[PATCH 4/9] Revert "x86: Switch QEMU over to use the bochs driver"

[Simon Glass]

Hi Bin,

On Mon, 31 Jul 2023 at 11:07, Simon Glass <sjg at chromium.org> wrote:
>
> Hi Bin,
>
> On Mon, 31 Jul 2023 at 08:46, Bin Meng <bmeng.cn at gmail.com> wrote:
> >
> > Hi Simon,
> >
> > On Mon, Jul 31, 2023 at 10:37 PM Simon Glass <sjg at chromium.org> wrote:
> > >
> > > Hi Bin,
> > >
> > > On Mon, 31 Jul 2023 at 08:32, Bin Meng <bmeng.cn at gmail.com> wrote:
> > > >
> > > > Hi Simon,
> > > >
> > > > On Sat, Jul 29, 2023 at 1:32 AM Simon Glass <sjg at chromium.org> wrote:
> > > > >
> > > > > Hi Bin,
> > > > >
> > > > > On Fri, 28 Jul 2023 at 09:46, Bin Meng <bmeng.cn at gmail.com> wrote:
> > > > > >
> > > > > > Hi Simon,
> > > > > >
> > > > > > On Mon, Jul 24, 2023 at 10:52 PM Simon Glass <sjg at chromium.org> wrote:
> > > > > > >
> > > > > > > Unfortunately the bochs driver does not currently work with distros. It
> > > > > > > causes a hang sometime between grub menu selection and the OS displaying
> > > > > > > something.
> > > > > >
> > > > > > Does this reproduce reliably?
> > > > >
> > > > > Yes, it does.
> > > > >
> > > > > BTW I've also hit another problem with '-M accel-kvm' which hangs on
> > > > > the move to 64-bit mode. Oddly if I boot U-Boot from coreboot, the
> > > > > problem goes away. So perhaps coreboot is doing some x86 init which
> > > > > U-Boot is missing.
> > > >
> > > > This indicates a kvm-x86 bug. I think you can simplify a test case and
> > > > report it to the KVM mailing list.
> > >
> > > Firstly it doesn't really matter, since it doesn't work, so we do need
> > > to revert.
> >
> > Agreed, but before we revert this, I would like to see we investigate
> > this issue a little bit further.
> >
> > I enabled the grub debug output today and not sure if grub jumps to
> > the kernel entry yet:
> >
> > kern/verifiers.c:214: string: boot, type: 2
> > loader/efi/linux_sb.c:55: kernel_addr: 0x1000000 handover_offset: 0x190 params:
> > 0x7ce63000
> >
> > >
> > > Secondly I am not sure it kvm's fault, since the problem does not
> > > happen with coreboot. I will see if I can isolate what coreboot is
> > > doing differently, but not sure that that fix will make the release.
> >
> > Does the 64-bit mode switch ever work on a real x86 board? If yes,
> > then it's very likely a KVM bug.
>
> Yes it works fine on chromebook_link64, for example.
>
> >
> > If not, then we can check the coreboot to find out the difference.
>
> Yes...do you have any ideas?
>
> If it helps, I am using something like this:
>
>  qemu-system-x86_64 -m 8G -smp 4  -bios /tmp/b/qemu-x86_64/u-boot.rom
> -drive id=fdisk,file=root.img,if=virtio,driver=raw  -drive \
>     file=/vid/software/linux/ubuntu/ubuntu-23.04-desktop-amd64.iso,if=virtio,driver=raw
>  \
>     -serial mon:stdio
>
> For coreboot my script is:
>
> UBTEST=/vid/software/devel/ubtest/
> BUILD_DIR=/tmp/b/coreboot64
> DISK=/vid/software/linux/ubuntu/ubuntu-23.04-desktop-amd64.iso
>
> cp $UBTEST/coreboot.rom.in $BUILD_DIR/coreboot.rom
> cbfstool $BUILD_DIR/coreboot.rom add-flat-binary \
>         -f $BUILD_DIR/u-boot-x86-with-spl.bin \
>         -n fallback/payload -c LZMA -l 0x1110000 -e 0x1110000
>
> Add this in below:
> #-M accel=kvm
>
> qemu-system-x86_64 -m 2G -smp 4 -bios $BUILD_DIR/coreboot.rom -serial
> mon:stdio \
>         -drive id=disk,file=$DISK,if=none \
>         -device ahci,id=ahci \
>         -device ide-hd,drive=disk,bus=ahci.0

I haven't looked into the problem here, but I did notice this coreboot
commit: https://review.coreboot.org/c/coreboot/+/49228

I am not sure it is related though, since we already have page tables in RAM.

Regards,
Simon