2009-04-29 13:55:02

by Michal Simek

[permalink] [raw]
Subject: Memory limits - mm_segment_t - MAKE_MM_SEG

Hi all,

I look at some things which I need to clear for MMU Microblaze patches and
I would like to know your opinion about.

First of all I found that almost all archs use MAKE_MM_SEG macro which should be good
to move to generic location (asm-generic/uaccess.h ? )
#define MAKE_MM_SEG(s) ((mm_segment_t) { (s) })


The second thing is about place where is stored limit for processes -> mm_segment_t structure

Where is the proper location for storing mm_segment_t? Some arch use thread_info some of them
thread_struct

For microblaze case are used both -> that's definitely not correct.

Here is the small table for cpus which are in linux kernel and location and type for them.
The most of them uses thread_info structure for it and name is addr_limit.
The location of mm_segment_t is different too -> we should move it to any generic location too.
What do you think?

location name location struct mm_segment_t
s390: processor.h/thread_struct mm_segment processor.h __u32 ar4
alpha: thread_info.h/thread_info addr_limit processor.h unsigned long seg
x86: thread_info.h/thread_info addr_limit processor.h unsigned long seg
ia64: thread_info.h/thread_info addr_limit processor.h unsigned long seg
mips: thread_info.h/thread_info addr_limit processor.h unsigned long seg
um: thread_info.h/thread_info addr_limit uaccess.h unsigned long seg
frv: thread_info.h/thread_info addr_limit segment.h unsigned long seg
cris: thread_info.h/thread_info addr_limit segment.h unsigned long seg
sh: thread_info.h/thread_info addr_limit segment.h unsigned long seg
h8300: not_used not_used segment.h unsigned long seg
nm10300:thread_info.h/thread_info addr_limit processor.h unsigned long seg
arm: thread_info.h/thread_info addr_limit thread_info.h typedef unsigned long mm_segment_t;
parisc: thread_info.h/thread_info addr_limit processor.h int seq
m32r: thread_info.h/thread_info addr_limit processor.h unsigned long seg
xtensa: thread_info.h/thread_info addr_limit segment.h unsigned long seg
avr32: not_used not_used uaccess.h unsigned int is_user_space;
sparc: processor_32.h/thread_struct current_ds processor_32.h int seg
m68k: not_used not_used segment.h unsigned long seg
blackfin:thread_info.h/thread_info addr_limit thread_info.h typedef unsigned long mm_segment_t;
powerpc: processor.h/thread_struct current_ds processor.h unsigned long seg


There is really mess in it and I would like to have good solution for Microblaze and I think
that will be good to have any generic solution and remove code duplication.

Thanks for your comments,
Michal


--
Michal Simek, Ing. (M.Eng)
w: http://www.monstr.eu p: +42-0-721842854


2009-04-29 14:03:32

by Michal Simek

[permalink] [raw]
Subject: Re: Memory limits - mm_segment_t - MAKE_MM_SEG

Michal Simek wrote:
> Hi all,
>
> I look at some things which I need to clear for MMU Microblaze patches and
> I would like to know your opinion about.
>
> First of all I found that almost all archs use MAKE_MM_SEG macro which should be good
> to move to generic location (asm-generic/uaccess.h ? )
> #define MAKE_MM_SEG(s) ((mm_segment_t) { (s) })
>
>
> The second thing is about place where is stored limit for processes -> mm_segment_t structure
>
> Where is the proper location for storing mm_segment_t? Some arch use thread_info some of them
> thread_struct
>
> For microblaze case are used both -> that's definitely not correct.
>
> Here is the small table for cpus which are in linux kernel and location and type for them.
> The most of them uses thread_info structure for it and name is addr_limit.
> The location of mm_segment_t is different too -> we should move it to any generic location too.
> What do you think?
>
> location name location struct mm_segment_t
> s390: processor.h/thread_struct mm_segment processor.h __u32 ar4
> alpha: thread_info.h/thread_info addr_limit processor.h unsigned long seg
> x86: thread_info.h/thread_info addr_limit processor.h unsigned long seg
> ia64: thread_info.h/thread_info addr_limit processor.h unsigned long seg
> mips: thread_info.h/thread_info addr_limit processor.h unsigned long seg
> um: thread_info.h/thread_info addr_limit uaccess.h unsigned long seg
> frv: thread_info.h/thread_info addr_limit segment.h unsigned long seg
> cris: thread_info.h/thread_info addr_limit segment.h unsigned long seg
> sh: thread_info.h/thread_info addr_limit segment.h unsigned long seg
> h8300: not_used not_used segment.h unsigned long seg
> nm10300:thread_info.h/thread_info addr_limit processor.h unsigned long seg
> arm: thread_info.h/thread_info addr_limit thread_info.h typedef unsigned long mm_segment_t;
> parisc: thread_info.h/thread_info addr_limit processor.h int seq
> m32r: thread_info.h/thread_info addr_limit processor.h unsigned long seg
> xtensa: thread_info.h/thread_info addr_limit segment.h unsigned long seg
> avr32: not_used not_used uaccess.h unsigned int is_user_space;
> sparc: processor_32.h/thread_struct current_ds processor_32.h int seg
> m68k: not_used not_used segment.h unsigned long seg
> blackfin:thread_info.h/thread_info addr_limit thread_info.h typedef unsigned long mm_segment_t;
> powerpc: processor.h/thread_struct current_ds processor.h unsigned long seg
>
>
> There is really mess in it and I would like to have good solution for Microblaze and I think
> that will be good to have any generic solution and remove code duplication.
>
> Thanks for your comments,
> Michal
>
>
+ move some macros to generic location too.
#define get_ds() (KERNEL_DS)

+ get_fs(), set_fs(), segment_eq() after sync changes above

Michal





--
Michal Simek, Ing. (M.Eng)
PetaLogix - Linux Solutions for a Reconfigurable World
w: http://www.petalogix.com p: +61-7-30090663,+42-0-721842854 f: +61-7-30090663

2009-04-30 13:52:44

by Arnd Bergmann

[permalink] [raw]
Subject: Re: Memory limits - mm_segment_t - MAKE_MM_SEG

On Wednesday 29 April 2009, Michal Simek wrote:
> I look at some things which I need to clear for MMU Microblaze patches and
> I would like to know your opinion about.
>
> First of all I found that almost all archs use MAKE_MM_SEG macro which
> should be good to move to generic location (asm-generic/uaccess.h ? )
> #define MAKE_MM_SEG(s) ((mm_segment_t) { (s) })

I have a generic uaccess.h queued up, which I'm planning to submit for
2.6.30 that includes this.

> The second thing is about place where is stored limit for processes -> mm_segment_t structure
>
> Where is the proper location for storing mm_segment_t? Some arch use
> thread_info some of them thread_struct

The method that is used here is different on some architectures.
Most of them use the address limit, which is a property of the
thread, and sensibly belongs into the thread_info. s390 and possibly
others have separate address spaces for user access and use a CPU
feature for this, which belongs into thread_struct.

> Here is the small table for cpus which are in linux kernel and location and type for them.
> The most of them uses thread_info structure for it and name is addr_limit.
> The location of mm_segment_t is different too -> we should move it to any generic location too.
> What do you think?

Sounds fair to me. I think uaccess.h is the best place for this, or maybe
a new mm_segment.h, but not segment.h which is used traditionally for
real segments on x86.

Arnd <><

2009-04-30 14:28:50

by Michal Simek

[permalink] [raw]
Subject: Re: Memory limits - mm_segment_t - MAKE_MM_SEG

Arnd Bergmann wrote:
> On Wednesday 29 April 2009, Michal Simek wrote:
>> I look at some things which I need to clear for MMU Microblaze patches and
>> I would like to know your opinion about.
>>
>> First of all I found that almost all archs use MAKE_MM_SEG macro which
>> should be good to move to generic location (asm-generic/uaccess.h ? )
>> #define MAKE_MM_SEG(s) ((mm_segment_t) { (s) })
>
> I have a generic uaccess.h queued up, which I'm planning to submit for
> 2.6.30 that includes this.

Please CC me for this patch. Thanks.

>
>> The second thing is about place where is stored limit for processes -> mm_segment_t structure
>>
>> Where is the proper location for storing mm_segment_t? Some arch use
>> thread_info some of them thread_struct
>
> The method that is used here is different on some architectures.
> Most of them use the address limit, which is a property of the
> thread, and sensibly belongs into the thread_info. s390 and possibly
> others have separate address spaces for user access and use a CPU
> feature for this, which belongs into thread_struct.

Great. I'll use only thread_info structure.

>
>> Here is the small table for cpus which are in linux kernel and location and type for them.
>> The most of them uses thread_info structure for it and name is addr_limit.
>> The location of mm_segment_t is different too -> we should move it to any generic location too.
>> What do you think?
>
> Sounds fair to me. I think uaccess.h is the best place for this, or maybe
> a new mm_segment.h, but not segment.h which is used traditionally for
> real segments on x86.

OK. I'll move all things from segment.h to uaccess.h in my next branch.

Michal

>
> Arnd <><


--
Michal Simek, Ing. (M.Eng)
w: http://www.monstr.eu p: +42-0-721842854