From: liuj97@gmail.com (Jiang Liu)
To: linux-arm-kernel@lists.infradead.org
Subject: [RFT PATCH v1 7/7] arm64: disable lazy load if FPSIMD registers are frequently used
Date: Fri, 27 Sep 2013 16:04:47 +0800 [thread overview]
Message-ID: <1380269087-8520-8-git-send-email-liuj97@gmail.com> (raw)
In-Reply-To: <1380269087-8520-1-git-send-email-liuj97@gmail.com>
From: Jiang Liu <jiang.liu@huawei.com>
Disable lazy load if FPSIMD registers are frequently used by the thread,
so we can reduce overhead of lazy FPSIMD restore for FPU extensive
applications. The thresholds are randomly chosen without thorough
benchmarks, we may need to tune it for really systems.
Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
Cc: Jiang Liu <liuj97@gmail.com>
---
arch/arm64/include/asm/fpsimd.h | 1 +
arch/arm64/kernel/fpsimd.c | 9 +++++++--
2 files changed, 8 insertions(+), 2 deletions(-)
diff --git a/arch/arm64/include/asm/fpsimd.h b/arch/arm64/include/asm/fpsimd.h
index 3490935..de6e877 100644
--- a/arch/arm64/include/asm/fpsimd.h
+++ b/arch/arm64/include/asm/fpsimd.h
@@ -37,6 +37,7 @@ struct fpsimd_state {
u32 fpcr;
bool on_hw; /* soft state: whether loaded onto hw */
int last_cpu;
+ int fpu_counter;
};
};
};
diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c
index a81af5f..03e96b8 100644
--- a/arch/arm64/kernel/fpsimd.c
+++ b/arch/arm64/kernel/fpsimd.c
@@ -107,7 +107,7 @@ static void fpsimd_load_state_lazy(struct fpsimd_state *state,
__this_cpu_read(fpsimd_owner) == state)
return;
- if (static_key_false(&fpsimd_lazy_mode)) {
+ if (state->fpu_counter < 5 && static_key_false(&fpsimd_lazy_mode)) {
fpsimd_clear_on_hw(state);
fpsimd_enable_trap();
} else if (tsk_used_math(tsk)) {
@@ -124,8 +124,11 @@ static void fpsimd_save_state_lazy(struct fpsimd_state *state)
return;
if (static_key_false(&fpsimd_lazy_mode)) {
- if (!fpsimd_is_on_hw(state))
+ if (!fpsimd_is_on_hw(state)) {
+ state->fpu_counter = clamp(state->fpu_counter - 1,
+ 0, 10);
return;
+ }
}
fpsimd_save_state(state);
@@ -144,6 +147,7 @@ void do_fpsimd_acc(unsigned int esr, struct pt_regs *regs)
fpsimd_load_state(state);
if (static_key_false(&fpsimd_lazy_mode)) {
fpsimd_set_on_hw(state);
+ state->fpu_counter++;
if (!used_math())
set_used_math();
} else {
@@ -213,6 +217,7 @@ void fpsimd_dup_state(struct fpsimd_state *src, struct fpsimd_state *dst)
preempt_disable();
if (fpsimd_is_on_hw(src))
fpsimd_save_state(dst);
+ dst->fpu_counter = 0;
fpsimd_clear_on_hw(dst);
preempt_enable();
} else if (used_math()) {
--
1.8.1.2
next prev parent reply other threads:[~2013-09-27 8:04 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-09-27 8:04 [RFT PATCH v1 0/7] enable FPSIMD lazy save and restore for arm64 Jiang Liu
2013-09-27 8:04 ` [RFT PATCH v1 1/7] arm64: fix possible invalid FPSIMD initialization state Jiang Liu
2013-09-27 10:59 ` Catalin Marinas
2013-09-27 8:04 ` [RFT PATCH v1 2/7] arm64: restore FPSIMD to default state for kernel and signal contexts Jiang Liu
2013-09-27 11:35 ` Catalin Marinas
2013-09-27 13:20 ` Jiang Liu
2013-09-27 8:04 ` [RFT PATCH v1 3/7] arm64: implement basic lazy save and restore for FPSIMD registers Jiang Liu
2013-09-27 8:04 ` [RFT PATCH v1 4/7] arm64: provide boot option "eagerfpu" to control FPSIMD restore policy Jiang Liu
2013-09-27 8:04 ` [RFT PATCH v1 5/7] arm64: reuse FPSIMD hardware context if possible Jiang Liu
2013-09-27 8:04 ` [RFT PATCH v1 6/7] amd64: avoid saving and restoring FPSIMD registers until threads access them Jiang Liu
2013-09-27 10:59 ` Will Deacon
2013-09-27 14:20 ` Jiang Liu
2013-09-30 9:34 ` Will Deacon
2013-09-27 8:04 ` Jiang Liu [this message]
2013-09-27 10:50 ` [RFT PATCH v1 0/7] enable FPSIMD lazy save and restore for arm64 Catalin Marinas
2013-09-27 11:23 ` Will Deacon
2013-09-27 15:20 ` Jiang Liu
2013-09-27 16:16 ` Catalin Marinas
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1380269087-8520-8-git-send-email-liuj97@gmail.com \
--to=liuj97@gmail.com \
--cc=linux-arm-kernel@lists.infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).