2020-11-14 15:24:34

by Jonathan Marek

Subject: [RESEND PATCH v2 3/5] drm/msm: call dma_direct_bypass()

Always use direct dma ops and no swiotlb.

Note: arm-smmu-qcom already avoids creating iommu dma ops, but not
everything uses arm-smmu-qcom and this also sets the dma mask.

Signed-off-by: Jonathan Marek <[email protected]>
---
drivers/gpu/drm/msm/Kconfig | 1 +
drivers/gpu/drm/msm/msm_drv.c | 8 +++++---
2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/msm/Kconfig b/drivers/gpu/drm/msm/Kconfig
index e5816b498494..07c50405970a 100644
--- a/drivers/gpu/drm/msm/Kconfig
+++ b/drivers/gpu/drm/msm/Kconfig
@@ -20,6 +20,7 @@ config DRM_MSM
select SND_SOC_HDMI_CODEC if SND_SOC
select SYNC_FILE
select PM_OPP
+ select DMA_OPS_BYPASS
help
DRM/KMS driver for MSM/snapdragon.

diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index 49685571dc0e..bae48afca82e 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -6,6 +6,7 @@
*/

#include <linux/dma-mapping.h>
+#include <linux/dma-direct.h>
#include <linux/kthread.h>
#include <linux/uaccess.h>
#include <uapi/linux/sched/types.h>
@@ -1288,10 +1289,11 @@ static int msm_pdev_probe(struct platform_device *pdev)
if (ret)
goto fail;

- /* on all devices that I am aware of, iommu's which can map
- * any address the cpu can see are used:
+ /* always use direct dma ops and no swiotlb
+ * note: arm-smmu-qcom already avoids creating iommu dma ops, but
+ * not everything uses arm-smmu-qcom and this also sets the dma mask
*/
- ret = dma_set_mask_and_coherent(&pdev->dev, ~0);
+ ret = dma_direct_bypass(&pdev->dev);
if (ret)
goto fail;

--
2.26.1

2020-11-14 16:25:14

by Christoph Hellwig

Subject: Re: [RESEND PATCH v2 3/5] drm/msm: call dma_direct_bypass()

On Sat, Nov 14, 2020 at 10:17:11AM -0500, Jonathan Marek wrote:
> Always use direct dma ops and no swiotlb.
>
> Note: arm-smmu-qcom already avoids creating iommu dma ops, but not
> everything uses arm-smmu-qcom and this also sets the dma mask.
>
> Signed-off-by: Jonathan Marek <[email protected]>
> ---
> drivers/gpu/drm/msm/Kconfig | 1 +
> drivers/gpu/drm/msm/msm_drv.c | 8 +++++---
> 2 files changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/msm/Kconfig b/drivers/gpu/drm/msm/Kconfig
> index e5816b498494..07c50405970a 100644
> --- a/drivers/gpu/drm/msm/Kconfig
> +++ b/drivers/gpu/drm/msm/Kconfig
> @@ -20,6 +20,7 @@ config DRM_MSM
> select SND_SOC_HDMI_CODEC if SND_SOC
> select SYNC_FILE
> select PM_OPP
> + select DMA_OPS_BYPASS
> help
> DRM/KMS driver for MSM/snapdragon.
>
> diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
> index 49685571dc0e..bae48afca82e 100644
> --- a/drivers/gpu/drm/msm/msm_drv.c
> +++ b/drivers/gpu/drm/msm/msm_drv.c
> @@ -6,6 +6,7 @@
> */
>
> #include <linux/dma-mapping.h>
> +#include <linux/dma-direct.h>
> #include <linux/kthread.h>
> #include <linux/uaccess.h>
> #include <uapi/linux/sched/types.h>
> @@ -1288,10 +1289,11 @@ static int msm_pdev_probe(struct platform_device *pdev)
> if (ret)
> goto fail;
>
> - /* on all devices that I am aware of, iommu's which can map
> - * any address the cpu can see are used:
> + /* always use direct dma ops and no swiotlb

Again, and implementation detail. Comments should not explain details
obvious from the code (especially layers away) but the intent.