[08/11] drm/nouveau: tegra: Skip IOMMU initialization if already attached

Submitted by Thierry Reding on Sept. 16, 2019, 3:04 p.m.

Details

Message ID 20190916150412.10025-9-thierry.reding@gmail.com
State New
Headers show
Series "drm/nouveau: Enable GP10B by default" ( rev: 1 ) in Nouveau

Not browsing as part of any series.

Commit Message

Thierry Reding Sept. 16, 2019, 3:04 p.m.
From: Thierry Reding <treding@nvidia.com>

If the GPU is already attached to an IOMMU, don't detach it and setup an
explicit IOMMU domain. Since Nouveau can now properly handle the case of
the DMA API being backed by an IOMMU, just continue using the DMA API.

Signed-off-by: Thierry Reding <treding@nvidia.com>
---
 .../drm/nouveau/nvkm/engine/device/tegra.c    | 19 +++++++------------
 1 file changed, 7 insertions(+), 12 deletions(-)

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
index d0d52c1d4aee..fc652aaa41c7 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
@@ -23,10 +23,6 @@ 
 #ifdef CONFIG_NOUVEAU_PLATFORM_DRIVER
 #include "priv.h"
 
-#if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
-#include <asm/dma-iommu.h>
-#endif
-
 static int
 nvkm_device_tegra_power_up(struct nvkm_device_tegra *tdev)
 {
@@ -109,14 +105,13 @@  nvkm_device_tegra_probe_iommu(struct nvkm_device_tegra *tdev)
 	unsigned long pgsize_bitmap;
 	int ret;
 
-#if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
-	if (dev->archdata.mapping) {
-		struct dma_iommu_mapping *mapping = to_dma_iommu_mapping(dev);
-
-		arm_iommu_detach_device(dev);
-		arm_iommu_release_mapping(mapping);
-	}
-#endif
+	/*
+	 * Skip explicit IOMMU initialization if the GPU is already attached
+	 * to an IOMMU domain. This can happen if the DMA API is backed by an
+	 * IOMMU.
+	 */
+	if (iommu_get_domain_for_dev(dev))
+		return;
 
 	if (!tdev->func->iommu_bit)
 		return;

Comments

Hi Thierry,

On 16/09/2019 16:04, Thierry Reding wrote:
> From: Thierry Reding <treding@nvidia.com>
> 
> If the GPU is already attached to an IOMMU, don't detach it and setup an
> explicit IOMMU domain. Since Nouveau can now properly handle the case of
> the DMA API being backed by an IOMMU, just continue using the DMA API.
> 
> Signed-off-by: Thierry Reding <treding@nvidia.com>
> ---
>   .../drm/nouveau/nvkm/engine/device/tegra.c    | 19 +++++++------------
>   1 file changed, 7 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
> index d0d52c1d4aee..fc652aaa41c7 100644
> --- a/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
> +++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
> @@ -23,10 +23,6 @@
>   #ifdef CONFIG_NOUVEAU_PLATFORM_DRIVER
>   #include "priv.h"
>   
> -#if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
> -#include <asm/dma-iommu.h>
> -#endif
> -
>   static int
>   nvkm_device_tegra_power_up(struct nvkm_device_tegra *tdev)
>   {
> @@ -109,14 +105,13 @@ nvkm_device_tegra_probe_iommu(struct nvkm_device_tegra *tdev)
>   	unsigned long pgsize_bitmap;
>   	int ret;
>   
> -#if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
> -	if (dev->archdata.mapping) {
> -		struct dma_iommu_mapping *mapping = to_dma_iommu_mapping(dev);
> -
> -		arm_iommu_detach_device(dev);
> -		arm_iommu_release_mapping(mapping);
> -	}
> -#endif
> +	/*
> +	 * Skip explicit IOMMU initialization if the GPU is already attached
> +	 * to an IOMMU domain. This can happen if the DMA API is backed by an
> +	 * IOMMU.
> +	 */
> +	if (iommu_get_domain_for_dev(dev))
> +		return;

Beware of "iommu.passthrough=1" - you could get a valid default domain 
here yet still have direct/SWIOTLB DMA ops. I guess you probably want to 
double-check the domain type as well.

Robin.

>   
>   	if (!tdev->func->iommu_bit)
>   		return;
>

On 16/09/2019 16:57, Thierry Reding wrote:
> On Mon, Sep 16, 2019 at 04:29:18PM +0100, Robin Murphy wrote:
>> Hi Thierry,
>>
>> On 16/09/2019 16:04, Thierry Reding wrote:
>>> From: Thierry Reding <treding@nvidia.com>
>>>
>>> If the GPU is already attached to an IOMMU, don't detach it and setup an
>>> explicit IOMMU domain. Since Nouveau can now properly handle the case of
>>> the DMA API being backed by an IOMMU, just continue using the DMA API.
>>>
>>> Signed-off-by: Thierry Reding <treding@nvidia.com>
>>> ---
>>>    .../drm/nouveau/nvkm/engine/device/tegra.c    | 19 +++++++------------
>>>    1 file changed, 7 insertions(+), 12 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
>>> index d0d52c1d4aee..fc652aaa41c7 100644
>>> --- a/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
>>> +++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
>>> @@ -23,10 +23,6 @@
>>>    #ifdef CONFIG_NOUVEAU_PLATFORM_DRIVER
>>>    #include "priv.h"
>>> -#if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
>>> -#include <asm/dma-iommu.h>
>>> -#endif
>>> -
>>>    static int
>>>    nvkm_device_tegra_power_up(struct nvkm_device_tegra *tdev)
>>>    {
>>> @@ -109,14 +105,13 @@ nvkm_device_tegra_probe_iommu(struct nvkm_device_tegra *tdev)
>>>    	unsigned long pgsize_bitmap;
>>>    	int ret;
>>> -#if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
>>> -	if (dev->archdata.mapping) {
>>> -		struct dma_iommu_mapping *mapping = to_dma_iommu_mapping(dev);
>>> -
>>> -		arm_iommu_detach_device(dev);
>>> -		arm_iommu_release_mapping(mapping);
>>> -	}
>>> -#endif
>>> +	/*
>>> +	 * Skip explicit IOMMU initialization if the GPU is already attached
>>> +	 * to an IOMMU domain. This can happen if the DMA API is backed by an
>>> +	 * IOMMU.
>>> +	 */
>>> +	if (iommu_get_domain_for_dev(dev))
>>> +		return;
>>
>> Beware of "iommu.passthrough=1" - you could get a valid default domain here
>> yet still have direct/SWIOTLB DMA ops. I guess you probably want to
>> double-check the domain type as well.
> 
> Good point. An earlier version of this patch had an additional check for
> IOMMU_DOMAIN_DMA, but then that failed on 32-bit ARM because there the
> DMA API can also use IOMMU_DOMAIN_UNMANAGED type domains. Checking for
> IOMMU_DOMAIN_IDENTIFY should be safe, though. That doesn't seem to
> appear in arch/arm, arch/arm64 or drivers/iommu/dma-iommu.c.

Right, "domain && domain->type != IOMMU_DOMAIN_IDENTITY" should be 
sufficient to answer "is the DMA layer managing my address space for 
me?" unless and until some massive API change happens (which I certainly 
don't foresee).

Robin.