[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[jc at eclis.ch]

From: Jean-Christian de Rivaz <jc@eclis.ch>

According to Russell King:
In any case, unless you have a Marvell platform, you should have this
disabled - it provides a performance boost for Marvell platforms, but
degrades performance everywhere else.  Please ensure that you disable
CONFIG_UACCESS_WITH_MEMCPY.

Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>
---
 arch/arm/configs/at91_dt_defconfig |    2 +-
 arch/arm/configs/sama5_defconfig   |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/arm/configs/at91_dt_defconfig b/arch/arm/configs/at91_dt_defconfig
index 090c5b2..1807d7f 100644
--- a/arch/arm/configs/at91_dt_defconfig
+++ b/arch/arm/configs/at91_dt_defconfig
@@ -21,7 +21,7 @@ CONFIG_SOC_SAM_V4_V5=y
 CONFIG_SOC_AT91RM9200=y
 CONFIG_SOC_AT91SAM9=y
 CONFIG_AEABI=y
-CONFIG_UACCESS_WITH_MEMCPY=y
+# CONFIG_UACCESS_WITH_MEMCPY is not set
 CONFIG_ZBOOT_ROM_TEXT=0x0
 CONFIG_ZBOOT_ROM_BSS=0x0
 CONFIG_ARM_APPENDED_DTB=y
diff --git a/arch/arm/configs/sama5_defconfig b/arch/arm/configs/sama5_defconfig
index 31eb951..471e214 100644
--- a/arch/arm/configs/sama5_defconfig
+++ b/arch/arm/configs/sama5_defconfig
@@ -19,7 +19,7 @@ CONFIG_SOC_SAM_V7=y
 CONFIG_SOC_SAMA5D3=y
 CONFIG_SOC_SAMA5D4=y
 CONFIG_AEABI=y
-CONFIG_UACCESS_WITH_MEMCPY=y
+# CONFIG_UACCESS_WITH_MEMCPY is not set
 CONFIG_ZBOOT_ROM_TEXT=0x0
 CONFIG_ZBOOT_ROM_BSS=0x0
 CONFIG_ARM_APPENDED_DTB=y
-- 
1.7.10.4

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Nicolas Pitre]

On Tue, 29 Sep 2015, jc at eclis.ch wrote:

> From: Jean-Christian de Rivaz <jc@eclis.ch>
> 
> According to Russell King:
> In any case, unless you have a Marvell platform, you should have this
> disabled - it provides a performance boost for Marvell platforms, but
> degrades performance everywhere else.  Please ensure that you disable
> CONFIG_UACCESS_WITH_MEMCPY.
> 
> Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>

Acked-by: Nicolas Pitre <nico@linaro.org>


> ---
>  arch/arm/configs/at91_dt_defconfig |    2 +-
>  arch/arm/configs/sama5_defconfig   |    2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm/configs/at91_dt_defconfig b/arch/arm/configs/at91_dt_defconfig
> index 090c5b2..1807d7f 100644
> --- a/arch/arm/configs/at91_dt_defconfig
> +++ b/arch/arm/configs/at91_dt_defconfig
> @@ -21,7 +21,7 @@ CONFIG_SOC_SAM_V4_V5=y
>  CONFIG_SOC_AT91RM9200=y
>  CONFIG_SOC_AT91SAM9=y
>  CONFIG_AEABI=y
> -CONFIG_UACCESS_WITH_MEMCPY=y
> +# CONFIG_UACCESS_WITH_MEMCPY is not set
>  CONFIG_ZBOOT_ROM_TEXT=0x0
>  CONFIG_ZBOOT_ROM_BSS=0x0
>  CONFIG_ARM_APPENDED_DTB=y
> diff --git a/arch/arm/configs/sama5_defconfig b/arch/arm/configs/sama5_defconfig
> index 31eb951..471e214 100644
> --- a/arch/arm/configs/sama5_defconfig
> +++ b/arch/arm/configs/sama5_defconfig
> @@ -19,7 +19,7 @@ CONFIG_SOC_SAM_V7=y
>  CONFIG_SOC_SAMA5D3=y
>  CONFIG_SOC_SAMA5D4=y
>  CONFIG_AEABI=y
> -CONFIG_UACCESS_WITH_MEMCPY=y
> +# CONFIG_UACCESS_WITH_MEMCPY is not set
>  CONFIG_ZBOOT_ROM_TEXT=0x0
>  CONFIG_ZBOOT_ROM_BSS=0x0
>  CONFIG_ARM_APPENDED_DTB=y
> -- 
> 1.7.10.4
> 
> 

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Uwe Kleine-König]

Hello,

On Tue, Sep 29, 2015 at 10:03:07PM +0200, jc at eclis.ch wrote:
> From: Jean-Christian de Rivaz <jc@eclis.ch>
> 
> According to Russell King:
> In any case, unless you have a Marvell platform, you should have this
> disabled - it provides a performance boost for Marvell platforms, but
> degrades performance everywhere else.  Please ensure that you disable
> CONFIG_UACCESS_WITH_MEMCPY.
> 
> Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>
> ---
>  arch/arm/configs/at91_dt_defconfig |    2 +-
>  arch/arm/configs/sama5_defconfig   |    2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm/configs/at91_dt_defconfig b/arch/arm/configs/at91_dt_defconfig
> index 090c5b2..1807d7f 100644
> --- a/arch/arm/configs/at91_dt_defconfig
> +++ b/arch/arm/configs/at91_dt_defconfig
> @@ -21,7 +21,7 @@ CONFIG_SOC_SAM_V4_V5=y
>  CONFIG_SOC_AT91RM9200=y
>  CONFIG_SOC_AT91SAM9=y
>  CONFIG_AEABI=y
> -CONFIG_UACCESS_WITH_MEMCPY=y
> +# CONFIG_UACCESS_WITH_MEMCPY is not set
>  CONFIG_ZBOOT_ROM_TEXT=0x0
>  CONFIG_ZBOOT_ROM_BSS=0x0
>  CONFIG_ARM_APPENDED_DTB=y
What is the reason to add the "# ... is not set" line? Given the default
line of UACCESS_WITH_MEMCPY I suggest to just drop the line without
replacement in the defconfigs.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Jean-Christian de Rivaz]

Le 29. 09. 15 22:35, Uwe Kleine-K?nig a ?crit :
> Hello,
>
> On Tue, Sep 29, 2015 at 10:03:07PM +0200, jc at eclis.ch wrote:
>> From: Jean-Christian de Rivaz <jc@eclis.ch>
>>
>> According to Russell King:
>> In any case, unless you have a Marvell platform, you should have this
>> disabled - it provides a performance boost for Marvell platforms, but
>> degrades performance everywhere else.  Please ensure that you disable
>> CONFIG_UACCESS_WITH_MEMCPY.
>>
>> Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>
>> ---
>>   arch/arm/configs/at91_dt_defconfig |    2 +-
>>   arch/arm/configs/sama5_defconfig   |    2 +-
>>   2 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/arm/configs/at91_dt_defconfig b/arch/arm/configs/at91_dt_defconfig
>> index 090c5b2..1807d7f 100644
>> --- a/arch/arm/configs/at91_dt_defconfig
>> +++ b/arch/arm/configs/at91_dt_defconfig
>> @@ -21,7 +21,7 @@ CONFIG_SOC_SAM_V4_V5=y
>>   CONFIG_SOC_AT91RM9200=y
>>   CONFIG_SOC_AT91SAM9=y
>>   CONFIG_AEABI=y
>> -CONFIG_UACCESS_WITH_MEMCPY=y
>> +# CONFIG_UACCESS_WITH_MEMCPY is not set
>>   CONFIG_ZBOOT_ROM_TEXT=0x0
>>   CONFIG_ZBOOT_ROM_BSS=0x0
>>   CONFIG_ARM_APPENDED_DTB=y
> What is the reason to add the "# ... is not set" line? Given the default
> line of UACCESS_WITH_MEMCPY I suggest to just drop the line without
> replacement in the defconfigs.
>

Hi Uwe,

Without the explicit disabling, the user will be asked and could 
possibly give the wrong answer as the help look as a cool feature and 
the condition is very hard to understand.

Regards,
Jean-Christian

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Alexandre Belloni]

Hi,

On 29/09/2015 at 22:59:46 +0200, Jean-Christian de Rivaz wrote :
> Without the explicit disabling, the user will be asked and could possibly
> give the wrong answer as the help look as a cool feature and the condition
> is very hard to understand.
> 

That is not true, if you remove it from the defconfig, it will simply be
disabled in the final config.

Moreover, I would prefer a better commit message than "Russell said so"

-- 
Alexandre Belloni, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Russell King - ARM Linux]

On Wed, Sep 30, 2015 at 12:01:40AM +0200, Alexandre Belloni wrote:
> Hi,
> 
> On 29/09/2015 at 22:59:46 +0200, Jean-Christian de Rivaz wrote :
> > Without the explicit disabling, the user will be asked and could possibly
> > give the wrong answer as the help look as a cool feature and the condition
> > is very hard to understand.
> > 
> 
> That is not true, if you remove it from the defconfig, it will simply be
> disabled in the final config.
> 
> Moreover, I would prefer a better commit message than "Russell said so"

Well, if people are going to take that kind of attitude, then I'm not
going to bother fixing it at all.  I've got better things to do than
chase after bugs in almost unused code and code which people shouldn't
be enabling on their platforms.

It's not "Russell said so" but "Nicolas, the author of the code, said so."
I'm merely passing the information along.  Search the archives if you don't
believe me, it's been discussed in the last couple of months, including
potentially removing this code.

I regard this code as a maintanence burden, one that I personally can do
without.

-- 
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[jc at eclis.ch]

From: Jean-Christian de Rivaz <jc@eclis.ch>

According to Russell King:
In any case, unless you have a Marvell platform, you should have this
disabled - it provides a performance boost for Marvell platforms, but
degrades performance everywhere else.  Please ensure that you disable
CONFIG_UACCESS_WITH_MEMCPY.

Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>
---
 arch/arm/configs/at91_dt_defconfig |    1 -
 arch/arm/configs/sama5_defconfig   |    1 -
 2 files changed, 2 deletions(-)

diff --git a/arch/arm/configs/at91_dt_defconfig b/arch/arm/configs/at91_dt_defconfig
index 090c5b2..8a374e9 100644
--- a/arch/arm/configs/at91_dt_defconfig
+++ b/arch/arm/configs/at91_dt_defconfig
@@ -21,7 +21,6 @@ CONFIG_SOC_SAM_V4_V5=y
 CONFIG_SOC_AT91RM9200=y
 CONFIG_SOC_AT91SAM9=y
 CONFIG_AEABI=y
-CONFIG_UACCESS_WITH_MEMCPY=y
 CONFIG_ZBOOT_ROM_TEXT=0x0
 CONFIG_ZBOOT_ROM_BSS=0x0
 CONFIG_ARM_APPENDED_DTB=y
diff --git a/arch/arm/configs/sama5_defconfig b/arch/arm/configs/sama5_defconfig
index 31eb951..84fc70f 100644
--- a/arch/arm/configs/sama5_defconfig
+++ b/arch/arm/configs/sama5_defconfig
@@ -19,7 +19,6 @@ CONFIG_SOC_SAM_V7=y
 CONFIG_SOC_SAMA5D3=y
 CONFIG_SOC_SAMA5D4=y
 CONFIG_AEABI=y
-CONFIG_UACCESS_WITH_MEMCPY=y
 CONFIG_ZBOOT_ROM_TEXT=0x0
 CONFIG_ZBOOT_ROM_BSS=0x0
 CONFIG_ARM_APPENDED_DTB=y
-- 
1.7.10.4

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Alexandre Belloni]

On 29/09/2015 at 23:18:01 +0100, Russell King - ARM Linux wrote :
> On Wed, Sep 30, 2015 at 12:01:40AM +0200, Alexandre Belloni wrote:
> > Hi,
> > 
> > On 29/09/2015 at 22:59:46 +0200, Jean-Christian de Rivaz wrote :
> > > Without the explicit disabling, the user will be asked and could possibly
> > > give the wrong answer as the help look as a cool feature and the condition
> > > is very hard to understand.
> > > 
> > 
> > That is not true, if you remove it from the defconfig, it will simply be
> > disabled in the final config.
> > 
> > Moreover, I would prefer a better commit message than "Russell said so"
> 
> Well, if people are going to take that kind of attitude, then I'm not
> going to bother fixing it at all.  I've got better things to do than
> chase after bugs in almost unused code and code which people shouldn't
> be enabling on their platforms.
> 
> It's not "Russell said so" but "Nicolas, the author of the code, said so."
> I'm merely passing the information along.  Search the archives if you don't
> believe me, it's been discussed in the last couple of months, including
> potentially removing this code.
> 
> I regard this code as a maintanence burden, one that I personally can do
> without.
> 

Come on, I never said it was not worth fixing, I'm just saying the
commit message doesn't carry the proper explanation. I would prefer
something a bit more technical.
A better commit message could point to the discussion you are referring
to for example so that the maintainers (btw they are not in copy) don't
have to look it up themselves.


-- 
Alexandre Belloni, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Nicolas Pitre]

On Wed, 30 Sep 2015, Alexandre Belloni wrote:

> On 29/09/2015 at 23:18:01 +0100, Russell King - ARM Linux wrote :
> > On Wed, Sep 30, 2015 at 12:01:40AM +0200, Alexandre Belloni wrote:
> > > Hi,
> > > 
> > > On 29/09/2015 at 22:59:46 +0200, Jean-Christian de Rivaz wrote :
> > > > Without the explicit disabling, the user will be asked and could possibly
> > > > give the wrong answer as the help look as a cool feature and the condition
> > > > is very hard to understand.
> > > > 
> > > 
> > > That is not true, if you remove it from the defconfig, it will simply be
> > > disabled in the final config.
> > > 
> > > Moreover, I would prefer a better commit message than "Russell said so"
> > 
> > Well, if people are going to take that kind of attitude, then I'm not
> > going to bother fixing it at all.  I've got better things to do than
> > chase after bugs in almost unused code and code which people shouldn't
> > be enabling on their platforms.
> > 
> > It's not "Russell said so" but "Nicolas, the author of the code, said so."
> > I'm merely passing the information along.  Search the archives if you don't
> > believe me, it's been discussed in the last couple of months, including
> > potentially removing this code.
> > 
> > I regard this code as a maintanence burden, one that I personally can do
> > without.
> > 
> 
> Come on, I never said it was not worth fixing, I'm just saying the
> commit message doesn't carry the proper explanation. I would prefer
> something a bit more technical.
> A better commit message could point to the discussion you are referring
> to for example so that the maintainers (btw they are not in copy) don't
> have to look it up themselves.

FYI: here is the thread:

http://article.gmane.org/gmane.linux.ports.arm.kernel/433489


Nicolas

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Uwe Kleine-König]

Hello,

On Wed, Sep 30, 2015 at 12:32:48AM +0200, jc at eclis.ch wrote:
> From: Jean-Christian de Rivaz <jc@eclis.ch>
> 
> According to Russell King:
> In any case, unless you have a Marvell platform, you should have this
> disabled - it provides a performance boost for Marvell platforms, but
> degrades performance everywhere else.  Please ensure that you disable
> CONFIG_UACCESS_WITH_MEMCPY.
> 
> Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>
the change is fine now. Feel free to add my Ack also if you do another
round to improve the changelog.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |

[PATCH 1/1] Disable CONFIG_UACCESS_WITH_MEMCPY in Atmel config.

[Jean-Christian de Rivaz]

Le 30. 09. 15 08:10, Uwe Kleine-K?nig a ?crit :
> Hello,
>
> On Wed, Sep 30, 2015 at 12:32:48AM +0200, jc at eclis.ch wrote:
>> From: Jean-Christian de Rivaz <jc@eclis.ch>
>>
>> According to Russell King:
>> In any case, unless you have a Marvell platform, you should have this
>> disabled - it provides a performance boost for Marvell platforms, but
>> degrades performance everywhere else.  Please ensure that you disable
>> CONFIG_UACCESS_WITH_MEMCPY.
>>
>> Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>
> the change is fine now. Feel free to add my Ack also if you do another
> round to improve the changelog.
>
> Best regards
> Uwe
>

Thanks Uwe,

Nicolas and Russell, for my own curiosity I tested the difference on 
this system:

# cat /proc/cpuinfo
processor    : 0
model name    : ARMv7 Processor rev 1 (v7l)
BogoMIPS    : 348.97
Features    : half thumb fastmult vfp edsp vfpv3 vfpv3d16 tls vfpv4
CPU implementer    : 0x41
CPU architecture: 7
CPU variant    : 0x0
CPU part    : 0xc05
CPU revision    : 1

Hardware    : Atmel SAMA5
Revision    : 0000
Serial        : 0000000000000000

# dd if=/dev/zero of=/dev/null bs=4k count=1M

Take about 10 seconds to complete on this system and report the 
following speed in MB/s in 10 rounds:

CONFIG_UACCESS_WITH_MEMCPY=y
470, 419, 440, 446, 434, 437, 418, 412, 431, 437

# CONFIG_UACCESS_WITH_MEMCPY disabled is not set
387, 387, 382, 383, 385, 388, 387, 387, 384, 386

So It look like this feature still give some advantage on system other 
than Orion CPU core. I still propose my patch to disable it until it's 
rock stable as it used to be a least with the kernel 3.19.0.

Regards,
Jean-Christian