On 11/12/2015 12:41, Borislav Petkov wrote: > On Fri, Dec 11, 2015 at 11:41:30AM +0100, Paolo Bonzini wrote: >> It would be a kvm hypervisor page, not a kvm guest page, hence unrelated >> to the zapping thing. > > Ah right, guest pages should be userspace addresses, come to think of > it. > >> Can you grab the kallsyms before making it crash? > > Attached. It was a different corruption this time, see below. This time > we don't even have a page table, PGD is 0, rIP is 1. (Fun :-)) Hmm, you had: - RIP=0 in the original report (start_this_handle) - RIP=0 in the second (mutex_lock_nested in ext4) - RIP=1 now The more interesting one is the other one which doesn't have a small RIP, because it has RIP that is slightly larger than the stack pointer, meaning it's likely a frame pointer. And this means in turn that the call trace is correct, and the bug might have happened closer to the actual corruption. [ 959.548625] RIP: 0010:[<ffff8800b9f9bdf0>] [<ffff8800b9f9bdf0>] 0xffff8800b9f9bdf0 [ 959.556338] RSP: 0018:ffff8800b9f9bde0 EFLAGS: 00010206 [ 959.618579] Stack: [ 959.620607] ffffffffa02d5e17 ffff8800b7d48000 ffff8800b9f9be08 ffffffffa02bdb1f [ 959.628104] 0000000000000000 ffff8800b9f9be98 ffffffffa02bdc7b ffff8804242a4400 [ 959.635601] 0000000000000070 0000000000004000 ffffffff81a3c1e0 ffff8800b7ca5e00 [ 959.643114] Call Trace: [ 959.645599] [<ffffffffa02d5e17>] ? kvm_arch_vcpu_put+0x17/0x40 [kvm] [ 959.652081] [<ffffffffa02bdb1f>] ? vcpu_put+0x1f/0x60 [kvm] [ 959.657782] [<ffffffffa02bdc7b>] ? kvm_vcpu_ioctl+0x11b/0x6f0 [kvm] [ 959.664169] [<ffffffff811a0930>] ? do_vfs_ioctl+0x2e0/0x540 [ 959.669855] [<ffffffff811ac8e9>] ? __fget_light+0x29/0x90 [ 959.675364] [<ffffffff811a0bdc>] ? SyS_ioctl+0x4c/0x90 [ 959.680618] [<ffffffff816e2d5b>] ? entry_SYSCALL_64_fastpath+0x16/0x6f My wild guess is that RSP is getting corrupted, but I guess I'll have to try to reproduce to figure out what happens. The last thing I need from you (hopefully) is a Kconfig. If you have some time, it would be great to check if you can reproduce it with an older kernel version---trying 4.4-rc1 and 4.3 would be great. Paolo -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
