summary refs log tree commit diff stats
path: root/results/classifier/105/graphic/2361
diff options
context:
space:
mode:
Diffstat (limited to 'results/classifier/105/graphic/2361')
-rw-r--r--results/classifier/105/graphic/236126
1 files changed, 26 insertions, 0 deletions
diff --git a/results/classifier/105/graphic/2361 b/results/classifier/105/graphic/2361
new file mode 100644
index 000000000..80a2aef6c
--- /dev/null
+++ b/results/classifier/105/graphic/2361
@@ -0,0 +1,26 @@
+graphic: 0.987
+instruction: 0.887
+device: 0.877
+boot: 0.820
+semantic: 0.733
+mistranslation: 0.544
+assembly: 0.465
+vnc: 0.418
+other: 0.382
+network: 0.152
+socket: 0.129
+KVM: 0.030
+
+-cpu host or -cpu max breaks GRUB on AMD
+Description of problem:
+I'm running the on an AMD Ryzen CPU host.  I am emulating a Debian Bookworm image stored in a raw disk.  It uses GRUB to load a large (400MB) initrd.  When ran with the flag -cpu host or -cpu max, GRUB throws an out of memory error while loading the initrd.  This doesn't occur when using -cpu kvm64 or excluding the -cpu flag.
+
+If I direct boot the initrd and kernel via -initrd and -kernel, it works fine.  The image also works with -cpu host on an Intel CPU host machine.  The image also works with -cpu EPYC.
+Steps to reproduce:
+1. Create a raw disk with a large initrd and GRUB boot loader
+2. Start a qemu machine on an AMD host 
+3. Receive an error: out of memory
+Additional information:
+I could try selectively enabling CPU features, but I was wondering if the maintainers knew of any feature that might be causing this or how to list the features -cpu host enables.
+
+I also am not 100% that this is a QEMU bug, but it seems the only way to fix it is changing the QEMU config.