On 10/14/25 23:07, Dmitry Osipenko wrote: > On 10/14/25 23:01, Dmitry Osipenko wrote: >> On 10/14/25 19:23, ~myrslint wrote: >>> From: myrslint <[email protected]> >>> >>> On x86_64, where most CPUs support self-snoop, it is preferrable to >>> always honor guest PAT. Not doing so is a quirk. There is a default >>> enabled KVM quirk flag which enforces not doing so due to a former bug >>> in Bochs display driver. >>> >>> The bug has been fixed but not enough has yet passed since so we only >>> disable said quirk flag if a Bochs display is not configured for the >>> virtual machine. >>> >>> This commit also moves around a bit of code that would be called when >>> the initialization of a VM is done. >>> >>> Signed-off-by: myrslint <[email protected]> >>> --- >>> accel/kvm/kvm-all.c | 1 + >>> accel/stubs/kvm-stub.c | 1 + >>> hw/display/bochs-display.c | 17 +++++++++++++++ >>> include/system/kvm.h | 9 ++++++++ >>> target/i386/kvm/kvm.c | 42 +++++++++++++++++++++++++++++++------- >>> 5 files changed, 63 insertions(+), 7 deletions(-) >> >> Looks perfect. Now up to Paolo to review further and apply the patch. >> >> Reviewed-by: Dmitry Osipenko <[email protected]> > > Ah, the commit title still isn't ideal. It should be prefixed with > "i386/kvm:" like other kvm commits are in QEMU. Likely Paolo could > correct it while applying if v9 won't be needed.
Tested on older Intel IVB machine where virtio-gpu hostmem doesn't work without this patch. Please add my r-b and t-b, specify your full name "Myrsky Lintu" in the Signed-off-by of the commit msg, correct the patch title prefix and send the v9. Tested-by: Dmitry Osipenko <[email protected]> # intel-ivb -- Best regards, Dmitry
