Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1569550 > unrolled thread

[PATCHv2] x86/mm/cpa: avoid wbinvd() for PREEMPT

Back to article view | Back to linux.kernel

Contents

  [PATCHv2] x86/mm/cpa: avoid wbinvd() for PREEMPT John Ogness <john.ogness@linutronix.de> - 2017-01-30 11:20 +0100
    Re: [PATCHv2] x86/mm/cpa: avoid wbinvd() for PREEMPT Peter Zijlstra <peterz@infradead.org> - 2017-01-30 13:40 +0100

#1569550 — [PATCHv2] x86/mm/cpa: avoid wbinvd() for PREEMPT

Although wbinvd() is faster than flushing many individual pages, it
blocks the memory bus for "long" periods of time (>100us), thus
directly causing unusually large latencies on all CPUs, regardless
of any CPU isolation features that may be active.

For 1024 pages, flushing those pages individually can take up to
2200us, but the task remains fully preemptible during that time.

Signed-off-by: John Ogness <john.ogness@linutronix.de>
---
 v1-v2: changed CONFIG_PREEMPT_RT_FULL to CONFIG_PREEMPT

 It was suggested that wbinvd() is removed altogether, but any
 kernel configured without CONFIG_PREEMPT probably doesn't care
 about latencies.

 arch/x86/mm/pageattr.c |    8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c
index 5a287e5..ba1393d 100644
--- a/arch/x86/mm/pageattr.c
+++ b/arch/x86/mm/pageattr.c
@@ -214,7 +214,15 @@ static void cpa_flush_array(unsigned long *start, int numpages, int cache,
 			    int in_flags, struct page **pages)
 {
 	unsigned int i, level;
+#ifdef CONFIG_PREEMPT
+	/*
+	 * Avoid wbinvd() because it causes latencies on all CPUs,
+	 * regardless of any CPU isolation that may be in effect.
+	 */
+	unsigned long do_wbinvd = 0;
+#else
 	unsigned long do_wbinvd = cache && numpages >= 1024; /* 4M threshold */
+#endif
 
 	BUG_ON(irqs_disabled());
 
-- 
1.7.10.4

[toc] | [next] | [standalone]

#1569695

On Mon, Jan 30, 2017 at 09:41:21AM +0100, John Ogness wrote:
> Although wbinvd() is faster than flushing many individual pages, it
> blocks the memory bus for "long" periods of time (>100us), thus
> directly causing unusually large latencies on all CPUs, regardless
> of any CPU isolation features that may be active.
> 
> For 1024 pages, flushing those pages individually can take up to
> 2200us, but the task remains fully preemptible during that time.
> 
> Signed-off-by: John Ogness <john.ogness@linutronix.de>

Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>

I think we need an additional patch to deal with CAT, but we can easily
do that later.

I've thought about that situation a bit and wbinvd cannot and must not
respect the CAT mask because while all the new lines brought in by the
current context are constrained, lines shared with another context, or
if the current context changed might well be outside the current mask.

In any case, this is a unprivileged interface, and therefore I feel we
need to remove the wbinvd usage entirely, esp. if CAT is enabled.

But like said, we can do that in a follow up patch.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web