Re: [PATCH] include: sbi_platform: Increase default stack size

On Thu, Jun 4, 2026 at 4:43 PM Aurelien Jarno <[email protected]> wrote:
>
> Hi Anirudh,
>
> On 2026-06-04 15:09, Anirudh Srinivasan wrote:
> > Hi Aurelien,
> >
> > On Tue, Jun 2, 2026 at 2:12 PM Aurelien Jarno <[email protected]> wrote:
> > >
> > > On 2026-05-24 13:53, Vivian Wang wrote:
> > > > sbi_misaligned_v_ld_emulator() has a local buffer (namely mask) which is
> > > > 8192 bytes in size, so the default stack size is not enough. Double it
> > > > to 16384.
> > > >
> > > > This stack overflow can be observed by any S-mode software and leads to
> > > > M-mode crash or unexpected behavior, as long as the S-mode software does
> > > > not enable FWFT misaligned delegation [1]. This may be considered a
> > > > temporary fix until further fixes are made to misaligned handling [2].
> > > >
> > > > Link: https://lore.kernel.org/linux-riscv/nrvt74qnojaubiwjo37ums4lnclu466hovwrhmtbag6f5uhrql@q6msoe2oto4b # [1]
> > > > Link: https://lore.kernel.org/opensbi/[email protected] # [2]
> > > > Signed-off-by: Vivian Wang <[email protected]>
> > > > ---
> > > >  include/sbi/sbi_platform.h | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > Tested-by: Aurelien Jarno <[email protected]>
> > >
> > > I can confirm that this fix is required to boot a 7.1-rc6 kernel on a
> > > CPU with vector instructions (in my case SpacemiT K1).
> >
> > Are you sure that this is needed for the K1? It seemed like this was
> > only an issue on boards with large vlen (512 like the x280)?
>
> Yes this is an issue. The stack allocation doesn't depend on the actual
> vector size, but on VLEN_MAX which is 65536. The alternative to increase
> the stack size is to reduce this value.

Thanks for clarifying. I must have misunderstood this.

>
> > Do you
> > mind sharing any kernel boot logs with/without this patch?
>
> Yep, please find them attached. Note that I have attached the useful
> failure, sometimes the kernel boots correctly and most of the time it
> just hangs.
>
> > I was able to boot 7.1.0-rc6-next-20260603 defconfig without issue on
> > a MilkV Jupiter. Only caveat is that my board is  running an old
> > vendor opensbi (version k1-bl-v2.2.4 that may be based on upstream
> > v1.3) that the board shipped with, and not sure if that's
> > helping/hurting in any manner.
>
> I am using the current OpenSBI HEAD. Support for misaligned vector
> load/store was added in version 1.6 in commit c2acc5e5b0d8 ("lib:
> sbi_misaligned_ldst: Add handling of vector load/store"), so it's not
> surprising that your version does not have the issue.

I wasn't aware that you could use upstream OpenSBI on the k1 boards.

I imagine that a lot of folks with older boards aren't gonna upgrade
OpenSBI, so won't run into this issue. Still useful to have a fix
though.


>
> Regards
> Aurelien
>
> --
> Aurelien Jarno                          GPG: 4096R/1DDD8C9B
> [email protected]                     http://aurel32.net

-- 
opensbi mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/opensbi