Bug 1656257

Created attachment 1511527 [details]
dmesg

Created attachment 1511528 [details]
system log

Proposed as a Blocker and Freeze Exception for 30-beta by Fedora user mikhail using the blocker tracking app because:

  Because of this bug impossible login to system with AMD GPU

Which kernel commit are corresponds to 4.20.0-0.rc5.git0.1.fc30.x86_64 package?

If I am understand right from change log:

* Mon Dec 03 2018 Justin M. Forbes <jforbes> - 4.20.0-0.rc5.git0.1
- Linux v4.20-rc5

* Mon Dec 03 2018 Justin M. Forbes <jforbes>
- Disable debugging options.

 this is tag v4.20-rc5 which is corresponds to commit 2595646791c319cadfdbf271563aac97d0843dc7 but when I builded vanilla kernel during bisecting this issue at this commit I am not experienced this issue.

So I suppose problem in some Fedora patch. 


$ git bisect log
git bisect start
# bad: [2595646791c319cadfdbf271563aac97d0843dc7] Linux 4.20-rc5
git bisect bad 2595646791c319cadfdbf271563aac97d0843dc7
# good: [94f371cb73944b410a269d570d6946c042f2ddd0] Merge tag 'acpi-4.20-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
git bisect good 94f371cb73944b410a269d570d6946c042f2ddd0
# good: [08be37b798921af207e78082fe261a6ca8be5024] mm/gup: finish consolidating error handling
git bisect good 08be37b798921af207e78082fe261a6ca8be5024
# good: [880584176ed7875117a5ba76cf316cb60f7ad30b] Merge tag 'for-linus-20181201' of git://git.kernel.dk/linux-block
git bisect good 880584176ed7875117a5ba76cf316cb60f7ad30b
# good: [292974c5acae330186cbf5a833385f666aeb12c0] Merge tag 'for-linus-4.20a-rc5-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip
git bisect good 292974c5acae330186cbf5a833385f666aeb12c0
# good: [292974c5acae330186cbf5a833385f666aeb12c0] Merge tag 'for-linus-4.20a-rc5-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip
git bisect good 292974c5acae330186cbf5a833385f666aeb12c0
# good: [adb97bcdbdb2d42c90b5f11e08a9b5fbc017e5d7] Merge tag 'v4.20-rockchip-dts64fixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git/mmind/linux-rockchip into fixes
git bisect good adb97bcdbdb2d42c90b5f11e08a9b5fbc017e5d7
# good: [89acb56db4979e55380839c815566ddb9a01949b] Merge tag 'davinci-fixes-for-v4.20' of git://git.kernel.org/pub/scm/linux/kernel/git/nsekhar/linux-davinci into fixes
git bisect good 89acb56db4979e55380839c815566ddb9a01949b
# good: [91e43395820baad80248987608216c35da9df65b] Merge branch 'fixes-dts' into omap-for-v4.20/fixes
git bisect good 91e43395820baad80248987608216c35da9df65b
# good: [7e76e65ce7e9405a9608e1b806be58a6cbf4a737] MAINTAINERS: Remove unused Qualcomm SoC mailing list
git bisect good 7e76e65ce7e9405a9608e1b806be58a6cbf4a737
# good: [bfed4d730823440d0da0cd21554efc2de831627d] Merge tag 'imx-fixes-4.20-2' of git://git.kernel.org/pub/scm/linux/kernel/git/shawnguo/linux into fixes
git bisect good bfed4d730823440d0da0cd21554efc2de831627d
# good: [bfed4d730823440d0da0cd21554efc2de831627d] Merge tag 'imx-fixes-4.20-2' of git://git.kernel.org/pub/scm/linux/kernel/git/shawnguo/linux into fixes
git bisect good bfed4d730823440d0da0cd21554efc2de831627d
# good: [6a512726090a5cfd8d5cd41652d2b98a222854b8] Merge tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
git bisect good 6a512726090a5cfd8d5cd41652d2b98a222854b8
# first bad commit: [2595646791c319cadfdbf271563aac97d0843dc7] Linux 4.20-rc5


But vanilla commit 2595646791c319cadfdbf271563aac97d0843dc7 also good for me.

Its probably the CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT=y  tunrned on after the rc5 fedora kernels.

mem_encrypt=off  on the kernel command line makes the boot work as usual here.

I am confirm that I able reproduce this issue with vanilla kernel after recompiling with option CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT=y
Also mem_encrypt=off works as workaround for me.

This commit are blame: https://src.fedoraproject.org/rpms/kernel/c/f8216ee47abe1997e9fc7efb3d819d6105d898f5?branch=master

Peter, why you enable option CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT=y ?

I asked AMD developers do they know about problem with option `CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT=y` with their GPU.

And I got short answer:

Yes, that is a known limitation.

Almost all GPU drivers are incompatible with memory encryption.

As a workaround you can either turn off memory encryption or force the slowpath for memory allocation in the GPU drivers.

Well, the commit message says:

"This makes it so users don't have to do mem_encrypt=1 to enable SEV VMs."

which is a reference to AMD's 'secure encrypted virtualization' tech: https://developer.amd.com/sev/

so...that's why. He wants to let people use that feature (without a special kernel arg).

However, if that's going to break most AMD GPUs (per the reply Mikhail got), I'd agree it's unacceptable and probably a release blocker.

CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT is off again starting in 4.20-rc6-82-g65e08c5e8631.