 lines · Mario Limonciello
Quoted 8 lines

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c index 05d6915f9a6b0..079d4ccc88da6 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
@@ -287,10 +287,19 @@ static inline int amdgpu_dm_crtc_set_vblank(struct drm_crtc *crtc, bool enable)
       * is enabled. On DCE, vupdate is only needed in VRR mode.
       */
      if (amdgpu_ip_version(adev, DCE_HWIP, 0) != 0) {

Quoted 8 lines

-        rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, enable); +        if (enable) {
+            rc = amdgpu_irq_get(adev, &adev->vupdate_irq, irq_type); +            drm_dbg_vbl(crtc->dev, "Get vupdate_irq ret=%d\n", rc); +        } else {
+            rc = amdgpu_irq_put(adev, &adev->vupdate_irq, irq_type); +            drm_dbg_vbl(crtc->dev, "Put vupdate_irq ret=%d\n", rc); +        }

Quoted 1 line

For completeness in error handling, shouldn't you pass up the return code on non-zero?  It looks like it could pass up to DRM core then.

It's not shown in the context here, but the rc is passed up later:
https://elixir.bootlin.com/linux/v7.2-rc4/source/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c#L295

- Leo

Quoted 11 lines
Quoted 11 lines

      } else if (dc_supports_vrr(dm->dc->ctx->dce_version)) { if (enable) {
-            /* vblank irq on -> Only need vupdate irq in vrr mode */ +            /* vblank irq on -> Only need vupdate irq in vrr mode +             * Not ref-counted since we need explicit enable/disable +             * for DCE VRR handling
+             */
              if (amdgpu_dm_crtc_vrr_active(acrtc_state)) rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, true); } else {

Reviewed-by
#3 lore raw
Quoted 20 lines · Leo Li
Quoted 17 lines · Mario Limonciello
Quoted 16 lines

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c index 05d6915f9a6b0..079d4ccc88da6 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
@@ -287,10 +287,19 @@ static inline int amdgpu_dm_crtc_set_vblank(struct drm_crtc *crtc, bool enable)
       * is enabled. On DCE, vupdate is only needed in VRR mode.
       */
      if (amdgpu_ip_version(adev, DCE_HWIP, 0) != 0) {
-        rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, enable);
+        if (enable) {
+            rc = amdgpu_irq_get(adev, &adev->vupdate_irq, irq_type);
+            drm_dbg_vbl(crtc->dev, "Get vupdate_irq ret=%d\n", rc);
+        } else {
+            rc = amdgpu_irq_put(adev, &adev->vupdate_irq, irq_type);
+            drm_dbg_vbl(crtc->dev, "Put vupdate_irq ret=%d\n", rc);
+        }

Quoted 7 lines
Quoted 1 line

For completeness in error handling, shouldn't you pass up the return code on non-zero?  It looks like it could pass up to DRM core then.

It's not shown in the context here, but the rc is passed up later:
https://elixir.bootlin.com/linux/v7.2-rc4/source/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c#L295

- Leo

Ah got it, thanks. No concerns.

Quoted 12 lines
Quoted 11 lines
Quoted 11 lines

      } else if (dc_supports_vrr(dm->dc->ctx->dce_version)) { if (enable) {
-            /* vblank irq on -> Only need vupdate irq in vrr mode */ +            /* vblank irq on -> Only need vupdate irq in vrr mode +             * Not ref-counted since we need explicit enable/disable +             * for DCE VRR handling
+             */
              if (amdgpu_dm_crtc_vrr_active(acrtc_state)) rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, true); } else {

author maintainer Applied
#4 lore raw
Quoted 23 lines · Mario Limonciello
Quoted 22 lines · Leo Li
Quoted 18 lines · Mario Limonciello
Quoted 8 lines

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c index 05d6915f9a6b0..079d4ccc88da6 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c
@@ -287,10 +287,19 @@ static inline int amdgpu_dm_crtc_set_vblank(struct drm_crtc *crtc, bool enable)
        * is enabled. On DCE, vupdate is only needed in VRR mode.
        */
       if (amdgpu_ip_version(adev, DCE_HWIP, 0) != 0) {

Quoted 8 lines

-        rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, enable); +        if (enable) {
+            rc = amdgpu_irq_get(adev, &adev->vupdate_irq, irq_type); +            drm_dbg_vbl(crtc->dev, "Get vupdate_irq ret=%d\n", rc); +        } else {
+            rc = amdgpu_irq_put(adev, &adev->vupdate_irq, irq_type); +            drm_dbg_vbl(crtc->dev, "Put vupdate_irq ret=%d\n", rc); +        }

For completeness in error handling, shouldn't you pass up the return code on non-zero?  It looks like it could pass up to DRM core then.

Quoted 8 lines
Quoted 5 lines

It's not shown in the context here, but the rc is passed up later:
https://elixir.bootlin.com/linux/v7.2-rc4/source/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c#L295

- Leo

Ah got it, thanks.  No concerns.

Merged into asdn, thanks!
- Leo

Quoted 14 lines
Quoted 13 lines
Quoted 12 lines
Quoted 12 lines

       } else if (dc_supports_vrr(dm->dc->ctx->dce_version)) { if (enable) {
-            /* vblank irq on -> Only need vupdate irq in vrr mode */ +            /* vblank irq on -> Only need vupdate irq in vrr mode +             * Not ref-counted since we need explicit enable/disable +             * for DCE VRR handling
+             */
               if (amdgpu_dm_crtc_vrr_active(acrtc_state)) rc = amdgpu_dm_crtc_set_vupdate_irq(crtc, true); } else {