Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
To:     Will Deacon <will@kernel.org>, Marc Zyngier <maz@kernel.org>,
        James Morse <james.morse@arm.com>,
        Julien Thierry <julien.thierry.kdev@gmail.com>,
        Suzuki K Poulose <suzuki.poulose@arm.com>
Cc:     suleiman@google.com, joelaf@google.com,
        linux-arm-kernel@lists.infradead.org, kvmarm@lists.cs.columbia.edu,
        linux-kernel@vger.kernel.org,
        Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Subject: [RFC][PATCH 4/4] arm64: do not use dummy vcpu_is_preempted() anymore
Date:   Tue, 21 Jul 2020 13:17:42 +0900
Message-Id: <20200721041742.197354-5-sergey.senozhatsky@gmail.com>
In-Reply-To: <20200721041742.197354-1-sergey.senozhatsky@gmail.com>
References: <20200721041742.197354-1-sergey.senozhatsky@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

vcpu_is_preempted() now can represent the actual state of
the VCPU, so the scheduler can make better decisions when
it picks the idle CPU to enqueue a task on. I executed a
whole bunch of scheduler tests [0]. One particular test
that shows the importance of vcpu_is_preempted() is AIO
stress-ng test:

x Disabled vcpu_is_preempted()
stress-ng: info:  [100] stressor       bogo ops real time  usr time  sys time   bogo ops/s   bogo ops/s
stress-ng: info:  [100]                           (secs)    (secs)    (secs)   (real time) (usr+sys time)
stress-ng: info:  [100] aio              222927     10.01      0.89     27.61     22262.04      7822.00
stress-ng: info:  [139] aio              217043     10.01      1.00     26.80     21685.46      7807.30
stress-ng: info:  [178] aio              217261     10.01      1.08     26.79     21709.36      7795.51

+ Enabled vcpu_is_preempted()
stress-ng: info:  [100] aio              432750     10.00      1.14     19.03     43264.33     21455.13
stress-ng: info:  [139] aio              426771     10.01      1.09     18.67     42629.13     21597.72
stress-ng: info:  [179] aio              533039     10.00      1.42     20.39     53281.70     24440.12

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
---
 arch/arm64/include/asm/spinlock.h | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)

diff --git a/arch/arm64/include/asm/spinlock.h b/arch/arm64/include/asm/spinlock.h
index 9083d6992603..6a390eeabe82 100644
--- a/arch/arm64/include/asm/spinlock.h
+++ b/arch/arm64/include/asm/spinlock.h
@@ -11,17 +11,20 @@
 /* See include/linux/spinlock.h */
 #define smp_mb__after_spinlock()	smp_mb()
 
-/*
- * Changing this will break osq_lock() thanks to the call inside
- * smp_cond_load_relaxed().
- *
- * See:
- * https://lore.kernel.org/lkml/20200110100612.GC2827@hirez.programming.kicks-ass.net
- */
 #define vcpu_is_preempted vcpu_is_preempted
+
+#ifdef CONFIG_PARAVIRT
+extern bool paravirt_vcpu_is_preempted(int cpu);
+
+static inline bool vcpu_is_preempted(int cpu)
+{
+	return paravirt_vcpu_is_preempted(cpu);
+}
+#else
 static inline bool vcpu_is_preempted(int cpu)
 {
 	return false;
 }
+#endif /* CONFIG_PARAVIRT */
 
 #endif /* __ASM_SPINLOCK_H */
-- 
2.27.0