diff mbox series

[v2,1/7] media: hantro: Set DMA max segment size

Message ID 20190725141756.2518-2-ezequiel@collabora.com (mailing list archive)
State New, archived
Headers show
Series hantro: Add RK3399 VP8 decoding support | expand

Commit Message

Ezequiel Garcia July 25, 2019, 2:17 p.m. UTC 
From: Francois Buergisser <fbuergisser@chromium.org>

The Hantro codec is typically used in platforms with an IOMMU,
so we need to set a proper DMA segment size. Devices without an
IOMMU will still fallback to default 64KiB segments.

Cc: stable@vger.kernel.org
Fixes: 775fec69008d3 ("media: add Rockchip VPU JPEG encoder driver")
Signed-off-by: Francois Buergisser <fbuergisser@chromium.org>
Signed-off-by: Ezequiel Garcia <ezequiel@collabora.com>
---
 drivers/staging/media/hantro/hantro_drv.c | 1 +
 1 file changed, 1 insertion(+)

Comments

Philipp Zabel July 25, 2019, 3:36 p.m. UTC | #1 
On Thu, 2019-07-25 at 11:17 -0300, Ezequiel Garcia wrote:
> From: Francois Buergisser <fbuergisser@chromium.org>
> 
> The Hantro codec is typically used in platforms with an IOMMU,
> so we need to set a proper DMA segment size.

... to make sure the DMA-mapping subsystem produces contiguous mappings?

> Devices without an
> IOMMU will still fallback to default 64KiB segments.

I don't understand this comment. The default max_seg_size may be 64 KiB,
but if we are always setting it to DMA_BUT_MASK(32), there is no falling
back.

> Cc: stable@vger.kernel.org
> Fixes: 775fec69008d3 ("media: add Rockchip VPU JPEG encoder driver")
> Signed-off-by: Francois Buergisser <fbuergisser@chromium.org>
> Signed-off-by: Ezequiel Garcia <ezequiel@collabora.com>
> ---
>  drivers/staging/media/hantro/hantro_drv.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/staging/media/hantro/hantro_drv.c b/drivers/staging/media/hantro/hantro_drv.c
> index b71a06e9159e..4eae1dbb1ac8 100644
> --- a/drivers/staging/media/hantro/hantro_drv.c
> +++ b/drivers/staging/media/hantro/hantro_drv.c
> @@ -731,6 +731,7 @@ static int hantro_probe(struct platform_device *pdev)
>  		dev_err(vpu->dev, "Could not set DMA coherent mask.\n");
>  		return ret;
>  	}
> +	vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));

This should be complemented by a call to
vb2_dma_contig_clear_max_seg_size() in _remove,
to avoid leaking dev->dma_parms.

>  
>  	for (i = 0; i < vpu->variant->num_irqs; i++) {
>  		const char *irq_name = vpu->variant->irqs[i].name;

regards
Philipp
Tomasz Figa July 25, 2019, 3:41 p.m. UTC | #2 
On Fri, Jul 26, 2019 at 12:36 AM Philipp Zabel <p.zabel@pengutronix.de> wrote:
>
> On Thu, 2019-07-25 at 11:17 -0300, Ezequiel Garcia wrote:
> > From: Francois Buergisser <fbuergisser@chromium.org>
> >
> > The Hantro codec is typically used in platforms with an IOMMU,
> > so we need to set a proper DMA segment size.
>
> ... to make sure the DMA-mapping subsystem produces contiguous mappings?
>
> > Devices without an
> > IOMMU will still fallback to default 64KiB segments.
>
> I don't understand this comment. The default max_seg_size may be 64 KiB,
> but if we are always setting it to DMA_BUT_MASK(32), there is no falling
> back.
>

DMA mask and segment size are two completely orthogonal parameters.
Please check https://elixir.bootlin.com/linux/v5.3-rc1/source/drivers/iommu/dma-iommu.c#L740
for an example of how the latter is used.

Best regards,
Tomasz
diff mbox series

Patch

diff --git a/drivers/staging/media/hantro/hantro_drv.c b/drivers/staging/media/hantro/hantro_drv.c
index b71a06e9159e..4eae1dbb1ac8 100644
--- a/drivers/staging/media/hantro/hantro_drv.c
+++ b/drivers/staging/media/hantro/hantro_drv.c
@@ -731,6 +731,7 @@  static int hantro_probe(struct platform_device *pdev)
 		dev_err(vpu->dev, "Could not set DMA coherent mask.\n");
 		return ret;
 	}
+	vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
 
 	for (i = 0; i < vpu->variant->num_irqs; i++) {
 		const char *irq_name = vpu->variant->irqs[i].name;