Re: [GIT PULL] KVM changes for Linux 5.6-rc4

From: Paolo Bonzini
Date: Tue Mar 03 2020 - 05:58:34 EST


On 02/03/20 19:40, Vitaly Kuznetsov wrote:
>
> qemu-system-x86-23579 [005] 22018.775584: kvm_exit: reason EPT_VIOLATION rip 0xfffff802987d6169 info 181 0
> qemu-system-x86-23579 [005] 22018.775584: kvm_nested_vmexit: rip fffff802987d6169 reason EPT_VIOLATION info1 181 info2 0 int_info 0 int_info_err 0
> qemu-system-x86-23579 [005] 22018.775585: kvm_page_fault: address febd0000 error_code 181
> qemu-system-x86-23579 [005] 22018.775592: kvm_emulate_insn: 0:fffff802987d6169: f3 a5
> qemu-system-x86-23579 [005] 22018.775593: kvm_emulate_insn: 0:fffff802987d6169: f3 a5 FAIL
> qemu-system-x86-23579 [005] 22018.775596: kvm_inj_exception: #UD (0x0)
>
> We probably need to re-enable instruction emulation for something...

This is a rep movsw instruction, it shouldn't be intercepted. I think
we have a stale ctxt->intercept because the

/* Fields above regs are cleared together. */

comment is not true anymore since

commit c44b4c6ab80eef3a9c52c7b3f0c632942e6489aa
Author: Bandan Das <bsd@xxxxxxxxxx>
Date: Wed Apr 16 12:46:12 2014 -0400

KVM: emulate: clean up initializations in init_decode_cache

A lot of initializations are unnecessary as they get set to
appropriate values before actually being used. Optimize
placement of fields in x86_emulate_ctxt

Signed-off-by: Bandan Das <bsd@xxxxxxxxxx>
Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>

Paolo