Re: [REGRESSION] Invalid gather when using Tegra210 media engines

[Jason Gunthorpe <jgg@ziepe.ca>]

On Mon, Feb 03, 2025 at 02:55:12PM +0000, Diogo Ivo wrote:
> Hello,
> 
> Commit c8cc2655cc6c introduced a regression when trying to use the media
> accelerators present on the Tegra X1 SoC.
> 
> I came across this regression when testing the branch [1] that leverages
> the NVJPG engine in the Tegra X1 for decoding a JPEG file. After commit
> c8cc2655cc6c we see the following error messages after submitting a job
> through the TEGRA_CHANNEL_SUBMIT IOCTL:
> 
> [   46.879757] tegra-nvjpg 54380000.nvjpg: invalid gather for push buffer
> 0x0000000108f08000

What driver is this? The message comes from 
   drivers/gpu/host1x/hw/channel_hw.c

But what driver is 'tegra-nvjpg' that is bound to 54380000.nvjpg ?

Is it the stuff in 
 drivers/gpu/drm/nouveau/nvkm/engine/nvjpg/

I don't see "tegra-nvjpg" in the kernel?

Can you share where the failing command was sent to the device?

> Please let me know if you need more information on my side and I'll be
> happy to provide it.

It is still ARM64 & CONFIG_ARM_DMA_USE_IOMMU=n

?

I'm guessing it is the same basic issue as fae6e669cdc5 ("drm/tegra:
Do not assume that a NULL domain means no DMA IOMMU"), except in the
host1x not DRM code. It looks to me like the same pattern was copied
there.

How about this:

diff --git a/drivers/gpu/host1x/dev.c b/drivers/gpu/host1x/dev.c
index be2ad7203d7b96..090b1fc97a7309 100644
--- a/drivers/gpu/host1x/dev.c
+++ b/drivers/gpu/host1x/dev.c
@@ -361,6 +361,10 @@ static bool host1x_wants_iommu(struct host1x *host1x)
 	return true;
 }
 
+/*
+ * Returns ERR_PTR on failure, NULL if the translation is IDENTITY, otherwise a
+ * valid paging domain.
+ */
 static struct iommu_domain *host1x_iommu_attach(struct host1x *host)
 {
 	struct iommu_domain *domain = iommu_get_domain_for_dev(host->dev);
@@ -385,6 +389,8 @@ static struct iommu_domain *host1x_iommu_attach(struct host1x *host)
 	 * Similarly, if host1x is already attached to an IOMMU (via the DMA
 	 * API), don't try to attach again.
 	 */
+	if (domain && domain->type == IOMMU_DOMAIN_IDENTITY)
+		domain = NULL;
 	if (!host1x_wants_iommu(host) || domain)
 		return domain;
 

(if not can you investigate this function's flow compared to a good
kernel?)

Jason