Description of problem: Booting on an EFI machine with radeon 6770m, with 3.8rc5 git 1.1. This is always reproducible, and essentially renders the machine useless. This may be a regression from the previously working VBIOS loading. Jan 30 23:33:07 localhost kernel: Linux version 3.8.0-0.rc5.git1.1.fc19.x86_64 (mockbuild@bkernel02) (gcc version 4.8.0 20130129 (Red Hat 4.8.0-0.7) (GCC) ) #1 SMP Tue Jan 29 18:34:46 UTC 2013 Jan 30 23:33:08 localhost kernel: radeon 0000:01:00.0: Invalid ROM contents Jan 30 23:33:08 localhost kernel: radeon 0000:01:00.0: Invalid ROM contents Jan 30 23:33:08 localhost kernel: [drm:radeon_get_bios] *ERROR* Unable to locate a BIOS ROM Jan 30 23:33:08 localhost kernel: radeon 0000:01:00.0: Fatal error during GPU init Jan 30 23:33:08 localhost kernel: [drm] radeon: finishing device. Jan 30 23:33:08 localhost kernel: [TTM] Memory type 2 has not been initialized Jan 30 23:33:08 localhost kernel: INFO: trying to register non-static key. Jan 30 23:33:08 localhost kernel: the code is fine but needs lockdep annotation. Jan 30 23:33:08 localhost kernel: turning off the locking correctness validator. Jan 30 23:33:08 localhost kernel: Pid: 238, comm: systemd-udevd Not tainted 3.8.0-0.rc5.git1.1.fc19.x86_64 #1 Jan 30 23:33:08 localhost kernel: Call Trace: Jan 30 23:33:08 localhost kernel: [<ffffffff816a7349>] register_lock_class.part.24+0x38/0x3c Jan 30 23:33:08 localhost kernel: [<ffffffff810c8cbf>] __lock_acquire+0x14ff/0x1910 Jan 30 23:33:08 localhost kernel: [<ffffffff810c729d>] ? mark_held_locks+0x9d/0x100 Jan 30 23:33:08 localhost kernel: [<ffffffff816b28f3>] ? retint_restore_args+0x13/0x13 Jan 30 23:33:08 localhost kernel: [<ffffffff810c9852>] lock_acquire+0xa2/0x120 Jan 30 23:33:08 localhost kernel: [<ffffffff8107ea85>] ? flush_work+0x5/0x2a0 Jan 30 23:33:08 localhost kernel: [<ffffffff8107eacc>] flush_work+0x4c/0x2a0 Jan 30 23:33:08 localhost kernel: [<ffffffff8107ea85>] ? flush_work+0x5/0x2a0 Jan 30 23:33:08 localhost kernel: [<ffffffff8101b4e0>] ? hw_breakpoint_pmu_read+0x10/0x10 Jan 30 23:33:08 localhost kernel: [<ffffffffa02087fb>] radeon_irq_kms_fini+0x2b/0x70 [radeon] Jan 30 23:33:08 localhost kernel: [<ffffffffa023c8da>] evergreen_fini+0x4a/0xb0 [radeon] Jan 30 23:33:08 localhost kernel: [<ffffffffa01dbc4e>] radeon_device_fini+0x3e/0x100 [radeon] Jan 30 23:33:08 localhost kernel: [<ffffffffa01dd5c3>] radeon_driver_unload_kms+0x33/0x50 [radeon] Jan 30 23:33:08 localhost kernel: [<ffffffffa01dd698>] radeon_driver_load_kms+0xb8/0x190 [radeon] Jan 30 23:33:08 localhost kernel: [<ffffffffa0061c16>] drm_get_pci_dev+0x176/0x2b0 [drm] Jan 30 23:33:08 localhost kernel: [<ffffffff8119f738>] ? kfree+0x188/0x1c0 Jan 30 23:33:08 localhost kernel: [<ffffffffa01c31d2>] radeon_pci_probe+0xb2/0xf0 [radeon] Jan 30 23:33:08 localhost kernel: [<ffffffff8135451e>] local_pci_probe+0x3e/0x70 Jan 30 23:33:08 localhost kernel: [<ffffffff81355811>] pci_device_probe+0x121/0x130 Jan 30 23:33:08 localhost kernel: [<ffffffff8141ad27>] driver_probe_device+0x87/0x390 Jan 30 23:33:08 localhost kernel: [<ffffffff8141b103>] __driver_attach+0x93/0xa0 Jan 30 23:33:08 localhost kernel: [<ffffffff8141b070>] ? __device_attach+0x40/0x40 Jan 30 23:33:08 localhost kernel: [<ffffffff81418ca3>] bus_for_each_dev+0x63/0xa0 Jan 30 23:33:08 localhost kernel: [<ffffffff8141a7ae>] driver_attach+0x1e/0x20 Jan 30 23:33:08 localhost kernel: [<ffffffff8141a20a>] bus_add_driver+0xda/0x2c0 Jan 30 23:33:08 localhost kernel: [<ffffffffa00cd000>] ? 0xffffffffa00ccfff Jan 30 23:33:08 localhost kernel: [<ffffffff8141b771>] driver_register+0x71/0x150 Jan 30 23:33:08 localhost kernel: [<ffffffffa00cd000>] ? 0xffffffffa00ccfff Jan 30 23:33:08 localhost kernel: [<ffffffff813543b0>] __pci_register_driver+0x60/0x70 Jan 30 23:33:08 localhost kernel: [<ffffffffa0061e6a>] drm_pci_init+0x11a/0x130 [drm] Jan 30 23:33:08 localhost kernel: [<ffffffffa00cd000>] ? 0xffffffffa00ccfff Jan 30 23:33:08 localhost kernel: [<ffffffffa00cd0ec>] radeon_init+0xec/0x1000 [radeon] Jan 30 23:33:08 localhost kernel: [<ffffffff8100210a>] do_one_initcall+0x10a/0x160 Jan 30 23:33:08 localhost kernel: [<ffffffff810d66ee>] load_module+0x198e/0x2190 Jan 30 23:33:08 localhost kernel: [<ffffffff81345ab0>] ? ddebug_proc_write+0xf0/0xf0 Jan 30 23:33:08 localhost kernel: [<ffffffff810d6fb1>] sys_init_module+0xc1/0x110 Jan 30 23:33:08 localhost kernel: [<ffffffff816bb499>] system_call_fastpath+0x16/0x1b Jan 30 23:33:08 localhost kernel: radeon: probe of 0000:01:00.0 failed with error -22
You build your kernel with CONFIG_ACPI enabled ?
Also please dump your vbios while booting in uefi mode with kms disabled (you can connect through ssh or log in level3). cd /sys/bus/pci/devices/0000:01:05.0 sudo sh -c "echo 1 > rom" sudo sh -c "cat rom > ~/bios.rom" replace 0000:01:05.0 by the pciid of your gpu and attach bios.rom to the bug Thanks
Also when you say it was working before, you mean it was working before when booting in bios legacy mode or when booting in uefi mode ?
CONFIG_ACPI should be enabled - this is the build from rawhide, I did not build it myself. It was working in kernel 3.7 series with the radeon card under EFI. Do you want the vbios rom from a working 3.7 boot and / or the non-working 3.8 boot?
Created attachment 690720 [details] Vbios rom from 6770m 3.7 kernel Booting 3.8, that command yields a blank file. Booting 3.7, the rom is present, and I have attached it.
Just for the record, these are both being booted from grub2-efi, and in efi mode (Not CSM)
So this sounds like a kernel regression. Did you tested other 3.8rc that would narrow down the list of commit ? If you are able to do kernel bisection with something like : git bisect start drivers/gpu/drm/radeon (to restrict bisection to radeon assuming the issue is a patch against radeon) that would be helpfull.
kernel 3.8.0-0-rc4.git1.1 is the "first" kernel with the issue. kernel 3.8.0-0-rc3.git1.1 is the "last" kernel that works. I am not sure how to "bisect" between the two of these, using the kernel git, but being able to produce rpms and initrd that works for fedora etc. Any tips and guidance would be appreciated.
kernel 3.8.0-0-rc6 corrects this issue.