Re: [RFC] memblock; Properly handle overlaps

[Yinghai Lu <yinghai@kernel.org>]

On 03/04/2011 08:05 PM, Benjamin Herrenschmidt wrote:
> Hi folks !
> 
> This is not fully tested yet (I'm toying with a little userspace
> test bench, it seems to work well so far but I haven't yet tested
> the cases with no-coalesce boundaries which at least ARM needs).
> 
> But it's good enough to get comments...
> 
> So currently, things like memblock_reserve() or memblock_free()
> don't deal well -at-all- with overlaps of all kinds. Some specific
> cases are handled but the code is clumsy and things will fall over
> in many cases.
> 
> This is annoying because typically memblock_reserve() is used to
> mark regions passed by the firmware as reserved and we all know
> how much we can trust our firmwares right ?
> 
> I have also a case I need to deal with on powerpc where the flat
> device-tree is fully enclosed within some other FW blob that has
> its own reserve map entry, so when I end up trying to reserve
> both, the current memblock code pukes.

did you try remove and add tricks?

diff --git a/mm/memblock.c b/mm/memblock.c
index 4618fda..ba4ffdc 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -453,6 +453,9 @@ long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
 
        BUG_ON(0 == size);
 
+       while (__memblock_remove(_rgn, base, size) >= 0)
+               ;
+
        return memblock_add_region(_rgn, base, size);
 }
 

Yinghai