Skip to contentSkip to conversation

DRM GEM fixes

PATCH Superseded Dmitry Osipenko 5 participants 32 messages Jul 1, 2022 – Sep 11, 2022

Subsystems (4)

v8 of 19 Superseded Latest is v19 →

lore
Share & tools
lore
#0 lore raw

v8 cover DRM GEM fixes

This patchset fixes two problems in the common GEM code. First fixed problem is the bogus lockdep splat that complicates debugging of DRM drivers. Second problem is the inconsistency in behaviour and improper handling of mapping the imported GEMs by some drivers, to fix it we will prohibit to map the imported GEMs like majority of drivers already do.

Changelog:

v8: - Dropped stable tag from the "Don't map imported GEMs" patch,
like was suggested by Thomas Hellström in the review comment to v7.

v7: - Factored out GEM patches from [1] since I'll be working on the
dma-buf locking in a separate patchset now.

[1] https://lore.kernel.org/all/[email protected]/

- Improved commit message and added fixes tag to the "Properly annotate WW context" patch.

- Replaced "Move mapping of imported dma-bufs to drm_gem_mmap_obj()" patch with "Don't map imported GEMs", like was suggested by Thomas Hellström.

- Added r-b and suggested-by from Thomas Hellström.

Dmitry Osipenko (2):
drm/gem: Properly annotate WW context on drm_gem_lock_reservations()
error
drm/gem: Don't map imported GEMs

 drivers/gpu/drm/drm_gem.c              | 8 ++++++--
 drivers/gpu/drm/drm_gem_shmem_helper.c | 9 ---------
 2 files changed, 6 insertions(+), 11 deletions(-)
Signature · 1 line
2.36.1
#1 lore raw

PATCH v8 1/2 drm/gem: Properly annotate WW context on drm_gem_lock_reservations() error

Use ww_acquire_fini() in the error code paths. Otherwise lockdep thinks that lock is held when lock's memory is freed after the drm_gem_lock_reservations() error. The ww_acquire_context needs to be annotated as "released", which fixes the noisy "WARNING: held lock freed!" splat of VirtIO-GPU driver with CONFIG_DEBUG_MUTEXES=y and enabled lockdep.

Fixes 7edc3e3b975b5 ("drm: Add helpers for locking an array of BO reservations.")
---
 drivers/gpu/drm/drm_gem.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/drm_gem.c +2 -2
index eb0c2d041f13..86d670c71286 100644
@@ -1226,7 +1226,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
 		ret = dma_resv_lock_slow_interruptible(obj->resv,
 								 acquire_ctx);
 		if (ret) {
-			ww_acquire_done(acquire_ctx);
+			ww_acquire_fini(acquire_ctx);
 			return ret;
 		}
 	}
@@ -1251,7 +1251,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
 				goto retry;
 			}
 
-			ww_acquire_done(acquire_ctx);
+			ww_acquire_fini(acquire_ctx);
 			return ret;
 		}
 	}
Signature · 1 line
2.36.1
Reviewed-by on v8 1/2
#2 lore raw

Re: [PATCH v8 1/2] drm/gem: Properly annotate WW context on drm_gem_lock_reservations() error

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 10 lines

Use ww_acquire_fini() in the error code paths. Otherwise lockdep thinks that lock is held when lock's memory is freed after the drm_gem_lock_reservations() error. The ww_acquire_context needs to be annotated as "released", which fixes the noisy "WARNING: held lock freed!" splat of VirtIO-GPU driver with CONFIG_DEBUG_MUTEXES=y and enabled lockdep.

Fixes 7edc3e3b975b5 ("drm: Add helpers for locking an array of BO reservations.")
Quoted 26 lines
---
  drivers/gpu/drm/drm_gem.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/drm_gem.c +2 -2
index eb0c2d041f13..86d670c71286 100644
@@ -1226,7 +1226,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
  		ret = dma_resv_lock_slow_interruptible(obj->resv,
  								 acquire_ctx);
  		if (ret) {
-			ww_acquire_done(acquire_ctx);
+			ww_acquire_fini(acquire_ctx);
  			return ret;
  		}
  	}
@@ -1251,7 +1251,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
  				goto retry;
  			}
 
-			ww_acquire_done(acquire_ctx);
+			ww_acquire_fini(acquire_ctx);
  			return ret;
  		}
  	}
#3 lore raw

Re: [PATCH v8 1/2] drm/gem: Properly annotate WW context on drm_gem_lock_reservations() error

Quoted 7 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 6 lines

Use ww_acquire_fini() in the error code paths. Otherwise lockdep thinks that lock is held when lock's memory is freed after the drm_gem_lock_reservations() error. The ww_acquire_context needs to be annotated as "released", which fixes the noisy "WARNING: held lock freed!" splat of VirtIO-GPU driver with CONFIG_DEBUG_MUTEXES=y and enabled lockdep.

Quoted 6 lines
Quoted 4 lines
Fixes 7edc3e3b975b5 ("drm: Add helpers for locking an array of BO reservations.")

Also added this r-b tag when merging to drm-misc-next-fixes. -Daniel

Quoted 27 lines
Quoted 26 lines
---
  drivers/gpu/drm/drm_gem.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/drm_gem.c +2 -2
index eb0c2d041f13..86d670c71286 100644
@@ -1226,7 +1226,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
  		ret = dma_resv_lock_slow_interruptible(obj->resv,
  								 acquire_ctx);
  		if (ret) {
-			ww_acquire_done(acquire_ctx);
+			ww_acquire_fini(acquire_ctx);
  			return ret;
  		}
  	}
@@ -1251,7 +1251,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
  				goto retry;
  			}
-			ww_acquire_done(acquire_ctx);
+			ww_acquire_fini(acquire_ctx);
  			return ret;
  		}
  	}
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
#4 lore raw

Re: [PATCH v8 1/2] drm/gem: Properly annotate WW context on drm_gem_lock_reservations() error

Am 09.08.22 um 18:44 schrieb Daniel Vetter:

Quoted 8 lines · Christian König
Quoted 7 lines

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 6 lines

Use ww_acquire_fini() in the error code paths. Otherwise lockdep thinks that lock is held when lock's memory is freed after the drm_gem_lock_reservations() error. The ww_acquire_context needs to be annotated as "released", which fixes the noisy "WARNING: held lock freed!" splat of VirtIO-GPU driver with CONFIG_DEBUG_MUTEXES=y and enabled lockdep.

Quoted 6 lines
Quoted 5 lines
Quoted 4 lines
Fixes 7edc3e3b975b5 ("drm: Add helpers for locking an array of BO reservations.")

Also added this r-b tag when merging to drm-misc-next-fixes.

IIRC I've already pushed this to drm-misc-fixes with a CC stable tag about 2 weeks ago.

Please double check, it probably just hasn't come down the stream again yet.

Christian.

Quoted 27 lines

-Daniel

Quoted 25 lines
Quoted 25 lines
---
   drivers/gpu/drm/drm_gem.c | 4 ++--
   1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/drm_gem.c +2 -2
index eb0c2d041f13..86d670c71286 100644
@@ -1226,7 +1226,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
   		ret = dma_resv_lock_slow_interruptible(obj->resv,
   								 acquire_ctx);
   		if (ret) {
-			ww_acquire_done(acquire_ctx);
+			ww_acquire_fini(acquire_ctx);
   			return ret;
   		}
   	}
@@ -1251,7 +1251,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
   				goto retry;
   			}
-			ww_acquire_done(acquire_ctx);
+			ww_acquire_fini(acquire_ctx);
   			return ret;
   		}
   	}
#5 lore raw

Re: [PATCH v8 1/2] drm/gem: Properly annotate WW context on drm_gem_lock_reservations() error

Quoted 14 lines · Christian König

Am 09.08.22 um 18:44 schrieb Daniel Vetter:

Quoted 12 lines · Christian König
Quoted 11 lines

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 10 lines

Use ww_acquire_fini() in the error code paths. Otherwise lockdep thinks that lock is held when lock's memory is freed after the drm_gem_lock_reservations() error. The ww_acquire_context needs to be annotated as "released", which fixes the noisy "WARNING: held lock freed!" splat of VirtIO-GPU driver with CONFIG_DEBUG_MUTEXES=y and enabled lockdep.

Fixes 7edc3e3b975b5 ("drm: Add helpers for locking an array of BO reservations.")
Quoted 7 lines
Quoted 2 lines
Quoted 1 line

Also added this r-b tag when merging to drm-misc-next-fixes.

IIRC I've already pushed this to drm-misc-fixes with a CC stable tag about 2 weeks ago.

Please double check, it probably just hasn't come down the stream again yet.

Hm quickly check and I didn't spot it? There's a few patches from Dmitry in the last few pulls, and some more stuff pending, but not these two afaics?
-Daniel

Quoted 31 lines

Christian.

Quoted 27 lines

-Daniel

Quoted 25 lines
Quoted 25 lines
---
   drivers/gpu/drm/drm_gem.c | 4 ++--
   1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/drm_gem.c +2 -2
index eb0c2d041f13..86d670c71286 100644
@@ -1226,7 +1226,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
            ret = dma_resv_lock_slow_interruptible(obj->resv,
                                                             acquire_ctx);
            if (ret) {
-                   ww_acquire_done(acquire_ctx);
+                   ww_acquire_fini(acquire_ctx);
                    return ret;
            }
    }
@@ -1251,7 +1251,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
                            goto retry;
                    }
-                   ww_acquire_done(acquire_ctx);
+                   ww_acquire_fini(acquire_ctx);
                    return ret;
            }
    }
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
#6 lore raw

Re: [PATCH v8 1/2] drm/gem: Properly annotate WW context on drm_gem_lock_reservations() error

Am 10.08.22 um 10:33 schrieb Daniel Vetter:

Quoted 15 lines · Christian König
Quoted 14 lines

Am 09.08.22 um 18:44 schrieb Daniel Vetter:

Quoted 13 lines · Christian König
Quoted 12 lines

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 10 lines

Use ww_acquire_fini() in the error code paths. Otherwise lockdep thinks that lock is held when lock's memory is freed after the drm_gem_lock_reservations() error. The ww_acquire_context needs to be annotated as "released", which fixes the noisy "WARNING: held lock freed!" splat of VirtIO-GPU driver with CONFIG_DEBUG_MUTEXES=y and enabled lockdep.

Fixes 7edc3e3b975b5 ("drm: Add helpers for locking an array of BO reservations.")
Quoted 8 lines
Quoted 5 lines
Quoted 1 line

Also added this r-b tag when merging to drm-misc-next-fixes.

IIRC I've already pushed this to drm-misc-fixes with a CC stable tag about 2 weeks ago.

Please double check, it probably just hasn't come down the stream again yet.

Hm quickly check and I didn't spot it? There's a few patches from Dmitry in the last few pulls, and some more stuff pending, but not these two afaics?

Mhm, there is some potential that I wanted to push it but got distracted by the re-occurring drm-tip build breakages.

Anyway what I wanted to say is that this stuff should probably go to drm-misc-fixes with a CC: stable tag :)

Christian.

Quoted 23 lines

-Daniel

Quoted 21 lines

Christian.

Quoted 19 lines

-Daniel

Quoted 17 lines
Quoted 17 lines
---
    drivers/gpu/drm/drm_gem.c | 4 ++--
    1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/drm_gem.c +1 -1
index eb0c2d041f13..86d670c71286 100644
@@ -1226,7 +1226,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
             ret = dma_resv_lock_slow_interruptible(obj->resv,
                                                              acquire_ctx);
             if (ret) {
-                   ww_acquire_done(acquire_ctx);
+                   ww_acquire_fini(acquire_ctx);
                     return ret;
             }
     }
Quoted 9 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

drivers/gpu/drm/drm_gem.c drm_gem_lock_reservations(struct drm_gem_object **objs, int count, · line 1256

Quoted 8 lines
drivers/gpu/drm/drm_gem.c +1 -1
@@ -1251,7 +1251,7 @@ drm_gem_lock_reservations(struct drm_gem_object **objs, int count,
                             goto retry;
                     }
-                   ww_acquire_done(acquire_ctx);
+                   ww_acquire_fini(acquire_ctx);
                     return ret;
             }
     }

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#7 lore raw

PATCH v8 2/2 drm/gem: Don't map imported GEMs

In reply to Dmitry Osipenko

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

---
 drivers/gpu/drm/drm_gem.c              | 4 ++++
 drivers/gpu/drm/drm_gem_shmem_helper.c | 9 ---------
 2 files changed, 4 insertions(+), 9 deletions(-)
drivers/gpu/drm/drm_gem.c +4 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object *obj, unsigned long obj_size,
 {
 	int ret;
 
+	/* Don't allow imported objects to be mapped */
+	if (obj->import_attach)
+		return -EINVAL;
+
 	/* Check for valid size. */
 	if (obj_size < vma->vm_end - vma->vm_start)
 		return -EINVAL;
drivers/gpu/drm/drm_gem_shmem_helper.c +0 -9
index 8ad0e02991ca..6190f5018986 100644
@@ -609,17 +609,8 @@ EXPORT_SYMBOL_GPL(drm_gem_shmem_vm_ops);
  */
 int drm_gem_shmem_mmap(struct drm_gem_shmem_object *shmem, struct vm_area_struct *vma)
 {
-	struct drm_gem_object *obj = &shmem->base;
 	int ret;
 
-	if (obj->import_attach) {
-		/* Drop the reference drm_gem_mmap_obj() acquired.*/
-		drm_gem_object_put(obj);
-		vma->vm_private_data = NULL;
-
-		return dma_buf_mmap(obj->dma_buf, vma, 0);
-	}
-
 	ret = drm_gem_shmem_get_pages(shmem);
 	if (ret) {
 		drm_gem_vm_close(vma);
Signature · 1 line
2.36.1
#8 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 6 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Quoted 8 lines

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

Anyway patch is Reviewed-by: Christian König <[email protected]> since you are really fixing a major stability problem here.

Regards,
Christian.

Quoted 42 lines
---
  drivers/gpu/drm/drm_gem.c              | 4 ++++
  drivers/gpu/drm/drm_gem_shmem_helper.c | 9 ---------
  2 files changed, 4 insertions(+), 9 deletions(-)
drivers/gpu/drm/drm_gem.c +4 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object *obj, unsigned long obj_size,
  {
  	int ret;
 
+	/* Don't allow imported objects to be mapped */
+	if (obj->import_attach)
+		return -EINVAL;
+
  	/* Check for valid size. */
  	if (obj_size < vma->vm_end - vma->vm_start)
  		return -EINVAL;
drivers/gpu/drm/drm_gem_shmem_helper.c +0 -9
index 8ad0e02991ca..6190f5018986 100644
@@ -609,17 +609,8 @@ EXPORT_SYMBOL_GPL(drm_gem_shmem_vm_ops);
   */
  int drm_gem_shmem_mmap(struct drm_gem_shmem_object *shmem, struct vm_area_struct *vma)
  {
-	struct drm_gem_object *obj = &shmem->base;
  	int ret;
 
-	if (obj->import_attach) {
-		/* Drop the reference drm_gem_mmap_obj() acquired.*/
-		drm_gem_object_put(obj);
-		vma->vm_private_data = NULL;
-
-		return dma_buf_mmap(obj->dma_buf, vma, 0);
-	}
-
  	ret = drm_gem_shmem_get_pages(shmem);
  	if (ret) {
  		drm_gem_vm_close(vma);
#9 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

BR,
-R

Quoted 49 lines

Anyway patch is Reviewed-by: Christian König <[email protected]> since you are really fixing a major stability problem here.

Regards,
Christian.

Quoted 42 lines
---
  drivers/gpu/drm/drm_gem.c              | 4 ++++
  drivers/gpu/drm/drm_gem_shmem_helper.c | 9 ---------
  2 files changed, 4 insertions(+), 9 deletions(-)
drivers/gpu/drm/drm_gem.c +4 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object *obj, unsigned long obj_size,
  {
      int ret;
 
+     /* Don't allow imported objects to be mapped */
+     if (obj->import_attach)
+             return -EINVAL;
+
      /* Check for valid size. */
      if (obj_size < vma->vm_end - vma->vm_start)
              return -EINVAL;
drivers/gpu/drm/drm_gem_shmem_helper.c +0 -9
index 8ad0e02991ca..6190f5018986 100644
@@ -609,17 +609,8 @@ EXPORT_SYMBOL_GPL(drm_gem_shmem_vm_ops);
   */
  int drm_gem_shmem_mmap(struct drm_gem_shmem_object *shmem, struct vm_area_struct *vma)
  {
-     struct drm_gem_object *obj = &shmem->base;
      int ret;
 
-     if (obj->import_attach) {
-             /* Drop the reference drm_gem_mmap_obj() acquired.*/
-             drm_gem_object_put(obj);
-             vma->vm_private_data = NULL;
-
-             return dma_buf_mmap(obj->dma_buf, vma, 0);
-     }
-
      ret = drm_gem_shmem_get_pages(shmem);
      if (ret) {
              drm_gem_vm_close(vma);
#10 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Signature · 2 lines
Best regards,
Dmitry
#11 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.
-Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
#12 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object

*obj, unsigned long obj_size,

 {
        int ret;

+ WARN_ON_ONCE(obj->import_attach);
+

        /* Check for valid size. */
        if (obj_size < vma->vm_end - vma->vm_start)
                return -EINVAL;
(quoted diff) +0 -0

b/drivers/gpu/drm/drm_gem_shmem_helper.c
index 8ad0e02991ca..6190f5018986 100644

drivers/gpu/drm/drm_gem_shmem_helper.c +0 -0
@@ -609,17 +609,8 @@ EXPORT_SYMBOL_GPL(drm_gem_shmem_vm_ops);
  */
 int drm_gem_shmem_mmap(struct drm_gem_shmem_object *shmem, struct

vm_area_struct *vma)
{

-       struct drm_gem_object *obj = &shmem->base;
        int ret;

        if (obj->import_attach) {
-               /* Drop the reference drm_gem_mmap_obj() acquired.*/
-               drm_gem_object_put(obj);
-               vma->vm_private_data = NULL;
-
-               return dma_buf_mmap(obj->dma_buf, vma, 0);
+               return -EINVAL;
        }

        ret = drm_gem_shmem_get_pages(shmem);
        if (ret) {
                drm_gem_vm_close(vma);

--
2.36.1

#13 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 46 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

Quoted 6 lines

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 13 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +1 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Signature · 2 lines
Best regards,
Dmitry
#14 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 71 lines · Dmitry Osipenko
Quoted 46 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

Quoted 6 lines

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

BR,
-R

#15 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 79 lines · Rob Clark
Quoted 71 lines · Dmitry Osipenko
Quoted 46 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

Quoted 6 lines

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Quoted 7 lines

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

It's not clear to me what bad Android does form yours comments. Does it export dma-buf from GPU and then import it to GPU? If yes, then DRM core has a check for the self-importing [1].

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_prime.c#L918

If you're meaning something else, then please explain in a more details.

Quoted 7 lines

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

That's what the first versions of the patch did and Christian suggested that it's not a good approach. In fact it should be not only Tegra that has a broken dma-buf mapping, but apparently OMAP driver too.

Signature · 2 lines
Best regards,
Dmitry
#16 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 89 lines · Dmitry Osipenko
Quoted 79 lines · Rob Clark
Quoted 71 lines · Dmitry Osipenko
Quoted 46 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

Quoted 6 lines

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Quoted 7 lines

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

Quoted 8 lines

It's not clear to me what bad Android does form yours comments. Does it export dma-buf from GPU and then import it to GPU? If yes, then DRM core has a check for the self-importing [1].

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_prime.c#L918

If you're meaning something else, then please explain in a more details.

So, android/gralloc allocates buffers externally to the driver and imports them into driver. (And that seems to not just be window surfaces, but in cases random textures, etc)

In the normal case these should be allocated from drm/msm so it should hit [1].. this is the "dumb luck" I mentioned earlier. But I'm not confident enough to say that there is no other case.

Quoted 9 lines
Quoted 8 lines

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

Quoted 3 lines

That's what the first versions of the patch did and Christian suggested that it's not a good approach. In fact it should be not only Tegra that has a broken dma-buf mapping, but apparently OMAP driver too.

Hmm, I guess I'm a bit more conservative when it comes to potentially breaking userspace.

BR,
-R

Quoted 4 lines

--
Best regards,
Dmitry

#17 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 61 lines · Rob Clark
Quoted 59 lines · Dmitry Osipenko
Quoted 57 lines · Rob Clark
Quoted 55 lines · Dmitry Osipenko
Quoted 52 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Upd: I re-checked Panfrost using today's linux-next and mapping of imported dma-buf works, I mapped imported buf from video decoder. Apparently previously I had some local kernel change that broke the mapping.

Quoted 52 lines
Quoted 26 lines
Quoted 18 lines
Quoted 12 lines
Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Quoted 7 lines

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

Quoted 8 lines

It's not clear to me what bad Android does form yours comments. Does it export dma-buf from GPU and then import it to GPU? If yes, then DRM core has a check for the self-importing [1].

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_prime.c#L918

If you're meaning something else, then please explain in a more details.

So, android/gralloc allocates buffers externally to the driver and imports them into driver. (And that seems to not just be window surfaces, but in cases random textures, etc)

In the normal case these should be allocated from drm/msm so it should hit [1].. this is the "dumb luck" I mentioned earlier. But I'm not confident enough to say that there is no other case.

Quoted 9 lines
Quoted 8 lines

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

Quoted 6 lines
Quoted 3 lines

That's what the first versions of the patch did and Christian suggested that it's not a good approach. In fact it should be not only Tegra that has a broken dma-buf mapping, but apparently OMAP driver too.

Hmm, I guess I'm a bit more conservative when it comes to potentially breaking userspace.

If such userspace exists, then of course the mapping should continue to work. Still will be great to know what that userpsace is.

Alright, let's keep the dma-buf mapping as-is for now. I'll fix just the Tegra driver then.

Signature · 2 lines
Best regards,
Dmitry
#18 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 126 lines · Dmitry Osipenko
Quoted 61 lines · Rob Clark
Quoted 59 lines · Dmitry Osipenko
Quoted 57 lines · Rob Clark
Quoted 55 lines · Dmitry Osipenko
Quoted 52 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Upd: I re-checked Panfrost using today's linux-next and mapping of imported dma-buf works, I mapped imported buf from video decoder. Apparently previously I had some local kernel change that broke the mapping.

Quoted 56 lines
Quoted 26 lines
Quoted 18 lines
Quoted 12 lines
Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Quoted 7 lines

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

Quoted 8 lines

It's not clear to me what bad Android does form yours comments. Does it export dma-buf from GPU and then import it to GPU? If yes, then DRM core has a check for the self-importing [1].

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_prime.c#L918

If you're meaning something else, then please explain in a more details.

So, android/gralloc allocates buffers externally to the driver and imports them into driver. (And that seems to not just be window surfaces, but in cases random textures, etc)

In the normal case these should be allocated from drm/msm so it should hit [1].. this is the "dumb luck" I mentioned earlier. But I'm not confident enough to say that there is no other case.

Quoted 12 lines
Quoted 8 lines

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

That's what the first versions of the patch did and Christian suggested that it's not a good approach. In fact it should be not only Tegra that has a broken dma-buf mapping, but apparently OMAP driver too.

Quoted 5 lines
Quoted 2 lines

Hmm, I guess I'm a bit more conservative when it comes to potentially breaking userspace.

If such userspace exists, then of course the mapping should continue to work. Still will be great to know what that userpsace is.

Definitely existing mesa does not have a special mmap path for imported dma-bufs (both in the case of panfrost and freedreno, I didn't check any others). The only question is whether there is a case where some app/test/etc imports a foreign dma-buf fd and then does something that would trigger mmap'ing, like readpix.

BR,
-R

Quoted 6 lines

Alright, let's keep the dma-buf mapping as-is for now. I'll fix just the Tegra driver then.

--
Best regards,
Dmitry

#19 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 132 lines · Rob Clark
Quoted 129 lines · Dmitry Osipenko
Quoted 61 lines · Rob Clark
Quoted 59 lines · Dmitry Osipenko
Quoted 57 lines · Rob Clark
Quoted 55 lines · Dmitry Osipenko
Quoted 52 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't handle imported dma-bufs properly, which results in mapping of something else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when userspace writes to the memory mapping of a dma-buf that was imported into Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects. Mapping of imported GEMs require special care from userspace since it should sync dma-buf because mapping coherency of the exporter device may not match the DRM device. Let's prohibit the mapping for all DRM drivers for consistency.

Quoted 5 lines
Quoted 2 lines

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Upd: I re-checked Panfrost using today's linux-next and mapping of imported dma-buf works, I mapped imported buf from video decoder. Apparently previously I had some local kernel change that broke the mapping.

Quoted 58 lines
Quoted 26 lines
Quoted 18 lines
Quoted 12 lines
Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Quoted 7 lines

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

Quoted 8 lines

It's not clear to me what bad Android does form yours comments. Does it export dma-buf from GPU and then import it to GPU? If yes, then DRM core has a check for the self-importing [1].

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_prime.c#L918

If you're meaning something else, then please explain in a more details.

So, android/gralloc allocates buffers externally to the driver and imports them into driver. (And that seems to not just be window surfaces, but in cases random textures, etc)

In the normal case these should be allocated from drm/msm so it should hit [1].. this is the "dumb luck" I mentioned earlier. But I'm not confident enough to say that there is no other case.

Quoted 9 lines
Quoted 8 lines

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

Quoted 3 lines

That's what the first versions of the patch did and Christian suggested that it's not a good approach. In fact it should be not only Tegra that has a broken dma-buf mapping, but apparently OMAP driver too.

Hmm, I guess I'm a bit more conservative when it comes to potentially breaking userspace.

Quoted 8 lines
Quoted 2 lines

If such userspace exists, then of course the mapping should continue to work. Still will be great to know what that userpsace is.

Definitely existing mesa does not have a special mmap path for imported dma-bufs (both in the case of panfrost and freedreno, I didn't check any others). The only question is whether there is a case where some app/test/etc imports a foreign dma-buf fd and then does something that would trigger mmap'ing, like readpix.

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle. And I could imagine that doing so could cause issues with too many fd's. So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

BR,
-R

Quoted 9 lines

BR,
-R

Quoted 6 lines

Alright, let's keep the dma-buf mapping as-is for now. I'll fix just the Tegra driver then.

--
Best regards,
Dmitry

#20 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 142 lines · Rob Clark
Quoted 132 lines · Rob Clark
Quoted 129 lines · Dmitry Osipenko
Quoted 61 lines · Rob Clark
Quoted 59 lines · Dmitry Osipenko
Quoted 57 lines · Rob Clark
Quoted 55 lines · Dmitry Osipenko
Quoted 52 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 14 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:

Quoted 12 lines

Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't
handle imported dma-bufs properly, which results in mapping of something
else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when
userspace writes to the memory mapping of a dma-buf that was imported into
Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects.
Mapping of imported GEMs require special care from userspace since it
should sync dma-buf because mapping coherency of the exporter device may
not match the DRM device. Let's prohibit the mapping for all DRM drivers
for consistency.

Quoted 5 lines
Quoted 2 lines

Suggested-by: Thomas Hellström <[email protected]>
Signed-off-by: Dmitry Osipenko <[email protected]>

I'm pretty sure that this is the right approach, but it's certainly more than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Upd: I re-checked Panfrost using today's linux-next and mapping of imported dma-buf works, I mapped imported buf from video decoder. Apparently previously I had some local kernel change that broke the mapping.

Quoted 58 lines
Quoted 26 lines
Quoted 18 lines
Quoted 12 lines
Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Quoted 7 lines

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

Quoted 8 lines

It's not clear to me what bad Android does form yours comments. Does it export dma-buf from GPU and then import it to GPU? If yes, then DRM core has a check for the self-importing [1].

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_prime.c#L918

If you're meaning something else, then please explain in a more details.

So, android/gralloc allocates buffers externally to the driver and imports them into driver. (And that seems to not just be window surfaces, but in cases random textures, etc)

In the normal case these should be allocated from drm/msm so it should hit [1].. this is the "dumb luck" I mentioned earlier. But I'm not confident enough to say that there is no other case.

Quoted 9 lines
Quoted 8 lines

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

Quoted 3 lines

That's what the first versions of the patch did and Christian suggested that it's not a good approach. In fact it should be not only Tegra that has a broken dma-buf mapping, but apparently OMAP driver too.

Hmm, I guess I'm a bit more conservative when it comes to potentially breaking userspace.

Quoted 8 lines
Quoted 2 lines

If such userspace exists, then of course the mapping should continue to work. Still will be great to know what that userpsace is.

Definitely existing mesa does not have a special mmap path for imported dma-bufs (both in the case of panfrost and freedreno, I didn't check any others). The only question is whether there is a case where some app/test/etc imports a foreign dma-buf fd and then does something that would trigger mmap'ing, like readpix.

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle. And I could imagine that doing so could cause issues with too many fd's. So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

Signature · 2 lines
Best regards,
Dmitry
#21 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 154 lines · Dmitry Osipenko
Quoted 142 lines · Rob Clark
Quoted 132 lines · Rob Clark
Quoted 129 lines · Dmitry Osipenko
Quoted 61 lines · Rob Clark
Quoted 59 lines · Dmitry Osipenko
Quoted 57 lines · Rob Clark
Quoted 55 lines · Dmitry Osipenko
Quoted 52 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 19 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:
Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't
handle imported dma-bufs properly, which results in mapping of something
else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when
userspace writes to the memory mapping of a dma-buf that was imported into
Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects.
Mapping of imported GEMs require special care from userspace since it
should sync dma-buf because mapping coherency of the exporter device may
not match the DRM device. Let's prohibit the mapping for all DRM drivers
for consistency.

Suggested-by: Thomas Hellström <[email protected]>
Signed-off-by: Dmitry Osipenko <[email protected]>

I'm pretty sure that this is the right approach, but it's certainly more
than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Upd: I re-checked Panfrost using today's linux-next and mapping of imported dma-buf works, I mapped imported buf from video decoder. Apparently previously I had some local kernel change that broke the mapping.

Quoted 58 lines
Quoted 26 lines
Quoted 18 lines
Quoted 12 lines
Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Quoted 7 lines

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

Quoted 8 lines

It's not clear to me what bad Android does form yours comments. Does it export dma-buf from GPU and then import it to GPU? If yes, then DRM core has a check for the self-importing [1].

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_prime.c#L918

If you're meaning something else, then please explain in a more details.

So, android/gralloc allocates buffers externally to the driver and imports them into driver. (And that seems to not just be window surfaces, but in cases random textures, etc)

In the normal case these should be allocated from drm/msm so it should hit [1].. this is the "dumb luck" I mentioned earlier. But I'm not confident enough to say that there is no other case.

Quoted 9 lines
Quoted 8 lines

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

Quoted 3 lines

That's what the first versions of the patch did and Christian suggested that it's not a good approach. In fact it should be not only Tegra that has a broken dma-buf mapping, but apparently OMAP driver too.

Hmm, I guess I'm a bit more conservative when it comes to potentially breaking userspace.

Quoted 8 lines
Quoted 2 lines

If such userspace exists, then of course the mapping should continue to work. Still will be great to know what that userpsace is.

Definitely existing mesa does not have a special mmap path for imported dma-bufs (both in the case of panfrost and freedreno, I didn't check any others). The only question is whether there is a case where some app/test/etc imports a foreign dma-buf fd and then does something that would trigger mmap'ing, like readpix.

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle. And I could imagine that doing so could cause issues with too many fd's. So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Quoted 7 lines

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

To make it more clear, I imported AMDGPU dma-buf to Intel iGPU and vice versa.

Quoted 4 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

Signature · 2 lines
Best regards,
Dmitry
#22 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

In reply to Dmitry Osipenko

Am 16.08.22 um 13:44 schrieb Dmitry Osipenko:

Quoted 11 lines

[SNIP]

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle. And I could imagine that doing so could cause issues with too many fd's. So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Quoted 7 lines

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

That's then certainly broken somehow. Amdgpu should nerve ever have allowed to mmap() imported DMA-bufs and the last time I check it didn't.

Regards,
Christian.

Quoted 5 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#23 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 15 lines · Christian König

Am 16.08.22 um 13:44 schrieb Dmitry Osipenko:

Quoted 14 lines

[SNIP]

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle.  And I could imagine that doing so could cause issues with too many fd's.  So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Quoted 7 lines
Quoted 4 lines

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

That's then certainly broken somehow. Amdgpu should nerve ever have allowed to mmap() imported DMA-bufs and the last time I check it didn't.

I'll take a closer look. So far I can only tell that it's a kernel driver issue because once I re-applied this "Don't map imported GEMs" patch, AMDGPU began to refuse mapping AMDGPU_GEM_DOMAIN_GTT.

Quoted 3 lines
Quoted 3 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

Actually, I'm now looking at Etnaviv and Nouveau and seems they should map imported dma-buf properly. I know that people ran Android on Etnaviv. So maybe devices with a separated GPU/display need to map imported display BO for Android support. Wish somebody who ran Android on one of these devices using upstream drivers could give a definitive answer. I may try to test Nouveau later on.

Signature · 2 lines
Best regards,
Dmitry
#24 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 32 lines · Dmitry Osipenko
Quoted 15 lines · Christian König

Am 16.08.22 um 13:44 schrieb Dmitry Osipenko:

Quoted 14 lines

[SNIP]

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle.  And I could imagine that doing so could cause issues with too many fd's.  So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Quoted 7 lines
Quoted 4 lines

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

That's then certainly broken somehow. Amdgpu should nerve ever have allowed to mmap() imported DMA-bufs and the last time I check it didn't.

I'll take a closer look. So far I can only tell that it's a kernel driver issue because once I re-applied this "Don't map imported GEMs" patch, AMDGPU began to refuse mapping AMDGPU_GEM_DOMAIN_GTT.

Quoted 3 lines
Quoted 3 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

Quoted 7 lines

Actually, I'm now looking at Etnaviv and Nouveau and seems they should map imported dma-buf properly. I know that people ran Android on Etnaviv. So maybe devices with a separated GPU/display need to map imported display BO for Android support. Wish somebody who ran Android on one of these devices using upstream drivers could give a definitive answer. I may try to test Nouveau later on.

Nouveau+Intel combo doesn't work because of [1] that says:

"Refuse to fault imported pages. This should be handled (if at all) by redirecting mmap to the exporter."

[1]
https://elixir.bootlin.com/linux/v5.19/source/drivers/gpu/drm/ttm/ttm_bo_vm.c#L154

Interestingly, I noticed that there are IGT tests which check prime mmaping of Nouveau+Intel [2] (added 9 years ago), but they fail as well, as expected. The fact that IGT has such tests is interesting because it suggests that the mapping worked in the past. It's also surprising that nobody cared to fix the failing tests. For the reference, I checked v5.18 and today's linux-next.

[2]
https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/blob/master/tests/prime_nv_test.c#L132

Starting subtest: nv_write_i915_cpu_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x406]
#3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_cpu_mmap_read: CRASH (0,005s)

Starting subtest: nv_write_i915_gtt_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x33d]
#3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_gtt_mmap_read: CRASH (0,004s)

I'm curious about the Etnaviv driver because it uses own shmem implementation and maybe it has a working mmaping of imported GEMs since it imports the dma-buf pages into Entaviv BO. Although, it should be risking to map pages using a different caching attributes (WC) from the exporter, which is prohibited on ARM ad then one may try to map imported udmabuf.

Apparently, the Intel DG TTM driver should be able to map imported dma-buf because it sets TTM_TT_FLAG_EXTERNAL_MAPPABLE.

Overall, it still questionable to me whether it's worthwhile to allow the mmaping of imported GEMs since only Panfrost/Lima can do it out of all drivers and h/w that I tested. Feels like drivers that can do the mapping have it just because they can and not because they need.

Signature · 2 lines
Best regards,
Dmitry
#25 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Am 22.08.22 um 19:26 schrieb Dmitry Osipenko:

Quoted 86 lines · Dmitry Osipenko
Quoted 29 lines · Christian König
Quoted 15 lines

Am 16.08.22 um 13:44 schrieb Dmitry Osipenko:

Quoted 14 lines

[SNIP]

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle.  And I could imagine that doing so could cause issues with too many fd's.  So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Quoted 6 lines
Quoted 4 lines

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

That's then certainly broken somehow. Amdgpu should nerve ever have allowed to mmap() imported DMA-bufs and the last time I check it didn't.

I'll take a closer look. So far I can only tell that it's a kernel driver issue because once I re-applied this "Don't map imported GEMs" patch, AMDGPU began to refuse mapping AMDGPU_GEM_DOMAIN_GTT.

Quoted 3 lines
Quoted 3 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

Quoted 7 lines

Actually, I'm now looking at Etnaviv and Nouveau and seems they should map imported dma-buf properly. I know that people ran Android on Etnaviv. So maybe devices with a separated GPU/display need to map imported display BO for Android support. Wish somebody who ran Android on one of these devices using upstream drivers could give a definitive answer. I may try to test Nouveau later on.

Nouveau+Intel combo doesn't work because of [1] that says:

"Refuse to fault imported pages. This should be handled (if at all) by redirecting mmap to the exporter."

[1]
https://elixir.bootlin.com/linux/v5.19/source/drivers/gpu/drm/ttm/ttm_bo_vm.c#L154

Interestingly, I noticed that there are IGT tests which check prime mmaping of Nouveau+Intel [2] (added 9 years ago), but they fail as well, as expected. The fact that IGT has such tests is interesting because it suggests that the mapping worked in the past. It's also surprising that nobody cared to fix the failing tests. For the reference, I checked v5.18 and today's linux-next.

[2]
https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/blob/master/tests/prime_nv_test.c#L132

Starting subtest: nv_write_i915_cpu_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x406] #3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_cpu_mmap_read: CRASH (0,005s)

Starting subtest: nv_write_i915_gtt_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x33d] #3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_gtt_mmap_read: CRASH (0,004s)

I'm curious about the Etnaviv driver because it uses own shmem implementation and maybe it has a working mmaping of imported GEMs since it imports the dma-buf pages into Entaviv BO. Although, it should be risking to map pages using a different caching attributes (WC) from the exporter, which is prohibited on ARM ad then one may try to map imported udmabuf.

Quoted 2 lines

Apparently, the Intel DG TTM driver should be able to map imported dma-buf because it sets TTM_TT_FLAG_EXTERNAL_MAPPABLE.

Even with that flag set it is illegal to map the pages directly by an importer.

If that ever worked then the only real solution is to redirect mmap() calls on importer BOs to dma_buf_mmap().

Regards,
Christian.

Quoted 6 lines

Overall, it still questionable to me whether it's worthwhile to allow the mmaping of imported GEMs since only Panfrost/Lima can do it out of all drivers and h/w that I tested. Feels like drivers that can do the mapping have it just because they can and not because they need.

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#26 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 88 lines · Christian König

Am 22.08.22 um 19:26 schrieb Dmitry Osipenko:

Quoted 86 lines · Dmitry Osipenko
Quoted 29 lines · Christian König
Quoted 15 lines

Am 16.08.22 um 13:44 schrieb Dmitry Osipenko:

Quoted 14 lines

[SNIP]

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle. And I could imagine that doing so could cause issues with too many fd's. So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Quoted 6 lines
Quoted 4 lines

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

That's then certainly broken somehow. Amdgpu should nerve ever have allowed to mmap() imported DMA-bufs and the last time I check it didn't.

I'll take a closer look. So far I can only tell that it's a kernel driver issue because once I re-applied this "Don't map imported GEMs" patch, AMDGPU began to refuse mapping AMDGPU_GEM_DOMAIN_GTT.

Quoted 3 lines
Quoted 3 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

Quoted 7 lines

Actually, I'm now looking at Etnaviv and Nouveau and seems they should map imported dma-buf properly. I know that people ran Android on Etnaviv. So maybe devices with a separated GPU/display need to map imported display BO for Android support. Wish somebody who ran Android on one of these devices using upstream drivers could give a definitive answer. I may try to test Nouveau later on.

Nouveau+Intel combo doesn't work because of [1] that says:

"Refuse to fault imported pages. This should be handled (if at all) by redirecting mmap to the exporter."

[1]
https://elixir.bootlin.com/linux/v5.19/source/drivers/gpu/drm/ttm/ttm_bo_vm.c#L154

Interestingly, I noticed that there are IGT tests which check prime mmaping of Nouveau+Intel [2] (added 9 years ago), but they fail as well, as expected. The fact that IGT has such tests is interesting because it suggests that the mapping worked in the past. It's also surprising that nobody cared to fix the failing tests. For the reference, I checked v5.18 and today's linux-next.

[2]
https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/blob/master/tests/prime_nv_test.c#L132

Starting subtest: nv_write_i915_cpu_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x406] #3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_cpu_mmap_read: CRASH (0,005s)

Starting subtest: nv_write_i915_gtt_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x33d] #3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_gtt_mmap_read: CRASH (0,004s)

I'm curious about the Etnaviv driver because it uses own shmem implementation and maybe it has a working mmaping of imported GEMs since it imports the dma-buf pages into Entaviv BO. Although, it should be risking to map pages using a different caching attributes (WC) from the exporter, which is prohibited on ARM ad then one may try to map imported udmabuf.

Quoted 8 lines
Quoted 2 lines

Apparently, the Intel DG TTM driver should be able to map imported dma-buf because it sets TTM_TT_FLAG_EXTERNAL_MAPPABLE.

Even with that flag set it is illegal to map the pages directly by an importer.

If that ever worked then the only real solution is to redirect mmap() calls on importer BOs to dma_buf_mmap().

Yeah, I think this is the best option. Forcing userspace to hang on to the fd just in case someone calls readpix would be pretty harsh.

BR,
-R

Quoted 10 lines

Regards,
Christian.

Quoted 6 lines

Overall, it still questionable to me whether it's worthwhile to allow the mmaping of imported GEMs since only Panfrost/Lima can do it out of all drivers and h/w that I tested. Feels like drivers that can do the mapping have it just because they can and not because they need.

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#27 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 83 lines · Rob Clark
Quoted 81 lines · Christian König

Am 22.08.22 um 19:26 schrieb Dmitry Osipenko:

Quoted 79 lines · Dmitry Osipenko
Quoted 29 lines · Christian König
Quoted 15 lines

Am 16.08.22 um 13:44 schrieb Dmitry Osipenko:

Quoted 14 lines

[SNIP]

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle. And I could imagine that doing so could cause issues with too many fd's. So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Quoted 6 lines
Quoted 4 lines

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

That's then certainly broken somehow. Amdgpu should nerve ever have allowed to mmap() imported DMA-bufs and the last time I check it didn't.

I'll take a closer look. So far I can only tell that it's a kernel driver issue because once I re-applied this "Don't map imported GEMs" patch, AMDGPU began to refuse mapping AMDGPU_GEM_DOMAIN_GTT.

Quoted 3 lines
Quoted 3 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

Quoted 7 lines

Actually, I'm now looking at Etnaviv and Nouveau and seems they should map imported dma-buf properly. I know that people ran Android on Etnaviv. So maybe devices with a separated GPU/display need to map imported display BO for Android support. Wish somebody who ran Android on one of these devices using upstream drivers could give a definitive answer. I may try to test Nouveau later on.

Nouveau+Intel combo doesn't work because of [1] that says:

"Refuse to fault imported pages. This should be handled (if at all) by redirecting mmap to the exporter."

[1]
https://elixir.bootlin.com/linux/v5.19/source/drivers/gpu/drm/ttm/ttm_bo_vm.c#L154

Interestingly, I noticed that there are IGT tests which check prime mmaping of Nouveau+Intel [2] (added 9 years ago), but they fail as well, as expected. The fact that IGT has such tests is interesting because it suggests that the mapping worked in the past. It's also surprising that nobody cared to fix the failing tests. For the reference, I checked v5.18 and today's linux-next.

[2]
https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/blob/master/tests/prime_nv_test.c#L132

Starting subtest: nv_write_i915_cpu_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x406] #3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_cpu_mmap_read: CRASH (0,005s)

Starting subtest: nv_write_i915_gtt_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x33d] #3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_gtt_mmap_read: CRASH (0,004s)

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

I'm curious about the Etnaviv driver because it uses own shmem implementation and maybe it has a working mmaping of imported GEMs since it imports the dma-buf pages into Entaviv BO. Although, it should be risking to map pages using a different caching attributes (WC) from the exporter, which is prohibited on ARM ad then one may try to map imported udmabuf.

I see now that Etnaviv uses dma_buf_mmap(), so it should be okay.

Quoted 11 lines
Quoted 8 lines
Quoted 2 lines

Apparently, the Intel DG TTM driver should be able to map imported dma-buf because it sets TTM_TT_FLAG_EXTERNAL_MAPPABLE.

Even with that flag set it is illegal to map the pages directly by an importer.

If that ever worked then the only real solution is to redirect mmap() calls on importer BOs to dma_buf_mmap().

Yeah, I think this is the best option. Forcing userspace to hang on to the fd just in case someone calls readpix would be pretty harsh.

Actually, I proposed this couple months ago [1].

[1] https://patchwork.freedesktop.org/patch/487481/

What's not clear to me is how userspace is supposed to sync CPU accesses for imported GEMs. Either userspace need to use dma_buf_sync UAPI for dmabuf of imported GEM or importer driver should do that, or there is some other option?

Signature · 2 lines
Best regards,
Dmitry
#28 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

Quoted 106 lines · Dmitry Osipenko
Quoted 83 lines · Rob Clark
Quoted 81 lines · Christian König

Am 22.08.22 um 19:26 schrieb Dmitry Osipenko:

Quoted 79 lines · Dmitry Osipenko
Quoted 29 lines · Christian König
Quoted 15 lines

Am 16.08.22 um 13:44 schrieb Dmitry Osipenko:

Quoted 14 lines

[SNIP]

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle. And I could imagine that doing so could cause issues with too many fd's. So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Quoted 6 lines
Quoted 4 lines

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

That's then certainly broken somehow. Amdgpu should nerve ever have allowed to mmap() imported DMA-bufs and the last time I check it didn't.

I'll take a closer look. So far I can only tell that it's a kernel driver issue because once I re-applied this "Don't map imported GEMs" patch, AMDGPU began to refuse mapping AMDGPU_GEM_DOMAIN_GTT.

Quoted 3 lines
Quoted 3 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

Quoted 7 lines

Actually, I'm now looking at Etnaviv and Nouveau and seems they should map imported dma-buf properly. I know that people ran Android on Etnaviv. So maybe devices with a separated GPU/display need to map imported display BO for Android support. Wish somebody who ran Android on one of these devices using upstream drivers could give a definitive answer. I may try to test Nouveau later on.

Nouveau+Intel combo doesn't work because of [1] that says:

"Refuse to fault imported pages. This should be handled (if at all) by redirecting mmap to the exporter."

[1]
https://elixir.bootlin.com/linux/v5.19/source/drivers/gpu/drm/ttm/ttm_bo_vm.c#L154

Interestingly, I noticed that there are IGT tests which check prime mmaping of Nouveau+Intel [2] (added 9 years ago), but they fail as well, as expected. The fact that IGT has such tests is interesting because it suggests that the mapping worked in the past. It's also surprising that nobody cared to fix the failing tests. For the reference, I checked v5.18 and today's linux-next.

[2]
https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/blob/master/tests/prime_nv_test.c#L132

Starting subtest: nv_write_i915_cpu_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x406] #3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_cpu_mmap_read: CRASH (0,005s)

Starting subtest: nv_write_i915_gtt_mmap_read Received signal SIGBUS.
Stack trace:
#0 [fatal_sig_handler+0x163]
#1 [__sigaction+0x50]
#2 [__igt_unique____real_main354+0x33d] #3 [main+0x23]
#4 [__libc_start_call_main+0x80]
#5 [__libc_start_main+0x89]
#6 [_start+0x25]
Subtest nv_write_i915_gtt_mmap_read: CRASH (0,004s)

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

I'm curious about the Etnaviv driver because it uses own shmem implementation and maybe it has a working mmaping of imported GEMs since it imports the dma-buf pages into Entaviv BO. Although, it should be risking to map pages using a different caching attributes (WC) from the exporter, which is prohibited on ARM ad then one may try to map imported udmabuf.

I see now that Etnaviv uses dma_buf_mmap(), so it should be okay.

Quoted 11 lines
Quoted 8 lines
Quoted 2 lines

Apparently, the Intel DG TTM driver should be able to map imported dma-buf because it sets TTM_TT_FLAG_EXTERNAL_MAPPABLE.

Even with that flag set it is illegal to map the pages directly by an importer.

If that ever worked then the only real solution is to redirect mmap() calls on importer BOs to dma_buf_mmap().

Yeah, I think this is the best option. Forcing userspace to hang on to the fd just in case someone calls readpix would be pretty harsh.

Quoted 8 lines

Actually, I proposed this couple months ago [1].

[1] https://patchwork.freedesktop.org/patch/487481/

What's not clear to me is how userspace is supposed to sync CPU accesses for imported GEMs. Either userspace need to use dma_buf_sync UAPI for dmabuf of imported GEM or importer driver should do that, or there is some other option?

For anything involving the GPU, userspace already needs to sync access (ie. GPUs are asynchronous).. this is done using drm_gem_object::resv (which comes from dma_buf::resv)..

Possibly we should require that dma_buf_begin_cpu_access()/dma_buf_end_cpu_access().. but in practice if the gl/vk driver needs CPU access it is to read the results of it's own GPU rendering. Disallow that and all of deqp/piglit/etc will break on systems where "window" surfaces are dma-buf imports (like android), which would be a rather unpopular outcome ;-)

BR,
-R

#29 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

In reply to Dmitry Osipenko
Quoted 150 lines · Dmitry Osipenko
Quoted 142 lines · Rob Clark
Quoted 132 lines · Rob Clark
Quoted 129 lines · Dmitry Osipenko
Quoted 61 lines · Rob Clark
Quoted 59 lines · Dmitry Osipenko
Quoted 57 lines · Rob Clark
Quoted 55 lines · Dmitry Osipenko
Quoted 52 lines · Rob Clark
Quoted 38 lines · Daniel Vetter
Quoted 32 lines · Dmitry Osipenko
Quoted 25 lines · Rob Clark
Quoted 19 lines · Christian König

Am 01.07.22 um 11:02 schrieb Dmitry Osipenko:
Drivers that use drm_gem_mmap() and drm_gem_mmap_obj() helpers don't
handle imported dma-bufs properly, which results in mapping of something
else than the imported dma-buf. On NVIDIA Tegra we get a hard lockup when
userspace writes to the memory mapping of a dma-buf that was imported into
Tegra's DRM GEM.

Majority of DRM drivers prohibit mapping of the imported GEM objects.
Mapping of imported GEMs require special care from userspace since it
should sync dma-buf because mapping coherency of the exporter device may
not match the DRM device. Let's prohibit the mapping for all DRM drivers
for consistency.

Suggested-by: Thomas Hellström <[email protected]>
Signed-off-by: Dmitry Osipenko <[email protected]>

I'm pretty sure that this is the right approach, but it's certainly more
than possible that somebody abused this already.

I suspect that this is abused if you run deqp cts on android.. ie. all winsys buffers are dma-buf imports from gralloc. And then when you hit readpix...

Quoted 5 lines

You might only hit this in scenarios with separate gpu and display (or dGPU+iGPU) because self-imports are handled differently in drm_gem_prime_import_dev().. and maybe not in cases where you end up with a blit from tiled/compressed to linear.. maybe that narrows the scope enough to just fix it in userspace?

Quoted 3 lines

Given that that only drivers which use DRM-SHMEM potentially could've map imported dma-bufs (Panfrost, Lima) and they already don't allow to do that, I think we're good.

Quoted 6 lines

So can I have an ack from Rob here or are there still questions that this might go boom?

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.

I *think* we'd be ok with this on msm, mostly just by dumb luck. Because the dma-buf's we import will be self-import. I'm less sure about panfrost (src/panfrost/lib/pan_bo.c doesn't seem to have a special path for imported dma-bufs either, and in that case they won't be self-imports.. but I guess no one has tried to run android cts on panfrost).

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

The last time I tried to mmap dma-buf imported to Panfrost didn't work because Panfrost didn't implement something needed for that. I'll need to take a look again because can't recall what it was.

Upd: I re-checked Panfrost using today's linux-next and mapping of imported dma-buf works, I mapped imported buf from video decoder. Apparently previously I had some local kernel change that broke the mapping.

Quoted 58 lines
Quoted 26 lines
Quoted 18 lines
Quoted 12 lines
Quoted 12 lines

What about something less drastic to start, like (apologies for hand-edited patch):

drivers/gpu/drm/drm_gem.c +0 -0
index 86d670c71286..fc9ec42fa0ab 100644
@@ -1034,6 +1034,10 @@ int drm_gem_mmap_obj(struct drm_gem_object
 *obj, unsigned long obj_size,
 {
        int ret;
 
Quoted 5 lines

drivers/gpu/drm/drm_gem.c int drm_gem_mmap_obj(struct drm_gem_object · line 1038

Quoted 1 line
drivers/gpu/drm/drm_gem.c +1 -0
+       WARN_ON_ONCE(obj->import_attach);

This will hang NVIDIA Tegra, which is what this patch fixed initially. If neither of upstream DRM drivers need to map imported dma-bufs and never needed, then why do we need this?

Quoted 7 lines

oh, tegra isn't using shmem helpers? I assumed it was. Well my point was to make a more targeted fail on tegra, and a WARN_ON for everyone else to make it clear that what they are doing is undefined behavior. Because so far existing userspace (or well, panfrost and freedreno at least, those are the two I know or checked) don't make special cases for mmap'ing against the dmabuf fd against the dmabuf fd instead of the drm device fd.

Quoted 8 lines

It's not clear to me what bad Android does form yours comments. Does it export dma-buf from GPU and then import it to GPU? If yes, then DRM core has a check for the self-importing [1].

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_prime.c#L918

If you're meaning something else, then please explain in a more details.

So, android/gralloc allocates buffers externally to the driver and imports them into driver. (And that seems to not just be window surfaces, but in cases random textures, etc)

In the normal case these should be allocated from drm/msm so it should hit [1].. this is the "dumb luck" I mentioned earlier. But I'm not confident enough to say that there is no other case.

Quoted 9 lines
Quoted 8 lines

I *think* it should work out that we don't hit this path with freedreno but on android I can't really guarantee or prove it. So your patch would potentially break existing working userspace. Maybe it is userspace that isn't portable (but OTOH it isn't like you are going to be using freedreno on tegra). So why don't you go for a more targeted fix that only returns an error on hw where this is problematic?

Quoted 3 lines

That's what the first versions of the patch did and Christian suggested that it's not a good approach. In fact it should be not only Tegra that has a broken dma-buf mapping, but apparently OMAP driver too.

Hmm, I guess I'm a bit more conservative when it comes to potentially breaking userspace.

Quoted 8 lines
Quoted 2 lines

If such userspace exists, then of course the mapping should continue to work. Still will be great to know what that userpsace is.

Definitely existing mesa does not have a special mmap path for imported dma-bufs (both in the case of panfrost and freedreno, I didn't check any others). The only question is whether there is a case where some app/test/etc imports a foreign dma-buf fd and then does something that would trigger mmap'ing, like readpix.

Quoted 5 lines

The other complication I noticed is that we don't seem to keep around the fd after importing to a GEM handle. And I could imagine that doing so could cause issues with too many fd's. So I guess the best thing is to keep the status quo and let drivers that cannot mmap imported buffers just fail mmap?

Quoted 4 lines

That actually should be all the drivers excluding those that use DRM-SHMEM because only DRM-SHMEM uses dma_buf_mmap(), that's why it works for Panfrost. I'm pretty sure mmaping of imported GEMs doesn't work for the MSM driver, isn't it?

Hmm, I would have expected it to work, at least prior to the mangling debug thing danvet added.

Actually, I suspect with the mangling enabled, all sorts of things would fall apart, I'm not seeing how things like sg_phys() continue to work. But not sure if we have that enabled in anything that is getting android CTS run on it.

BR,
-R

Quoted 8 lines

Intel and AMD drivers don't allow to map the imported dma-bufs. Both refuse to do the mapping.

Although, AMDGPU "succeeds" to do the mapping using AMDGPU_GEM_DOMAIN_GTT, but then touching the mapping causes bus fault, hence mapping actually fails. I think it might be the AMDGPU driver/libdrm bug, haven't checked yet.

Quoted 7 lines

So we're back to the point that neither of DRM drivers need to map imported dma-bufs and this was never tested. In this case this patch is valid, IMO.

--
Best regards,
Dmitry

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#30 lore raw

Re: [PATCH v8 2/2] drm/gem: Don't map imported GEMs

In reply to Daniel Vetter
Quoted 4 lines · Daniel Vetter

Dmitry, since you have a bunch of patches merged now I think would also be good to get commit rights so you can drive this more yourself. I've asked Daniel Stone to help you out with getting that.
-Daniel

Thank you!

Signature · 2 lines
Best regards,
Dmitry
#31 lore raw
In reply to Dmitry Osipenko

Hi,

Quoted 25 lines · Dmitry Osipenko

This patchset fixes two problems in the common GEM code. First fixed problem is the bogus lockdep splat that complicates debugging of DRM drivers. Second problem is the inconsistency in behaviour and improper handling of mapping the imported GEMs by some drivers, to fix it we will prohibit to map the imported GEMs like majority of drivers already do.

Changelog:

v8: - Dropped stable tag from the "Don't map imported GEMs" patch,
like was suggested by Thomas Hellström in the review comment to v7.

v7: - Factored out GEM patches from [1] since I'll be working on the
dma-buf locking in a separate patchset now.

[1] https://lore.kernel.org/all/[email protected]/

- Improved commit message and added fixes tag to the "Properly annotate WW context" patch.

- Replaced "Move mapping of imported dma-bufs to drm_gem_mmap_obj()" patch with "Don't map imported GEMs", like was suggested by Thomas Hellström.

- Added r-b and suggested-by from Thomas Hellström.

Quoted 9 lines

Dmitry Osipenko (2):
drm/gem: Properly annotate WW context on drm_gem_lock_reservations()
error
drm/gem: Don't map imported GEMs

 drivers/gpu/drm/drm_gem.c              | 8 ++++++--
 drivers/gpu/drm/drm_gem_shmem_helper.c | 9 ---------
 2 files changed, 6 insertions(+), 11 deletions(-)

Could anyone please take this into drm-misc-next? Maybe Thomas?

Thanks in advance

Signature · 2 lines
Best regards,
Dmitry