2019-02-12 20:07:22

by Mathieu Desnoyers

Subject: [PATCH 2/4] glibc: sched_getcpu(): use rseq cpu_id TLS on Linux

When available, use the cpu_id field from __rseq_abi on Linux to
implement sched_getcpu(). Fall-back on the vgetcpu vDSO if unavailable.

Benchmarks:

x86-64: Intel E5-2630 [email protected], 16-core, hyperthreading

glibc sched_getcpu(): 13.7 ns (baseline)
glibc sched_getcpu() using rseq: 2.5 ns (speedup: 5.5x)
inline load cpuid from __rseq_abi TLS: 0.8 ns (speedup: 17.1x)

Signed-off-by: Mathieu Desnoyers <[email protected]>
CC: Carlos O'Donell <[email protected]>
CC: Florian Weimer <[email protected]>
CC: Joseph Myers <[email protected]>
CC: Szabolcs Nagy <[email protected]>
CC: Thomas Gleixner <[email protected]>
CC: Ben Maurer <[email protected]>
CC: Peter Zijlstra <[email protected]>
CC: "Paul E. McKenney" <[email protected]>
CC: Boqun Feng <[email protected]>
CC: Will Deacon <[email protected]>
CC: Dave Watson <[email protected]>
CC: Paul Turner <[email protected]>
CC: [email protected]
CC: [email protected]
CC: [email protected]
---
sysdeps/unix/sysv/linux/sched_getcpu.c | 25 +++++++++++++++++++++++--
1 file changed, 23 insertions(+), 2 deletions(-)

diff --git a/sysdeps/unix/sysv/linux/sched_getcpu.c b/sysdeps/unix/sysv/linux/sched_getcpu.c
index fb0d317f83..8bfb03778b 100644
--- a/sysdeps/unix/sysv/linux/sched_getcpu.c
+++ b/sysdeps/unix/sysv/linux/sched_getcpu.c
@@ -24,8 +24,8 @@
#endif
#include <sysdep-vdso.h>

-int
-sched_getcpu (void)
+static int
+vsyscall_sched_getcpu (void)
{
#ifdef __NR_getcpu
unsigned int cpu;
@@ -37,3 +37,24 @@ sched_getcpu (void)
return -1;
#endif
}
+
+#ifdef __NR_rseq
+#include <linux/rseq.h>
+
+extern __attribute__ ((tls_model ("initial-exec")))
+__thread volatile struct rseq __rseq_abi;
+
+int
+sched_getcpu (void)
+{
+ int cpu_id = __rseq_abi.cpu_id;
+
+ return cpu_id >= 0 ? cpu_id : vsyscall_sched_getcpu ();
+}
+#else
+int
+sched_getcpu (void)
+{
+ return vsyscall_sched_getcpu ();
+}
+#endif
--
2.17.1

2019-03-22 20:14:48

by Carlos O'Donell

Subject: Re: [PATCH 2/4] glibc: sched_getcpu(): use rseq cpu_id TLS on Linux

On 2/12/19 2:42 PM, Mathieu Desnoyers wrote:
> When available, use the cpu_id field from __rseq_abi on Linux to
> implement sched_getcpu(). Fall-back on the vgetcpu vDSO if unavailable.
>
> Benchmarks:
>
> x86-64: Intel E5-2630 [email protected], 16-core, hyperthreading

This patch looks good to me for master, but is blocked on patch 1/4
being reworked.

Reviewed-by: Carlos O'Donell <[email protected]>

> glibc sched_getcpu(): 13.7 ns (baseline)
> glibc sched_getcpu() using rseq: 2.5 ns (speedup: 5.5x)
> inline load cpuid from __rseq_abi TLS: 0.8 ns (speedup: 17.1x)
>
> Signed-off-by: Mathieu Desnoyers <[email protected]>
> CC: Carlos O'Donell <[email protected]>
> CC: Florian Weimer <[email protected]>
> CC: Joseph Myers <[email protected]>
> CC: Szabolcs Nagy <[email protected]>
> CC: Thomas Gleixner <[email protected]>
> CC: Ben Maurer <[email protected]>
> CC: Peter Zijlstra <[email protected]>
> CC: "Paul E. McKenney" <[email protected]>
> CC: Boqun Feng <[email protected]>
> CC: Will Deacon <[email protected]>
> CC: Dave Watson <[email protected]>
> CC: Paul Turner <[email protected]>
> CC: [email protected]
> CC: [email protected]
> CC: [email protected]
> ---
> sysdeps/unix/sysv/linux/sched_getcpu.c | 25 +++++++++++++++++++++++--
> 1 file changed, 23 insertions(+), 2 deletions(-)
>
> diff --git a/sysdeps/unix/sysv/linux/sched_getcpu.c b/sysdeps/unix/sysv/linux/sched_getcpu.c
> index fb0d317f83..8bfb03778b 100644
> --- a/sysdeps/unix/sysv/linux/sched_getcpu.c
> +++ b/sysdeps/unix/sysv/linux/sched_getcpu.c
> @@ -24,8 +24,8 @@
> #endif
> #include <sysdep-vdso.h>
>
> -int
> -sched_getcpu (void)
> +static int
> +vsyscall_sched_getcpu (void)

OK.

> {
> #ifdef __NR_getcpu
> unsigned int cpu;
> @@ -37,3 +37,24 @@ sched_getcpu (void)
> return -1;
> #endif
> }
> +
> +#ifdef __NR_rseq
> +#include <linux/rseq.h>
> +
> +extern __attribute__ ((tls_model ("initial-exec")))
> +__thread volatile struct rseq __rseq_abi;

OK.

> +
> +int
> +sched_getcpu (void)
> +{
> + int cpu_id = __rseq_abi.cpu_id;
> +
> + return cpu_id >= 0 ? cpu_id : vsyscall_sched_getcpu ();

OK. Impressive :-)

> +}
> +#else
> +int
> +sched_getcpu (void)
> +{
> + return vsyscall_sched_getcpu ();

OK.

> +}
> +#endif
>


--
Cheers,
Carlos.