Re: [PATCH v3 00/11] tcg/riscv64: Fix AUIPC pair range validation

[Max Chou <max.chou@sifive.com>]

On 2026-09-27 10:30, Richard Henderson wrote:
> Ping.
> 
> On 9/13/26 17:08, Richard Henderson wrote:
> > Hi Max,
> > 
> > In v2, the change from int32_t to uint32_t is pretty much
> > the opposite of what should have been done, since all of
> > these are logically signed values.  I've adjusted all of
> > the encoders to take signed inputs and assert in range.
> > 
> > In the process, I noticed tha the branch range tests were
> > off by one.  Now all performed in exactly one place.
> > 
> > Finally, tcg_out_ldst needed updates for AUIPC as well.
> > 
> > I've tested this lightly so far; full testing on a rather
> > slow riscv machine is still on-going.
> > 
> > 
> > r~
> > 

Hi Richard,

I’ve discovered a verification scenario that could trigger an unexpected
segfault due to the AUIPC issue.

The scenario involves running a debug cross-build (riscv64) of
qemu-system-riscv64 with a plugin that instruments callback (e.g.,
libhotblocks.so) in a host riscv machine. (I used qemu-system-riscv64
with the same images here :p)

Before applying this patchset, the internal qemu-system-riscv64
encountered an unexpected segfault during bootup.

(P.S.: I built the riscv64 system mode image using buildroot
qemu_riscv64_virt_defconfig.)

Thanks,
rnax