public inbox for kvm@vger.kernel.org
 help / color / mirror / Atom feed
From: Marcelo Tosatti <mtosatti@redhat.com>
To: Michael Tokarev <mjt@tls.msk.ru>
Cc: KVM list <kvm@vger.kernel.org>
Subject: Re: *insane* amount of host_state_reloads for openbsd guest in 0.14
Date: Wed, 16 Mar 2011 16:48:11 -0300	[thread overview]
Message-ID: <20110316194811.GB23920@amt.cnet> (raw)
In-Reply-To: <4D7A0F59.1060604@msgid.tls.msk.ru>

On Fri, Mar 11, 2011 at 03:02:33PM +0300, Michael Tokarev wrote:
> I tried Openbsd 4.8 in kvm today and immediately noticed
> that it is running _very_ slow, boot takes several minutes
> to complete to the "login:" prompt.  After investigating
> I found that kvm does really insane amount of host_state_reloads.
> Here are typical kvm_stats:
> 
> During bootup while checking/mounting filesystems etc:
> 
>  host_state_reload                         13782074  671689
>  exits                                     24071297   45751
>  insn_emulation                            21707073   45353
>  io_exits                                   6900215     288
>  mmio_exits                                   20057     226
>  irq_injections                                6533     223
>  pf_fixed                                      9801     176
>  mmu_pte_write                                17781     111
>  irq_exits                                     6490      20
>  fpu_reload                                     133       1
>  halt_exits                                     422       0
>  mmu_cache_miss                                 333       0
>  halt_wakeup                                     30       0
>  mmu_shadow_zapped                              358       0
>  remote_tlb_flush                                 3       0
>  signal_exits                                     1       0
> 
> 
> After booting, idle guest sitting at login: prompt:
> 
>  host_state_reload                         28675531   44171
>  exits                                     25184108    3379
>  insn_emulation                            22784490    3266
>  irq_injections                               54139     200
>  mmio_exits                                   52748     122
>  mmu_pte_write                                41584     100
>  halt_exits                                   22731      99
>  irq_exits                                     9220      12
>  io_exits                                   6912620       9
>  fpu_reload                                     765       1
>  pf_fixed                                     12419       0
>  mmu_shadow_zapped                              358       0
>  mmu_cache_miss                                 339       0
>  halt_wakeup                                     38       0
>  remote_tlb_flush                                 3       0
>  signal_exits                                     1       0
> 
> 
> So I played with various options for a bit and found that:
> 
> o 0.14 -no-kvm-irqchip gets rid of host_state_reloads, making
>  them to be the same (+/-100) as insn_emulations, and speeds
>  up guest significantly (and lowers CPU usage on host during
>  guest idle time).
> 
> o 0.14 -no-kvm makes the guest *faster* than w/o any additional
>  options, reducing boot time for about 50% (but increasing host
>  CPU usage to maximum)
> 
> o 0.12 without any options is the fastest, and it keeps number
>  of reloads to be within 3k/sec range during all time. here's
>  typical kvm_stats of idle guest with 0.12:
> 
> kvm 0.12 and idle guest:
> 
>  exits                                     25975600    3329
>  insn_emulation                            23555272    3229
>  host_state_reload                          1898575    3117
>  irq_injections                               13125     200
>  mmio_exits                                   26603     122
>  mmu_pte_write                                61515     100
>  halt_exits                                    5173     100
>  irq_exits                                     6927       1
>  fpu_reload                                     499       1
>  io_exits                                   7058999       1
>  pf_fixed                                     11902       0
>  mmu_shadow_zapped                              634       0
>  mmu_cache_miss                                 558       0
>  halt_wakeup                                    166       0
>  remote_tlb_flush                                 2       0
>  signal_exits                                     1       0
> 
> 
> Again, running on the same AthlonII machine with 2.6.37 host
> kernel, freshly installed openbsd 4.3 32bit guest (with all
> defaults).
> 
> FreeBSD does not show this behavour.
> 
> What it can be?

Perhaps OpenBSD is accessing the HPET frequently, which is emulated 
in userspace. Try disabling it with "-no-hpet".


  parent reply	other threads:[~2011-03-16 19:49 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2011-03-11 12:02 *insane* amount of host_state_reloads for openbsd guest in 0.14 Michael Tokarev
2011-03-14 10:00 ` Daniel Bareiro
2011-03-16 19:48 ` Marcelo Tosatti [this message]
2011-03-16 19:55   ` Michael Tokarev
2011-03-17  7:24     ` Gleb Natapov

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20110316194811.GB23920@amt.cnet \
    --to=mtosatti@redhat.com \
    --cc=kvm@vger.kernel.org \
    --cc=mjt@tls.msk.ru \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox