Karmic fsl-imx51 SRU: only export NEON flag to userspace on Freescale iMX51 rev3.x or later silicon

[Bryan Wu]

Andy Whitcroft 写道:
> On Thu, Feb 11, 2010 at 02:08:44PM -0700, Tim Gardner wrote:
>> The following changes since commit 4b1196706272826e74e35e9ebf2cc94ba8d4062c:
>>   Leann Ogasawara (1):
>>         UBUNTU: Ubuntu-2.6.31-108.21
>>
>> are available in the git repository at:
>>
>>   git://kernel.ubuntu.com/rtg/ubuntu-karmic fsl-imx51-neon
>>
>> Bryan Wu (1):
>>       UBUNTU: SAUCE: IMX51: only export NEON flag to userspace on Freescale iMX51 rev3.x or later silicon
>>
>>  arch/arm/vfp/vfpmodule.c |    7 ++++++-
>>  1 files changed, 6 insertions(+), 1 deletions(-)
>>
>> From d77e3635c6d402cebdda49f654402c67460acfab Mon Sep 17 00:00:00 2001
>> From: Bryan Wu <bryan.wu at canonical.com>
>> Date: Tue, 26 Jan 2010 19:00:22 +0000
>> Subject: [PATCH] UBUNTU: SAUCE: IMX51: only export NEON flag to userspace on Freescale iMX51 rev3.x or later silicon
>>
>> BugLink: http://bugs.launchpad.net/bugs/507416
>>
>> NEON function is broken on old Freescale iMX51 silicons, such as rev1.x and
>> rev2.x. Since more and more iMX51 based products will move to rev3.x or later
>> silicon, we need to enable NEON on these newer silicons and disable NEON on
>> those old ones.
>>
>> This patch will detect the silicon revision dynamically and only export NEON
>> flag to userspace on Freescale iMX51 rev3.x or later silicon. It was tested on
>> Babbage 3.0 board and Babbage 2.x board.
>>
>> Signed-off-by: Bryan Wu <bryan.wu at canonical.com>
>> Signed-off-by: Andy Whitcroft <apw at canonical.com>
>> Signed-off-by: Tim Gardner <tim.gardner at canonical.com>
>> ---
>>  arch/arm/vfp/vfpmodule.c |    7 ++++++-
>>  1 files changed, 6 insertions(+), 1 deletions(-)
>>
>> diff --git a/arch/arm/vfp/vfpmodule.c b/arch/arm/vfp/vfpmodule.c
>> index 2d7423a..e0d6b01 100644
>> --- a/arch/arm/vfp/vfpmodule.c
>> +++ b/arch/arm/vfp/vfpmodule.c
>> @@ -432,6 +432,10 @@ out:
>>  
>>  #include <linux/smp.h>
>>  
>> +#if defined(CONFIG_ARCH_MX51) && defined(CONFIG_NEON)
>> +#include <mach/hardware.h>
>> +#endif
>> +
>>  /*
>>   * VFP support code initialisation.
>>   */
>> @@ -498,7 +502,8 @@ static int __init vfp_init(void)
>>  		 * load/store instructions, integer and single
>>  		 * precision floating point operations.
>>  		 */
>> -		if ((fmrx(MVFR1) & 0x000fff00) == 0x00011100)
>> +		if (((fmrx(MVFR1) & 0x000fff00) == 0x00011100)
>> +			&& cpu_is_mx51_rev(CHIP_REV_3_0) > 0)
>>  			elf_hwcap |= HWCAP_NEON;
>>  #endif
> 
> Although this looks perfectly good for the fsl-imx51 branch, this
> appears wrong for non-imx51 use.  I think we should be programming with
> a view to this being merged into mainline at some point and not be
> adding anything which makes that harder.  Something more like this (with
> the appropriate incantation in the first stanza):
> 
>  && (!IMX51 || cpu_is_mx51_rev(CHIP_REV_3_0) > 0)
> 
> -apw
> 

Actually, at the beginning I just wanna to provide a workaround for this dirty 
silicon issue. I was not very sure about our upstream approach at that time. It 
is pretty sure the compiling will fail on other !ARCH_MX51.

If we just keep it locally, it is OK just for fsl-imx51 branch. But if we are 
going to push it to upstream since currently Amit's imx51 patches will be merged 
into mainline soon, I do like to rework it to make it in right way.

Cheers,
-- 
Bryan Wu <bryan.wu at canonical.com>
Kernel Developer    +86.138-1617-6545 Mobile
Ubuntu Kernel Team | Hardware Enablement Team
Canonical Ltd.      www.canonical.com
Ubuntu - Linux for human beings | www.ubuntu.com