Linux Virtualization - Jan 2016 - [PATCH v4 5/5] x86: drop mfence in favor of lock+addl

mb() typically uses mfence on modern x86, but a micro-benchmark shows that
it's
2 to 3 times slower than lock; addl that we use on older CPUs.

So we really should use the locked variant everywhere, except that intel manual
says that clflush is only ordered by mfence, so we can't.
Note: some callers of clflush seems to assume sfence will
order it, so there could be existing bugs around this code.

Fortunately no callers of clflush (except one) order it using smp_mb(), so
after fixing that one caller, it seems safe to override smp_mb straight away.

Down the road, it might make sense to introduce clflush_mb() and switch
to that for clflush callers.

While I was at it, I found some inconsistencies in comments in
arch/x86/include/asm/barrier.h

The documentation fixes are included first - I verified that
they do not change the generated code at all. Borislav Petkov
said they will appear in tip eventually, included here for
completeness.

The last patch changes __smp_mb() to lock addl. I was unable to
measure a speed difference on a macro benchmark,
but I noted that even doing
	#define mb() barrier()
seems to make no difference for most benchmarks
(it causes hangs sometimes, of course).

HPA asked that the last patch is deferred until we hear back from
intel, which makes sense of course. So it needs HPA's ack.

Changes from v3:
	Leave mb() alone for now since it's used to order
	clflush, which requires mfence. Optimize smp_mb instead.

Changes from v2:
	add patch adding cc clobber for addl
	tweak commit log for patch 2
	use addl at SP-4 (as opposed to SP) to reduce data dependencies

Michael S. Tsirkin (5):
  x86: add cc clobber for addl
  x86: drop a comment left over from X86_OOSTORE
  x86: tweak the comment about use of wmb for IO
  x86: use mb() around clflush
  x86: drop mfence in favor of lock+addl

 arch/x86/include/asm/barrier.h | 17 ++++++++---------
 arch/x86/kernel/process.c      |  4 ++--
 2 files changed, 10 insertions(+), 11 deletions(-)

-- 
MST

addl clobbers flags (such as CF) but barrier.h didn't tell this to gcc.
Historically, gcc doesn't need one on x86, and always considers flags
clobbered. We are probably missing the cc clobber in a *lot* of places
for this reason.

But even if not necessary, it's probably a good thing to add for
documentation, and in case gcc semantcs ever change.

Reported-by: Borislav Petkov <bp at alien8.de>
Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
---
 arch/x86/include/asm/barrier.h | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/arch/x86/include/asm/barrier.h b/arch/x86/include/asm/barrier.h
index a584e1c..a65bdb1 100644
--- a/arch/x86/include/asm/barrier.h
+++ b/arch/x86/include/asm/barrier.h
@@ -15,9 +15,12 @@
  * Some non-Intel clones support out of order store. wmb() ceases to be a
  * nop for these.
  */
-#define mb() alternative("lock; addl $0,0(%%esp)",
"mfence", X86_FEATURE_XMM2)
-#define rmb() alternative("lock; addl $0,0(%%esp)",
"lfence", X86_FEATURE_XMM2)
-#define wmb() alternative("lock; addl $0,0(%%esp)",
"sfence", X86_FEATURE_XMM)
+#define mb() asm volatile(ALTERNATIVE("lock; addl $0,0(%%esp)",
"mfence", \
+				      X86_FEATURE_XMM2) ::: "memory", "cc")
+#define rmb() asm volatile(ALTERNATIVE("lock; addl $0,0(%%esp)",
"lfence", \
+				       X86_FEATURE_XMM2) ::: "memory", "cc")
+#define wmb() asm volatile(ALTERNATIVE("lock; addl $0,0(%%esp)",
"sfence", \
+				       X86_FEATURE_XMM2) ::: "memory", "cc")
 #else
 #define mb() 	asm volatile("mfence":::"memory")
 #define rmb()	asm volatile("lfence":::"memory")
-- 
MST

The comment about wmb being non-nop to deal with non-intel CPUs is a
left over from before commit 09df7c4c8097 ("x86: Remove
CONFIG_X86_OOSTORE").

It makes no sense now: in particular, wmb is not a nop even for regular
intel CPUs because of weird use-cases e.g. dealing with WC memory.

Drop this comment.

Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
---
 arch/x86/include/asm/barrier.h | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/arch/x86/include/asm/barrier.h b/arch/x86/include/asm/barrier.h
index a65bdb1..a291745 100644
--- a/arch/x86/include/asm/barrier.h
+++ b/arch/x86/include/asm/barrier.h
@@ -11,10 +11,6 @@
  */
 
 #ifdef CONFIG_X86_32
-/*
- * Some non-Intel clones support out of order store. wmb() ceases to be a
- * nop for these.
- */
 #define mb() asm volatile(ALTERNATIVE("lock; addl $0,0(%%esp)",
"mfence", \
 				      X86_FEATURE_XMM2) ::: "memory", "cc")
 #define rmb() asm volatile(ALTERNATIVE("lock; addl $0,0(%%esp)",
"lfence", \
-- 
MST

On x86, we *do* still use the non-nop rmb/wmb for IO barriers, but even
that is generally questionable.

Leave them around as historial unless somebody can point to a case where
they care about the performance, but tweak the comment so people
don't think they are strictly required in all cases.

Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
---
 arch/x86/include/asm/barrier.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/include/asm/barrier.h b/arch/x86/include/asm/barrier.h
index a291745..bfb28ca 100644
--- a/arch/x86/include/asm/barrier.h
+++ b/arch/x86/include/asm/barrier.h
@@ -6,7 +6,7 @@
 
 /*
  * Force strict CPU ordering.
- * And yes, this is required on UP too when we're talking
+ * And yes, this might be required on UP too when we're talking
  * to devices.
  */
 
-- 
MST

commit f8e617f4582995f7c25ef25b4167213120ad122b ("sched/idle/x86:
Optimize unnecessary mwait_idle() resched IPIs") adds
memory barriers around clflush, but this seems wrong
for UP since barrier() has no effect on clflush.
We really want mfence so switch to mb() instead.

Cc: Mike Galbraith <bitbucket at online.de>
Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
---
 arch/x86/kernel/process.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index 9f7c21c..9decee2 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -418,9 +418,9 @@ static void mwait_idle(void)
 	if (!current_set_polling_and_test()) {
 		trace_cpu_idle_rcuidle(1, smp_processor_id());
 		if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
-			smp_mb(); /* quirk */
+			mb(); /* quirk */
 			clflush((void *)&current_thread_info()->flags);
-			smp_mb(); /* quirk */
+			mb(); /* quirk */
 		}
 
 		__monitor((void *)&current_thread_info()->flags, 0, 0);
-- 
MST

mfence appears to be way slower than a locked instruction - let's use
lock+add unconditionally, as we always did on old 32-bit.

Just poking at SP would be the most natural, but if we
then read the value from SP, we get a false dependency
which will slow us down.

This was noted in this article:
http://shipilev.net/blog/2014/on-the-fence-with-dependencies/

And is easy to reproduce by sticking a barrier in a small non-inline
function.

So let's use a negative offset - which avoids this problem since we
build with the red zone disabled.

Unfortunately there's some code that wants to order clflush instructions
using mb(), so we can't replace that - but smp_mb should be safe
to replace.

Update mb/rmb/wmb on 32 bit to use the negative offset, too, for
consistency.

Suggested-by: Andy Lutomirski <luto at amacapital.net>
Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
---
 arch/x86/include/asm/barrier.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/arch/x86/include/asm/barrier.h b/arch/x86/include/asm/barrier.h
index bfb28ca..7ab9581 100644
--- a/arch/x86/include/asm/barrier.h
+++ b/arch/x86/include/asm/barrier.h
@@ -11,11 +11,11 @@
  */
 
 #ifdef CONFIG_X86_32
-#define mb() asm volatile(ALTERNATIVE("lock; addl $0,0(%%esp)",
"mfence", \
+#define mb() asm volatile(ALTERNATIVE("lock; addl $0,-4(%%esp)",
"mfence", \
 				      X86_FEATURE_XMM2) ::: "memory", "cc")
-#define rmb() asm volatile(ALTERNATIVE("lock; addl $0,0(%%esp)",
"lfence", \
+#define rmb() asm volatile(ALTERNATIVE("lock; addl $0,-4(%%esp)",
"lfence", \
 				       X86_FEATURE_XMM2) ::: "memory", "cc")
-#define wmb() asm volatile(ALTERNATIVE("lock; addl $0,0(%%esp)",
"sfence", \
+#define wmb() asm volatile(ALTERNATIVE("lock; addl $0,-4(%%esp)",
"sfence", \
 				       X86_FEATURE_XMM2) ::: "memory", "cc")
 #else
 #define mb() 	asm volatile("mfence":::"memory")
@@ -30,7 +30,7 @@
 #endif
 #define dma_wmb()	barrier()
 
-#define __smp_mb()	mb()
+#define __smp_mb()	asm volatile("lock; addl $0,-4(%%esp)" :::
"memory", "cc")
 #define __smp_rmb()	dma_rmb()
 #define __smp_wmb()	barrier()
 #define __smp_store_mb(var, value) do { (void)xchg(&var, value); } while
(0)
-- 
MST

On Wed, Jan 27, 2016 at 7:10 AM, Michael S. Tsirkin <mst at redhat.com>
wrote:
>
> -#define __smp_mb()     mb()
> +#define __smp_mb()     asm volatile("lock; addl $0,-4(%%esp)"
::: "memory", "cc")
So this doesn't look right for x86-64. Using %esp rather than %rsp.
How did that even work for you?

                Linus