linux-kernel - Re: [PATCH] Try to avoid a pessimistic vmalloc() recursion

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

Date:	Mon, 09 Oct 2006 21:43:15 +0200
From:	Eric Dumazet <dada1@...mosbay.com>
To:	Nick Piggin <nickpiggin@...oo.com.au>
Cc:	Andrew Morton <akpm@...l.org>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] Try to avoid a pessimistic vmalloc() recursion

Nick Piggin a écrit :
> Eric Dumazet wrote:
>> __vmalloc_area_node() is a litle bit pessimist when allocating space 
>> for storing struct page pointers.
>>
>> When allocating more than 4 MB on ia32, or 2 MB on x86_64,  
>> __vmalloc_area_node() has to allocate more than PAGE_SIZE bytes to 
>> store pointers to  page structs. This means that two TLB translations 
>> are needed to access data.
>>
>> This patch tries a kmalloc() call, then only if this first attempt 
>> failed, a vmalloc() is performed. (Later, at vfree() time we chose 
>> kfree() or vfree() with a test on flags & VM_VPAGES : no change is 
>> needed)
>> Most of the time, the first kmalloc() should be OK, so we reduce TLB 
>> usage.
> 
> But this is only TLB usage when managing (read: freeing) the vmalloc pages,
> isn't it? Not when actually accessing the data.

Yes indeed...
I was trying to reduce time taken by a processes handling lot of files (thus 
vmalloc()ing fdtables and fdset). I noticed a high oprofile hit in 
fget_light(). I suspected overhead caused by vmalloc(), but obviously, once 
the vmalloc() mapping is done, the array of pointers wont be used until vfree().

> 
> I'd be inclined to NACK this, unless you can show an improvement somewhere:
> it is suboptimal to even _try_ allocating higher order pages.
> 

Your point is valid. And it seems there is not much cpu used to linearly scan 
vmlist (to find the vm_struct), at least on my little servers.

Eric
-
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/