Date: Tue, 23 Jun 2015 09:45:59 +0300 From: Andriy Gapon <avg@FreeBSD.org> To: Neel Natu <neelnatu@gmail.com> Cc: "freebsd-virtualization@freebsd.org" <freebsd-virtualization@FreeBSD.org> Subject: Re: bhyve: centos 7.1 with multiple virtual processors Message-ID: <558900A7.40609@FreeBSD.org> In-Reply-To: <CAFgRE9Hpxm7pC_ETdQJKNk7FwbGvYjd60D0bnoOC=t46aJvusQ@mail.gmail.com> References: <5587EE05.2020001@FreeBSD.org> <CAFgRE9Hpxm7pC_ETdQJKNk7FwbGvYjd60D0bnoOC=t46aJvusQ@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 23/06/2015 05:37, Neel Natu wrote:
> Hi Andriy,
>
> FWIW I can boot up a Centos 7.1 virtual machine with 2 and 4 vcpus
> fine on my host with 8 physical cores.
>
> I have some questions about your setup inline.
>
> On Mon, Jun 22, 2015 at 4:14 AM, Andriy Gapon <avg@freebsd.org> wrote:
>>
>> If I run a CentOS 7.1 VM with more than one CPU more often than not it would
>> hang on startup and bhyve would start spinning.
>>
>> The following are the last messages seen in the VM:
>>
>> Switching to clocksource hpet
>> ------------[ cut here ]------------
>> WARNING: at kernel/time/clockevents.c:239 clockevents_program_event+0xdb/0xf0()
>> Modules linked in:
>> CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.10.0-229.4.2.el7.x86_64 #1
>> Hardware name: BHYVE, BIOS 1.00 03/14/2014
>> 0000000000000000 00000000cab5bdb6 ffff88003fc03e08 ffffffff81604eaa
>> ffff88003fc03e40 ffffffff8106e34b 80000000000f423f 80000000000f423f
>> ffffffff81915440 0000000000000000 0000000000000000 ffff88003fc03e50
>> Call Trace:
>> <IRQ> [<ffffffff81604eaa>] dump_stack+0x19/0x1b
>> [<ffffffff8106e34b>] warn_slowpath_common+0x6b/0xb0
>> [<ffffffff8106e49a>] warn_slowpath_null+0x1a/0x20
>> [<ffffffff810ce6eb>] clockevents_program_event+0xdb/0xf0
>> [<ffffffff810cf211>] tick_handle_periodic_broadcast+0x41/0x50
>> [<ffffffff81016525>] timer_interrupt+0x15/0x20
>> [<ffffffff8110b5ee>] handle_irq_event_percpu+0x3e/0x1e0
>> [<ffffffff8110b7cd>] handle_irq_event+0x3d/0x60
>> [<ffffffff8110e467>] handle_edge_irq+0x77/0x130
>> [<ffffffff81015cff>] handle_irq+0xbf/0x150
>> [<ffffffff81077df7>] ? irq_enter+0x17/0xa0
>> [<ffffffff816172af>] do_IRQ+0x4f/0xf0
>> [<ffffffff8160c4ad>] common_interrupt+0x6d/0x6d
>> <EOI> [<ffffffff8126e359>] ? selinux_inode_alloc_security+0x59/0xa0
>> [<ffffffff811de58f>] ? __d_instantiate+0xbf/0x100
>> [<ffffffff811de56f>] ? __d_instantiate+0x9f/0x100
>> [<ffffffff811de60d>] d_instantiate+0x3d/0x70
>> [<ffffffff8124d748>] debugfs_mknod.isra.5.part.6.constprop.15+0x98/0x130
>> [<ffffffff8124da82>] __create_file+0x1c2/0x2c0
>> [<ffffffff81a6c6bf>] ? set_graph_function+0x1f/0x1f
>> [<ffffffff8124dbcb>] debugfs_create_dir+0x1b/0x20
>> [<ffffffff8112c1ce>] tracing_init_dentry_tr+0x7e/0x90
>> [<ffffffff8112c250>] tracing_init_dentry+0x10/0x20
>> [<ffffffff81a6c6d2>] ftrace_init_debugfs+0x13/0x1fd
>> [<ffffffff81a6c6bf>] ? set_graph_function+0x1f/0x1f
>> [<ffffffff810020e8>] do_one_initcall+0xb8/0x230
>> [<ffffffff81a45203>] kernel_init_freeable+0x18b/0x22a
>> [<ffffffff81a449db>] ? initcall_blacklist+0xb0/0xb0
>> [<ffffffff815f33f0>] ? rest_init+0x80/0x80
>> [<ffffffff815f33fe>] kernel_init+0xe/0xf0
>> [<ffffffff81614d3c>] ret_from_fork+0x7c/0xb0
>> [<ffffffff815f33f0>] ? rest_init+0x80/0x80
>> ---[ end trace d5caa1cab8e7e98d ]---
>>
>
> A few questions to narrow this down:
> - Is the host very busy when the VM is started (or what is the host
> doing when this happened)?
The host typically is not heavily loaded. There is X server running and some
applications. I'd imagine that those could cause some additional latency but
not CPU starvation.
> - How many vcpus are you giving to the VM?
> - How many cores on the host?
I tried only 2 / 2.
>>
>> At the same time sometimes there is one or more of spurious NMIs on the _host_
>> system:
>> NMI ISA c, EISA ff
>> NMI ... going to debugger
>>
>
> Hmm, that's interesting. Are you using hwpmc to do instruction sampling?
hwpmc driver is in the kernel, but it was not used.
>> bhyve seems to spin here:
>> vmm.ko`svm_vmrun+0x894
>> vmm.ko`vm_run+0xbb7
>> vmm.ko`vmmdev_ioctl+0x5a4
>> kernel`devfs_ioctl_f+0x13b
>> kernel`kern_ioctl+0x1e1
>> kernel`sys_ioctl+0x16a
>> kernel`amd64_syscall+0x3ca
>> kernel`0xffffffff8088997b
>>
>> (kgdb) list *svm_vmrun+0x894
>> 0xffffffff813c9194 is in svm_vmrun
>> (/usr/src/sys/modules/vmm/../../amd64/vmm/amd/svm.c:1895).
>> 1890
>> 1891 static __inline void
>> 1892 enable_gintr(void)
>> 1893 {
>> 1894
>> 1895 __asm __volatile("stgi");
>> 1896 }
>> 1897
>> 1898 /*
>> 1899 * Start vcpu with specified RIP.
>>
>
> Yeah, that's not surprising because host interrupts are blocked when
> the cpu is executing in guest context. The 'enable_gintr()' enables
> interrupts so it gets blamed by the interrupt-based sampling.
>
> In this case it just means that the cpu was in guest context when a
> host-interrupt fired.
I see. FWIW, that was captured with DTrace.
--
Andriy Gapon
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?558900A7.40609>