[PATCH v12 2/7] iommu/mediatek: Fix two IOMMU share pagetable issue

Laura Nao laura.nao at collabora.com
Fri Aug 18 08:41:56 PDT 2023


Hello,

This patch is causing fluster tests to fail on MT8192, IOMMU read and write faults are reported in the log. Here's an extract is from the VP9 tests:

mtk-iommu 1401d000.m4u: fault type=0x280 iova=0x1ff7c0000 pa=0x0 master=0x5000480(larb=5 port=0) layer=0 read
mtk-iommu 1401d000.m4u: fault type=0x280 iova=0x1fe3ee000 pa=0x0 master=0x5000492(larb=5 port=4) layer=0 write

Tests are failing for H264, VP8 and VP9 decoding on next-20230817 with fluster in the same way, complete logs can be found here: 
- H264: https://storage.kernelci.org/next/master/next-20230817/arm64/defconfig+arm64-chromebook+videodec/gcc-10/lab-collabora/v4l2-decoder-conformance-h264-mt8192-asurada-spherion-r0.html
- VP8: https://storage.kernelci.org/next/master/next-20230817/arm64/defconfig+arm64-chromebook+videodec/gcc-10/lab-collabora/v4l2-decoder-conformance-vp8-mt8192-asurada-spherion-r0.html
- VP9: https://storage.kernelci.org/next/master/next-20230817/arm64/defconfig+arm64-chromebook+videodec/gcc-10/lab-collabora/v4l2-decoder-conformance-vp9-mt8192-asurada-spherion-r0.html

Reverting this patch fixes the issue. 

>From my understanding, on MT8192 the 16GB iova space is partitioned between display, vcodec and camera. The iommu domain configuration for vcodec was loaded from frstdata->bank[0]->m4u_dom (vcodec is preassigned to a specific iova range), while after this patch a new pagetable is created instead. 

Do you have any insight on how to fix this while keeping compatibility with MT8188 and MT8195?

Best,

Laura



More information about the Linux-mediatek mailing list