[Bug 2142389] Re: amdgpu (R9 380) fails to resume from suspend (deep sleep) – black screen, requires hard reboot

Subject: FINAL EVIDENCE: 3rd Git Bisect confirms Kernel Regression (AMD Tonga) Hi everyone, I’m posting this update to bring maximum visibility to this issue. After completing my third full git bisect to eliminate any margin of error, I have officially isolated the "first bad commit" responsible for the regressions on my AMD R9 380 (Tonga) system. The Smoking Gun: Commit: da1a055d01ed0c18402dd1f1934096ac4bb36ada Subject: lib/test_bpf: Call page_address() on page acquired with GFP_KERNEL flag Technical Context: The regression stems from replacing kmap() with page_address() in lib/test_bpf.c and removing the subsequent NULL checks. While this cleanup might be valid for most, it directly breaks stability for the Tonga architecture on kernels 6.6+. I have already engaged the upstream experts on GitLab (Ticket #5123), including @karolherbst, @daniel-schuermann, @agd5f (Alex Deucher), @sageofredondo, and @Venemo. I’m also calling the attention of @michaellarabel (Phoronix) to this case, as it highlights how a "minor" BPF cleanup can have unintended consequences for specific GCN 3.0 hardware generations. My Ask to the Ubuntu Kernel Team: I have done the heavy lifting of identifying the root cause through three separate bisections. We need reciprocity here. The community cannot be ignored when the evidence is this precise. To deny this regression now would be to invalidate the git bisect process itself. I am attaching the full technical report (Relatorio_Final_Bisect_Danilo.pdf) and bisection logs for verification. I am ready to test any patches to confirm the fix for Ubuntu users. Best regards, Danilo."The Tonga Guy" ** Attachment added: "[BOMBSHELL] Tonga R9 380 HDMI Dead on 6.8 LTS - Proof inside (Git Bisect + Logs)" https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2142389/+attachment/5967460/+files/3%C2%BA%20Git%20Bisect.png -- You received this bug notification because you are subscribed to linux in Ubuntu. Matching subscriptions: Bgg, Bmail, Nb https://bugs.launchpad.net/bugs/2142389 Title: amdgpu (R9 380) fails to resume from suspend (deep sleep) – black screen, requires hard reboot Status in linux package in Ubuntu: New Bug description: AMDGPU suspend → display black / no video after resume on Radeon R9 380 (No EDID read) Summary: After system suspend from Zorin OS 18 (Ubuntu 24.10 base, kernel 6.17.0-14), the system sometimes resumes but the display remains black (no signal). System continues running (fans/LEDs active), but monitor shows no output. Only hard reboot restores video. Steps to reproduce: Boot Zorin OS 18 (Ubuntu 24.10 kernel 6.17). Suspend system (e.g., via GNOME “Suspend”). Wait short period. Attempt to resume (mouse/keyboard). System wakes but display either shows garbled video or no output. Observed behavior: System appears not crashed (fans/LEDs/keyboard continue). Screen stays black or displays remnants but no usable video. Sometimes resume works, sometimes fails. Relevant log excerpt: amdgpu 0000:01:00.0: [drm] *ERROR* No EDID read. Hardware: Motherboard: Gigabyte B450 AORUS PRO WIFI CPU: AMD Ryzen 5 5500 GPU: AMD Radeon R9 380 Series (Tonga, amdgpu driver) Software environment: Zorin OS 18 Core (Ubuntu 24.10 base) kernel: 6.17.0-14-generic X11 session Workaround currently applied: Suspend disabled. System remains stable without suspend. Note: Bug appears related to video resume rather than system freeze; display subsystem (EDID handshake) may fail after suspend. Additional info: Similar reports of amdgpu black screen / suspend issues exist (e.g., Launchpad #2141216) and community discussions on black screen resume after suspend for AMD GPUs. To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2142389/+subscriptions