[Bug 221073] xHCI host controller dies on resume from s2idle on AMD Strix Halo [1022:1587]

[bugzilla-daemon@kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=221073

mattwork (mrh@frame.work) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|https://github.com/Framewor |
                   |kComputer/SoftwareFirmwareI |
                   |ssueTracker/issues/163      |

--- Comment #1 from mattwork (mrh@frame.work) ---
Update: new finding from GitHub #163.

@davidhubbard isolated iGPU memory allocation as a variable:
- iGPU Memory Size = 96GB: 3/4 suspends = xHCI dead
- iGPU Memory Size = Auto (0.5GB): cannot reproduce

However, my system reproduces at 16GB iGPU allocation. Higher iGPU
memory makes it more likely, but does not eliminate it at lower
settings. This is a severity dial, not an on/off switch.

This makes sense if iGPU memory allocation changes how the PCI root
complex manages power states during s2idle — the xHCI controller
at c1:00.4 shares the same PCI root as the amdgpu at c1:00.0.

See:
https://github.com/FrameworkComputer/SoftwareFirmwareIssueTracker/issues/163#issuecomment-3874652761

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.