Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Eric Dumazet <edumazet@google.com>
To:     x86 <x86@kernel.org>
Cc:     lkml <linux-kernel@vger.kernel.org>,
        Eric Dumazet <edumazet@google.com>,
        Eric Dumazet <eric.dumazet@gmail.com>,
        "H. Peter Anvin" <hpa@zytor.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        Hugh Dickins <hughd@google.com>
Subject: [PATCH v3 1/2] x86, msr: allow rdmsr_safe_on_cpu() to schedule
Date:   Fri, 23 Mar 2018 14:58:17 -0700
Message-Id: <20180323215818.127774-1-edumazet@google.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

I noticed high latencies caused by a daemon periodically reading
various MSR on all cpus. KASAN kernels would see ~10ms latencies
simply reading one MSR. Even without KASAN, sending IPI to CPU
in deep sleep state or blocking hard IRQ in a a long section,
then waiting for the answer can consume hundreds of usec.

Converts rdmsr_safe_on_cpu() to use a completion instead
of busy polling.

Overall daemon cpu usage was reduced by 35 %,
and latencies caused by msr_read() disappeared.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Hugh Dickins <hughd@google.com>
---
 arch/x86/lib/msr-smp.c | 32 ++++++++++++++++++++++++--------
 1 file changed, 24 insertions(+), 8 deletions(-)

diff --git a/arch/x86/lib/msr-smp.c b/arch/x86/lib/msr-smp.c
index 693cce0be82dffb822cecd0c7e38d2821aff896c..761ba062afdaf7f7d0603ed94ed6cc3e46b37f76 100644
--- a/arch/x86/lib/msr-smp.c
+++ b/arch/x86/lib/msr-smp.c
@@ -2,6 +2,7 @@
 #include <linux/export.h>
 #include <linux/preempt.h>
 #include <linux/smp.h>
+#include <linux/completion.h>
 #include <asm/msr.h>
 
 static void __rdmsr_on_cpu(void *info)
@@ -143,13 +144,19 @@ void wrmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs)
 }
 EXPORT_SYMBOL(wrmsr_on_cpus);
 
+struct msr_info_completion {
+	struct msr_info		msr;
+	struct completion	done;
+};
+
 /* These "safe" variants are slower and should be used when the target MSR
    may not actually exist. */
 static void __rdmsr_safe_on_cpu(void *info)
 {
-	struct msr_info *rv = info;
+	struct msr_info_completion *rv = info;
 
-	rv->err = rdmsr_safe(rv->msr_no, &rv->reg.l, &rv->reg.h);
+	rv->msr.err = rdmsr_safe(rv->msr.msr_no, &rv->msr.reg.l, &rv->msr.reg.h);
+	complete(&rv->done);
 }
 
 static void __wrmsr_safe_on_cpu(void *info)
@@ -161,17 +168,26 @@ static void __wrmsr_safe_on_cpu(void *info)
 
 int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
 {
+	struct msr_info_completion rv;
+	call_single_data_t csd = {
+		.func	= __rdmsr_safe_on_cpu,
+		.info	= &rv,
+	};
 	int err;
-	struct msr_info rv;
 
 	memset(&rv, 0, sizeof(rv));
+	init_completion(&rv.done);
+	rv.msr.msr_no = msr_no;
 
-	rv.msr_no = msr_no;
-	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
-	*l = rv.reg.l;
-	*h = rv.reg.h;
+	err = smp_call_function_single_async(cpu, &csd);
+	if (!err) {
+		wait_for_completion(&rv.done);
+		err = rv.msr.err;
+	}
+	*l = rv.msr.reg.l;
+	*h = rv.msr.reg.h;
 
-	return err ? err : rv.err;
+	return err;
 }
 EXPORT_SYMBOL(rdmsr_safe_on_cpu);
 
-- 
2.17.0.rc0.231.g781580f067-goog