linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] powerpc32: memcpy: only use dcbz once cache is enabled
@ 2015-09-07 14:24 Christophe Leroy
  2015-09-07 14:50 ` Michal Sojka
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Christophe Leroy @ 2015-09-07 14:24 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	scottwood, sojkam1
  Cc: linux-kernel, linuxppc-dev

memcpy() uses instruction dcbz to speed up copy by not wasting time
loading cache line with data that will be overwritten.
Some platform like mpc52xx do no have cache active at startup and
can therefore not use memcpy(). Allthough no part of the code
explicitly uses memcpy(), GCC makes calls to it.

This patch modifies memcpy() such that at startup, the 'dcbz'
instruction is replaced by 'dcbt' which is harmless if cache is not
enabled, and which helps a bit (allthough not as much as dcbz) if
cache is already enabled.

Once the initial MMU is setup, in machine_init() we patch memcpy()
by replacing the temporary 'dcbt' by 'dcbz'

Reported-by: Michal Sojka <sojkam1@fel.cvut.cz>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
@Michal, can you please test it ?

 arch/powerpc/kernel/setup_32.c | 12 ++++++++++++
 arch/powerpc/lib/copy_32.S     | 11 ++++++++++-
 2 files changed, 22 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/kernel/setup_32.c b/arch/powerpc/kernel/setup_32.c
index 07831ed..93715f3 100644
--- a/arch/powerpc/kernel/setup_32.c
+++ b/arch/powerpc/kernel/setup_32.c
@@ -39,6 +39,7 @@
 #include <asm/udbg.h>
 #include <asm/mmu_context.h>
 #include <asm/epapr_hcalls.h>
+#include <asm/code-patching.h>
 
 #define DBG(fmt...)
 
@@ -108,6 +109,15 @@ notrace unsigned long __init early_init(unsigned long dt_ptr)
 	return KERNELBASE + offset;
 }
 
+extern unsigned int ppc32_memcpy_dcbz;
+notrace void __init patch_memcpy(void)
+{
+	unsigned int instr = ppc32_memcpy_dcbz;
+
+	instr &= 0x001ff800; /* keep only RA and RB */
+	instr |= 0x7c0007ec; /* dcbz */
+	patch_instruction(&ppc32_memcpy_dcbz, instr);
+}
 
 /*
  * Find out what kind of machine we're on and save any data we need
@@ -122,6 +132,8 @@ notrace void __init machine_init(u64 dt_ptr)
 	/* Enable early debugging if any specified (see udbg.h) */
 	udbg_early_init();
 
+	patch_memcpy();
+
 	/* Do some early initialization based on the flat device tree */
 	early_init_devtree(__va(dt_ptr));
 
diff --git a/arch/powerpc/lib/copy_32.S b/arch/powerpc/lib/copy_32.S
index 2ef50c6..05b3096 100644
--- a/arch/powerpc/lib/copy_32.S
+++ b/arch/powerpc/lib/copy_32.S
@@ -172,7 +172,16 @@ _GLOBAL(memcpy)
 	mtctr	r0
 	beq	63f
 53:
-	dcbz	r11,r6
+	/*
+	 * During early init, cache might not be active yet, so dcbz cannot be
+	 * used. We put dcbt instead of dcbz. If cache is not active, it's just
+	 * like a not. If cache is active, at least it prefetchs the line to be
+	 * overwritten.
+	 * Will be replaced by dcbz in machine_init()
+	 */
+_GLOBAL(ppc32_memcpy_dcbz)
+	dcbt	r11,r6
+
 	COPY_16_BYTES
 #if L1_CACHE_BYTES >= 32
 	COPY_16_BYTES
-- 
2.1.0


^ permalink raw reply related	[flat|nested] 4+ messages in thread

* Re: [PATCH] powerpc32: memcpy: only use dcbz once cache is enabled
  2015-09-07 14:24 [PATCH] powerpc32: memcpy: only use dcbz once cache is enabled Christophe Leroy
@ 2015-09-07 14:50 ` Michal Sojka
  2015-09-07 15:04 ` David Laight
  2015-09-08  3:56 ` Michael Ellerman
  2 siblings, 0 replies; 4+ messages in thread
From: Michal Sojka @ 2015-09-07 14:50 UTC (permalink / raw)
  To: Christophe Leroy, Benjamin Herrenschmidt, Paul Mackerras,
	Michael Ellerman, scottwood
  Cc: linux-kernel, linuxppc-dev

On Mon, Sep 07 2015, Christophe Leroy wrote:
> memcpy() uses instruction dcbz to speed up copy by not wasting time
> loading cache line with data that will be overwritten.
> Some platform like mpc52xx do no have cache active at startup and
> can therefore not use memcpy(). Allthough no part of the code
> explicitly uses memcpy(), GCC makes calls to it.
>
> This patch modifies memcpy() such that at startup, the 'dcbz'
> instruction is replaced by 'dcbt' which is harmless if cache is not
> enabled, and which helps a bit (allthough not as much as dcbz) if
> cache is already enabled.
>
> Once the initial MMU is setup, in machine_init() we patch memcpy()
> by replacing the temporary 'dcbt' by 'dcbz'
>
> Reported-by: Michal Sojka <sojkam1@fel.cvut.cz>
> Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
> ---
> @Michal, can you please test it ?

Yes, it works.

Tested-by: Michal Sojka <sojkam1@fel.cvut.cz>

-Michal

^ permalink raw reply	[flat|nested] 4+ messages in thread

* RE: [PATCH] powerpc32: memcpy: only use dcbz once cache is enabled
  2015-09-07 14:24 [PATCH] powerpc32: memcpy: only use dcbz once cache is enabled Christophe Leroy
  2015-09-07 14:50 ` Michal Sojka
@ 2015-09-07 15:04 ` David Laight
  2015-09-08  3:56 ` Michael Ellerman
  2 siblings, 0 replies; 4+ messages in thread
From: David Laight @ 2015-09-07 15:04 UTC (permalink / raw)
  To: 'Christophe Leroy',
	Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	scottwood, sojkam1
  Cc: linuxppc-dev, linux-kernel

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 955 bytes --]

From: Christophe Leroy
> Sent: 07 September 2015 15:25
...
> diff --git a/arch/powerpc/lib/copy_32.S b/arch/powerpc/lib/copy_32.S
> index 2ef50c6..05b3096 100644
> --- a/arch/powerpc/lib/copy_32.S
> +++ b/arch/powerpc/lib/copy_32.S
> @@ -172,7 +172,16 @@ _GLOBAL(memcpy)
>  	mtctr	r0
>  	beq	63f
>  53:
> -	dcbz	r11,r6
> +	/*
> +	 * During early init, cache might not be active yet, so dcbz cannot be
> +	 * used. We put dcbt instead of dcbz. If cache is not active, it's just
> +	 * like a not. If cache is active, at least it prefetchs the line to be
                ^^^ nop ??

	David

> +	 * overwritten.
> +	 * Will be replaced by dcbz in machine_init()
> +	 */
> +_GLOBAL(ppc32_memcpy_dcbz)
> +	dcbt	r11,r6
> +
>  	COPY_16_BYTES
>  #if L1_CACHE_BYTES >= 32
>  	COPY_16_BYTES
> --
> 2.1.0

ÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a­¢f£¢·hšïêÿ‘êçz_è®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨è­Ú&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] powerpc32: memcpy: only use dcbz once cache is enabled
  2015-09-07 14:24 [PATCH] powerpc32: memcpy: only use dcbz once cache is enabled Christophe Leroy
  2015-09-07 14:50 ` Michal Sojka
  2015-09-07 15:04 ` David Laight
@ 2015-09-08  3:56 ` Michael Ellerman
  2 siblings, 0 replies; 4+ messages in thread
From: Michael Ellerman @ 2015-09-08  3:56 UTC (permalink / raw)
  To: Christophe Leroy
  Cc: Benjamin Herrenschmidt, Paul Mackerras, scottwood, sojkam1,
	linux-kernel, linuxppc-dev

On Mon, 2015-09-07 at 16:24 +0200, Christophe Leroy wrote:
> memcpy() uses instruction dcbz to speed up copy by not wasting time
> loading cache line with data that will be overwritten.
> Some platform like mpc52xx do no have cache active at startup and
> can therefore not use memcpy(). Allthough no part of the code
> explicitly uses memcpy(), GCC makes calls to it.
> 
> This patch modifies memcpy() such that at startup, the 'dcbz'
> instruction is replaced by 'dcbt' which is harmless if cache is not
> enabled, and which helps a bit (allthough not as much as dcbz) if
> cache is already enabled.
> 
> Once the initial MMU is setup, in machine_init() we patch memcpy()
> by replacing the temporary 'dcbt' by 'dcbz'
> 
> Reported-by: Michal Sojka <sojkam1@fel.cvut.cz>
> Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>


Is there some reason you can't use the normal cpu feature sections? See
arch/powerpc/lib/memcpy_64.S for an example of what I mean.

cheers



^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2015-09-08  3:56 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-09-07 14:24 [PATCH] powerpc32: memcpy: only use dcbz once cache is enabled Christophe Leroy
2015-09-07 14:50 ` Michal Sojka
2015-09-07 15:04 ` David Laight
2015-09-08  3:56 ` Michael Ellerman

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).