This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


RE: [Xen-devel] RE: Kernel BUG at arch/x86/mm/tlb.c:61

To: MaoXiaoyun <tinnycloud@xxxxxxxxxxx>, "jeremy@xxxxxxxx" <jeremy@xxxxxxxx>
Subject: RE: [Xen-devel] RE: Kernel BUG at arch/x86/mm/tlb.c:61
From: "Tian, Kevin" <kevin.tian@xxxxxxxxx>
Date: Tue, 26 Apr 2011 13:52:11 +0800
Accept-language: en-US
Acceptlanguage: en-US
Cc: xen devel <xen-devel@xxxxxxxxxxxxxxxxxxx>, "giamteckchoon@xxxxxxxxx" <giamteckchoon@xxxxxxxxx>, "konrad.wilk@xxxxxxxxxx" <konrad.wilk@xxxxxxxxxx>
Delivery-date: Mon, 25 Apr 2011 22:53:28 -0700
Envelope-to: www-data@xxxxxxxxxxxxxxxxxxx
In-reply-to: <BLU157-w51A8A73D5A656542F9AB13DA960@xxxxxxx>
List-help: <mailto:xen-devel-request@lists.xensource.com?subject=help>
List-id: Xen developer discussion <xen-devel.lists.xensource.com>
List-post: <mailto:xen-devel@lists.xensource.com>
List-subscribe: <http://lists.xensource.com/mailman/listinfo/xen-devel>, <mailto:xen-devel-request@lists.xensource.com?subject=subscribe>
List-unsubscribe: <http://lists.xensource.com/mailman/listinfo/xen-devel>, <mailto:xen-devel-request@lists.xensource.com?subject=unsubscribe>
References: <COL0-MC1-F14hmBzxHs00230882@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx>, <BLU157-w488E5FEBD5E2DBC0666EF1DAA70@xxxxxxx>, <BLU157-w5025BFBB4B1CDFA7AA0966DAA90@xxxxxxx>, <BLU157-w540B39FBA137B4D96278D2DAA90@xxxxxxx>, <BANLkTimgh_iip27zkDPNV9r7miwbxHmdVg@xxxxxxxxxxxxxx>, <BANLkTimkMgYNyANcKiZu5tJTL4==zdP3xg@xxxxxxxxxxxxxx>, <BLU157-w116F1BB57ABFDE535C7851DAA80@xxxxxxx>, <4DA3438A.6070503@xxxxxxxx>, <BLU157-w2C6CD57CEA345B8D115E8DAAB0@xxxxxxx>, <BLU157-w36F4E0A7503A357C9DE6A3DAAB0@xxxxxxx>, <20110412100000.GA15647@xxxxxxxxxxxx>, <BLU157-w14B84A51C80B41AB72B6CBDAAD0@xxxxxxx>, <BANLkTinNxLnJxtZD68ODLSJqafq0tDRPfw@xxxxxxxxxxxxxx>, <BLU157-w30A1A208238A9031F0D18EDAAD0@xxxxxxx>, <BLU157-w383D1A2536480BCD4C0E0EDAAD0@xxxxxxx> <BLU157-w42DAD248C94153635E9749DAAC0@xxxxxxx>, <4DA8B715.9080508@xxxxxxxx> <BLU157-w51A8A73D5A656542F9AB13DA960@xxxxxxx>
Sender: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx
Thread-index: AcwC9yV9lyho/SbnS+KWtULtUHnTyQA3ccCg
Thread-topic: [Xen-devel] RE: Kernel BUG at arch/x86/mm/tlb.c:61
>From: MaoXiaoyun
>Sent: Monday, April 25, 2011 11:15 AM
>> Date: Fri, 15 Apr 2011 14:22:29 -0700
>> From: jeremy@xxxxxxxx
>> To: tinnycloud@xxxxxxxxxxx
>> CC: giamteckchoon@xxxxxxxxx; xen-devel@xxxxxxxxxxxxxxxxxxx; 
>> konrad.wilk@xxxxxxxxxx
>> Subject: Re: Kernel BUG at arch/x86/mm/tlb.c:61
>> On 04/15/2011 05:23 AM, MaoXiaoyun wrote:
>> > Hi:
>> >
>> > Could the crash related to this patch ?
>> > http://git.kernel.org/?p=linux/kernel/git/jeremy/xen.git;a=commitdiff;h=45bfd7bfc6cf32f8e60bb91b32349f0b5090eea3
>> >
>> > Since now TLB state change to TLBSTATE_OK(mmu_context.h:40) is before
>> > cpumask_clear_cpu(line 49).
>> > Could it possible that right after execute line 40 of mmu_context.h,
>> > CPU revice IPI from other CPU to
>> > flush the mm, and when in interrupt, find the TLB state happened to be
>> > TLBSTATE_OK. Which conflicts.
>> Does reverting it help?
>> J
>Hi Jeremy:
>    The lastest test result shows the reverting didn't help.
>    Kernel panic exactly at the same place in tlb.c.
>    I have question about TLB state, from the stack, 
>    xen_do_hypervisor_callback-> xen_evtchn_do_upcall->... ->drop_other_mm_ref
>    What  cpu_tlbstate.state should be,  could  TLBSTATE_OK or TLBSTATE_LAZY 
>all be possible? 
>    That is after a hypercall from userspace, state will be TLBSTATE_OK, and
>      if from kernel space, state will be TLBSTATE_LAZE ? 
>       thanks.

it looks a bug in drop_other_mm_ref implementation, that current TLB state 
should be checked
before invoking leave_mm(). There's a window between below lines of code:

       /* Get the "official" set of cpus referring to our pagetable. */
        if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
                for_each_online_cpu(cpu) {
                        if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
                            && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
                        smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);

there's chance that when smp_call_function_single is invoked, actual TLB state 
has been
updated in the other cpu. The upstream kernel patch you referred to earlier 
just makes
this bug exposed more easily. But even without this patch, you may still suffer 
such issue
which is why reverting the patch doesn't help.

Could you try adding a check in drop_other_mm_ref?

        if (active_mm == mm && percpu_read(cpu_tlbstate.state) != TLBSTATE_OK)

once the interrupted context has TLBSTATE_OK, it implicates that later it will 
the TLB flush and thus no need for leave_mm from interrupt handler, and that's 
assumption of doing leave_mm.

Xen-devel mailing list
<Prev in Thread] Current Thread [Next in Thread>