[Bug 2158236] Re: Fix amdgpu hang in amdgpu_ctx_mgr_entity_flush during exec

** Also affects: linux (Ubuntu) Importance: Undecided Status: New -- You received this bug notification because you are subscribed to linux in Ubuntu. Matching subscriptions: Bgg, Bmail, Nb https://bugs.launchpad.net/bugs/2158236 Title: Fix amdgpu hang in amdgpu_ctx_mgr_entity_flush during exec Status in HWE Next: In Progress Status in linux package in Ubuntu: New Status in linux-oem-6.17 package in Ubuntu: Invalid Status in linux source package in Noble: New Status in linux-oem-6.17 source package in Noble: Fix Committed Status in linux source package in Resolute: New Status in linux-oem-6.17 source package in Resolute: Fix Released Status in linux source package in Stonking: New Status in linux-oem-6.17 source package in Stonking: Invalid Bug description: [Impact] System hangs on amdgpu during exec. systemd blocked for 368s in D state in amdgpu_ctx_mgr_entity_flush (mutex_lock) via filp_close -> amdgpu_flush during do_close_on_exec -> load_elf_binary. Reproduces on 6.17.0-1020-oem.  kernel: INFO: task systemd:16426 blocked for more than 368 seconds.  kernel: Tainted: G O 6.17.0-1020-oem #20-Ubuntu  kernel: task:systemd state:D pid:16426  kernel: mutex_lock  kernel: amdgpu_ctx_mgr_entity_flush+0x46/0x1f0 [amdgpu]  kernel: amdgpu_flush+0x26/0x50 [amdgpu]  kernel: filp_flush+0x6f/0xb0  kernel: filp_close+0x14/0x30  kernel: do_close_on_exec+0xe7/0x140  kernel: begin_new_exec+0x1ab/0x420  kernel: load_elf_binary+0x32d/0xf40 [Fix] Cherry-pick upstream commits: - b18fc0ab837381 "drm/amdgpu: fix sync handling in amdgpu_dma_buf_move_notify" - 930595df251c "drm/amdgpu: remove check for BO reservation add assert instead" Pass NULL ticket to amdgpu_vm_handle_moved in move_notify so the clear=true path is used, avoiding a PT update while another process's job is still running — the contention that blocks amdgpu_ctx_mgr_entity_flush. [Test Plan] Boot oem-6.17 on dual-GPU AMD system without P2P PCI; glxgears on GPU0, Xorg on GPU1 sharing a dmabuf. Without fix: systemd hung >368s in amdgpu_ctx_mgr_entity_flush. With fix: no hang. [Where problems could occur] changes only the ticket arg in amdgpu_dma_buf_move_notify; normal VM update path untouched. Risk limited to dmabuf move-notify sync on multi-GPU shared-BO setups. It may break amdgpu driver. To manage notifications about this bug go to: https://bugs.launchpad.net/hwe-next/+bug/2158236/+subscriptions