Description of problem: Module kvm_amd fails to load on module ccp Attempting to rmmod -f ccp provides a Resource busy message. kvm_amd loaded correctly under 4.16.7 kernel # modprobe -v kvm_amd Description of Hardware: AMD Ryzen 1950X Gigabyte x399 motherboard (April 2018 BIOS) SVM and SMT enabled (confirmed that these show enabled in lscpu and 32 CPU's are visible) Version-Release number of selected component (if applicable): Linux localhost.localdomain 4.17.11-200.fc28.x86_64 #1 SMP Mon Jul 30 15:21:09 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux How reproducible: Steps to Reproduce: 1. Use 1950X AMD system and X399 motherboard (April/2018 2. Upgrade to kernel 4.17.11-200 3. Reboot the system 4. Attempt to use any KVM function (errors) 5. Check that kvm_amd loaded (it does not) 6. Attempt to load kvm_amd module (fails by hanging on loading ccp module) Actual results: virsh start vm_1 fails because /dev/kvm does not exist kvm_amd fails to load on cpp module [root] modprobe -v kvm_amd insmod /lib/modules/4.17.11-200.fc28.x86_64/kernel/virt/lib/irqbypass.ko.xz insmod /lib/modules/4.17.11-200.fc28.x86_64/kernel/arch/x86/kvm/kvm.ko.xz insmod /lib/modules/4.17.11-200.fc28.x86_64/kernel/kernel/drivers/crypto/ccp/ccp.ko.xz <hangs here until timeout occurs> Expected results: Kernel boots and system loads correctly, with kvm loading (as it was prior to the upgrade) Additional info: Module was loading correctly under previous kernel 4.16.7.
Seems very similar to this thread: https://www.reddit.com/r/VFIO/comments/8yux5r/error_loading_kvmamd/
Also tested with previous 4.17.3 kernel - same results with ccp hanging on loading with kvm_amd module.
This bug still occurs on 4.17.12-200.fc28.x86_64 packages - tested this morning.
Realized I put in the wrong component. Changing kernel to qemu.
I've hunted down the issue: I upgraded to a BIOS version with AGESA 1.0.0.5 When I dropped back to the previous BIOS level using AGESA 1.0.0.3 Patch 4, everything started working again and ccp and kvm_amd loading correctly.
Disabling CONFIG_CRYPTO_DEV_SP_PSP in the kernel 4.17 configuration (as suggested in that Reddit thread above) and recompiling fixes the hangup. - Gigabyte X399 Designare EX with F10 bios (AGESA 1.0.0.6 / 1.1.0.0) - Threadripper 1900X - kernel 4.17.12 Kernel 4.15.18 works fine.
kvm module is kernel territory, changing component back Alex, I see you commented in the reddit thread, positing this is a kernel misconfiguration... is this something that can/should be fixed in fedora kernel config or is this something else?
Same problem with gigabyte x470 board / ryzen 5 1400 after bios update. AGESA 1.0.0.2c --> AGESA 1.0.0.4 [root@mother rave]# uname -r 4.17.18-200.fc28.x86_64 Beside from that the kvm_amd module doesnt load, shutting down the box take more than 5 min. now. Systemd-shutdown hangs but i don't see any useful infos in logs. Resetting the bios back to AGESA 1.0.0.2c solved all problems.
I just rebuild current f28 kernel with CONFIG_CRYPTO_DEV_SP_PSP disabled, and it fixes the ugly shutdown issues but not the broken kvm-amd modules load. I added this content to kernel-local file in fedora git repo. [rave@mother ~]$ cat kernel/kernel-local # This file is intentionally left empty in the stock kernel. Its a nicety # added for those wanting to do custom rebuilds with altered config opts. # CONFIG_CRYPTO_DEV_SP_PSP is not set The resulting scratch build is here https://koji.fedoraproject.org/koji/taskinfo?taskID=29425990 Did i something wrong? I am not a specialist with rebuilding a kernel. Would be nice if someone from kernel team can take a look at it. Amd ryzen is i a very nice and common proc, and AMD did an amazing job with the AGESA 1.0.0.4 bios update, eg. speed up booting and memory stability. I am happy to provide more infos if needed. Thank you.
Ok, i did a mistake during testing. Latest fedora kernel rebuild with # CONFIG_CRYPTO_DEV_SP_PSP is not set fixes the kvm-modul load and system shutdown issues. https://koji.fedoraproject.org/koji/taskinfo?taskID=29425990 But i don't wanna rebuild the kernel every week in fedora ;)
I tested a bit more and the issue was introduced during early kernel-4.16.0-rc1 developing. Last good kernel is: kernel-4.16.0-0.rc0.git9.1.fc28 First bad kernel is: kernel-4.16.0-0.rc1.git0.1.fc28 In other words, first bad commit is https://src.fedoraproject.org/cgit/rpms/kernel.git/commit/?h=f28&id=94e04cd684a20d47e9529936b8beea179faaee81 commit limited to x86_64 https://src.fedoraproject.org/cgit/rpms/kernel.git/diff/kernel-x86_64.config?h=f28&id=94e04cd684a20d47e9529936b8beea179faaee81 diff --git a/kernel-x86_64.config b/kernel-x86_64.config index ffb2bbe..bf1c130 100644 --- a/kernel-x86_64.config +++ b/kernel-x86_64.config @@ -519,6 +519,7 @@ CONFIG_BOOT_PRINTK_DELAY=y CONFIG_BOUNCE=y CONFIG_BPF_JIT_ALWAYS_ON=y CONFIG_BPF_JIT=y +# CONFIG_BPF_KPROBE_OVERRIDE is not set CONFIG_BPF_STREAM_PARSER=y CONFIG_BPF_SYSCALL=y CONFIG_BPQETHER=m @@ -909,6 +910,7 @@ CONFIG_CRYPTO_DEV_QAT_C62XVF=m CONFIG_CRYPTO_DEV_QAT_DH895xCC=m CONFIG_CRYPTO_DEV_QAT_DH895xCCVF=m # CONFIG_CRYPTO_DEV_SP_CCP is not set +CONFIG_CRYPTO_DEV_SP_PSP=y CONFIG_CRYPTO_DEV_VIRTIO=m CONFIG_CRYPTO_DH=y CONFIG_CRYPTO_DRBG_CTR=y @@ -2642,6 +2644,7 @@ CONFIG_KS0108_PORT=0x378 CONFIG_KSM=y CONFIG_KSZ884X_PCI=m CONFIG_KVM_AMD=m +CONFIG_KVM_AMD_SEV=y # CONFIG_KVM_DEBUG_FS is not set CONFIG_KVM_DEVICE_ASSIGNMENT=y CONFIG_KVM_GUEST=y I don't know if there is a related change in patch-4.16-rc1.xz I hope this will help kernel developers to fix this annoying issue.
*** This bug has been marked as a duplicate of bug 1608242 ***
The needinfo request[s] on this closed bug have been removed as they have been unresolved for 1000 days