linux-kernel - Re: percpu related boot crash on x86

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]

Message-ID: <20110120111927.GE6036@htj.dyndns.org>
Date:	Thu, 20 Jan 2011 12:19:27 +0100
From:	Tejun Heo <tj@...nel.org>
To:	Eric Dumazet <eric.dumazet@...il.com>
Cc:	Peter Zijlstra <a.p.zijlstra@...llo.nl>,
	Pekka Enberg <penberg@...nel.org>, Ingo Molnar <mingo@...e.hu>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Thomas Gleixner <tglx@...utronix.de>,
	"H. Peter Anvin" <hpa@...or.com>,
	Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: percpu related boot crash on x86

Hello,

On Thu, Jan 20, 2011 at 12:12:47PM +0100, Eric Dumazet wrote:
> Le jeudi 20 janvier 2011 à 11:47 +0100, Peter Zijlstra a écrit :
> > On Thu, 2011-01-20 at 10:31 +0200, Pekka Enberg wrote:
> > >    if (!irqs_disabled())
> > > 
> > > guard around the debugging code to make vfree() always work with IRQs 
> > > disabled. 
> > 
> > That won't be sufficient, __purge_vmap_area_lazy() needs to call
> > flush_tlb_kernel_range() occasionally.
> > 
> > There really is a good reason vmalloc()/vfree() don't work with
> > interrupts disabled.
> 
> Hmm, I would try to increase PERCPU_DYNAMIC_RESERVE, or pcpu_mem_alloc()
> could try to use kzalloc() even for allocations bigger than one PAGE,
> and fallback to vzalloc().

It already handles all those stuff.  As I wrote in the other message,
I don't think this problem has much to do with percpu allocator
itself.

> We have more alloc_percpu() uses, we also should try to use first chunk
> if possible.

Yeah, but that's true.  It's probably too small at this point.  Maybe
there should be a way to monitor percpu memory status, a proc entry or
something so that we can have more data on sizing.

Thanks.

-- 
tejun
--
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/