Re: Re: [PATCH] blkfront: Move blkif_interrupt into a tasklet.

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Wed, Aug 24, 2011 at 06:36:58PM +0200, Igor Mammedov wrote:
> On 08/24/2011 05:36 PM, Konrad Rzeszutek Wilk wrote:
> >On Wed, Aug 17, 2011 at 11:07:19AM +0200, Igor Mammedov wrote:
> >>On 08/17/2011 04:38 AM, Konrad Rzeszutek Wilk wrote:
> >>>On Tue, Aug 16, 2011 at 04:26:55AM -0700, imammedo wrote:
> >>>>
> >>>>Jeremy Fitzhardinge wrote:
> >>>>>
> >>>>>Have you tried bisecting to see when this particular problem appeared?
> >>>>>It looks to me like something is accidentally re-enabling interrupts -
> >>>>>perhaps a stack overrun is corrupting the "flags" argument between a
> >>>>>spin_lock_irqsave()/restore pair.
> >>>>>
> >>>>>Is it only on 32-bit kernels?
> >>>>>
> >>>>  ------------[ cut here ]------------
> >>>>[604001.659925] WARNING: at block/blk-core.c:239 blk_start_queue+0x70/0x80()
> >>>>[604001.659964] Modules linked in: nfs lockd fscache auth_rpcgss nfs_acl
> >>>>sunrpc ip6t_REJECT nf_conntrack_ipv6 nf_conntrack_ipv4 nf_defrag_ipv4
> >>>>nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables xen_netfront
> >>>>pcspkr [last unloaded: scsi_wait_scan]
> >>>>[604001.660147] Pid: 336, comm: udevd Tainted: G        W   3.0.0+ #50
> >>>>[604001.660181] Call Trace:
> >>>>[604001.660209]  [<c045c512>] warn_slowpath_common+0x72/0xa0
> >>>>[604001.660243]  [<c06643a0>] ? blk_start_queue+0x70/0x80
> >>>>[604001.660275]  [<c06643a0>] ? blk_start_queue+0x70/0x80
> >>>>[604001.660310]  [<c045c562>] warn_slowpath_null+0x22/0x30
> >>>>[604001.660343]  [<c06643a0>] blk_start_queue+0x70/0x80
> >>>>[604001.660379]  [<c075e231>] kick_pending_request_queues+0x21/0x30
> >>>>[604001.660417]  [<c075e42f>] blkif_interrupt+0x19f/0x2b0
> >>>>...
> >>>>  ------------[ cut here ]------------
> >>>>
> >>>>I've debugged a bit blk-core warning and can say:
> >>>>   - Yes, It is 32-bit PAE kernel and happens only with it so far.
> >>>>   - Affects PV xen guest, bare-metal and kvm configs are not affected.
> >>>>   - Upstream kernel is affected as well.
> >>>>   - Reproduces on xen 4.1.1 and 3.1.2 hosts
> >>>
> >>>And the dom0 is 2.6.18 right? This problem is not present
> >>>when you use a 3.0 dom0?
> >>
> >>For xen 4.1.1 testing, I've used as dom0 Jeremy's 2.6.32.43
> >
> >Jeremy pointed me to this:
> >https://patchwork.kernel.org/patch/1091772/
> >(and http://groups.google.com/group/linux.kernel/browse_thread/thread/39a397566cafc979)
> >which looks to have a similar  backtrack.
> >
> >Perhaps Peter's fix solves the issue?
> 
> 
> I've applied patches:
> sched-separate-the-scheduler-entry-for-preemption.patch
> sched-move-blk_schedule_flush_plug-out-of-__schedule.patch
> block-shorten-interrupt-disabled-regions.patch
> 
> Unfortunately these patches don't help, the problem is still there.

Those patches were a bit fresh. Both Peter and Thomas have some updated ones:

http://git.kernel.org/?p=linux/kernel/git/x86/linux-2.6-tip.git;a=commitdiff;h=9c40cef2b799f9b5e7fa5de4d2ad3a0168ba118c
http://git.kernel.org/?p=linux/kernel/git/x86/linux-2.6-tip.git;a=commitdiff;h=c259e01a1ec90063042f758e409cd26b2a0963c8

Please try those out