I have these installed:
kernel-modules-5.18.19-200.fc36.x86_64 kernel-modules-5.19.16-200.fc36.x86_64 kernel-modules-5.19.17-200.fc36.x86_64
on:
# inxi -Gxx Graphics: Device-1: AMD Kaveri [Radeon R7 Graphics] vendor: ASRock driver: amdgpu v: kernel arch: GCN-2 ports: active: DP-1,DVI-D-1,HDMI-A-1 empty: VGA-1 bus-ID: 00:01.0 chip-ID: 1002:130f Display: x11 server: X.org v: 1.20.14 driver: X: loaded: amdgpu unloaded: modesetting alternate: fbdev,vesa dri: radeonsi gpu: amdgpu
producing while booted to either newer kernel in Xorg.0.log:
# grep (EE Xorg.0.log ... [ 18.020] (EE) open /dev/dri/card0: No such file or directory [ 18.020] (EE) open /dev/dri/card0: No such file or directory # ls -gG /dev/dr* ls: cannot access '/dev/dr*': No such file or directory # lsmod | egrep 'amd|eon|vid' | sort amdgpu 8544256 0 drm_display_helper 172032 2 amdgpu,radeon drm_ttm_helper 16384 2 amdgpu,radeon edac_mce_amd 40960 0 gpu_sched 49152 1 amdgpu iommu_v2 24576 1 amdgpu radeon 1654784 0 ttm 90112 3 amdgpu,radeon,drm_ttm_helper video 61440 0 # dmesg | grep aile | egrep 'amd|eon' [ 4.839738] amdgpu 0000:00:01.0: Direct firmware load for amdgpu/kaveri_pfp.bin failed with error -2 [ 4.839743] amdgpu: gfx7: Failed to load firmware "amdgpu/kaveri_pfp.bin" [ 4.839744] [drm:gfx_v7_0_sw_init.cold [amdgpu]] *ERROR* Failed to load gfx firmware! [ 4.840065] [drm:amdgpu_device_init.cold [amdgpu]] *ERROR* sw_init of IP block <gfx_v7_0> failed -2 [ 4.840322] amdgpu 0000:00:01.0: amdgpu: amdgpu_device_ip_init failed [ 4.840480] amdgpu: probe of 0000:00:01.0 failed with error -2 # rpm -qa | grep mwar amd-gpu-firmware-20221012-141.fc36.noarch linux-firmware-20221012-141.fc36.noarch linux-firmware-whence-20221012-141.fc36.noarch
Various versions of linux-firmware, linux-firmware-whence and amd-gpu-firmware don't seem to have any impact on this. I tried going all the way back to July's linux-firmwar* and no amd-gpu-firmware.
Using old kernel, all is good, while with newer kernels, nothing is possible except via remote login - no video, no keyboard.
I reported similar trouble in Bugzilla month ago with much older Radeon: https://bugzilla.redhat.com/show_bug.cgi?id=2130843
Are others with GCN2 graphics having this trouble?
On Wed, 26 Oct 2022 00:55:51 -0400 Felix Miata mrmazda@earthlink.net wrote: [remove problem exploration]
Using old kernel, all is good, while with newer kernels, nothing is possible except via remote login - no video, no keyboard.
I reported similar trouble in Bugzilla month ago with much older Radeon: https://bugzilla.redhat.com/show_bug.cgi?id=2130843
Are others with GCN2 graphics having this trouble?
I don't think I have GCN2 graphics, though I do have an older radeon that is working without issue. I'm not sure what GCN2 is.
Can you run grep -i amd /boot/config[kernel version] on the kernel that is working and the kernel that isn't and check if there are CONFIG options that are different?
It might be a regression if everything is the same.
stan via users composed on 2022-10-26 09:13 (UTC-0700):
On Wed, 26 Oct 2022 00:55:51 -0400 Felix Miata wrote:
Using old kernel, all is good, while with newer kernels, nothing is possible except via remote login - no video, no keyboard.
I reported similar trouble in Bugzilla month ago with much older Radeon: https://bugzilla.redhat.com/show_bug.cgi?id=2130843
Are others with GCN2 graphics having this trouble?
I don't think I have GCN2 graphics, though I do have an older radeon that is working without issue. I'm not sure what GCN2 is.
Sea Islands is GCN2 initially released 201309: https://en.wikipedia.org/wiki/Graphics_Core_Next#second
Temash Kabini Kaveri Godavari Mullins Beema Carrizo-L
Can you run grep -i amd /boot/config[kernel version] on the kernel that is working and the kernel that isn't and check if there are CONFIG options that are different?
It might be a regression if everything is the same.
# diff -u kcfg51819.txt kcfg51917.txt --- kcfg51819.txt 2022-10-26 13:45:37.000000000 -0400 +++ kcfg51917.txt 2022-10-26 13:45:58.000000000 -0400 @@ -3,6 +3,7 @@ CONFIG_X86_MCE_AMD=y CONFIG_PERF_EVENTS_AMD_POWER=m CONFIG_PERF_EVENTS_AMD_UNCORE=y +CONFIG_PERF_EVENTS_AMD_BRS=y CONFIG_MICROCODE_AMD=y CONFIG_AMD_MEM_ENCRYPT=y # CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT is not set
BRS is "AMD Branch Sampling support". My Google-fu hasn't found a way to disable it as yet, but maxcpus=1 works.
On Wed, 26 Oct 2022 14:06:04 -0400 Felix Miata mrmazda@earthlink.net wrote:
Sea Islands is GCN2 initially released 201309: https://en.wikipedia.org/wiki/Graphics_Core_Next#second
Don't have it, but I had no idea I was so out of date. That is almost ten years ago. My card does everything I need, so I suppose I can live with using a dinosaur.
# diff -u kcfg51819.txt kcfg51917.txt --- kcfg51819.txt 2022-10-26 13:45:37.000000000 -0400 +++ kcfg51917.txt 2022-10-26 13:45:58.000000000 -0400 @@ -3,6 +3,7 @@ CONFIG_X86_MCE_AMD=y CONFIG_PERF_EVENTS_AMD_POWER=m CONFIG_PERF_EVENTS_AMD_UNCORE=y +CONFIG_PERF_EVENTS_AMD_BRS=y CONFIG_MICROCODE_AMD=y CONFIG_AMD_MEM_ENCRYPT=y # CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT is not set
BRS is "AMD Branch Sampling support". My Google-fu hasn't found a way to disable it as yet, but maxcpus=1 works.
Dealing with this extra config solved the issue? Or just prevented another issue?
Maybe you can open a ticket asking that this be a module instead of built in, if it is possible, in the fedora kernels. I compile a custom kernel locally, and have that turned off since I use the radeon driver instead of the amd driver. I've read that the amd driver can replace the radeon driver, even for old cards, but the radeon driver has been rock solid and done everything I want, so I haven't bothered.
stan via users composed on 2022-10-26 20:07 (UTC-0700):
On Wed, 26 Oct 2022 14:06:04 -0400 Felix Miata wrote:
# diff -u kcfg51819.txt kcfg51917.txt --- kcfg51819.txt 2022-10-26 13:45:37.000000000 -0400 +++ kcfg51917.txt 2022-10-26 13:45:58.000000000 -0400 @@ -3,6 +3,7 @@ CONFIG_X86_MCE_AMD=y CONFIG_PERF_EVENTS_AMD_POWER=m CONFIG_PERF_EVENTS_AMD_UNCORE=y +CONFIG_PERF_EVENTS_AMD_BRS=y CONFIG_MICROCODE_AMD=y CONFIG_AMD_MEM_ENCRYPT=y # CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT is not set
BRS is "AMD Branch Sampling support". My Google-fu hasn't found a way to disable it as yet, but maxcpus=1 works.
Dealing with this extra config solved the issue? Or just prevented another issue?
I simply tried something that seemed would obviate BRS, but disabling cores is a performance killer, so only done as test.
Maybe you can open a ticket asking that this be a module instead of built in, if it is possible, in the fedora kernels.
I copied 5.19.17-300.fc37.x86_64 from the F37 installation to the F36 installation:
# ls -Gg /boot/ini*19.17* -rw------- 1 19015128 Oct 25 19:11 /boot/initramfs-5.19.17-200.fc36.x86_64.img -rw------- 1 34070416 Oct 25 23:02 /boot/initramfs-5.19.17-300.fc37.x86_64.img # The F37 version boots normally. Both installations have identical /etc/dracut.conf.d/*, so I have no idea why the F37 is so much heavier, unless there's extra stuff in it because F37 is still not yet released.
# rpm -qa | grep mwar amd-gpu-firmware-20221012-141.fc36.noarch linux-firmware-20221012-141.fc36.noarch linux-firmware-whence-20221012-141.fc36.noarch
# rpm -qa | grep mwar linux-firmware-whence-20221012-142.fc37.noarch linux-firmware-20221012-142.fc37.noarch
# rpm -e amd-gpu-firmware error: Failed dependencies: amd-gpu-firmware is needed by (installed) linux-firmware-20221012-141.fc36.noarch # rpm -e --nodeps amd-gpu-firmware # <reboot using F36 kernel> black # <reboot using F37 kernel> succeeds # dnf install amd-gpu-firmware # echo 'blacklist radeon' > /etc/modprobe.d/0radeon.conf # <reboot using F36 kernel> black <remove blacklist,reboot with F36 kernel> black # lsmod | sort | grep -E 'vid|eon|amd' amdgpu 8544256 0 drm_display_helper 172032 2 amdgpu,radeon drm_ttm_helper 16384 2 amdgpu,radeon edac_mce_amd 40960 0 gpu_sched 49152 1 amdgpu iommu_v2 24576 1 amdgpu radeon 1654784 0 ttm 90112 3 amdgpu,radeon,drm_ttm_helper video 61440 0 #
So, amdgpu module loads, but: # grep (EE /var/log/Xorg.0.log ... [ 319.021] (EE) open /dev/dri/card0: No such file or directory ... # dmesg | grep -i amd | grep aile [ 4.771180] amdgpu 0000:00:01.0: Direct firmware load for amdgpu/kaveri_pfp.bin failed with error -2 [ 4.771187] amdgpu: gfx7: Failed to load firmware "amdgpu/kaveri_pfp.bin" [ 4.771189] [drm:gfx_v7_0_sw_init.cold [amdgpu]] *ERROR* Failed to load gfx firmware! [ 4.771559] [drm:amdgpu_device_init.cold [amdgpu]] *ERROR* sw_init of IP block <gfx_v7_0> failed -2 [ 4.771868] amdgpu 0000:00:01.0: amdgpu: amdgpu_device_ip_init failed [ 4.772081] amdgpu: probe of 0000:00:01.0 failed with error -2 #
What to do next? Bugzilla?
Felix Miata composed on 2022-10-27 00:56 (UTC-0400):
# dnf install amd-gpu-firmware
...
So, amdgpu module loads, but: # grep (EE /var/log/Xorg.0.log ... [ 319.021] (EE) open /dev/dri/card0: No such file or directory ... # dmesg | grep -i amd | grep aile [ 4.771180] amdgpu 0000:00:01.0: Direct firmware load for amdgpu/kaveri_pfp.bin failed with error -2 [ 4.771187] amdgpu: gfx7: Failed to load firmware "amdgpu/kaveri_pfp.bin" [ 4.771189] [drm:gfx_v7_0_sw_init.cold [amdgpu]] *ERROR* Failed to load gfx firmware! [ 4.771559] [drm:amdgpu_device_init.cold [amdgpu]] *ERROR* sw_init of IP block <gfx_v7_0> failed -2 [ 4.771868] amdgpu 0000:00:01.0: amdgpu: amdgpu_device_ip_init failed [ 4.772081] amdgpu: probe of 0000:00:01.0 failed with error -2 #
What to do next? Bugzilla?
# dracut -f ... # ls -Gg .ini*5.19.17* initramfs-5.19.17-200.fc36.x86_64.img -rw------- 1 34233108 Oct 27 01:49 initramfs-5.19.17-200.fc36.x86_64.img -rw------- 1 19015128 Oct 25 19:11 .initramfs-5.19.17-200.fc36.x86_64.img1 -rw------- 1 34233108 Oct 27 01:49 .initramfs-5.19.17-200.fc36.x86_64.img2
With the big fat new initrd all is as expected. Apparently the only way firmware will load is if in the initrd, and the last initrd builds didn't include it. Or, is it?
lsinitrd <kernel-initrd-filename> | grep amdgpu and same with the name of the firmware. If the module is there and the firmware is not then the module would load and the firmware would not exist so cannot load. The firmware gets loaded or not when the module loads.
On Thu, Oct 27, 2022 at 1:08 AM Felix Miata mrmazda@earthlink.net wrote:
Felix Miata composed on 2022-10-27 00:56 (UTC-0400):
# dnf install amd-gpu-firmware
...
So, amdgpu module loads, but: # grep (EE /var/log/Xorg.0.log ... [ 319.021] (EE) open /dev/dri/card0: No such file or directory ... # dmesg | grep -i amd | grep aile [ 4.771180] amdgpu 0000:00:01.0: Direct firmware load for amdgpu/kaveri_pfp.bin failed with error -2 [ 4.771187] amdgpu: gfx7: Failed to load firmware "amdgpu/kaveri_pfp.bin" [ 4.771189] [drm:gfx_v7_0_sw_init.cold [amdgpu]] *ERROR* Failed to load gfx firmware! [ 4.771559] [drm:amdgpu_device_init.cold [amdgpu]] *ERROR* sw_init of IP block <gfx_v7_0> failed -2 [ 4.771868] amdgpu 0000:00:01.0: amdgpu: amdgpu_device_ip_init failed [ 4.772081] amdgpu: probe of 0000:00:01.0 failed with error -2 #
What to do next? Bugzilla?
# dracut -f ... # ls -Gg .ini*5.19.17* initramfs-5.19.17-200.fc36.x86_64.img -rw------- 1 34233108 Oct 27 01:49 initramfs-5.19.17-200.fc36.x86_64.img -rw------- 1 19015128 Oct 25 19:11 .initramfs-5.19.17-200.fc36.x86_64.img1 -rw------- 1 34233108 Oct 27 01:49 .initramfs-5.19.17-200.fc36.x86_64.img2
With the big fat new initrd all is as expected. Apparently the only way firmware will load is if in the initrd, and the last initrd builds didn't include it. Or, is it? -- Evolution as taught in public schools is, like religion, based on faith, not based on science.
Team OS/2 ** Reg. Linux User #211409 ** a11y rocks!
Felix Miata _______________________________________________ users mailing list -- users@lists.fedoraproject.org To unsubscribe send an email to users-leave@lists.fedoraproject.org Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/users@lists.fedoraproject.org Do not reply to spam, report it: https://pagure.io/fedora-infrastructure/new_issue