Re: [PATCH] Simple brk allocator for very early allocations

[Jeremy Fitzhardinge <jeremy@goop.org>]

Andrew Morton wrote:
> On Fri, 27 Feb 2009 17:51:17 -0800 Jeremy Fitzhardinge <jeremy@goop.org> wrote:
>
>   
>> This series adds a very simple brk-like allocator for very early
>> allocations.  By default it extends the bss segment, starting at _end.
>>
>> This is used to allocate x86-32's initial head_32.S pagetable, removing
>> init_pg_table_start/end and pg0, replacing them with brk allocations.
>>     
>
> Changelog fails to provide a reason for this?
>   

Well, the whole thing is just a generalization of what the 32-bit 
pagetable builder does anyway, to make it more useful.  With the brk 
allocator in place, there's no reason for head_32.S to do it again.

>> dmi_alloc() is also changed to use extend_brk.
>>     
>
> Seems a large patchset just to clean up DMI a bit ;)
>   

A few patches could be folded together.  But its pretty small really...

> I assume that xen needs this?  domU or dom0

I can make use of it in Xen to remove a bunch of static arrays.  I'm 
pretty sure there's quite a few places around the kernel which could 
make use of this facility.  My kernel has 900k of bss; how much of that 
is stuff that 1) could be allocated, and 2) not actually being used at 
runtime?  A lot of things which are compile-time sized, hash tables, log 
buffers, etc, could be runtime sized instead.

    J