Re: [PATCH v3 10/13] target/hexagon: add COREDUMP semihosting operation

On Mon, 17 Aug 2026 06:34:54 +0200 =?UTF-8?Q?Philippe_Mathieu-Daud=C3=A9?= <[email protected]> wrote:
>
> On 20/7/26 19:41, Matheus Tavares Bernardino wrote:
> > Baremetal Hexagon programs invoke SYS_COREDUMP (0xCD)
> > through semihosting to dump CPU state on a fatal exception.
> > Implement the handler to decode the SSR cause field and
> > print the full register file, matching hexagon-sim behavior.
> > 
> > Signed-off-by: Brian Cain <[email protected]>
> > Reviewed-by: Pierrick Bouvier <[email protected]>
> > Signed-off-by: Matheus Tavares Bernardino <[email protected]>
> > ---
> >   target/hexagon/internal.h |   1 +
> >   target/hexagon/cpu.c      |   2 +-
> >   target/hexagon/hexswi.c   | 143 ++++++++++++++++++++++++++++++++++++++
> >   3 files changed, 145 insertions(+), 1 deletion(-)
> 
> 
> > diff --git a/target/hexagon/cpu.c b/target/hexagon/cpu.c
> > index e6b2ac1939d..06c378c5e02 100644
> > --- a/target/hexagon/cpu.c
> > +++ b/target/hexagon/cpu.c
> > @@ -232,7 +232,7 @@ void hexagon_debug_qreg(CPUHexagonState *env, int regnum)
> >       print_qreg(stdout, env, regnum, false);
> >   }
> >   
> > -static void hexagon_dump(CPUHexagonState *env, FILE *f, int flags)
> > +void hexagon_dump(CPUHexagonState *env, FILE *f, int flags)
> >   {
> >       HexagonCPU *cpu = env_archcpu(env);
> >   
> > diff --git a/target/hexagon/hexswi.c b/target/hexagon/hexswi.c
> > index 6efc00fedf9..564a11e557a 100644
> > --- a/target/hexagon/hexswi.c
> > +++ b/target/hexagon/hexswi.c
> > @@ -162,6 +162,145 @@ static void common_semi_ftell_cb(CPUState *cs, uint64_t ret, int err)
> >       common_semi_cb(cs, ret, err);
> >   }
> >   
> > +static void coredump(CPUHexagonState *env)
> > +{
> > +    uint32_t ssr = arch_get_system_reg(env, HEX_SREG_SSR);
> > +    FILE *f = qemu_log_trylock();
> > +
> > +    if (!f) {
> > +        return;
> > +    }
> > +
> > +    fprintf(f, "CRASH!\n");
> > +    fprintf(f, "I think the exception was: ");
> > +    switch (GET_SSR_FIELD(SSR_CAUSE, ssr)) {
> > +    case 0x43:
> > +        fprintf(f, "0x43, NMI");
> > +        break;
> > +    case 0x42:
> > +        fprintf(f, "0x42, Data abort");
> > +        break;
> > +    case 0x44:
> > +        fprintf(f, "0x44, Multi TLB match");
> > +        break;
> > +    case HEX_CAUSE_BIU_PRECISE:
> > +        fprintf(f, "0x%x, Bus Error (Precise BIU error)",
> > +                HEX_CAUSE_BIU_PRECISE);
> > +        break;
> [...]
> 
> Should this be moved within hexagon_dump() instead? That information
> seems available and useful there.

Hmm, we could extract this out of coredump() if there are other users, but I don't
think it would be useful at hexagon_dump()... This info is only meaningful right
after an exception. In a generic cpu_dump_state() call, SSR could be 0 or even
hold the value of the last exception, so "CRASH, I think the exception was..."
would be misleading there.