Re: [patch 00/21] 2.6.19-stable review

From: Zwane Mwaikambo <zwane@infradead.org>
To: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Chuck Ebbert <cebbert@redhat.com>, Andi Kleen <ak@suse.de>,
	Greg KH <greg@kroah.com>,
	linux-kernel@vger.kernel.org, stable@kernel.org,
	Justin Forbes <jmforbes@linuxtx.org>,
	"Theodore Ts'o" <tytso@mit.edu>,
	Randy Dunlap <rdunlap@xenotime.net>,
	Dave Jones <davej@redhat.com>,
	Chuck Wolber <chuckw@quantumlinux.com>,
	Chris Wedgwood <reviews@ml.cw.f00f.org>,
	Michael Krufky <mkrufky@linuxtv.org>,
	alan@lxorguk.ukuu.org.uk
Subject: Re: [patch 00/21] 2.6.19-stable review
Date: Wed, 28 Feb 2007 00:51:18 -0800 (PST)	[thread overview]
Message-ID: <Pine.LNX.4.64.0702272312580.2632@montezuma.windriver.com> (raw)
In-Reply-To: <m1lkiikcth.fsf@ebiederm.dsl.xmission.com>

On Tue, 27 Feb 2007, Eric W. Biederman wrote:

> > http://www.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.20/2.6.20-mm2/broken-out/x86_64-mm-handle-irqs-pending-in-irr-during-irq-migration.patch
> >> 
> >> That's mainly an Andi decision.  Let's cc him.
> >
> > Would be good to have Eric also ack them as safe and obvious.
> >
> > Btw, that latter one has corrupted sign-offs from Andi (it's in the middle 
> > of the text, very confusing).
> 
> There are two questions.
> 1) What can we do to make the situation better.
> 2) Is the hole completely plugged.
> 
> When I wrote the patch I had the local apic priorities backwards in my
> head.  So apic_in_service_vector can return the wrong value if two
> irqs are in service.  Now I don't think we allows ourselves to enable
> interrupts in an interrupt service routing until after we have acked
> the local apic so this should be harmless.  The fix is also trivial
> of just having apic_in_service_vector return: "~get_irq_regs()->orig_rax".
> 
> Except for that one possible problem everything I can think of are
> just theoretical cracks at this point, and they don't make the
> situation any worse.
> 
> Given that this patch has appears to have undergone a noticeable
> amount of testing, by people other than myself, and clears up the
> symptoms.  I have no problem 

Hi Eric,
	Thanks for getting this cruft cleaned up. I have a few comments 
regarding;

handle-irqs-pending-in-irr-during-irq-migration.patch

1) It relies on checking the IRR, this could race with the corresponding 
vector bit being set by hardware.

2) apic_handle_pending_vector is oddly named since it doesn't actually 
handle a pending vector but drops it if it has been freed.

3) It looks complex

So how about the following scheme. Add a check in do_IRQ whether the irq's 
domain contains the current cpu, if not we free the vector upon handler 
completion.

Cheers,
	Zwane