On 04.10.19 14:11, Peter Maydell wrote:
> On Fri, 4 Oct 2019 at 09:02, David Hildenbrand <da...@redhat.com> wrote:
>> So shall we leave this patch as-is (adding a summary of what you
>> explained to the description) or shall we somehow factor out the
>> TCG-internal-thingy check?
>
> Nothing else in target code touches the icount data structures,
> so if this s390 insn needs to make this check I think it ought
> to do it by calling a function implemented by the tcg code;
> that can then have a good name that describes what it's doing
> and a doc comment explaining the reason we need to have it.
>
> thanks
> -- PMM
>
I can offer something like this:
diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h
index 49db07ba0b..d370ac0134 100644
--- a/include/exec/exec-all.h
+++ b/include/exec/exec-all.h
@@ -72,6 +72,26 @@ void QEMU_NORETURN cpu_loop_exit(CPUState *cpu);
void QEMU_NORETURN cpu_loop_exit_restore(CPUState *cpu, uintptr_t pc);
void QEMU_NORETURN cpu_loop_exit_atomic(CPUState *cpu, uintptr_t pc);
+/**
+ * cpu_cond_loop_exit_restore:
+ * @cpu: the vCPU state to be restored
+ * @pc: the host PC
+ *
+ * Trigger a cpu_loop_exit_restore() in case somebody asked for a return
+ * to the main loop (e.g. cpu_exit() or cpu_interrupt()).
+ *
+ * This is helpful for architectures that support interruptible
+ * instructions. After writing back all state to registers/memory, this
+ * call can be used to conditionally return back to the main loop or to
+ * continue executing the interruptible instruction.
+ */
+static inline void cpu_cond_loop_exit_restore(CPUState *cpu, uintptr_t pc)
+{
+ if (unlikely((int32_t)atomic_read(&cpu_neg(cs)->icount_decr.u32) < 0)) {
+ cpu_loop_exit_restore(cs, ra);
+ }
+}
+
#if !defined(CONFIG_USER_ONLY)
void cpu_reloading_memory_map(void);
/**
Or, as alternative, something like "cpu_shall_exit()" which only
wraps the single check.
--
Thanks,
David / dhildenb
