On Thu, Jul 24, 2025 at 10:36:57AM -0700, Eric Biggers wrote: > > > > Configuration file and decoded stacktrace are at > > > > http://server.roeck-us.net/qemu/crypto/ > > > > Please let me know if you need anything else. > > Thanks! It seems the kconfig is actually unrelated; what matters is how > the kernel is being run. It reproduces only in QEMU with KVM disabled. > It doesn't reproduce with KVM enabled or on bare metal. > > The crashed instruction is 'vinserti128 $0x1,0x30(%rsi),%ymm8,%ymm8'. > > I think it's a QEMU bug: its implementation of vinserti128 reads 32 > bytes from memory instead of 16. I'll report it to qemu-devel once I > have a bit more information. > > It will be the second QEMU bug found by the crypto tests this month.... > https://lore.kernel.org/r/20250710052824.GA608727@sol/ was the first. > Oh well :-(. Thanks for tracking it down. I'll disable this test for the time being for x86 tests. I'll let you know if I see it or something similar for other architectures. Thanks, Guenter
