On Wed, Jan 22, 2025 at 8:56 AM Jim Mattson <jmattson@xxxxxxxxxx> wrote: > > On Wed, Jan 22, 2025 at 1:01 AM Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote: > > > > Il mer 22 gen 2025, 07:07 John Stultz <jstultz@xxxxxxxxxx> ha scritto: > > > > > > I then cut down and ported the bionic test out so it could build under > > > a standard debian environment: > > > https://github.com/johnstultz-work/bionic-ptrace-reproducer > > > > > > Where I was able to reproduce the same problem in a debian VM (after > > > running the test in a loop for a short while). > > > > > > Thanks, that's nice to have. > > > > > Now, here's where it is odd. I could *not* reproduce the problem on > > > bare metal hardware, *nor* could I reproduce the problem in a virtual > > > environment. I can *only* reproduce the problem with nested > > > virtualization (running the VM inside a VM). > > > > Typically in that case the best thing to do is turn it into a > > kvm-unit-test or selftest (though that's often an endeavor of its own, > > as it requires distilling the Linux kernel and userspace code into a > > guest that runs without an OS). But what you've done is already a good > > step. > > Just run the kvm-unit-tests 'x86/debug' test in a loop inside an L1 > VM. It will eventually fail. Maybe not the same bug, but we can hope. > :) Thanks Jim, I've just reproduced this as well, after running the debug test in a loop. The one odd bit is that it's not always the same subtest that fails. I've seen: FAIL: Single-step #DB w/ STI blocking FAIL: Usermode Single-step #DB w/ STI blocking FAIL: Single-step #DB on emulated instructions FAIL: Single-step #DB w/ MOVSS blocking FAIL: Usermode Single-step #DB basic test FAIL: Single-step #DB w/ MOVSS blocking and DR7.GD=1 FAIL: hw breakpoint (test that dr6.BS is not cleared) So yeah, hopefully all the same bug? :) thanks -john
