[KJ] clear_page() and copy_page()

[KJ] clear_page() and copy_page()

[Robert P. J. Day]

  there are numerous instances of calls to memset() of the form

  memset(<addr>, 0, PAGE_SIZE);

to clear a single page, as you can see sprinkled throughout the output
here (obviously, not every line of output represents one of those
cases, the grep pattern is overly general):

  $ grep -r "memset.*PAGE_SIZE" .

  however, most architectures define a more convenient clear_page()
macro in a "page.h" header file thusly:

  $ grep -r "#define clear_page" include
  include/asm-frv/page.h:#define clear_page(pgaddr) memset((pgaddr), 0, PAGE_SIZE)
  include/asm-v850/page.h:#define clear_page(page)  memset ((void *)(page), 0, PAGE_SIZE)
  include/asm-parisc/page.h:#define clear_page(page) memset((void *)(page), 0, PAGE_SIZE)
  ...

and so on, which suggests the obvious cleanup that those calls to
memset() should be replaced by calls to clear_page().  is it really
that simple?  well, maybe not.

  from include/asm-i386/page.h:

=============================
#ifdef CONFIG_X86_USE_3DNOW

#include <asm/mmx.h>

#define clear_page(page)        mmx_clear_page((void *)(page))
#define copy_page(to,from)      mmx_copy_page(to,from)

#else

/*
 *      On older X86 processors it's not a win to use MMX here it seems.
 *      Maybe the K6-III ?
 */

#define clear_page(page)        memset((void *)(page), 0, PAGE_SIZE)
#define copy_page(to,from)      memcpy((void *)(to), (void *)(from),
PAGE_SIZE)

#endif
=============================

so what are the issues involved with that kind of cleanup?

  similarly, there are a (smaller) number of calls of the form

  memcpy(<src>, <dest>, PAGE_SIZE);

as you can see:

  $ grep -r "memcpy.*PAGE_SIZE"

even though most arches also define the equivalent "copy_page" macro:

  $ grep -r "#define copy_page" include

same questions here -- any issues with standardizing on calls to
copy_page()?

rday

p.s.  obviously, rewriting to use those simpler macros would imply
that *every* architecture would need to define those macros, and i
don't think that's the case at the moment.

-- 

====================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry
Waterloo, Ontario, CANADA

http://www.fsdev.dreamhosters.com/wiki/index.php?title=Main_Page
====================================
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] clear_page() and copy_page()

[Richard Knutsson]

Robert P. J. Day wrote:
>   there are numerous instances of calls to memset() of the form
>
>   memset(<addr>, 0, PAGE_SIZE);
>
> to clear a single page, as you can see sprinkled throughout the output
> here (obviously, not every line of output represents one of those
> cases, the grep pattern is overly general):
>   
[snip]
> same questions here -- any issues with standardizing on calls to
> copy_page()?
>   
> rday
>
> p.s.  obviously, rewriting to use those simpler macros would imply
> that *every* architecture would need to define those macros, and i
> don't think that's the case at the moment.
>   

include]$ grep -L clear_page asm-*/page.h
asm-generic/page.h
asm-powerpc/page.h
include]$ grep -L copy_page asm-*/page.h
asm-generic/page.h
asm-powerpc/page.h

so it seems it only needs to be defined on PowerPC. But what confuses me is:
...
asm-x86_64/page.h:48:void clear_page(void *);
asm-x86_64/page.h:51:#define clear_user_page(page, vaddr, pg)   clear_page(page)
...

Is x86_64 setting all its memory to zero before giving it to the caller, or what?

Well, just my 2 cents...
Richard Knutsson

_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] clear_page() and copy_page()

[Robert P. J. Day]

On Sun, 28 Jan 2007, Richard Knutsson wrote:

> Robert P. J. Day wrote:
> >   there are numerous instances of calls to memset() of the form
> >
> >   memset(<addr>, 0, PAGE_SIZE);
> >
> > to clear a single page, as you can see sprinkled throughout the output
> > here (obviously, not every line of output represents one of those
> > cases, the grep pattern is overly general):
> >
> [snip]
> > same questions here -- any issues with standardizing on calls to
> > copy_page()?
> >   rday
> >
> > p.s.  obviously, rewriting to use those simpler macros would imply
> > that *every* architecture would need to define those macros, and i
> > don't think that's the case at the moment.
> >
>
> include]$ grep -L clear_page asm-*/page.h
> asm-generic/page.h
> asm-powerpc/page.h
> include]$ grep -L copy_page asm-*/page.h
> asm-generic/page.h
> asm-powerpc/page.h
>
> so it seems it only needs to be defined on PowerPC. But what confuses me is:
> ...
> asm-x86_64/page.h:48:void clear_page(void *);
> asm-x86_64/page.h:51:#define clear_user_page(page, vaddr, pg)
> clear_page(page)
> ...
>
> Is x86_64 setting all its memory to zero before giving it to the caller, or
> what?

i can check with the appropriate people.  i guess the major issue is
whether the semantics of clear_page() are *meant* to be identical to a
single page-sized call to memset().  technically, of course, they
don't *have* to be -- clear_page() could always be defined to do
something extra, but it doesn't seem like that's happening here, so a
simple substitution would appear to be a valid cleanup, yes?

rday

-- 
====================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry
Waterloo, Ontario, CANADA

http://www.fsdev.dreamhosters.com/wiki/index.php?title=Main_Page
====================================
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] clear_page() and copy_page()

[Robert P. J. Day]

On Sun, 28 Jan 2007, Richard Knutsson wrote:

> include]$ grep -L clear_page asm-*/page.h
> asm-generic/page.h
> asm-powerpc/page.h
> include]$ grep -L copy_page asm-*/page.h
> asm-generic/page.h
> asm-powerpc/page.h
>
> so it seems it only needs to be defined on PowerPC.

powerpc should be OK since the files are named page_32.h and
page_64.h, not just page.h.  so i think we're good there.

> But what confuses me is:
> ...
> asm-x86_64/page.h:48:void clear_page(void *);
> asm-x86_64/page.h:51:#define clear_user_page(page, vaddr, pg)
> clear_page(page)
> ...
>
> Is x86_64 setting all its memory to zero before giving it to the
> caller, or what?

i'm not sure what the problem is here.  clear_page() still exists and
accepts a single page address.  what exactly is your concern?

rday

-- 
====================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry
Waterloo, Ontario, CANADA

http://www.fsdev.dreamhosters.com/wiki/index.php?title=Main_Page
====================================
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] clear_page() and copy_page()

[Richard Knutsson]

Robert P. J. Day wrote:
> On Sun, 28 Jan 2007, Richard Knutsson wrote:
>
>   
>> Robert P. J. Day wrote:
>>     
>>>   there are numerous instances of calls to memset() of the form
>>>
>>>   memset(<addr>, 0, PAGE_SIZE);
>>>
>>> to clear a single page, as you can see sprinkled throughout the output
>>> here (obviously, not every line of output represents one of those
>>> cases, the grep pattern is overly general):
>>>
>>>       
>> [snip]
>>     
>>> same questions here -- any issues with standardizing on calls to
>>> copy_page()?
>>>   rday
>>>
>>> p.s.  obviously, rewriting to use those simpler macros would imply
>>> that *every* architecture would need to define those macros, and i
>>> don't think that's the case at the moment.
>>>
>>>       
>> include]$ grep -L clear_page asm-*/page.h
>> asm-generic/page.h
>> asm-powerpc/page.h
>> include]$ grep -L copy_page asm-*/page.h
>> asm-generic/page.h
>> asm-powerpc/page.h
>>
>> so it seems it only needs to be defined on PowerPC. But what confuses me is:
>> ...
>> asm-x86_64/page.h:48:void clear_page(void *);
>> asm-x86_64/page.h:51:#define clear_user_page(page, vaddr, pg)
>> clear_page(page)
>> ...
>>
>> Is x86_64 setting all its memory to zero before giving it to the caller, or
>> what?
>>     
>
> i can check with the appropriate people.  i guess the major issue is
> whether the semantics of clear_page() are *meant* to be identical to a
> single page-sized call to memset().  technically, of course, they
>   
The only reason against it, would be if it is meant the address has to 
be a multiple of the PAGE_SIZE, but I find that unlikely.
> don't *have* to be -- clear_page() could always be defined to do
> something extra, but it doesn't seem like that's happening here, so a
> simple substitution would appear to be a valid cleanup, yes?
>   
I would think so. Just a bit concerned that PowerPC does not have it 
defined. I would rather see it defined in asm-generic and then if the 
arch has a better solution, it just undef the generic one. But that is 
just my opinion :)

Richard Knutsson

_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

[KJ] clear_page() and copy_page()

[Robert P. J. Day]

  i've updated the wiki page related to rewriting source in terms of
the above two macros.

http://www.fsdev.dreamhosters.com/wiki/index.php?title=Clear_page_copy_page

rday

-- 
====================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry
Waterloo, Ontario, CANADA

http://www.fsdev.dreamhosters.com/wiki/index.php?title=Main_Page
====================================
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

[KJ] "clear_page" and "copy_page"?

[Robert P. J. Day]

  i asked about this once upon a time, but can anyone clarify whether
it's possible to effect the following replacements:

  1) memset(addr, 0, PAGE_SIZE) --> clear_page(addr)
  2) memcpy(to, from, PAGE_SIZE) --> copy_page(to, from)

the only issues i can see is if clear_page() and/or copy_page() have
some special semantics above and beyond simple memset and/or memcpy.
perhaps an alignment requirement or something.

  it's fairly clear that every architecture *must* implement both
clear_page() and copy_page().  so is this a possible cleanup?  if so,
i can add a page to the wiki.

rday
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors