| Message ID | 4250629DA95C6D4F+20231204022303.528-2-luming.yu@shingroup.cn (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [1/2] powerpc/locking: implement this_cpu_cmpxchg local API | expand |
| Context | Check | Description |
|---|---|---|
| snowpatch_ozlabs/github-powerpc_selftests | success | Successfully ran 8 jobs. |
| snowpatch_ozlabs/github-powerpc_sparse | fail | sparse (pmac32, fedora-38, ppc64) failed at step Build. |
| snowpatch_ozlabs/github-powerpc_kernel_qemu | fail | 2 of 18 jobs failed. |
| snowpatch_ozlabs/github-powerpc_ppctests | success | Successfully ran 8 jobs. |
| snowpatch_ozlabs/github-powerpc_clang | fail | 3 of 6 jobs failed. |
Hi, specific for local cmpxchg enabled on p8 powernv platform on which the patch enabled vm_state update path, ftrace data below indicates it is at the level ofabout 4us or 5us latency, for such a big cache cold operations. <...>-277787 [008] ..... 88366.233643: refresh_cpu_vm_stats <-vmstat_update <idle>-0 [008] d.... 88366.233648: tick_nohz_idle_stop_tick <-do_idle I have no diff data against the latency without the patch. But based on the estimation of the pipeline overhead for such a big operation incurred by the vm stats update, it looks not bad. I will do more testing and collect more histogram for different scenarios to ensure that the latency level of understanding is something statistically make sense. Just for the record that the small patch works as expected. :-) BR Luming ------------------ Original ------------------ From: "虞陆铭"<luming.yu@shingroup.cn>; Date: Mon, Dec 4, 2023 10:23 AM To: "linuxppc-dev"<linuxppc-dev@lists.ozlabs.org>; "linux-kernel"<linux-kernel@vger.kernel.org>; "mpe"<mpe@ellerman.id.au>; "npiggin"<npiggin@gmail.com>; "christophe.leroy"<christophe.leroy@csgroup.eu>; Cc: "luming.yu"<luming.yu@gmail.com>; "ke.zhao"<ke.zhao@shingroup.cn>; "dawei.li"<dawei.li@shingroup.cn>; "shenghui.qu"<shenghui.qu@shingroup.cn>; "虞陆铭"<luming.yu@shingroup.cn>; Subject: [PATCH 2/2] powerpc/locking: enable HAVE_CMPXCHG_LOCAL in kconfig enable arch feature HAVE_CMPXCHG_LOCAL for ppc by default Signed-off-by: Luming Yu <luming.yu@shingroup.cn> --- arch/powerpc/Kconfig | 1 + 1 file changed, 1 insertion(+) diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig index 6f105ee4f3cf..c10229c0243c 100644 --- a/arch/powerpc/Kconfig +++ b/arch/powerpc/Kconfig @@ -226,6 +226,7 @@ config PPC select HAVE_ARCH_SECCOMP_FILTER select HAVE_ARCH_TRACEHOOK select HAVE_ASM_MODVERSIONS + select HAVE_CMPXCHG_LOCAL select HAVE_CONTEXT_TRACKING_USER select HAVE_C_RECORDMCOUNT select HAVE_DEBUG_KMEMLEAK
(in text)
Hi,
specific for local cmpxchg enabled on p8 powernv platform on which the patch enabled vm_state update path,
ftrace data below indicates it is at the level ofabout 4us or 5us latency, for such a big cache cold operations.
<...>-277787 [008] ..... 88366.233643: refresh_cpu_vm_stats <-vmstat_update
<idle>-0 [008] d.... 88366.233648: tick_nohz_idle_stop_tick <-do_idle
I have no diff data against the latency without the patch. But based on the estimation of the pipeline overhead for
such a big operation incurred by the vm stats update, it looks not bad.
I will do more testing and collect more histogram for different scenarios to ensure that
the latency level of understanding is something statistically make sense.
Just for the record that the small patch works as expected. :-)
BR
Luming
------------------ Original ------------------
From: "虞陆铭"<luming.yu@shingroup.cn>;
Date: Mon, Dec 4, 2023 10:23 AM
To: "linuxppc-dev"<linuxppc-dev@lists.ozlabs.org>; "linux-kernel"<linux-kernel@vger.kernel.org>; "mpe"<mpe@ellerman.id.au>; "npiggin"<npiggin@gmail.com>; "christophe.leroy"<christophe.leroy@csgroup.eu>;
Cc: "luming.yu"<luming.yu@gmail.com>; "ke.zhao"<ke.zhao@shingroup.cn>; "dawei.li"<dawei.li@shingroup.cn>; "shenghui.qu"<shenghui.qu@shingroup.cn>; "虞陆铭"<luming.yu@shingroup.cn>;
Subject: [PATCH 2/2] powerpc/locking: enable HAVE_CMPXCHG_LOCAL in kconfig
enable arch feature HAVE_CMPXCHG_LOCAL for ppc by default
Signed-off-by: Luming Yu <luming.yu@shingroup.cn>
---
arch/powerpc/Kconfig | 1 +
1 file changed, 1 insertion(+)
diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index 6f105ee4f3cf..c10229c0243c 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -226,6 +226,7 @@ config PPC
select HAVE_ARCH_SECCOMP_FILTER
select HAVE_ARCH_TRACEHOOK
select HAVE_ASM_MODVERSIONS
+ select HAVE_CMPXCHG_LOCAL
select HAVE_CONTEXT_TRACKING_USER
select HAVE_C_RECORDMCOUNT
select HAVE_DEBUG_KMEMLEAK
diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig index 6f105ee4f3cf..c10229c0243c 100644 --- a/arch/powerpc/Kconfig +++ b/arch/powerpc/Kconfig @@ -226,6 +226,7 @@ config PPC select HAVE_ARCH_SECCOMP_FILTER select HAVE_ARCH_TRACEHOOK select HAVE_ASM_MODVERSIONS + select HAVE_CMPXCHG_LOCAL select HAVE_CONTEXT_TRACKING_USER select HAVE_C_RECORDMCOUNT select HAVE_DEBUG_KMEMLEAK
enable arch feature HAVE_CMPXCHG_LOCAL for ppc by default Signed-off-by: Luming Yu <luming.yu@shingroup.cn> --- arch/powerpc/Kconfig | 1 + 1 file changed, 1 insertion(+)