From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from sc8-sf-mx2-b.sourceforge.net ([10.3.1.92] helo=mail.sourceforge.net) by sc8-sf-list1.sourceforge.net with esmtp (Exim 4.30) id 1EIBVk-0005TL-5a for user-mode-linux-devel@lists.sourceforge.net; Wed, 21 Sep 2005 13:48:12 -0700 Received: from smtp.osdl.org ([65.172.181.4]) by mail.sourceforge.net with esmtps (TLSv1:DES-CBC3-SHA:168) (Exim 4.44) id 1EIBVj-00029i-Vu for user-mode-linux-devel@lists.sourceforge.net; Wed, 21 Sep 2005 13:48:12 -0700 From: Andrew Morton Subject: Re: [uml-devel] Re: [PATCH 07/10] uml: avoid fixing faults while atomic Message-Id: <20050921134724.52603016.akpm@osdl.org> In-Reply-To: <200509212222.50653.blaisorblade@yahoo.it> References: <200509211923.21861.blaisorblade@yahoo.it> <20050921172908.10219.57644.stgit@zion.home.lan> <20050921124957.437cf069.akpm@osdl.org> <200509212222.50653.blaisorblade@yahoo.it> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: user-mode-linux-devel-admin@lists.sourceforge.net Errors-To: user-mode-linux-devel-admin@lists.sourceforge.net List-Unsubscribe: , List-Id: The user-mode Linux development list List-Post: List-Help: List-Subscribe: , List-Archive: Date: Wed, 21 Sep 2005 13:47:24 -0700 To: Blaisorblade Cc: user-mode-linux-devel@lists.sourceforge.net, torvalds@osdl.org, jdike@addtoit.com, linux-kernel@vger.kernel.org Blaisorblade wrote: > > On Wednesday 21 September 2005 21:49, Andrew Morton wrote: > > "Paolo 'Blaisorblade' Giarrusso" wrote: > > > From: Paolo 'Blaisorblade' Giarrusso > > > The in_atomic() test in x86's do_page_fault() is in fact a message passed > > into it from filemap.c's kmap_atomic(). > Ok, this can be ok, but: > > It has accidental side-effects, > > such as making copy_to_user() fail if inside spinlocks when > > CONFIG_PREEMPT=y. > Sorry, but should it ever succeed inside spinlocks? I mean, should it ever > call down() inside spinlocks? (We never do down_trylock, and ever if we did > the x86 trick, that wouldn't make the whole thing safe at all - they still > take the spinlock and potentially sleep. And it's legal only if no spinlock > is held). Not sure what you're asking here. copy_to/from_user() will fail inside spinlock if CONFIG_PREMPT=y and if the copy happens to cause a fault. Otherwise it will succeed inside spinlock, and it won't spew a sleeping-while-atomic warning, because that uses in_atomic() too. It might deadlock if we schedule away and try to retake the same lock. > Even if spinlocks don't always trigger in_atomic() - which means that we'd > need to have a better fix for this. The patch you have will correctly cause copy_*_user()->pagefault to fail the copy if the caller has run inc_preempt_count(). It will not cause copy_*_user()->pagefault to fail inside spinlocks unless UML does inc_preempt_count() in its spinlock implementation. > (Btw, I took the above reasoning from something said, as an aside, on LWN.net > kernel page, about the FUTEX deadlock on mm->mmap_sem of ~ 2.6.8 - yes, it > wasn't the full truth, but not totally dumb). > > > So I think this change is only needed if UML implements kmap_atomic, as in > > arch/i386/mm/highmem.c, which it surely does not do? > NACK, see above. Yup, the patch is needed for the futex code, and for general correct implementation of inc_preempt_count()'s intended effect. ------------------------------------------------------- SF.Net email is sponsored by: Tame your development challenges with Apache's Geronimo App Server. Download it for free - -and be entered to win a 42" plasma tv or your very own Sony(tm)PSP. Click here to play: http://sourceforge.net/geronimo.php _______________________________________________ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel