Skip to contentSkip to conversation

drm/msm: Fix MSM_INFO_GET_IOVA with carveout

PATCH Applied Iskren Chernev 5 participants 7 messages Jan 2, 2021 – Mar 1, 2021

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/msm: Fix MSM_INFO_GET_IOVA with carveout

The msm_gem_get_iova should be guarded with gpu != NULL and not aspace != NULL, because aspace is NULL when using vram carveout.

Fixes 933415e24bd0d ("drm/msm: Add support for private address space instances")
---
 drivers/gpu/drm/msm/msm_drv.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/msm/msm_drv.c +2 -1
index c5e61cb3356df..c1953fb079133 100644
@@ -775,9 +775,10 @@ static int msm_ioctl_gem_info_iova(struct drm_device *dev,
 		struct drm_file *file, struct drm_gem_object *obj,
 		uint64_t *iova)
 {
+	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_file_private *ctx = file->driver_priv;
 
-	if (!ctx->aspace)
+	if (!priv->gpu)
 		return -EINVAL;
 
 	/*
Signature · 1 line
2.29.2
#1 lore raw

Kind reminder that MSM8974, 8994, 8992 and friends are held back by the lack of IOMMU support upstream. There has been an attempt back in 2014(!) [1], but it was either overlooked or forgotten about ever since. I'd be more than happy to see someone look into this, as I have some other bits (almost) ready for both 8974 and 94, but MMUs aren't something I understand well enough yet.

Konrad

[1] https://lists.linuxfoundation.org/pipermail/iommu/2014-June/008993.html

Tested-by on v1 1/1
#2 lore raw
In reply to Iskren Chernev

I've tested all recent GPU bring-up patches on msm8974pro samsung-klte (a330v2) and with this patch everything is OK. But without this we're getting the following in dmesg while running kmscube (which is rendering nothing except black screen):

[   94.969272] msm fd900000.mdss: [drm:hangcheck_handler [msm]] *ERROR* A330: hangcheck detected gpu lockup rb 0!
[   94.970184] msm fd900000.mdss: [drm:hangcheck_handler [msm]] *ERROR* A330:     completed fence: 0
[   94.970873] msm fd900000.mdss: [drm:hangcheck_handler [msm]] *ERROR* A330:     submitted fence: 1
[   94.971600] msm fd900000.mdss: [drm:recover_worker [msm]] *ERROR* A330: hangcheck recover!
[   94.972329] msm fd900000.mdss: [drm:recover_worker [msm]] *ERROR* A330: offending task: kmscube (kmscube)
[   94.974101] revision: 330 (3.3.0.2)
[   94.974117] rb 0: fence:    0/1
[   94.974129] rptr:     36
[   94.974139] rb wptr:  36
[   94.974148] CP_SCRATCH_REG0: 0
[   94.974159] CP_SCRATCH_REG1: 0
[   94.974169] CP_SCRATCH_REG2: 0
[   94.974178] CP_SCRATCH_REG3: 0
[   94.974188] CP_SCRATCH_REG4: 0
[   94.974198] CP_SCRATCH_REG5: 0
[   94.974208] CP_SCRATCH_REG6: 10
[   94.974218] CP_SCRATCH_REG7: 12

So indeed partial revert of "if" condition fixes gpu at least on msm8974.

Quoted 27 lines · Iskren Chernev

The msm_gem_get_iova should be guarded with gpu != NULL and not aspace != NULL, because aspace is NULL when using vram carveout.

Fixes 933415e24bd0d ("drm/msm: Add support for private address space instances")
---
  drivers/gpu/drm/msm/msm_drv.c | 3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/msm/msm_drv.c +2 -1
index c5e61cb3356df..c1953fb079133 100644
@@ -775,9 +775,10 @@ static int msm_ioctl_gem_info_iova(struct drm_device *dev,
  		struct drm_file *file, struct drm_gem_object *obj,
  		uint64_t *iova)
  {
+	struct msm_drm_private *priv = dev->dev_private;
  	struct msm_file_private *ctx = file->driver_priv;
 
-	if (!ctx->aspace)
+	if (!priv->gpu)
  		return -EINVAL;
 
  	/*
#3 lore raw
In reply to Iskren Chernev
Quoted 16 lines · Iskren Chernev

The msm_gem_get_iova should be guarded with gpu != NULL and not aspace != NULL, because aspace is NULL when using vram carveout.

Fixes 933415e24bd0d ("drm/msm: Add support for private address space instances")
---
 drivers/gpu/drm/msm/msm_drv.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/msm/msm_drv.c +0 -0
index c5e61cb3356df..c1953fb079133 100644

drivers/gpu/drm/msm/msm_drv.c static int msm_ioctl_gem_info_iova(struct drm_device *dev, · line 782

Quoted 9 lines
drivers/gpu/drm/msm/msm_drv.c +2 -1
@@ -775,9 +775,10 @@ static int msm_ioctl_gem_info_iova(struct drm_device *dev,
                struct drm_file *file, struct drm_gem_object *obj,
                uint64_t *iova)
 {
+       struct msm_drm_private *priv = dev->dev_private;
        struct msm_file_private *ctx = file->driver_priv;
 
-       if (!ctx->aspace)
+       if (!priv->gpu)
                return -EINVAL;

Does this actually work? It seems like you would hit a null ptr deref in msm_gem_init_vma().. and in general I think a lot of code paths would be surprised by a null address space, so this seems like a risky idea.

Maybe instead we should be creating an address space for the vram carveout?

BR,
-R

Quoted 4 lines
drivers/gpu/drm/msm/msm_drv.c +0 -1
        /*
--

2.29.2

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#4 lore raw
Quoted 24 lines · Rob Clark
Quoted 22 lines · Iskren Chernev

The msm_gem_get_iova should be guarded with gpu != NULL and not aspace != NULL, because aspace is NULL when using vram carveout.

Fixes 933415e24bd0d ("drm/msm: Add support for private address space instances")
---
 drivers/gpu/drm/msm/msm_drv.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/msm/msm_drv.c +1 -0
index c5e61cb3356df..c1953fb079133 100644
@@ -775,9 +775,10 @@ static int msm_ioctl_gem_info_iova(struct drm_device *dev,
                struct drm_file *file, struct drm_gem_object *obj,
                uint64_t *iova)
 {
+       struct msm_drm_private *priv = dev->dev_private;
        struct msm_file_private *ctx = file->driver_priv;
 
Quoted 8 lines

drivers/gpu/drm/msm/msm_drv.c static int msm_ioctl_gem_info_iova(struct drm_device *dev, · line 782

Quoted 3 lines
drivers/gpu/drm/msm/msm_drv.c +1 -1
-       if (!ctx->aspace)
+       if (!priv->gpu)
                return -EINVAL;

Does this actually work? It seems like you would hit a null ptr deref in msm_gem_init_vma().. and in general I think a lot of code paths would be surprised by a null address space, so this seems like a risky idea.

oh, actually, I suppose it is ok, since in the vram carveout case we create the vma up front when the gem obj is created..

(still, it does seem a bit fragile.. and easy for folks testing on devices not using vram carvout to break.. hmm..)

BR,
-R

Quoted 10 lines

Maybe instead we should be creating an address space for the vram carveout?

BR,
-R

Quoted 4 lines
drivers/gpu/drm/msm/msm_drv.c +0 -1
        /*
--

2.29.2

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#5 lore raw
Quoted 34 lines · Rob Clark
Quoted 24 lines · Rob Clark
Quoted 22 lines · Iskren Chernev

The msm_gem_get_iova should be guarded with gpu != NULL and not aspace != NULL, because aspace is NULL when using vram carveout.

Fixes 933415e24bd0d ("drm/msm: Add support for private address space instances")
---
 drivers/gpu/drm/msm/msm_drv.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/msm/msm_drv.c +1 -0
index c5e61cb3356df..c1953fb079133 100644
@@ -775,9 +775,10 @@ static int msm_ioctl_gem_info_iova(struct drm_device *dev,
                struct drm_file *file, struct drm_gem_object *obj,
                uint64_t *iova)
 {
+       struct msm_drm_private *priv = dev->dev_private;
        struct msm_file_private *ctx = file->driver_priv;
 
Quoted 8 lines

drivers/gpu/drm/msm/msm_drv.c static int msm_ioctl_gem_info_iova(struct drm_device *dev, · line 782

Quoted 3 lines
drivers/gpu/drm/msm/msm_drv.c +1 -1
-       if (!ctx->aspace)
+       if (!priv->gpu)
                return -EINVAL;

Does this actually work? It seems like you would hit a null ptr deref in msm_gem_init_vma().. and in general I think a lot of code paths would be surprised by a null address space, so this seems like a risky idea.

Quoted 5 lines

oh, actually, I suppose it is ok, since in the vram carveout case we create the vma up front when the gem obj is created..

(still, it does seem a bit fragile.. and easy for folks testing on devices not using vram carvout to break.. hmm..)

In _msm_gem_new add_vma is called with NULL, so consequently lookup_vma finds it when aspace is NULL.

Also, this is how the code was before the "breaking" change, so it should not be worse.

I'll be happy to work on refactoring this a bit, but some some documentation about the different gpu/mdp pieces and how they fit together won't hurt.

Regards,
Iskren

Quoted 13 lines

BR,
-R

Quoted 10 lines

Maybe instead we should be creating an address space for the vram carveout?

BR,
-R

Quoted 4 lines
drivers/gpu/drm/msm/msm_drv.c +0 -1
        /*
--

2.29.2

#6 lore raw
In reply to Iskren Chernev
This patch was applied to qcom/linux.git (refs/heads/for-next):

Hello:

This patch was applied to qcom/linux.git (refs/heads/for-next):

Quoted 8 lines · On Sat, 2 Jan 2021 22:24:37 +0200 you wrote:

The msm_gem_get_iova should be guarded with gpu != NULL and not aspace != NULL, because aspace is NULL when using vram carveout.

Fixes 933415e24bd0d ("drm/msm: Add support for private address space instances")

[...]

Here is the summary with links:
- drm/msm: Fix MSM_INFO_GET_IOVA with carveout

    https://git.kernel.org/qcom/c/6cefa31e8104

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html