Re: [PATCH v4 1/4] u-boot: disable CONFIG_BLOBLIST on genericarm64 and qemuarm64

[Mikko Rapeli <mikko.rapeli@linaro.org>]

Hi,

On Wed, May 28, 2025 at 08:18:08AM -0600, Tom Rini wrote:
> On Wed, May 28, 2025 at 09:40:59AM +0300, Mikko Rapeli wrote:
> > On Tue, May 27, 2025 at 06:51:33PM -0600, Tom Rini wrote:
> > > On Mon, May 26, 2025 at 11:35:44AM +0300, Mikko Rapeli wrote:
> > > 
> > > > Booting u-boot on qemu with kvm is currently hanging on aarch64
> > > > build host. Root cause is in u-boot and CONFIG_BLOBLIST can be
> > > > disabled as a workaround.
> > > > 
> > > > To reproduce, build on kvm enabled host where "kvm-ok"
> > > > succeeds. For example genericarm64 machine and core-image-base
> > > > should then boot with:
> > > > 
> > > > $ runqemu slirp nographic novga snapshot kvm
> > > > 
> > > > On qemuarm64, default kvm setup will boot directly to kernel
> > > > and is not affected by this. If build enables u-boot as bios
> > > > then the same issue happens.
> > > > 
> > > > Without this config workaround, the boot hangs without
> > > > any messages in qemu output but ctrl-a-c to qemu console
> > > > can shutdown the emulated machine.
> > > > 
> > > > This seems to have regressed after u-boot 2025.04 update.
> > > > KVM boot can be detected from speed, for example genericarm64
> > > > boots in 550 ms with KVM and without in over 5 seconds.
> > > > 
> > > > Fixes: [YOCTO #15872]
> > > > 
> > > > Upstream u-boot discussion:
> > > > https://lists.denx.de/pipermail/u-boot/2025-May/590101.html
> > > > 
> > > > Cc: Ilias Apalodimas <ilias.apalodimas@linaro.org>
> > > > Signed-off-by: Mikko Rapeli <mikko.rapeli@linaro.org>
> > > 
> > > I think it's a tad early to disable this. On the U-Boot side, it's being
> > > looked in to. In fact, you're saying v2025.04 is working (current
> > > release) and v2025.07-rcX is failing. At this point in the U-Boot cycle
> > > I think we'll figure out the problem and fix it. If we can't figure out
> > > how to keep all the use cases working for v2025.07 then disabling
> > > BLOBLIST here in OE can be thought on, I would expect. Thanks.
> > 
> > 2025.04 is affected by this. My wording for this is not accurate enough, sorry.
> 
> Do you have a known good version where it does work? I assume it's
> v2025.01 that does work, and an OE-specific revert of commit
> 53d5a221632e ("emulation: Use bloblist to hold tables") with appropriate
> status tags about it being discussed upstream would make the most sense.

As you suspected and Ilias bisected this to:

53d5a221632eeef7483d250fdde09bde6cb54df9 is the first bad commit
commit 53d5a221632eeef7483d250fdde09bde6cb54df9
Author: Simon Glass <sjg@chromium.org>
Date:   Fri Jan 10 17:00:17 2025 -0700
    
    emulation: Use bloblist to hold tables
    
    QEMU can have its own internal ACPI and SMBIOS tables. At present U-Boot
    copies out the SMBIOS tables but points directly to the ACPI ones.

    The ACPI tables are not aligned on a 4KB boundary, which means that UPL
    cannot use them directly, since it uses a reserved-memory node for the
    tables and that it assumed (by EDK2) to be 4KB-aligned.

    On x86, QEMU provides the tables in a mapped memory region and U-Boot
    makes use of these directly, thus making it difficult to use any common
    code.
    
    Adjust the logic to fit within the existing table-generation code. Use a
    bloblist always and ensure that the ACPI tables is placed in an aligned
    region. Set a size of 8K for QEMU. This does not actually put all the
    tables in one place, for QEMU, since it currently adds a pointer to the
    tables in QFW.
    
    On ARM, enable bloblist so that SMBIOS tables can be added to the
    bloblist.

    Signed-off-by: Simon Glass <sjg@chromium.org>

So tags v2025.01-rc4 and earlier are fine and v2025.04-rc1 and newer
are affected.

Reverting this is a bit hard due to other changes so for oe-core it's
easier to disable CONFIG_BLOBLIST where possible, on qemuarm64 and
genericarm64 (uses same qemu defconfig) yocto target machines for
example, which for testing purposes can be used with qemu and KVM.

Cheers,

-Mikko