On 09/06/2016 20:08, Andy Lutomirski wrote:
> > __pvclock_read_cycles is pretty much the same as the code that is being
> > inlined. Thus the only change is that __pvclock_read_cycles is called
> > inside the loop rather than outside, but the loop really is expected to
> > never roll so why make a copy in the first place?
>
> I feel like I had a reason, but I don't remember what it was.
I cannot see any substantial difference in the generated code with
this patch. On the other hand, if I add this:
diff --git a/arch/x86/include/asm/pvclock.h b/arch/x86/include/asm/pvclock.h
index 0ee92db1e9f3..d019f0cc80ec 100644
--- a/arch/x86/include/asm/pvclock.h
+++ b/arch/x86/include/asm/pvclock.h
@@ -40,7 +40,7 @@ bool pvclock_read_retry(const struct pvclock_vcpu_time_info
*src,
{
/* Make sure that the version is re-read after the data. */
virt_rmb();
- return version != src->version;
+ return unlikely(version != src->version);
}
/*
then vread_pvclock is inlined everywhere, which bloats the code somewhat (+80
code bytes). Do you want me to mark vread_pvclock as noinline too?
Paolo
