Public bug reported:

BugLink: https://bugs.launchpad.net/bugs/2165889

[ Impact ]

We found when some AMD eGPUs are connected via thunderbolt (thus through
PCIe tunneling), pure compute workload via ROCm won't launch without
also connecting the said eGPU to a display. This is due to mismatched
expectation from runtime PM when dGPU is not connected to root port but
is instead external.

[ Fix ]

Cherry-pick the following commit, which is currently in 7.3-rc1
(c52feb436539):

- drm/amdgpu: Disable runtime PM for externally attached dGPUs

[ Test plan ]

- Connect the eGPU with thunderbolt
- Without connecting the eGPU to any display, launch compute workload with ROCm
- The workload should run.

[ Where problems could occur ]

This impacts runtime power management behavior for GPUs connected
externally via thunderbolt. This in theory doesn't impact dGPUs and
iGPUs.

** Affects: linux (Ubuntu)
     Importance: Undecided
     Assignee: Leo Lin (0xff07)
         Status: New

** Affects: linux-oem-7.0 (Ubuntu)
     Importance: Undecided
     Assignee: Leo Lin (0xff07)
         Status: New


** Tags: originate-from-2163510

** Changed in: linux-oem-7.0 (Ubuntu)
     Assignee: (unassigned) => Leo Lin (0xff07)

** Also affects: linux (Ubuntu)
   Importance: Undecided
       Status: New

** Description changed:

+ BugLink: https://bugs.launchpad.net/bugs/2165889
+ 
  [ Impact ]
  
  We found when some AMD eGPUs are connected via thunderbolt (thus through
  PCIe tunneling), pure compute workload via ROCm won't launch without
  also connecting the said eGPU to a display. This is due to mismatched
  expectation from runtime PM when dGPU is not connected to root port but
  is instead external.
  
  [ Fix ]
  
  Cherry-pick the following commit, which is currently in 7.3-rc1
  (c52feb436539):
  
  - drm/amdgpu: Disable runtime PM for externally attached dGPUs
  
- 
  [ Test plan ]
  
  - Connect the eGPU with thunderbolt
  - Without connecting the eGPU to any display, launch compute workload with 
ROCm
  - The workload should run.
  
- 
  [ Where problems could occur ]
  
  This impacts runtime power management behavior for GPUs connected
  externally via thunderbolt. This in theory doesn't impact dGPUs and
  iGPUs.

** Tags added: originate-from-2163510

** Changed in: linux (Ubuntu)
     Assignee: (unassigned) => Leo Lin (0xff07)

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2165889

Title:
  [SRU] Fix issues with dGPUs in eGPU enclosure

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2165889/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to