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 <[email protected]> --- 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
