Skip to contentSkip to conversation

drm/msm: Handle component bind failures a bit better

PATCH In review Jordan Crouse 2 participants 5 messages May 7, 2019 – May 9, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

cover drm/msm: Handle component bind failures a bit better

I somewhat accidently injected an error in the DPU KMS init that caused it to fail and a handful of NULL deferences and errors ended up popping out. Here are some fixes in the interest of robustness.

Jordan Crouse (3):
drm/msm/dpu: Fix error recovery after failing to enable clocks drm/msm/dpu: Avoid a null de-ref while recovering from kms init fail drm/msm/adreno: Call pm_runtime_force_suspend() during unbind

 drivers/gpu/drm/msm/adreno/a6xx_gmu.c       | 4 +---
 drivers/gpu/drm/msm/adreno/adreno_device.c  | 2 +-
 drivers/gpu/drm/msm/disp/dpu1/dpu_io_util.c | 6 +++---
 drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c     | 9 ++++++---
 4 files changed, 11 insertions(+), 10 deletions(-)
Signature · 1 line
2.7.4
#1 lore raw

PATCH 1/3 drm/msm/dpu: Fix error recovery after failing to enable clocks

If enabling clocks fails in msm_dss_enable_clk() the code to unwind the settings starts at 'i' which is the clock that just failed. While this isn't harmful it does result in a number of warnings from the clock subsystem while trying to unpreare/disable the very clock that had just failed to prepare/enable. Skip the current failed clock during the unwind to to avoid the extra log spew.

---

 drivers/gpu/drm/msm/disp/dpu1/dpu_io_util.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/msm/disp/dpu1/dpu_io_util.c +3 -3
index 78833c2..a40a630 100644
@@ -114,9 +114,9 @@ int msm_dss_enable_clk(struct dss_clk *clk_arry, int num_clk, int enable)
 				rc = -EPERM;
 			}
 
-			if (rc) {
-				msm_dss_enable_clk(&clk_arry[i],
-					i, false);
+			if (rc && i) {
+				msm_dss_enable_clk(&clk_arry[i - 1],
+					i - 1, false);
 				break;
 			}
 		}
Signature · 6 lines
2.7.4

_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno
#2 lore raw

PATCH 2/3 drm/msm/dpu: Avoid a null de-ref while recovering from kms init fail

In reply to Jordan Crouse

In the failure path for dpu_kms_init() it is possible to get to the MMU destroy function with uninitialized MMU structs. Check for NULl and skip if needed.

---

 drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +6 -3
index 885bf88..1beaf29 100644
@@ -56,7 +56,7 @@ static const char * const iommu_ports[] = {
 #define DPU_DEBUGFS_HWMASKNAME "hw_log_mask"
 
 static int dpu_kms_hw_init(struct msm_kms *kms);
-static int _dpu_kms_mmu_destroy(struct dpu_kms *dpu_kms);
+static void _dpu_kms_mmu_destroy(struct dpu_kms *dpu_kms);
 
 static unsigned long dpu_iomap_size(struct platform_device *pdev,
 				    const char *name)
@@ -725,17 +725,20 @@ static const struct msm_kms_funcs kms_funcs = {
 #endif
 };
 
-static int _dpu_kms_mmu_destroy(struct dpu_kms *dpu_kms)
+static void _dpu_kms_mmu_destroy(struct dpu_kms *dpu_kms)
 {
 	struct msm_mmu *mmu;
 
+	if (!dpu_kms->base.aspace)
+		return;
+
 	mmu = dpu_kms->base.aspace->mmu;
 
 	mmu->funcs->detach(mmu, (const char **)iommu_ports,
 			ARRAY_SIZE(iommu_ports));
 	msm_gem_address_space_put(dpu_kms->base.aspace);
 
-	return 0;
+	dpu_kms->base.aspace = NULL;
 }
 
 static int _dpu_kms_mmu_init(struct dpu_kms *dpu_kms)
Signature · 6 lines
2.7.4

_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno
Reviewed-by Kristian H. Kristensen on v1 2/3
#3 lore raw

Re: [Freedreno] [PATCH 2/3] drm/msm/dpu: Avoid a null de-ref while recovering from kms init fail

Quoted 3 lines · Jordan Crouse

In the failure path for dpu_kms_init() it is possible to get to the MMU destroy function with uninitialized MMU structs. Check for NULl and skip

s/NULl/NULL

Quoted 3 lines

if needed.

Quoted 48 lines

---

 drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +6 -4
index 885bf88..1beaf29 100644
@@ -56,7 +56,7 @@ static const char * const iommu_ports[] = {
 #define DPU_DEBUGFS_HWMASKNAME "hw_log_mask"
 
 static int dpu_kms_hw_init(struct msm_kms *kms);
-static int _dpu_kms_mmu_destroy(struct dpu_kms *dpu_kms);
+static void _dpu_kms_mmu_destroy(struct dpu_kms *dpu_kms);
 
 static unsigned long dpu_iomap_size(struct platform_device *pdev,
                                    const char *name)
@@ -725,17 +725,20 @@ static const struct msm_kms_funcs kms_funcs = {
 #endif
 };
 
-static int _dpu_kms_mmu_destroy(struct dpu_kms *dpu_kms)
+static void _dpu_kms_mmu_destroy(struct dpu_kms *dpu_kms)
 {
        struct msm_mmu *mmu;
 
+       if (!dpu_kms->base.aspace)
+               return;
+
        mmu = dpu_kms->base.aspace->mmu;
 
        mmu->funcs->detach(mmu, (const char **)iommu_ports,
                        ARRAY_SIZE(iommu_ports));
        msm_gem_address_space_put(dpu_kms->base.aspace);
 
-       return 0;
+       dpu_kms->base.aspace = NULL;
 }
 
 static int _dpu_kms_mmu_init(struct dpu_kms *dpu_kms)
--

2.7.4

_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno

#4 lore raw

PATCH 3/3 drm/msm/adreno: Call pm_runtime_force_suspend() during unbind

In reply to Jordan Crouse

The GPU specific pm_suspend code assumes that the hardware is active when the function is called, which it usually is when called as part of pm_runtime. But during unbind, the pm_suspend functions are called blindly resulting in a bit of a when the hardware wasn't already active (or booted, in the case of the GMU).

Instead of calling the pm_suspend function directly, use pm_runtime_force_suspend() which should check the correct state of runtime and call the functions on our behalf or skip them if they are not needed.

---

 drivers/gpu/drm/msm/adreno/a6xx_gmu.c      | 4 +---
 drivers/gpu/drm/msm/adreno/adreno_device.c | 2 +-
 2 files changed, 2 insertions(+), 4 deletions(-)
drivers/gpu/drm/msm/adreno/a6xx_gmu.c +1 -3
index 9155daf..447706d 100644
@@ -1230,9 +1230,7 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
 	if (IS_ERR_OR_NULL(gmu->mmio))
 		return;
 
-	a6xx_gmu_stop(a6xx_gpu);
-
-	pm_runtime_disable(gmu->dev);
+	pm_runtime_force_suspend(gmu->dev);
 
 	if (!IS_ERR(gmu->gxpd)) {
 		pm_runtime_disable(gmu->gxpd);
drivers/gpu/drm/msm/adreno/adreno_device.c +1 -1
index b907245..3dd90df 100644
@@ -351,7 +351,7 @@ static void adreno_unbind(struct device *dev, struct device *master,
 {
 	struct msm_gpu *gpu = dev_get_drvdata(dev);
 
-	gpu->funcs->pm_suspend(gpu);
+	pm_runtime_force_suspend(dev);
 	gpu->funcs->destroy(gpu);
 
 	set_gpu_pdev(dev_get_drvdata(master), NULL);
Signature · 6 lines
2.7.4

_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno