iommu/exynos: add missing set_platform_dma_ops callback

PATCH Applied Marek Szyprowski 6 participants 10 messages Jan 23, 2023 – Apr 12, 2023 1 message not in the archive

Subsystems (3)

Arm Samsung Exynos IOMMU Exynos sysmmu (IOMMU) driver

Share & tools

A message that is not in the archive (sent off-list or not yet received): CGME20230123093116eucas1p19b8fe8afc4b631debbdc5321c53009e9@eucas1p1.samsung.com

Marek Szyprowski

author maintainer

2023-01-23 09:31 UTC #1 lore raw

PATCH 1/1 iommu/exynos: add missing set_platform_dma_ops callback

Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).

Fixes c1fe9119ee70 ("iommu: Add set_platform_dma_ops callbacks")

Signed-off-by Marek Szyprowski <[email protected]>

---
 drivers/iommu/exynos-iommu.c | 5 +++++
 1 file changed, 5 insertions(+)

drivers/iommu/exynos-iommu.c +5 -0

index 6fc58e89712f..c4955d045855 100644
@@ -1346,8 +1346,10 @@ static void exynos_iommu_release_device(struct device *dev)
 		struct iommu_group *group = iommu_group_get(dev);
 
 		if (group) {
+#ifndef CONFIG_ARM
 			WARN_ON(owner->domain !=
 				iommu_group_default_domain(group));
+#endif
 			exynos_iommu_detach_device(owner->domain, dev);
 			iommu_group_put(group);
 		}

@@ -1398,6 +1400,9 @@ static int exynos_iommu_of_xlate(struct device *dev,
 static const struct iommu_ops exynos_iommu_ops = {
 	.domain_alloc = exynos_iommu_domain_alloc,
 	.device_group = generic_device_group,
+#ifdef CONFIG_ARM
+	.set_platform_dma_ops = exynos_iommu_release_device,
+#endif
 	.probe_device = exynos_iommu_probe_device,
 	.release_device = exynos_iommu_release_device,
 	.pgsize_bitmap = SECT_SIZE | LPAGE_SIZE | SPAGE_SIZE,

Signature · 1 line

2.39.0

Jason Gunthorpe

2023-01-23 21:00 UTC #2 lore raw

Quoted 3 lines · Marek Szyprowski

Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).

Thanks for looking into this!

Can you explain more about how this actually solves the problem in the commit message? I don't get it.

Quoted 16 lines

drivers/iommu/exynos-iommu.c +2 -0

index 6fc58e89712f..c4955d045855 100644
@@ -1346,8 +1346,10 @@ static void exynos_iommu_release_device(struct device *dev)
 		struct iommu_group *group = iommu_group_get(dev);
 
 		if (group) {
+#ifndef CONFIG_ARM
 			WARN_ON(owner->domain !=
 				iommu_group_default_domain(group));
+#endif
 			exynos_iommu_detach_device(owner->domain, dev);
 			iommu_group_put(group);
 		}

drivers/iommu/exynos-iommu.c static int exynos_iommu_of_xlate(struct device *dev, · line 1405

Quoted 6 lines

drivers/iommu/exynos-iommu.c +3 -0

@@ -1398,6 +1400,9 @@ static int exynos_iommu_of_xlate(struct device *dev,
 static const struct iommu_ops exynos_iommu_ops = {
 	.domain_alloc = exynos_iommu_domain_alloc,
 	.device_group = generic_device_group,
+#ifdef CONFIG_ARM
+	.set_platform_dma_ops = exynos_iommu_release_device,
+#endif

This is ugly, if you need a set_platform_dma_ops it should not be called release... Release is supposed to be about putting the HW back to some idle state because we are unplugging the struct device.

Jason

Marek Szyprowski

author maintainer

2023-02-17 11:10 UTC #3 lore raw

Hi,

I'm sorry for a delay in replying, but I was busy with other stuff.

Quoted 8 lines · Jason Gunthorpe

Quoted 3 lines · Marek Szyprowski
Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).
Thanks for looking into this!
Can you explain more about how this actually solves the problem in the commit message? I don't get it.

Exynos DRM driver calls arm_iommu_detach_device(), then arm_iommu_attach_device() with a difrent 'mapping', see drivers/gpu/drm/exynos/exynos_drm_dma.c Lack of set_platform_dma_ops leads to a warning in iommu_group_do_set_platform_dma(). The other case of calling arm_iommu_detach_device() is after unsuccessful probe of the platform device.

Quoted 17 lines

drivers/iommu/exynos-iommu.c +2 -0

index 6fc58e89712f..c4955d045855 100644
@@ -1346,8 +1346,10 @@ static void exynos_iommu_release_device(struct device *dev)
  		struct iommu_group *group = iommu_group_get(dev);
 
  		if (group) {
+#ifndef CONFIG_ARM
  			WARN_ON(owner->domain !=
  				iommu_group_default_domain(group));
+#endif
  			exynos_iommu_detach_device(owner->domain, dev);
  			iommu_group_put(group);
  		}

@@ -1398,6 +1400,9 @@ static int exynos_iommu_of_xlate(struct device *dev,
  static const struct iommu_ops exynos_iommu_ops = {

Quoted 8 lines

drivers/iommu/exynos-iommu.c static int exynos_iommu_of_xlate(struct device *dev, · line 1405
Quoted 5 lines
drivers/iommu/exynos-iommu.c +3 -0
  	.domain_alloc = exynos_iommu_domain_alloc,
  	.device_group = generic_device_group,
+#ifdef CONFIG_ARM
+	.set_platform_dma_ops = exynos_iommu_release_device,
+#endif
This is ugly, if you need a set_platform_dma_ops it should not be called release... Release is supposed to be about putting the HW back to some idle state because we are unplugging the struct device.

Indeed this patch was a bit ugly, I've did it in a bit of hurry. That time I've simply checked how it is implemented in other drivers and found that it very similar to the release operation, so I did it in exynos-iommu that way. I've checked again and indeed there are some differences, so I will send a fix in a few minutes.

Best regards

Signature · 2 lines

Marek Szyprowski, PhD
Samsung R&D Institute Poland

Baolu Lu

2023-02-17 11:18 UTC #4 lore raw

Quoted 11 lines · Marek Szyprowski

Hi,
I'm sorry for a delay in replying, but I was busy with other stuff.
Quoted 6 lines · Jason Gunthorpe
Quoted 3 lines · Marek Szyprowski
Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).
Thanks for looking into this!

Quoted 8 lines

Quoted 2 lines
Can you explain more about how this actually solves the problem in the commit message? I don't get it.
Exynos DRM driver calls arm_iommu_detach_device(), then arm_iommu_attach_device() with a difrent 'mapping', see drivers/gpu/drm/exynos/exynos_drm_dma.c Lack of set_platform_dma_ops leads to a warning in iommu_group_do_set_platform_dma(). The other case of calling arm_iommu_detach_device() is after unsuccessful probe of the platform device.

Do we really need to call iommu_detach_device() in arm_iommu_detach_device() for 32-bit ARM? The dma-iommu is disabled for 32-bit ARM, hence calling .set_platform_dma_ops makes no sense.

In below patch,

https://lore.kernel.org/linux-iommu/[email protected]/

I removed iommu_detach_device() from arm_iommu_detach_device().

Best regards,
baolu

Jason Gunthorpe

2023-02-17 12:36 UTC #5 lore raw

In reply to Marek Szyprowski

Quoted 14 lines · Marek Szyprowski

Hi,
I'm sorry for a delay in replying, but I was busy with other stuff.
Quoted 8 lines · Jason Gunthorpe
Quoted 3 lines · Marek Szyprowski
Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).
Thanks for looking into this!
Can you explain more about how this actually solves the problem in the commit message? I don't get it.

Quoted 6 lines

Exynos DRM driver calls arm_iommu_detach_device(), then arm_iommu_attach_device() with a difrent 'mapping', see drivers/gpu/drm/exynos/exynos_drm_dma.c Lack of set_platform_dma_ops leads to a warning in iommu_group_do_set_platform_dma(). The other case of calling arm_iommu_detach_device() is after unsuccessful probe of the platform device.

Why can't this just use the normal iommu path in all cases?

It looks like it is trying to copy the DMA API domain from a parent device to a sub device.

Even when using arm_iommu an iommu_domain is still present, so the copy code should work?

Though I'm still not really sure how this arm_iommu stuff works..

eg if a driver does iommu_device_claim_dma_owner() how does the iommu_domain get set back to the arm_iommu's mapping's iommu_domain?

According to the API that is what set_platform is supposed to do (eg it is what s390 does), but I don't see any code like that in any of the ARM32 drivers..

Jason

Robin Murphy

reviewer

2023-02-20 13:58 UTC #6 lore raw

Quoted 22 lines · Jason Gunthorpe

Quoted 14 lines · Marek Szyprowski
Hi,
I'm sorry for a delay in replying, but I was busy with other stuff.
Quoted 8 lines · Jason Gunthorpe
Quoted 3 lines · Marek Szyprowski
Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).
Thanks for looking into this!
Can you explain more about how this actually solves the problem in the commit message? I don't get it.
Quoted 6 lines
Exynos DRM driver calls arm_iommu_detach_device(), then arm_iommu_attach_device() with a difrent 'mapping', see drivers/gpu/drm/exynos/exynos_drm_dma.c Lack of set_platform_dma_ops leads to a warning in iommu_group_do_set_platform_dma(). The other case of calling arm_iommu_detach_device() is after unsuccessful probe of the platform device.

Quoted 7 lines

Why can't this just use the normal iommu path in all cases?
It looks like it is trying to copy the DMA API domain from a parent device to a sub device.
Even when using arm_iommu an iommu_domain is still present, so the copy code should work?

The ARM DMA domain is a regular unmanaged domain owned by the ARM DMA code - in order to use any *other* domain, a user has to detach from that first (wrapped up in arm_iommu_detach_device() which also swizzles the DMA ops at the same time). Without set_platform_dma, that detach is now impossible (because no IOMMU API default domain exists either).

Quoted 4 lines

Though I'm still not really sure how this arm_iommu stuff works..
eg if a driver does iommu_device_claim_dma_owner() how does the iommu_domain get set back to the arm_iommu's mapping's iommu_domain?

It doesn't. Fact is that VFIO has only ever worked on Arm with arm-smmu and its old deprecated DT bindings which don't interact with any of the DMA ops stuff. And ownership has always been inherently enforced here since the DMA ops are just another external user from the IOMMU API's PoV.

Quoted 3 lines

According to the API that is what set_platform is supposed to do (eg it is what s390 does), but I don't see any code like that in any of the ARM32 drivers..

As above, it's in the relevant users, not the drivers. For what you're thinking of, the "platform" state is only ever transient.

(for the purposes of this discussion, ignore the MTK and Renesas drivers bodging around to implement pseudo-default-domain support on *top* of the user-centric arm_iommu_* APIs; that's a whole other mess to unravel)

Thanks,
Robin.

Jason Gunthorpe

2023-04-06 23:37 UTC #7 lore raw

Quoted 28 lines · Robin Murphy

Quoted 27 lines · Jason Gunthorpe
Quoted 14 lines · Marek Szyprowski
Hi,
I'm sorry for a delay in replying, but I was busy with other stuff.
Quoted 8 lines · Jason Gunthorpe
Quoted 3 lines · Marek Szyprowski
Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).
Thanks for looking into this!
Can you explain more about how this actually solves the problem in the commit message? I don't get it.
Quoted 6 lines
Exynos DRM driver calls arm_iommu_detach_device(), then arm_iommu_attach_device() with a difrent 'mapping', see drivers/gpu/drm/exynos/exynos_drm_dma.c Lack of set_platform_dma_ops leads to a warning in iommu_group_do_set_platform_dma(). The other case of calling arm_iommu_detach_device() is after unsuccessful probe of the platform device.
Why can't this just use the normal iommu path in all cases?
It looks like it is trying to copy the DMA API domain from a parent device to a sub device.

Quoted 8 lines

Quoted 2 lines
Even when using arm_iommu an iommu_domain is still present, so the copy code should work?
The ARM DMA domain is a regular unmanaged domain owned by the ARM DMA code - in order to use any *other* domain, a user has to detach from that first (wrapped up in arm_iommu_detach_device() which also swizzles the DMA ops at the same time). Without set_platform_dma, that detach is now impossible (because no IOMMU API default domain exists either).

I was looking at this again, and for completeness, the reason it doesn't have a default_domain is a bit subtle.

The driver does support IOMMU_DOMAIN_DMA, and it will go through all the iommu_group_alloc_default_domain() stuff with that..

But... __iommu_domain_alloc() calls iommu_get_dma_cookie() which will be wired to fail on ARM32, so the core code nixes the IOMMU_DOMAIN_DMA after the driver successfully created it.

I wonder if that is actually what we want? As it seems like it would be OK for ARM32 to have, effectively, a permanently empty unmanaged domain as the default_domain?

If instead of failing due to iommu_get_dma_cookie() we set the type to UNMANAGED and make that the default domain it could fix all of this, even for all the other ARM32 drivers that haven't reported this yet?

Alternatively since we taught these drivers to support IDENTITY, we should be able to remove the set_platform_dma_ops and instead implement for ARM32 def_domain_type fixed to return IDENTITY?

Jason

Marek Szyprowski

author maintainer

2023-04-12 22:04 UTC #8 lore raw

Quoted 46 lines · Jason Gunthorpe

Quoted 24 lines · Robin Murphy
Quoted 23 lines · Jason Gunthorpe
Quoted 9 lines · Marek Szyprowski
I'm sorry for a delay in replying, but I was busy with other stuff.
Quoted 6 lines · Jason Gunthorpe
Quoted 3 lines · Marek Szyprowski
Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).
Thanks for looking into this!
Quoted 8 lines
Quoted 2 lines
Can you explain more about how this actually solves the problem in the commit message? I don't get it.
Exynos DRM driver calls arm_iommu_detach_device(), then arm_iommu_attach_device() with a difrent 'mapping', see drivers/gpu/drm/exynos/exynos_drm_dma.c Lack of set_platform_dma_ops leads to a warning in iommu_group_do_set_platform_dma(). The other case of calling arm_iommu_detach_device() is after unsuccessful probe of the platform device.
Why can't this just use the normal iommu path in all cases?
It looks like it is trying to copy the DMA API domain from a parent device to a sub device.
Quoted 7 lines
Quoted 2 lines
Even when using arm_iommu an iommu_domain is still present, so the copy code should work?
The ARM DMA domain is a regular unmanaged domain owned by the ARM DMA code - in order to use any *other* domain, a user has to detach from that first (wrapped up in arm_iommu_detach_device() which also swizzles the DMA ops at the same time). Without set_platform_dma, that detach is now impossible (because no IOMMU API default domain exists either).
I was looking at this again, and for completeness, the reason it doesn't have a default_domain is a bit subtle.
The driver does support IOMMU_DOMAIN_DMA, and it will go through all the iommu_group_alloc_default_domain() stuff with that..
But... __iommu_domain_alloc() calls iommu_get_dma_cookie() which will be wired to fail on ARM32, so the core code nixes the IOMMU_DOMAIN_DMA after the driver successfully created it.
I wonder if that is actually what we want? As it seems like it would be OK for ARM32 to have, effectively, a permanently empty unmanaged domain as the default_domain?

Quoted 7 lines

If instead of failing due to iommu_get_dma_cookie() we set the type to UNMANAGED and make that the default domain it could fix all of this, even for all the other ARM32 drivers that haven't reported this yet?
Alternatively since we taught these drivers to support IDENTITY, we should be able to remove the set_platform_dma_ops and instead implement for ARM32 def_domain_type fixed to return IDENTITY?

Maybe it would be easier to simply switch ARM32 to generic IOMMU-DMA layer?

Best regards

Signature · 2 lines

Marek Szyprowski, PhD
Samsung R&D Institute Poland

Jason Gunthorpe

2023-04-12 22:40 UTC #9 lore raw

Quoted 51 lines · Marek Szyprowski

Quoted 50 lines · Jason Gunthorpe
Quoted 24 lines · Robin Murphy
Quoted 23 lines · Jason Gunthorpe
Quoted 9 lines · Marek Szyprowski
I'm sorry for a delay in replying, but I was busy with other stuff.
Quoted 6 lines · Jason Gunthorpe
Quoted 3 lines · Marek Szyprowski
Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).
Thanks for looking into this!
Quoted 8 lines
Quoted 2 lines
Can you explain more about how this actually solves the problem in the commit message? I don't get it.
Exynos DRM driver calls arm_iommu_detach_device(), then arm_iommu_attach_device() with a difrent 'mapping', see drivers/gpu/drm/exynos/exynos_drm_dma.c Lack of set_platform_dma_ops leads to a warning in iommu_group_do_set_platform_dma(). The other case of calling arm_iommu_detach_device() is after unsuccessful probe of the platform device.
Why can't this just use the normal iommu path in all cases?
It looks like it is trying to copy the DMA API domain from a parent device to a sub device.
Quoted 7 lines
Quoted 2 lines
Even when using arm_iommu an iommu_domain is still present, so the copy code should work?
The ARM DMA domain is a regular unmanaged domain owned by the ARM DMA code - in order to use any *other* domain, a user has to detach from that first (wrapped up in arm_iommu_detach_device() which also swizzles the DMA ops at the same time). Without set_platform_dma, that detach is now impossible (because no IOMMU API default domain exists either).
I was looking at this again, and for completeness, the reason it doesn't have a default_domain is a bit subtle.
The driver does support IOMMU_DOMAIN_DMA, and it will go through all the iommu_group_alloc_default_domain() stuff with that..
But... __iommu_domain_alloc() calls iommu_get_dma_cookie() which will be wired to fail on ARM32, so the core code nixes the IOMMU_DOMAIN_DMA after the driver successfully created it.
I wonder if that is actually what we want? As it seems like it would be OK for ARM32 to have, effectively, a permanently empty unmanaged domain as the default_domain?
If instead of failing due to iommu_get_dma_cookie() we set the type to UNMANAGED and make that the default domain it could fix all of this, even for all the other ARM32 drivers that haven't reported this yet?

Quoted 5 lines

Quoted 3 lines
Alternatively since we taught these drivers to support IDENTITY, we should be able to remove the set_platform_dma_ops and instead implement for ARM32 def_domain_type fixed to return IDENTITY?
Maybe it would be easier to simply switch ARM32 to generic IOMMU-DMA layer?

It is the long term goal, yes

In the short term it would be nice if we could at least keep the hacks out of the drivers

Jason

Joerg Roedel

maintainer Applied

2023-02-03 09:41 UTC #10 lore raw

In reply to Marek Szyprowski

Quoted 6 lines · Marek Szyprowski

Add set_platform_dma_ops() required for proper driver operation on ARM 32bit arch after recent changes in the IOMMU framework (detach ops removal).
Fixes c1fe9119ee70 ("iommu: Add set_platform_dma_ops callbacks")
Signed-off-by Marek Szyprowski <[email protected]>

Applied, thanks.