[U-Boot] [PATCH] arm: lib: memcpy: Do not copy to same address
In some cases (e.g. bootm with a elf payload which is already at the right position) there is a in place copy of data to the same address. Catching this saves some ms while booting.
Signed-off-by: Matthias Weisser weisserm@arcor.de --- arch/arm/lib/memcpy.S | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/arch/arm/lib/memcpy.S b/arch/arm/lib/memcpy.S index 3b5aeec..f655256 100644 --- a/arch/arm/lib/memcpy.S +++ b/arch/arm/lib/memcpy.S @@ -60,6 +60,9 @@ .globl memcpy memcpy:
+ cmp r0, r1 + moveq pc, lr + enter r4, lr
subs r2, r2, #4
Am 23.05.2011 11:06, schrieb Matthias Weisser:
In some cases (e.g. bootm with a elf payload which is already at the right position) there is a in place copy of data to the same address. Catching this saves some ms while booting.
Signed-off-by: Matthias Weisserweisserm@arcor.de
arch/arm/lib/memcpy.S | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/arch/arm/lib/memcpy.S b/arch/arm/lib/memcpy.S index 3b5aeec..f655256 100644 --- a/arch/arm/lib/memcpy.S +++ b/arch/arm/lib/memcpy.S @@ -60,6 +60,9 @@ .globl memcpy memcpy:
cmp r0, r1moveq pc, lrenter r4, lr
subs r2, r2, #4
The standard clearly say to both memory regions should not overlap when memcpy() is used, so I would say this is the wrong place to fix that.
Regards,
Alexander
Am 23.05.2011 11:30, schrieb Alexander Holler:
Am 23.05.2011 11:06, schrieb Matthias Weisser:
In some cases (e.g. bootm with a elf payload which is already at the right position) there is a in place copy of data to the same address. Catching this saves some ms while booting.
Signed-off-by: Matthias Weisserweisserm@arcor.de
arch/arm/lib/memcpy.S | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/arch/arm/lib/memcpy.S b/arch/arm/lib/memcpy.S index 3b5aeec..f655256 100644 --- a/arch/arm/lib/memcpy.S +++ b/arch/arm/lib/memcpy.S @@ -60,6 +60,9 @@ .globl memcpy memcpy:
cmp r0, r1moveq pc, lrenter r4, lr subs r2, r2, #4The standard clearly say to both memory regions should not overlap when memcpy() is used, so I would say this is the wrong place to fix that.
Well, real world applications do this. And these two instructions shouldn't hurt a lot.
I first send a patch fixing only "my" problem in cmd_elf.c but Wolfgang suggested to do this globally. Please see http://www.mail-archive.com/u-boot@lists.denx.de/msg50612.html as reference.
Matthias
Am 23.05.2011 11:46, schrieb Matthias Weißer:
The standard clearly say to both memory regions should not overlap when memcpy() is used, so I would say this is the wrong place to fix that.
Well, real world applications do this. And these two instructions shouldn't hurt a lot.
Real bugs to this. Just see e.g the long discussion about some changes fo memcpy done in the glibc lately and what that did for flash-users because flash seemed to the same stupid stuff.
Regards,
Alexander
Le 23/05/2011 11:30, Alexander Holler a écrit :
Am 23.05.2011 11:06, schrieb Matthias Weisser:
In some cases (e.g. bootm with a elf payload which is already at the right position) there is a in place copy of data to the same address. Catching this saves some ms while booting.
Signed-off-by: Matthias Weisserweisserm@arcor.de
arch/arm/lib/memcpy.S | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/arch/arm/lib/memcpy.S b/arch/arm/lib/memcpy.S index 3b5aeec..f655256 100644 --- a/arch/arm/lib/memcpy.S +++ b/arch/arm/lib/memcpy.S @@ -60,6 +60,9 @@ .globl memcpy memcpy:
cmp r0, r1moveq pc, lrenter r4, lr subs r2, r2, #4The standard clearly say to both memory regions should not overlap when memcpy() is used, so I would say this is the wrong place to fix that.
I think the intent here is not to enforce the standard but to handle an actual, and degenerate, copy request in the most efficient manner, and in that respect, the patch does its job.
Besides, if the patch was about enforcing the standard, then I would consider it highly more efficient to check the areas once in the memcpy function than duplicating this check before each call to the function, considering that the place where the copy happens is not necessarily the one where the source and destination were computed.
Regards,
Alexander
Amicalement,
Hello,
Am 23.05.2011 11:49, schrieb Albert ARIBAUD:
The standard clearly say to both memory regions should not overlap when memcpy() is used, so I would say this is the wrong place to fix that.
I think the intent here is not to enforce the standard but to handle an actual, and degenerate, copy request in the most efficient manner, and in that respect, the patch does its job.
Besides, if the patch was about enforcing the standard, then I would consider it highly more efficient to check the areas once in the memcpy function than duplicating this check before each call to the function, considering that the place where the copy happens is not necessarily the one where the source and destination were computed.
A fool proof solution would be to always use memmove() and get rid of memcpy(). But checking for overlapped regions in memcpy() is imho the wrong way. This just leads to more possible wrong code which uses memcpy() when it should use memmove().
Regards,
Alexander
Dear Albert
Am 23.05.2011 11:49, schrieb Albert ARIBAUD:
Le 23/05/2011 11:30, Alexander Holler a écrit :
Am 23.05.2011 11:06, schrieb Matthias Weisser:
In some cases (e.g. bootm with a elf payload which is already at the right position) there is a in place copy of data to the same address. Catching this saves some ms while booting.
Signed-off-by: Matthias Weisserweisserm@arcor.de
arch/arm/lib/memcpy.S | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-)diff --git a/arch/arm/lib/memcpy.S b/arch/arm/lib/memcpy.S index 3b5aeec..f655256 100644 --- a/arch/arm/lib/memcpy.S +++ b/arch/arm/lib/memcpy.S @@ -60,6 +60,9 @@ .globl memcpy memcpy:
cmp r0, r1moveq pc, lrenter r4, lr subs r2, r2, #4The standard clearly say to both memory regions should not overlap when memcpy() is used, so I would say this is the wrong place to fix that.
I think the intent here is not to enforce the standard but to handle an actual, and degenerate, copy request in the most efficient manner, and in that respect, the patch does its job.
Can this patch go in or do I need to change anything? I really would like to see it in mainline.
Regards, Matthias
On 23/05/2011 11:06, Matthias Weisser wrote:
In some cases (e.g. bootm with a elf payload which is already at the right position) there is a in place copy of data to the same address. Catching this saves some ms while booting.
Signed-off-by: Matthias Weisserweisserm@arcor.de
arch/arm/lib/memcpy.S | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/arch/arm/lib/memcpy.S b/arch/arm/lib/memcpy.S index 3b5aeec..f655256 100644 --- a/arch/arm/lib/memcpy.S +++ b/arch/arm/lib/memcpy.S @@ -60,6 +60,9 @@ .globl memcpy memcpy:
cmp r0, r1moveq pc, lrenter r4, lr
subs r2, r2, #4
Applied to u-boot-arm/master, thanks.
Amicalement,
participants (4)
-
Albert ARIBAUD -
Alexander Holler -
Matthias Weisser -
Matthias Weißer