amdgpu: crash when PCI rescan discovers new card
Submitted by jim..@..il.com
Assigned to Xorg Project Team
This is semi-related to another bug: https://bugzilla.kernel.org/show_bug.cgi?id=150731 And by semi, I mean it's either completely related or completely not related, and I don't know which yet.
If I boot my computer with my AMD video card bound to a driver besides amdgpu (like vfio-pci for my Windows virtual machine), then unbind it from that driver, intending to bind to to amdgpu for Linux gaming...
EXPECTED BEHAVIOR: I can unbind, remove the card, rescan (echo 1 > /sys/bus/pci/rescan), then bind the card to amdgpu, using DRI3 and the DRI_PRIME variable for games. This is a process that some successfully do with the radeon driver on older AMD cards.
ACTUAL BEHAVIOR: At the moment that I rescan (echo 1 > /sys/bus/pci/rescan), X crashes and restarts, booting me back to the login screen, and something (I guess either the kernel or X) has automatically bound the card to amdgpu on its own, without me entering those echo commands (echo "
<card's vendor ID>" "
<card's device ID>" > /sys/bus/pci/drivers/amdgpu/new_id) to bind it.
DESIRED BEHAVIOR: I can get my card bound to amdgpu, whether it's automatic or by my own typed bind commands, WITHOUT a boot-me-to-login-screen X crash.
The X crash log is at https://bugzilla.kernel.org/attachment.cgi?id=228411 The crash, starting at [456.336], has no errors. It seems to be just reconfiguring the graphics because it noticed a new available card, and somehow that resulted in me being booted back to the login screen. The log that was made after the crash, when I logged back in, is at https://bugzilla.kernel.org/attachment.cgi?id=228421