linux-kernel - Re: [patch 4/4] mm: bootmem: try harder to free pages in bulk

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20111214204210.GF3047@cmpxchg.org>
Date:	Wed, 14 Dec 2011 21:42:10 +0100
From:	Johannes Weiner <hannes@...xchg.org>
To:	Uwe Kleine-König 
	<u.kleine-koenig@...gutronix.de>
Cc:	Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org,
	linux-kernel@...r.kernel.org
Subject: Re: [patch 4/4] mm: bootmem: try harder to free pages in bulk

On Wed, Dec 14, 2011 at 09:20:32PM +0100, Uwe Kleine-König wrote:
> On Tue, Dec 13, 2011 at 02:58:31PM +0100, Johannes Weiner wrote:
> > The loop that frees pages to the page allocator while bootstrapping
> > tries to free higher-order blocks only when the starting address is
> > aligned to that block size.  Otherwise it will free all pages on that
> > node one-by-one.
> > 
> > Change it to free individual pages up to the first aligned block and
> > then try higher-order frees from there.
> > 
> > Signed-off-by: Johannes Weiner <hannes@...xchg.org>
> I gave all four patches a try now on my ARM machine and it still works
> fine. But note that this patch isn't really tested, because for me
> free_all_bootmem_core is only called once and that with an aligned
> address.
> But at least you didn't broke that case :-)
> Having said that, I wonder if the code does the right thing for
> unaligned start. (That is, it's wrong to start testing for bit 0 of
> map[idx / BITS_PER_LONG], isn't it?) But if that's the case that's not
> something you introduced in this series.

We round up and cover area beyond the end of the node to the next
alignment boundary, but don't do the same for the beginning of the
node.  So map[0] is the first BITS_PER_LONG pages starting at start,
even when start is not aligned.

> > @@ -196,12 +189,17 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
> >  		map = bdata->node_bootmem_map;
> >  		idx = start - bdata->node_min_pfn;
> >  		vec = ~map[idx / BITS_PER_LONG];
> > -
> > -		if (aligned && vec == ~0UL) {
> > +		/*
> > +		 * If we have a properly aligned and fully unreserved
> > +		 * BITS_PER_LONG block of pages in front of us, free
> > +		 * it in one go.
> > +		 */
> > +		if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
> >  			int order = ilog2(BITS_PER_LONG);
> >  
> >  			__free_pages_bootmem(pfn_to_page(start), order);
> >  			count += BITS_PER_LONG;
> > +			start += BITS_PER_LONG;
> >  		} else {
> >  			unsigned long off = 0;
> >  
> > @@ -214,8 +212,8 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
> >  				vec >>= 1;
> >  				off++;
> >  			}
> > +			start = ALIGN(start + 1, BITS_PER_LONG);
> >  		}
> > -		start += BITS_PER_LONG;
> I don't know if the compiler would be more happy if you would just use
> 
> 	start = ALIGN(start + 1, BITS_PER_LONG);
> 
> unconditionally and drop
> 
> 	start += BITS_PER_LONG
> 
> in the if block?!

I thought it would be beneficial to have the simpler version for the
common case, which is freeing a full block.  Have you looked at the
object code?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/