Skip to contentSkip to conversation

drm/amdgpu: skip fence GFX interrupts disable/enable for S0ix

PATCH New Tim Huang 1 participant 2 messages Aug 15, 2023

Subsystems (2)
lore
Share & tools
lore

Tim Huang

author
#0 lore raw

PATCH 1/2 drm/amdgpu: skip fence GFX interrupts disable/enable for S0ix

GFX v11.0.1 reported fence fallback timer expired issue on SDMA and GFX rings after S0ix resume. This is generated by EOP interrupts are disabled when S0ix suspend but fails to re-enable when resume because of the GFX is in GFXOFF.

[  203.349571] [drm] Fence fallback timer expired on ring sdma0
[  203.349572] [drm] Fence fallback timer expired on ring gfx_0.0.0
[  203.861635] [drm] Fence fallback timer expired on ring gfx_0.0.0

For S0ix, GFX is in GFXOFF state, avoid to touch the GFX registers to configure the fence driver interrupts for rings that belong to GFX. The interrupts configuration will be restored by GFXOFF exit.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 41 +++++++++++++++++++++--
 1 file changed, 39 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +39 -2
index c694b41f6461..7537f5aa76f0 100644
@@ -551,6 +551,41 @@ int amdgpu_fence_driver_sw_init(struct amdgpu_device *adev)
 	return 0;
 }
 
+/**
+ * amdgpu_fence_need_ring_interrupt_restore - helper function to check whether
+ * fence driver interrupts need to be restored.
+ *
+ * @ring: ring that to be checked
+ *
+ * Interrupts for rings that belong to GFX IP don't need to be restored
+ * when the target power state is s0ix.
+ *
+ * Return true if need to restore interrupts, false otherwise.
+ */
+static bool amdgpu_fence_need_ring_interrupt_restore(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	bool is_gfx_power_domain = false;
+
+	switch (ring->funcs->type) {
+	case AMDGPU_RING_TYPE_SDMA:
+	/* SDMA 5.x+ is part of GFX power domain so it's covered by GFXOFF */
+		if (adev->ip_versions[SDMA0_HWIP][0] >= IP_VERSION(5, 0, 0))
+			is_gfx_power_domain = true;
+		break;
+	case AMDGPU_RING_TYPE_GFX:
+	case AMDGPU_RING_TYPE_COMPUTE:
+	case AMDGPU_RING_TYPE_KIQ:
+	case AMDGPU_RING_TYPE_MES:
+		is_gfx_power_domain = true;
+		break;
+	default:
+		break;
+	}
+
+	return !(adev->in_s0ix && is_gfx_power_domain);
+}
+
 /**
  * amdgpu_fence_driver_hw_fini - tear down the fence driver
  * for all possible rings.
@@ -579,7 +614,8 @@ void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
 			amdgpu_fence_driver_force_completion(ring);
 
 		if (!drm_dev_is_unplugged(adev_to_drm(adev)) &&
-		    ring->fence_drv.irq_src)
+		    ring->fence_drv.irq_src &&
+		    amdgpu_fence_need_ring_interrupt_restore(ring))
 			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
 				       ring->fence_drv.irq_type);
 
@@ -655,7 +691,8 @@ void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev)
 			continue;
 
 		/* enable the interrupt */
-		if (ring->fence_drv.irq_src)
+		if (ring->fence_drv.irq_src &&
+		    amdgpu_fence_need_ring_interrupt_restore(ring))
 			amdgpu_irq_get(adev, ring->fence_drv.irq_src,
 				       ring->fence_drv.irq_type);
 	}
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH 2/2 drm/amd: flush any delayed gfxoff on suspend entry

From: Mario Limonciello <[email protected]>

DCN 3.1.4 is reported to hang on s2idle entry if graphics activity is happening during entry. This is because GFXOFF was scheduled as delayed but RLC gets disabled in s2idle entry sequence which will hang GFX IP if not already in GFXOFF.

To help this problem, flush any delayed work for GFXOFF early in s2idle entry sequence to ensure that it's off when RLC is changed.

commit 3964b0c2e843 ("drm/amdgpu: complete gfxoff allow signal during suspend without delay") modified power gating flow so that if called in s0ix that it ensured that GFXOFF wasn't put in work queue but instead processed immediately.

This is dead code due to commit 5d70a549d00d ("drm/amdgpu: skip CG/PG for gfx during S0ix") because GFXOFF will now not be explicitly called as part of the suspend entry code. Remove that dead code.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c    | 9 +--------
 2 files changed, 2 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -0
index 1b3bd77d82da..e77f048c99d8 100644
@@ -4160,6 +4160,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool fbcon)
 		drm_fb_helper_set_suspend_unlocked(adev_to_drm(adev)->fb_helper, true);
 
 	cancel_delayed_work_sync(&adev->delayed_init_work);
+	flush_delayed_work(&adev->gfx.gfx_off_delay_work);
 
 	amdgpu_ras_suspend(adev);
 
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +1 -8
index c76b6bfc4dab..2382921710ec 100644
@@ -700,15 +700,8 @@ void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable)
 
 		if (adev->gfx.gfx_off_req_count == 0 &&
 		    !adev->gfx.gfx_off_state) {
-			/* If going to s2idle, no need to wait */
-			if (adev->in_s0ix) {
-				if (!amdgpu_dpm_set_powergating_by_smu(adev,
-						AMD_IP_BLOCK_TYPE_GFX, true))
-					adev->gfx.gfx_off_state = true;
-			} else {
-				schedule_delayed_work(&adev->gfx.gfx_off_delay_work,
+			schedule_delayed_work(&adev->gfx.gfx_off_delay_work,
 					      delay);
-			}
 		}
 	} else {
 		if (adev->gfx.gfx_off_req_count == 0) {
Signature · 1 line
2.34.1