linuxppc-dev.lists.ozlabs.org archive mirror
 help / color / mirror / Atom feed
From: Nicholas Piggin <npiggin@gmail.com>
To: linuxppc-dev@lists.ozlabs.org
Cc: Nicholas Piggin <npiggin@gmail.com>, Anton Blanchard <anton@samba.org>
Subject: [PATCH 02/14] powerpc/64: syscall avoid restore_math call if possible
Date: Fri,  2 Jun 2017 17:39:34 +1000	[thread overview]
Message-ID: <20170602073946.8983-3-npiggin@gmail.com> (raw)
In-Reply-To: <20170602073946.8983-1-npiggin@gmail.com>

The syscall exit code that branches to restore_math is quite
heavyweight on Book3S, consisting of 2 mtmsr instructions. This
case can happen even if restore_math decides there is nothing
to do due to lazy math restore.

Check for lazy restore before calling restore_math. Move most
of that case out of line.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
---
 arch/powerpc/kernel/entry_64.S | 62 +++++++++++++++++++++++++++++-------------
 arch/powerpc/kernel/process.c  |  4 +++
 2 files changed, 47 insertions(+), 19 deletions(-)

diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
index bfbad08a1207..019a6322b982 100644
--- a/arch/powerpc/kernel/entry_64.S
+++ b/arch/powerpc/kernel/entry_64.S
@@ -210,27 +210,17 @@ system_call:			/* label this so stack traces look sane */
 	andi.	r0,r9,(_TIF_SYSCALL_DOTRACE|_TIF_SINGLESTEP|_TIF_USER_WORK_MASK|_TIF_PERSYSCALL_MASK)
 	bne-	syscall_exit_work
 
-	andi.	r0,r8,MSR_FP
-	beq 2f
+	/* If MSR_FP and MSR_VEC are set in user msr, then no need to restore */
+	li	r0,MSR_FP
 #ifdef CONFIG_ALTIVEC
-	andis.	r0,r8,MSR_VEC@h
-	bne	3f
+	oris	r0,r0,MSR_VEC@h
 #endif
-2:	addi    r3,r1,STACK_FRAME_OVERHEAD
-#ifdef CONFIG_PPC_BOOK3S
-	li	r10,MSR_RI
-	mtmsrd	r10,1		/* Restore RI */
-#endif
-	bl	restore_math
-#ifdef CONFIG_PPC_BOOK3S
-	li	r11,0
-	mtmsrd	r11,1
-#endif
-	ld	r8,_MSR(r1)
-	ld	r3,RESULT(r1)
-	li	r11,-MAX_ERRNO
+	andc	r7,r8,r0
+	cmpd	r7,r0
+	bne	syscall_restore_math
+.Lsyscall_restore_math_cont:
 
-3:	cmpld	r3,r11
+	cmpld	r3,r11
 	ld	r5,_CCR(r1)
 	bge-	syscall_error
 .Lsyscall_error_cont:
@@ -263,7 +253,41 @@ syscall_error:
 	neg	r3,r3
 	std	r5,_CCR(r1)
 	b	.Lsyscall_error_cont
-	
+
+syscall_restore_math:
+	/*
+	 * Some initial tests from restore_math to avoid the heavyweight
+	 * C code entry and MSR manipulations.
+	 */
+	LOAD_REG_IMMEDIATE(r0, MSR_TS_MASK)
+	and.	r0,r0,r8
+	bne	1f
+
+	ld	r7,PACACURRENT(r13)
+	lbz	r0,THREAD+THREAD_LOAD_FP(r7)
+#ifdef CONFIG_ALTIVEC
+	lbz	r6,THREAD+THREAD_LOAD_VEC(r7)
+	add	r0,r0,r6
+#endif
+	cmpdi	r0,0
+	beq	.Lsyscall_restore_math_cont
+
+1:	addi    r3,r1,STACK_FRAME_OVERHEAD
+#ifdef CONFIG_PPC_BOOK3S
+	li	r10,MSR_RI
+	mtmsrd	r10,1		/* Restore RI */
+#endif
+	bl	restore_math
+#ifdef CONFIG_PPC_BOOK3S
+	li	r11,0
+	mtmsrd	r11,1
+#endif
+	/* Restore volatiles, reload MSR from updated one */
+	ld	r8,_MSR(r1)
+	ld	r3,RESULT(r1)
+	li	r11,-MAX_ERRNO
+	b	.Lsyscall_restore_math_cont
+
 /* Traced system call support */
 syscall_dotrace:
 	bl	save_nvgprs
diff --git a/arch/powerpc/kernel/process.c b/arch/powerpc/kernel/process.c
index baae104b16c7..5cbb8b1faf7e 100644
--- a/arch/powerpc/kernel/process.c
+++ b/arch/powerpc/kernel/process.c
@@ -511,6 +511,10 @@ void restore_math(struct pt_regs *regs)
 {
 	unsigned long msr;
 
+	/*
+	 * Syscall exit makes a similar initial check before branching
+	 * to restore_math. Keep them in synch.
+	 */
 	if (!msr_tm_active(regs->msr) &&
 		!current->thread.load_fp && !loadvec(current->thread))
 		return;
-- 
2.11.0

  parent reply	other threads:[~2017-06-02  7:40 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-06-02  7:39 [RFC][PATCH 00/14] syscall, context switch, idle performance stuff Nicholas Piggin
2017-06-02  7:39 ` [PATCH 01/14] powerpc/64s: optimize hypercall/syscall Nicholas Piggin
2017-06-02  7:39 ` Nicholas Piggin [this message]
2017-06-02  7:39 ` [PATCH 03/14] powerpc/64s: idle move soft interrupt mask logic into C code Nicholas Piggin
2017-06-02  7:39 ` [PATCH 04/14] powerpc/64s: process interrupts from system reset wakeup Nicholas Piggin
2017-06-02  7:39 ` [PATCH 05/14] powerpc/64s: msgclr when handling doorbell exceptions Nicholas Piggin
2017-06-02  7:39 ` [PATCH 06/14] powerpc/64s: branch to idle handler with virtual mode offset Nicholas Piggin
2017-06-02  7:39 ` [PATCH 07/14] powerpc/64s: idle avoid SRR usage in idle sleep/wake paths Nicholas Piggin
2017-06-02 13:37   ` kbuild test robot
2017-06-02  7:39 ` [PATCH 08/14] powerpc/64s: idle set polling before enabling irqs Nicholas Piggin
2017-06-02  7:39 ` [PATCH 09/14] powerpc/64s: idle read mostly for common globals Nicholas Piggin
2017-06-02  7:39 ` [PATCH 10/14] powerpc/64: CTRL[RUN] run-latch setting optimisation Nicholas Piggin
2017-06-02  7:39 ` [PATCH 11/14] powerpc/64s: idle no memory barrier after break from idle Nicholas Piggin
2017-06-02  7:39 ` [PATCH 12/14] powerpc/64s: Leave IRQs hard enabled over context switch for radix Nicholas Piggin
2017-06-02  7:39 ` [PATCH 13/14] powerpc/64: context switch can avoid reservation clear Nicholas Piggin
2017-06-02  7:39 ` [PATCH 14/14] powerpc/64: context switch additional hwsync can be avoided Nicholas Piggin

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=20170602073946.8983-3-npiggin@gmail.com \
    --to=npiggin@gmail.com \
    --cc=anton@samba.org \
    --cc=linuxppc-dev@lists.ozlabs.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).