Boot failures in -next on Jetson TK1

Boot failures in -next on Jetson TK1

[Mark Brown]

Currently -next is failing to boot on Jetson TK1.  The problem looks to
be the Nouveau driver, during initialization it reports an address
decode error then starts printing error messages saying "nouveau
57000000.gpu: fifo: SCHED_ERROR 20 []" over and over again.

I've pasted the start of the errors below, you can see a full log and
more details at:

   https://kernelci.org/boot/id/5b0882a259b514339779a881/

The warnings about Spectre are a separate issue and don't seem to affect
the boot.

[ 15.194484] nouveau 57000000.gpu: NVIDIA GK20A (0ea000a1)
[   15.200109] udevd[109]: could not rename interface '3' from 'eth0' to 'enp1s0': Device or resource busy
[   15.206399] nouveau 57000000.gpu: imem: using IOMMU
[   15.315122] CPU2: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
[   15.320021] nouveau 57000000.gpu: Direct firmware load for nvidia/gk20a/fecs_inst.bin failed with error -2
[   15.384841] nouveau 57000000.gpu: Direct firmware load for nouveau/nvea_fuc409c failed with error -2
[   15.393972] nouveau 57000000.gpu: Direct firmware load for nouveau/fuc409c failed with error -2
[   15.402679] nouveau 57000000.gpu: gr: failed to load fuc409c
[   15.409434] CPU1: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
[   15.419398] CPU1: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
[   15.482568] tegra-mc 70019000.memory-controller: gpusrd: read @0x00041200: EMEM address decode error (EMEM decode error)
[   15.491232] [TTM] Zone  kernel: Available graphics memory: 375202 kiB
[   15.502768] [TTM] Zone highmem: Available graphics memory: 1030050 kiB
[   15.509290] [TTM] Initializing pool allocator
[   15.513658] nouveau 57000000.gpu: DRM: VRAM: 0 MiB
[   15.518451] nouveau 57000000.gpu: DRM: GART: 1048576 MiB
[   15.526546] CPU1: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
[   15.527290] tegra-mc 70019000.memory-controller: gpusrd: read @0x00072000: EMEM address decode error (EMEM decode error)
[   15.537050] CPU1: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
[   15.546928] nouveau 57000000.gpu: fifo: SCHED_ERROR 20 []
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20180526/689741b6/attachment.sig>

Boot failures in -next on Jetson TK1

[Thierry Reding]

On Sat, May 26, 2018 at 11:36:29AM +0100, Mark Brown wrote:
> Currently -next is failing to boot on Jetson TK1.  The problem looks to
> be the Nouveau driver, during initialization it reports an address
> decode error then starts printing error messages saying "nouveau
> 57000000.gpu: fifo: SCHED_ERROR 20 []" over and over again.
> 
> I've pasted the start of the errors below, you can see a full log and
> more details at:
> 
>    https://kernelci.org/boot/id/5b0882a259b514339779a881/
> 
> The warnings about Spectre are a separate issue and don't seem to affect
> the boot.
> 
> [ 15.194484] nouveau 57000000.gpu: NVIDIA GK20A (0ea000a1)
> [   15.200109] udevd[109]: could not rename interface '3' from 'eth0' to 'enp1s0': Device or resource busy
> [   15.206399] nouveau 57000000.gpu: imem: using IOMMU
> [   15.315122] CPU2: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
> [   15.320021] nouveau 57000000.gpu: Direct firmware load for nvidia/gk20a/fecs_inst.bin failed with error -2
> [   15.384841] nouveau 57000000.gpu: Direct firmware load for nouveau/nvea_fuc409c failed with error -2
> [   15.393972] nouveau 57000000.gpu: Direct firmware load for nouveau/fuc409c failed with error -2
> [   15.402679] nouveau 57000000.gpu: gr: failed to load fuc409c
> [   15.409434] CPU1: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
> [   15.419398] CPU1: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
> [   15.482568] tegra-mc 70019000.memory-controller: gpusrd: read @0x00041200: EMEM address decode error (EMEM decode error)
> [   15.491232] [TTM] Zone  kernel: Available graphics memory: 375202 kiB
> [   15.502768] [TTM] Zone highmem: Available graphics memory: 1030050 kiB
> [   15.509290] [TTM] Initializing pool allocator
> [   15.513658] nouveau 57000000.gpu: DRM: VRAM: 0 MiB
> [   15.518451] nouveau 57000000.gpu: DRM: GART: 1048576 MiB
> [   15.526546] CPU1: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
> [   15.527290] tegra-mc 70019000.memory-controller: gpusrd: read @0x00072000: EMEM address decode error (EMEM decode error)
> [   15.537050] CPU1: Spectre v2: firmware did not set auxiliary control register IBE bit, system vulnerable
> [   15.546928] nouveau 57000000.gpu: fifo: SCHED_ERROR 20 []

This is a known issue that was introduced in v4.16 by a combination of
the 32-bit ARM DMA/IOMMU glue and an Tegra SMMU driver change.

There is a fix here:

	http://patchwork.ozlabs.org/patch/902830/

Which got remotely NAK'ed by the DMA API maintainer. I then came up with
this, based on feedback from Christoph:

	http://patchwork.ozlabs.org/project/linux-tegra/list/?series=40853

But that's kind of blocked right now, awaiting feedback. I'll send out
another version, which will hopefully strike the right balance.

Thierry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20180529/1f1a3b2f/attachment-0001.sig>