Bug 1047061

Description of problem:
After upgrading the kernel and linux-firmware to F20,  I started to get error messages about hardware errors.  The host has been running fine since March, so I didn't immediately believe in those error messages.  But something was apparently wrong now, the host started to crash intermittently.

I dug a bit in the history, and experimented a bit trying older kernels and packages.  It seems if I run the latest F19 kernel I used, and if /usr/lib/firmware has the contents of the F19 linux-firmware I use, then things are back to stable and no crash.  But if I try an F20 kernel, or the F19 kernel with firmware from F20, then I get error messages and crashes.

Version-Release number of selected component (if applicable):
These fails:
  linux-firmware-20130724-31.git31f6b30.fc20.noarch
  kernel-3.12.5-302.fc20.x86_64
Using these the machine is stable:
  linux-firmware-20130418-0.1.gitb584174.fc19.src.rpm (only replacing /usr/lib/firmware, otherwise F20 package)
  kernel-3.10.11-200.fc19.x86_64


How reproducible:
The error messages start more or less immediately, and comes with irregular intervals.  The crash can come within an hour, or can take a day.  I guess it is somehow related to how the machine is used, but I haven't figured out any pattern.


Additional info:
The typical error messages I get looks like this:

dec 28 15:44:18 mimmi kernel: mce: [Hardware Error]: Machine check events logged

I tried mcelog to see if I could get more information.  When run, it says the CPU is unsupported, and "AMD Processor family 21: Please load edac_mce_amd module."  Doing a modprobe on that, I get error messages like this instead:

dec 27 23:12:17 mimmi kernel: [Hardware Error]: MC2 Error: WCC Data ECC error.
dec 27 23:12:18 mimmi kernel: [Hardware Error]: Error Status: Corrected error, no action required.
dec 27 23:12:18 mimmi kernel: [Hardware Error]: CPU:2 (15:10:1) MC2_STATUS[Over|CE|MiscV|-|AddrV|-|-|CECC]: 0xdc054000010a0145
dec 27 23:12:18 mimmi kernel: [Hardware Error]: MC2_ADDR: 0x0000000000000040
dec 27 23:12:18 mimmi kernel: [Hardware Error]: cache level: L1, tx: DATA, mem-tx: DWR

These are the ones that the system survives.  I don't know what they would look like when it crashes.

The CPU is an AMD with combined CPU and GPU, AMD A8-5600K APU.  So when it comes to firmware, both CPU and GPU firmware might be involved.

I do get some error messages in the log when booting the working configuration.  I am slightly confused here, I am using the F19 kernel and F19 firmware, but they don't quite seem to match.  The missing TAHITI_uvd.bin and microcode_amd_fam15h.bin both exist in the newer linux-firmware.  But the machine is stable!

dec 28 16:25:20 mimmi kernel: [drm] radeon: 512M of VRAM memory ready
dec 28 16:25:20 mimmi kernel: [drm] radeon: 512M of GTT memory ready.
dec 28 16:25:20 mimmi kernel: radeon 0000:00:01.0: radeon_uvd: Can't load firmware "radeon/TAHITI_uvd.bin"
dec 28 16:25:20 mimmi kernel: [drm] GART: num cpu pages 131072, num gpu pages 131072
dec 28 16:25:20 mimmi kernel: [drm] Loading ARUBA Microcode
dec 28 16:25:20 mimmi kernel: [drm] PCIE GART of 512M enabled (table at 0x0000000000040000).

dec 28 16:25:29 mimmi kernel: microcode: failed to load file amd-ucode/microcode_amd_fam15h.bin
dec 28 16:25:29 mimmi kernel: microcode: CPU1: patch_level=0x06001116
dec 28 16:25:29 mimmi kernel: microcode: CPU2: patch_level=0x06001116
dec 28 16:25:29 mimmi kernel: microcode: CPU3: patch_level=0x06001116

When starting a newer kernel with the old firmware, I don't see these error messages.  (Which is also confusing me.  I thought the new kernel would try to use them.)