On Sun, 11 Jul 2021 06:22:51 -0700 stan via test test@lists.fedoraproject.org wrote:
I normally run custom compiled rawhide kernels tuned to my hardware. In those I turn debugging off. I haven't been able to build one for 5.14 yet because of a problem in the build process. I will try again with the 20210709 kernel because it seems the fix for the error is in that package.
I was able to compile a custom kernel with the 20200709 kernel src.rpm package. So the perf fix is working; the link that used to cause an error now causes a warning, warning: absolute symlink: /lib/modules/5.14.0-0.rc0.20210709gitf55966571d5e.14.20210711.fc35.x86_64/build -> /usr/src/kernels/5.14.0-0.rc0.20210709gitf55966571d5e.14.20210711.fc35.x86_64
Unfortunately, it doesn't boot correctly because it fails to get the edid from the monitor. The rest of boot goes off without a hitch, I can login (without seeing anything) and run commands. When I get to X, it is in some sort of vga mode with large text and flaky colors and it is very fragile and hangs after a short time. But I can switch from there to a virtual console, login as root, and shutdown normally.
The Fedora rawhide kernels do not have this issue. I can see them probe the EDID in the messages, whereas the custom kernel doesn't. The custom kernel built with this config file had no problem booting into this same hardware configuration with a 5.10 kernel, so something has changed in the settings for getting edid between then and 5.14.
Anyone have a suggestion for where the problem might be?
On Sun, 11 Jul 2021 12:46:28 -0700 stan upaitag@zoho.com wrote:
Unfortunately, it doesn't boot correctly because it fails to get the edid from the monitor. The rest of boot goes off without a hitch, I can login (without seeing anything) and run commands. When I get to X, it is in some sort of vga mode with large text and flaky colors and it is very fragile and hangs after a short time. But I can switch from there to a virtual console, login as root, and shutdown normally.
The Fedora rawhide kernels do not have this issue. I can see them probe the EDID in the messages, whereas the custom kernel doesn't. The custom kernel built with this config file had no problem booting into this same hardware configuration with a 5.10 kernel, so something has changed in the settings for getting edid between then and 5.14.
Anyone have a suggestion for where the problem might be?
I think I have found the problem. After many failed attempts to set configuration options. :-)
It seems that the custom kernel fails to load microcode that is required. Has this mechanism changed in recent kernels? Is there a way to fix this? I never had to worry about microcode in the past.
This is a fedora stock kernel boot message series:
Jul 15 11:48:34 fedora kernel: [drm] radeon kernel modesetting enabled. Jul 15 11:48:34 fedora kernel: radeon 0000:01:00.0: vgaarb: deactivate vga console Jul 15 11:48:34 fedora kernel: Console: switching to colour dummy device 80x25 Jul 15 11:48:34 fedora kernel: [drm] initializing kernel modesetting (CAICOS 0x1002:0x6779 0x174B:0xE164 0x00). Jul 15 11:48:34 fedora kernel: ATOM BIOS: C26401 Jul 15 11:48:34 fedora kernel: radeon 0000:01:00.0: VRAM: 1024M 0x0000000000000000 - 0x000000003FFFFFFF (1024M used) Jul 15 11:48:34 fedora kernel: radeon 0000:01:00.0: GTT: 1024M 0x0000000040000000 - 0x000000007FFFFFFF Jul 15 11:48:34 fedora kernel: [drm] Detected VRAM RAM=1024M, BAR=256M Jul 15 11:48:34 fedora kernel: [drm] RAM width 64bits DDR Jul 15 11:48:34 fedora kernel: [drm] radeon: 1024M of VRAM memory ready Jul 15 11:48:34 fedora kernel: [drm] radeon: 1024M of GTT memory ready. Jul 15 11:48:34 fedora kernel: [drm] Loading CAICOS Microcode Jul 15 11:48:34 fedora kernel: [drm] Internal thermal controller without fan control Jul 15 11:48:34 fedora kernel: [drm] radeon: dpm initialized
This is a custom kernel boot message series:
Jul 15 11:44:57 fedora kernel: [drm] radeon kernel modesetting enabled. Jul 15 11:44:57 fedora kernel: radeon 0000:01:00.0: vgaarb: deactivate vga console Jul 15 11:44:57 fedora kernel: Console: switching to colour dummy device 240x68 Jul 15 11:44:57 fedora kernel: [drm] initializing kernel modesetting (CAICOS 0x1002:0x6779 0x174B:0xE164 0x00). Jul 15 11:44:57 fedora kernel: ATOM BIOS: C26401 Jul 15 11:44:57 fedora kernel: radeon 0000:01:00.0: VRAM: 1024M 0x0000000000000000 - 0x000000003FFFFFFF (1024M used) Jul 15 11:44:57 fedora kernel: radeon 0000:01:00.0: GTT: 1024M 0x0000000040000000 - 0x000000007FFFFFFF Jul 15 11:44:57 fedora kernel: [drm] Detected VRAM RAM=1024M, BAR=256M Jul 15 11:44:57 fedora kernel: [drm] RAM width 64bits DDR Jul 15 11:44:57 fedora kernel: [drm] radeon: 1024M of VRAM memory ready Jul 15 11:44:57 fedora kernel: [drm] radeon: 1024M of GTT memory ready. Jul 15 11:44:57 fedora kernel: [drm] Loading CAICOS Microcode Jul 15 11:44:57 fedora kernel: radeon 0000:01:00.0: Direct firmware load for radeon/CAICOS_pfp.bin failed with error -2 Jul 15 11:44:57 fedora kernel: ni_cp: Failed to load firmware "radeon/CAICOS_pfp.bin" Jul 15 11:44:57 fedora kernel: [drm:evergreen_init [radeon]] *ERROR* Failed to load firmware! Jul 15 11:44:57 fedora kernel: radeon 0000:01:00.0: Fatal error during GPU init Jul 15 11:44:57 fedora kernel: [drm] radeon: finishing device. Jul 15 11:44:57 fedora kernel: [drm] radeon: ttm finalized Jul 15 11:44:57 fedora kernel: radeon: probe of 0000:01:00.0 failed with error -2
Thanks.
The rc1 series of 5.14 kernels do not have the lockups and failed boots.
On Thu, 15 Jul 2021 12:03:10 -0700 stan via test test@lists.fedoraproject.org wrote:
This is a custom kernel boot message series:
Jul 15 11:44:57 fedora kernel: [drm] Loading CAICOS Microcode Jul 15 11:44:57 fedora kernel: radeon 0000:01:00.0: Direct firmware load for radeon/CAICOS_pfp.bin failed with error -2 Jul 15 11:44:57 fedora kernel: ni_cp: Failed to load firmware "radeon/CAICOS_pfp.bin" Jul 15 11:44:57 fedora kernel: [drm:evergreen_init [radeon]] *ERROR* Failed to load firmware! Jul 15 11:44:57 fedora kernel: radeon 0000:01:00.0: Fatal error during GPU init Jul 15 11:44:57 fedora kernel: [drm] radeon: finishing device. Jul 15 11:44:57 fedora kernel: [drm] radeon: ttm finalized Jul 15 11:44:57 fedora kernel: radeon: probe of 0000:01:00.0 failed with error -2
The problem was that firmware is now compressed, and there is a configuration option to allow that so it is decoded before being sent to the device. I didn't have that option set, so the compressed firmware failed to load because it was compressed when the device was expecting valid binary. No gpu device, no scan of the EDID.
kernel@lists.fedoraproject.org