Re: [patch] mm: Reimplementation of dynamic percpu memory allocator

[Ravikiran G Thirumalai]

On Fri, Jan 14, 2005 at 01:34:25AM -0800, Andrew Morton wrote:
> Ravikiran G Thirumalai <kiran@xxxxxxxxxx> wrote:
> 
> >  > 
> >  > Why cannot the code simply call vmalloc rather than copying its internals?
> > 
> >  Node local allocation. vmalloc cannot ensure pages for correspomding
> >  cpus are node local.  Also, design goal was to allocate pages for 
> >  cpu_possible cpus only.  With plain vmalloc, we will end up allocating 
> >  pages for NR_CPUS.
> 
> So...  is it not possible to enhance vmalloc() for node-awareness, then
> just use it?
> 

Memory for block management (free lists, bufctl lists) is also resident 
in one block.  A typical block in this allocator looks like this:


VMALLOC_ADDR 	PAGE_ADDR	BLOCK			
============ 	=========	========

0xa0000		0x10100		-----------------	^	^
  .				|		|	|	|
  .				|   cpu 0	| PCPU_BLKSIZE	|
				|		|	|	|
0xa0100		0x30100		-----------------	v	|
				|		|		|
				|   cpu 1	|		|
				|		|		|
0xa0200		   -		-----------------	     NR_CPUS		
				|		|		|
				| !cpu_possible	|		|
				|		|		|
0xa0300		   -		-----------------		|
				|		|		|
				| !cpu_possible	|		|
				|		|		|
0xa0400		0x10300		-----------------	  ^	v
				|		|	  |
				|  Block mgmt	|BLOCK_MANAGEMENT_SIZE	
				|		|	  |
0xa05ff				-----------------	  v



This block is setup by valloc_percpu in the allocator code.  There is lot 
of allocator specific stuff like the PCPU_BLKSIZE, BLOCK_MANAGEMENT_SIZE 
used here.  I thought it was not appropriate to put them in vmalloc.c.  
A common vmalloc_percpu which can take arguments for PCPU_BLKSIZE and 
BLOCK_MANAGEMENT_SIZE is not useful anywhere else. 

Changed patchset with other modifications suggested will follow.

Thanks,
Kiran
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/