Re: [patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Vivek Goyal <vgoyal@in.ibm.com>]

On Wed, Oct 17, 2007 at 01:36:51PM +0200, Bernhard Walle wrote:
[..]
> > > +static int __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
> > > +					unsigned long size, int flags)
> > >  {
> > >  	unsigned long sidx, eidx;
> > >  	unsigned long i;
> > > @@ -133,7 +133,11 @@ static void __init reserve_bootmem_core(
> > >  #ifdef CONFIG_DEBUG_BOOTMEM
> > >  			printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
> > >  #endif
> > > +			if (flags & BOOTMEM_EXCLUSIVE)
> > > +				return -EBUSY;
> > 
> > I think we should unreserve the chunks of memory we have reserved so
> > far (Memory reserved from sidx to i), in case of error.
> 
> Unfortunately, that's not possible without using a lock (or counters
> instead of a bitmap) any more. If we just do
> 
> 	for (i--; i >= sidx; i--)
> 		clear_bit(i, bdata->node_bootmem_map);
> 
> then another thread of execution could reserve the memory (without
> BOOTMEM_EXCLUSIVE) in between -- and the code would free the memory
> which is already reserved.
> 
> I think that could be modelled with a rwlock, not changing the default
> case where BOOTMEM_EXCLUSIVE is not specified.

SMP initialization takes place after bootmem allocator has retired. That
would mean only one thread will be using bootmem allocator. Hence I think
unreserving memory without any kind of locking should be safe.

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