linux-kernel - Re: [PATCH 5/8] vmalloc: simplify vread()/vwrite()

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: <20100119013303.GA12513@localhost>
Date:	Tue, 19 Jan 2010 09:33:03 +0800
From:	Wu Fengguang <fengguang.wu@...el.com>
To:	Nick Piggin <npiggin@...e.de>
Cc:	Andrew Morton <akpm@...ux-foundation.org>,
	LKML <linux-kernel@...r.kernel.org>, Tejun Heo <tj@...nel.org>,
	Ingo Molnar <mingo@...e.hu>, Andi Kleen <andi@...stfloor.org>,
	Hugh Dickins <hugh.dickins@...cali.co.uk>,
	Christoph Lameter <cl@...ux-foundation.org>,
	KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>,
	Linux Memory Management List <linux-mm@...ck.org>
Subject: Re: [PATCH 5/8] vmalloc: simplify vread()/vwrite()

On Mon, Jan 18, 2010 at 07:23:59AM -0700, Nick Piggin wrote:
> On Mon, Jan 18, 2010 at 09:35:12PM +0800, Wu Fengguang wrote:
> > On Thu, Jan 14, 2010 at 05:45:26AM -0700, Nick Piggin wrote:
> > > On Wed, Jan 13, 2010 at 09:53:10PM +0800, Wu Fengguang wrote:
> > > > vread()/vwrite() is only called from kcore/kmem to access one page at a time.
> > > > So the logic can be vastly simplified.
> > > > 
> > > > The changes are:
> > > > - remove the vmlist walk and rely solely on vmalloc_to_page()
> > > > - replace the VM_IOREMAP check with (page && page_is_ram(pfn))
> > > > - rename to vread_page()/vwrite_page()
> > > > 
> > > > The page_is_ram() check is necessary because kmap_atomic() is not
> > > > designed to work with non-RAM pages.
> > > 
> > > I don't know if you can really do this. Previously vmlist_lock would be
> > > taken, which will prevent these vm areas from being freed.
> > >  
> > > > Note that even for a RAM page, we don't own the page, and cannot assume
> > > > it's a _PAGE_CACHE_WB page.
> > > 
> > > So why is this not a problem for your patch? I don't see how you handle
> > > it.
> > 
> > Sorry I didn't handle it. Just hope to catch attentions from someone
> > (ie. you :).
> > 
> > It's not a problem for x86_64 at all. For others I wonder if any
> > driver will vmalloc HIGHMEM pages with !_PAGE_CACHE_WB attribute..
> > 
> > So I noted the possible problem and leave it alone.
> 
> Well it doesn't need to be vmalloc. Any kind of vmap like ioremap. And
> these can be accompanied by changing the caching attribute. Like agp
> code, for an example. But I don't know if that ever becomes a problem
> in practice.

Yes vmap in general can change caching attribute. However I only care
about vmap that maps RAM pages, since my patch treats non-RAM pages as
hole and won't access them.

> > > What's the problem with the current code, exactly? I would prefer that
> > 
> > - unnecessary complexity to handle multi-page case, since it's always
> >   called to access one single page;
> 
> Fair point there. It just wasn't clear what exactly is your rationale
> because this was in a set of other patches.
>  
> > - the kmap_atomic() cache consistency problem, which I expressed some
> >   concern (without further action)
> 
> Which kmap_atomic problem? Can you explain again? Virtual cache aliasing
> problem you mean? Or caching attribute conflicts?

kmap_atomic() assumes you own the page and always use _PAGE_CACHE_WB.
So there may be conflicts if the page was !_PAGE_CACHE_WB.

> The whole thing looks stupid though, apparently kmap is used to avoid "the
> lock". But the lock is already held. We should just use the vmap
> address.

Yes. I wonder why Kame introduced kmap_atomic() in d0107eb07 -- given
that he at the same time fixed the order of removing vm_struct and
vmap in dd32c279983b.

> > > you continue using the same vmlist locking and checking for validating
> > > addresses.
> > 
> > It's a reasonable suggestion. Kame, would you agree on killing the
> > kmap_atomic() and revert to the vmlist walk?
> 
> Yes, vmlist locking is always required to have a pin on the pages, and
> IMO it should be quite easy to check for IOREMAP, so we should leave
> that check there to avoid the possibility of regressions.

I have no problem if Kame could dismiss my question :)

Thanks,
Fengguang
--
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/