[RESOLVED] aarch64 edk2 + linux kernel crash with gicv4

[RESOLVED] aarch64 edk2 + linux kernel crash with gicv4

[Richard Henderson]

Hi Luis,

Cc'ing qemu-devel just in case someone else runs into this problem:

The symptom is an early-ish crash booting the linux kernel -- without earlycon there is no 
output at all.  Reproducing it requires both -bios QEMU_EFI.fd and -kernel Image.gz. 
Using only one of the two options works.

I tracked the crash to the qemu commit that enabled GICv4.  So I tried using -M 
gic-version=3, and that works.

I wondered if the problem is that the edk2 image from Ubuntu 20.04 is too old and can't 
properly handle GICv4, leaving the device in an inconsistent state for the kernel.  So I 
tried an image from https://retrage.github.io/edk2-nightly/ with gic-version=max, and that 
works.

So... solved.


r~

Re: [RESOLVED] aarch64 edk2 + linux kernel crash with gicv4

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> Hi Luis,
>
> Cc'ing qemu-devel just in case someone else runs into this problem:
>
> The symptom is an early-ish crash booting the linux kernel -- without
> earlycon there is no output at all.  Reproducing it requires both
> -bios QEMU_EFI.fd and -kernel Image.gz. Using only one of the two
> options works.

What should the combination of the two options mean? Usually the
firmware loads the kernel and -kernel is a shortcut hack to avoid the
firmware setup. If you specify both does the firmware still run but
somehow find the kernel mapped in memory?

>
> I tracked the crash to the qemu commit that enabled GICv4.  So I tried
> using -M gic-version=3, and that works.
>
> I wondered if the problem is that the edk2 image from Ubuntu 20.04 is
> too old and can't properly handle GICv4, leaving the device in an
> inconsistent state for the kernel.  So I tried an image from
> https://retrage.github.io/edk2-nightly/ with gic-version=max, and that
> works.
>
> So... solved.
>
>
> r~


-- 
Alex Bennée

Re: [RESOLVED] aarch64 edk2 + linux kernel crash with gicv4

[Peter Maydell]

On Wed, 22 Jun 2022 at 06:07, Alex Bennée <alex.bennee@linaro.org> wrote:
> Richard Henderson <richard.henderson@linaro.org> writes:
> > The symptom is an early-ish crash booting the linux kernel -- without
> > earlycon there is no output at all.  Reproducing it requires both
> > -bios QEMU_EFI.fd and -kernel Image.gz. Using only one of the two
> > options works.
>
> What should the combination of the two options mean? Usually the
> firmware loads the kernel and -kernel is a shortcut hack to avoid the
> firmware setup. If you specify both does the firmware still run but
> somehow find the kernel mapped in memory?

It's like x86 -- it means "load the BIOS image, and provide the
kernel/initrd via the fw_cfg device, so that a QEMU-aware BIOS
can read the supplied kernel/initrd from fw_cfg and boot them".

thanks
-- PMM