Re: [PATCH 5.15 00/93] 5.15.107-rc1 review (possible amdgpu regression)

[Greg Kroah-Hartman]

On Wed, Apr 12, 2023 at 10:47:13PM +0100, Eddie Chapman wrote:
> Greg Kroah-Hartman wrote:
> > This is the start of the stable review cycle for the 5.15.107 release.
> > There are 93 patches in this series, all will be posted as a response
> > to this one.  If anyone has any issues with these being applied, please let
> > me know.
> > 
> > Responses should be made by Fri, 14 Apr 2023 08:28:02 +0000.
> > Anything received after that time might be too late.
> > 
> 
> I think I'm seeing a regression here in the amdgpu driver, though not being
> a kernel dev I could be wrong.
> 
> I built and booted this today on an x86_64 machine (AMD Ryzen 7 3700X,
> Gigabyte X570 UD motherboard) with 3 x AMD graphics cards (using names from
> lspci output):
> - Cape Verde GL [FirePro W4100]
> - Oland XT [Radeon HD 8670 / R5 340X OEM / R7 250/350/350X OEM] (rev 83)
> - Bonaire [Radeon R7 200 Series]
> 
> All three using the amdgpu driver (radeon module blacklisted).
> 
> This machine has been running vanilla 5.15 stable releases for a good while,
> with the kernel updated with whatever the latest 5.15 release is every 6
> weeks or so. Never had any amdgpu problems.
> 
> To build 5.15.107-rc1 I applied the contents of the queue-5.15 directory on
> top of 5.15.106, having synced the stable queue git repo up until commit
> 344d8ad1b5dde387d1ce4d1be2641753b89dd10d (still the latest commit as a
> type). This is what I have done for years running vanilla stable kernels.
> 
> There was nothing out of the ordinary in the build output, but on rebooting
> into 5.15.107-rc1 I had the following error in dmesg from 1 card only:
> 
> amdgpu 0000:0d:00.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test
> failed on vce0 (-110).
> [drm:process_one_work] *ERROR* ib ring test failed (-110).
> 
> This was during bootup immediately after driver loading. X is not running.
> 0000:0d:00.0 is the Bonaire card.
> 
> I then shutdown and fully powered off for a few minutes, booted 5.15.107-rc1
> again, but the error on that card persisted exactly the same.
> 
> This was a regression for me as I've never had that error before on any
> kernel release (I grepped through old kernel logs to check).
> 
> I then rebuilt 5.15.107-rc1 but without applying the following 4 patches:
> 
> drm-panfrost-fix-the-panfrost_mmu_map_fault_addr-error-path.patch
> drm-amdgpu-fix-amdgpu_job_free_resources-v2.patch
> drm-amdgpu-prevent-race-between-late-signaled-fences.patch
> drm-bridge-lt9611-fix-pll-being-unable-to-lock.patch
> 
> On booting into the newly built kernel there was no error anymore, amdgpu
> dmesg output was as normal, and the machine is running fine now on that.
> 
> So I'm quite confident one of those patches introduced the error for me.
> Having now looked at the contents of them I see the lt9611 is entirely
> different hardware and I'm guessing the panfrost one probably is as well, so
> most likely I didn't need to remove those 2.
> 
> This is not a great report and maybe not helpful (sorry) as unfortunately I
> cannot try and narrow it down further to a single patch as this machine has
> to stay running now for a while. I just crudely tried yanking those 4 to
> hopefully get rid of the error and get the machine running again. Also I
> didn't go on to test whether the card actually worked as expected, maybe the
> error is harmless after all, though it doesn't look insignificant.
> 
> As the error was only output for the Bonaire card (the other two were fine),
> below is lspci -vvv output for that card only in case it helps. If anyone
> would like further info just let me know.

Ok, I've dropped the two drm-amdgpu-* patches from the tree for now.
Let me know if the release works properly for you or not.

thanks for testing and letting us know!

greg k-h