Re: [opensuse-virtual] Xen live migration and xen-kmp issues
Danilo Godec 07/14/11 3:14 PM >>> This is what xenctx shows after unsuccessful migration: ... The last part - call trace - is considerably longer then for a normally running DomU.
Hopefully someone will make some sense out of that, cause I sure can't.
Without knowing the exact kernel version your using we can't either. You may be able to read something from it (and make our life easier) if you used xenctx's -s option (passing in the System.map that matches the kernel used in the guest).
It seems that HVM emulates a 'PIIX3 IDE' controller and the driver seems to be built-in (i.e. not a kernel module):
Hmm, yes, that's ugly. I have always been advocating for modular drivers...
Things go south when 'xen-platform-pci' or 'xen-vbd' module is loaded...
Is there a way to disable a built-in driver??
I'm not aware of one, but we surely had found some way to use PV drivers in 11.4 guests, just that I don't recall the details - hopefully others reading the list will. Jan
On 07/14/2011 03:28 PM, Jan Beulich wrote:
Danilo Godec <danilo.godec@agenda.si> 07/14/11 3:14 PM >>> This is what xenctx shows after unsuccessful migration: ... The last part - call trace - is considerably longer then for a normally running DomU.
Hopefully someone will make some sense out of that, cause I sure can't.
Without knowing the exact kernel version your using we can't either. You may be able to read something from it (and make our life easier) if you used xenctx's -s option (passing in the System.map that matches the kernel used in the guest).
Here it goes - this particular DomU I'm (ab)using for testing is OpenSuSE 11.2 running kernel 2.6.31.14-0.6-xen #1 SMP. I copied the appropriate System.map file to my Xen server. The DomU is started with 'vcpus=2' and I manually removed one VCPU with 'xm vcpu-set DomU 1': # /usr/lib64/xen/bin/xenctx -s /root/ab1-system.map 180 0 rip: ffffffff800083aa hypercall_page+0x3aa flags: 00001246 i z p rsp: ffff88001f933470 rax: 0000000000000000 rcx: ffffffff800083aa rdx: 0000000000000007 rbx: 0000000000000125 rsi: ffff88001f933488 rdi: 0000000000000003 rbp: ffff88001f9334d8 r8: 0000000000000000 r9: ffff88001f933c58 r10: 0000000000000001 r11: 0000000000000246 r12: 0000000000009d20 r13: ffffc90000000000 r14: 0000000000000004 r15: 0000000000000000 cs: e033 ss: e02b ds: 0000 es: 0000 fs: 0000 @ 00007fe7c14f1710 gs: 0000 @ ffffc90000000000/0000000000000000 Code (instr addr ffffffff800083aa) cc cc cc cc cc cc cc cc cc cc cc 51 41 53 b8 1d 00 00 00 0f 05 <41> 5b 59 c3 cc cc cc cc cc cc cc Stack: 0000000000000001 0000000000000000 ffffffff802feb51 ffff88001f9334ac 0000000000000001 0000000000000000 ffffffff802fedca 0000000700000000 00000000e49a7d90 0000000000000000 ffffffff80803490 0000000000000125 0000000000000001 ffff88001f933538 ffffffff803042e4 0000000000000000 Call Trace: [<ffffffff800083aa>] hypercall_page+0x3aa <-- [<ffffffff802feb51>] xen_poll_irq+0xc1 [<ffffffff802fedca>] xen_clear_irq_pending+0xda [<ffffffff803042e4>] xen_spin_wait+0x94 [<ffffffff80470a00>] _spin_lock+0x70 [<ffffffff80050567>] vprintk+0x67 [<ffffffff8000f8eb>] get_runstate_snapshot+0x9b [<ffffffff8046cb88>] printk+0x4b [<ffffffff8000f8eb>] get_runstate_snapshot+0x9b [<ffffffff8004f5af>] warn_slowpath_common+0x3f [<ffffffff8004f662>] warn_slowpath_null+0x22 [<ffffffff8000f8eb>] get_runstate_snapshot+0x9b [<ffffffff8000fde9>] sched_clock+0x29 [<ffffffff80077316>] sched_clock_cpu+0x86 [<ffffffff800774ce>] cpu_clock+0x2e [<ffffffff80245560>] vscnprintf+0x20 [<ffffffff8005071b>] vprintk+0x21b [<ffffffff8000f8eb>] get_runstate_snapshot+0x9b [<ffffffff8046cb88>] printk+0x4b [<ffffffff8000f8eb>] get_runstate_snapshot+0x9b [<ffffffff8004f5af>] warn_slowpath_common+0x3f [<ffffffff8004f662>] warn_slowpath_null+0x22 [<ffffffff8000f8eb>] get_runstate_snapshot+0x9b [<ffffffff8000fde9>] sched_clock+0x29 [<ffffffff80077316>] sched_clock_cpu+0x86 [<ffffffff800774ce>] cpu_clock+0x2e [<ffffffff80245560>] vscnprintf+0x20 [<ffffffff8005071b>] vprintk+0x21b [<ffffffff80302ba8>] take_machine_down+0x168 [<ffffffff8046cb88>] printk+0x4b [<ffffffff80239e7d>] report_bug+0x6d [<ffffffff80302ba8>] take_machine_down+0x168 [<ffffffff80011487>] die+0x97 [<ffffffff804711a4>] do_trap+0xe4 [<ffffffff80473948>] atomic_notifier_call_chain+0x28 [<ffffffff8000e2a2>] do_invalid_op+0xc2 [<ffffffff80302ba8>] take_machine_down+0x168 [<ffffffff800f4fbe>] apply_to_page_range+0x24e [<ffffffff80300940>] unmap_pte_fn [<ffffffff8000d175>] invalid_op+0x25 [<ffffffff80302ba8>] take_machine_down+0x168 [<ffffffff8009ec10>] stop_cpu [<ffffffff8009ecc1>] stop_cpu+0xb1 [<ffffffff80069443>] run_workqueue+0x83 [<ffffffff800696a4>] worker_thread+0xb4 [<ffffffff800701d0>] autoremove_wake_function [<ffffffff800695f0>] worker_thread [<ffffffff8006fb16>] kthread+0xb6 [<ffffffff8000d3ea>] child_rip+0xa [<ffffffff8006fa60>] kthread [<ffffffff8000d3e0>] child_rip # /usr/lib64/xen/bin/xenctx -s /root/ab1-system.map 180 1 xc_vcpu_getcontext: No data available # /usr/lib64/xen/bin/xenctx -s /root/ab1-system.map 180 2 xc_vcpu_getcontext: Invalid argument Hope this helps. Danilo -- To unsubscribe, e-mail: opensuse-virtual+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-virtual+help@opensuse.org
Jan Beulich wrote:
It seems that HVM emulates a 'PIIX3 IDE' controller and the driver seems to be built-in (i.e. not a kernel module):
Hmm, yes, that's ugly. I have always been advocating for modular drivers...
Things go south when 'xen-platform-pci' or 'xen-vbd' module is loaded...
Is there a way to disable a built-in driver??
I'm not aware of one, but we surely had found some way to use PV drivers in 11.4 guests, just that I don't recall the details - hopefully others reading the list will.
IIRC, use 'hda=noprobe' or similar on guest kernel cmdline to avoid the built-in driver. Jim -- To unsubscribe, e-mail: opensuse-virtual+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-virtual+help@opensuse.org
On 14.07.11 at 18:02, Jim Fehlig <jfehlig@novell.com> wrote: Jan Beulich wrote: It seems that HVM emulates a 'PIIX3 IDE' controller and the driver seems to be built-in (i.e. not a kernel module):
Hmm, yes, that's ugly. I have always been advocating for modular drivers...
Things go south when 'xen-platform-pci' or 'xen-vbd' module is loaded...
Is there a way to disable a built-in driver??
I'm not aware of one, but we surely had found some way to use PV drivers in 11.4 guests, just that I don't recall the details - hopefully others reading the list will.
IIRC, use 'hda=noprobe' or similar on guest kernel cmdline to avoid the built-in driver.
No, that's only applicable to the old IDE stack in the kernel (which shouldn't be used in newer kernels at all). Jan -- To unsubscribe, e-mail: opensuse-virtual+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-virtual+help@opensuse.org
participants (4)
-
Danilo Godec -
Jan Beulich -
Jan Beulich
-
Jim Fehlig