← back

Intel Arc B580 firmware debugging (GSC recovery)

Notes from debugging a Linux bring-up failure involving Intel Arc B580 (BMG G21 / Battlemage)

Spent the day trying to untangle firmware weirdness on my Arc B580.

Arc splits its firmware into several layers: PCode, GSC, OpROM code/data, GuC, and HuC. If one of them fails early enough, the whole graphics stack dies before /dev/dri even exists. (New GPU firmware stacks are annoyingly complicated, but Intel exposing these layers publicly made debugging possible.)

In my case, it looked like the previous owner corrupted the GSC firmware somehow. Every cold boot caused the xe driver to fail during early initialization while interacting with PCode, so the card never came online properly.

Debugging path

• Intermittent /dev/dri creation
• LightDM crashing with "no screens found"
• PCode initialization timeouts
• "Survivability mode"
• Silent xe hangs immediately after "deactivate VGA console"
• Different kernel behavior (6.8, 6.17 generic, 6.17 OEM)
• Card occasionally disappearing from lspci after reseating

At one point it looked more like hardware damage than firmware corruption. I checked for obvious shorts on 12V and 3.3V rails, and measured resistance/voltages on Vcore, memory, and memory controller rails. Nothing abnormal showed up, but that's the extent of my EE knowledge.

Recovery

Using Intel's igsc tool compiled from source, I was able to access the Graphics Security Controller through /dev/mei1 and reflash the GSC firmware from 21.1137 to 21.1182.

After firmware update

• Mesa detected: Intel Arc B580 Graphics (BMG G21)
• Vulkan initialized normally
• No more LLVM software rendering for OpenGL

Remaining issues

There's also a separate upstream bug in the Linux 6.17 xe driver where the GuC scheduler crashes at idle on Battlemage. As far as I can tell it has nothing to do with the firmware problem.

Between xe, Mesa, fwupd and igsc, I could recover the card without switching to Windows, which wouldn't have been possible on Linux not long ago.

Firmware distribution is still a mess, though. The firmware for my ASRock card isn't on LVFS yet, so some regions may still need the vendor's Windows flashing tools.

References

• Intel GSC tooling (igsc)
• fwupd project
• Linux xe driver documentation
• Igor's Lab teardown context
• Linux Mint discussion 1
• Linux Mint discussion 2

Also collected VRM rail resistance measurements during debugging if someone needs it(provided i didnt delete the pics from my phone lol).

Images

Board / teardown / reference images: