On 24/07/2026 09:48, Peter Maydell wrote:
On Tue, 21 Jul 2026 at 22:01, Mark Cave-Ayland
<[email protected]> wrote:
On 18/07/2026 09:46, Peter Maydell wrote:
Are you OK with this patch as it stands, or would you like me to
change it?
Sorry about the long delay getting to this one. Looking through the migration
code,
the current approach uses get/put handlers for registers that need to do more
than
just update value in the state. Longer term that code could do with an update
to use
.post_load, but for now I think it makes sense to use the existing pattern,
particularly as it keeps the separation between 32-bit and 64-bit.
The following diff gives the same output in analyze-migration.py and I think
should
solve the issue based upon your analysis:
Yes, that should also work. There's a minor bug in it, though:
+static int put_cwp(QEMUFile *f, void *opaque, size_t size,
+ const VMStateField *field, JSONWriter *vmdesc)
+{
+ SPARCCPU *cpu = opaque;
+ CPUSPARCState *env = &cpu->env;
+ uint32_t val = cpu_get_cwp64(env);
This should be "val = env->cwp;" (cpu_get_cwp64() returns
"env->nwindows - 1 - env->cwp", but we just want the plain env->cwp
value because that's what we used to migrate and what the get_cwp()
function wants to see.)
With that change, the sanitizer is happy with the migration-test.
Do you plan to send your version as a proper patch?
Thanks for confirming. I'll tidy it up and send it through as a proper patch
later.
ATB,
Mark.
