[Qemu-devel] KVM call agenda for October 25

[Qemu-devel] KVM call agenda for October 25

[Juan Quintela]

Hi

Please send in any agenda items you are interested in covering.

Thanks, Juan.

Re: [Qemu-devel] KVM call agenda for October 25

[Paolo Bonzini]

On 10/24/2011 01:04 PM, Juan Quintela wrote:
>
> Hi
>
> Please send in any agenda items you are interested in covering.

- What's left to merge for 1.0.

- What kind of patch after the end of the freeze

Paolo

Re: [Qemu-devel] KVM call agenda for October 25

[Peter Maydell]

On 24 October 2011 12:35, Paolo Bonzini <pbonzini@redhat.com> wrote:
> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>> Please send in any agenda items you are interested in covering.
>
> - What's left to merge for 1.0.

Things on my list, FWIW:
 * current target-arm pullreq
 * PL041 support (needs another patch round to fix a minor bug
   Andrzej spotted)
 * cpu_single_env must be thread-local

I also think that it's somewhat unfortunate that we now will
compile on ARM hosts so that we always abort on startup (due
to the reliance on a working makecontext()) but I'm not really
sure how to deal with that one.

-- PMM

Re: [Qemu-devel] KVM call agenda for October 25

[Andreas Färber]

Am 24.10.2011 14:02, schrieb Peter Maydell:
> On 24 October 2011 12:35, Paolo Bonzini <pbonzini@redhat.com> wrote:
>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>> Please send in any agenda items you are interested in covering.
>>
>> - What's left to merge for 1.0.

> I also think that it's somewhat unfortunate that we now will
> compile on ARM hosts so that we always abort on startup (due
> to the reliance on a working makecontext()) but I'm not really
> sure how to deal with that one.

FWIW we're also not working / not building on Darwin ppc+Intel, which is
related to a) softfloat integer types, b) GThread initialization, c)
unknown issues. Bisecting did not work well and I am lacking time and
ideas to investigate and fix this. For softfloat there are several
solutions around, in need of a decision.

Nice to merge would be the Cocoa sheet issue, once verified.

Andreas

Re: [Qemu-devel] KVM call agenda for October 25

[Luiz Capitulino]

On Mon, 24 Oct 2011 13:02:05 +0100
Peter Maydell <peter.maydell@linaro.org> wrote:

> On 24 October 2011 12:35, Paolo Bonzini <pbonzini@redhat.com> wrote:
> > On 10/24/2011 01:04 PM, Juan Quintela wrote:
> >> Please send in any agenda items you are interested in covering.
> >
> > - What's left to merge for 1.0.
> 
> Things on my list, FWIW:
>  * current target-arm pullreq
>  * PL041 support (needs another patch round to fix a minor bug
>    Andrzej spotted)
>  * cpu_single_env must be thread-local

I submitted today the second round of QAPI conversions, which converts all
existing QMP query commands to the QAPI (plus some fixes).

I expect that to make 1.0.

Re: [Qemu-devel] KVM call agenda for October 25

[Kevin Wolf]

Am 24.10.2011 13:35, schrieb Paolo Bonzini:
> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>
>> Hi
>>
>> Please send in any agenda items you are interested in covering.
> 
> - What's left to merge for 1.0.

I would still like to cache the default cache mode (probably to
cache=writeback). We don't allow guests to toggle WCE yet which Anthony
would have liked to see before doing the change. Is it a strict requirement?

Kevin

Re: [Qemu-devel] KVM call agenda for October 25

[Anthony Liguori]

On 10/25/2011 07:35 AM, Kevin Wolf wrote:
> Am 24.10.2011 13:35, schrieb Paolo Bonzini:
>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>>
>>> Hi
>>>
>>> Please send in any agenda items you are interested in covering.
>>
>> - What's left to merge for 1.0.
>
> I would still like to cache the default cache mode (probably to
> cache=writeback). We don't allow guests to toggle WCE yet which Anthony
> would have liked to see before doing the change. Is it a strict requirement?

I don't see a way around it.  If the default mode is cache=writeback, then we're 
open to data corruption in any guest where barrier=0.  With guest togglable WCE, 
it ends up being a guest configuration issue so we can more or less defer 
responsibility.

Do you think it's a good idea to change the default mode w/o guest WCE toggle 
support?  What's your view about older guests if we change the default mode? 
What's your main motivation for wanting to change the default mode?

I'd be much more open to changing the default mode to cache=none FWIW since the 
risk of data loss there is much, much lower.

Regards,

Anthony Liguori

>
> Kevin
>

Re: [Qemu-devel] KVM call agenda for October 25

[Dor Laor]

On 10/25/2011 03:05 PM, Anthony Liguori wrote:
> On 10/25/2011 07:35 AM, Kevin Wolf wrote:
>> Am 24.10.2011 13:35, schrieb Paolo Bonzini:
>>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>>>
>>>> Hi
>>>>
>>>> Please send in any agenda items you are interested in covering.
>>>
>>> - What's left to merge for 1.0.
>>
>> I would still like to cache the default cache mode (probably to
>> cache=writeback). We don't allow guests to toggle WCE yet which Anthony
>> would have liked to see before doing the change. Is it a strict
>> requirement?
>
> I don't see a way around it. If the default mode is cache=writeback,
> then we're open to data corruption in any guest where barrier=0. With
> guest togglable WCE, it ends up being a guest configuration issue so we
> can more or less defer responsibility.
>
> Do you think it's a good idea to change the default mode w/o guest WCE
> toggle support? What's your view about older guests if we change the
> default mode? What's your main motivation for wanting to change the
> default mode?
>
> I'd be much more open to changing the default mode to cache=none FWIW
> since the risk of data loss there is much, much lower.

A bit related to this, it would be nice to mark a VM un-migratable if 
cache!=none. Juan reports that currently it such VMs are exposed to data 
integrity issues so we need to fail migrating them automatically.

>
> Regards,
>
> Anthony Liguori
>
>>
>> Kevin
>>
>
>

Re: [Qemu-devel] KVM call agenda for October 25

[Anthony Liguori]

On 10/25/2011 08:18 AM, Dor Laor wrote:
> On 10/25/2011 03:05 PM, Anthony Liguori wrote:
>> On 10/25/2011 07:35 AM, Kevin Wolf wrote:
>>> Am 24.10.2011 13:35, schrieb Paolo Bonzini:
>>>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>>>>
>>>>> Hi
>>>>>
>>>>> Please send in any agenda items you are interested in covering.
>>>>
>>>> - What's left to merge for 1.0.
>>>
>>> I would still like to cache the default cache mode (probably to
>>> cache=writeback). We don't allow guests to toggle WCE yet which Anthony
>>> would have liked to see before doing the change. Is it a strict
>>> requirement?
>>
>> I don't see a way around it. If the default mode is cache=writeback,
>> then we're open to data corruption in any guest where barrier=0. With
>> guest togglable WCE, it ends up being a guest configuration issue so we
>> can more or less defer responsibility.
>>
>> Do you think it's a good idea to change the default mode w/o guest WCE
>> toggle support? What's your view about older guests if we change the
>> default mode? What's your main motivation for wanting to change the
>> default mode?
>>
>> I'd be much more open to changing the default mode to cache=none FWIW
>> since the risk of data loss there is much, much lower.
>
> A bit related to this, it would be nice to mark a VM un-migratable if
> cache!=none. Juan reports that currently it such VMs are exposed to data
> integrity issues so we need to fail migrating them automatically.

That's not correct.  cache!=none is perfectly safe *if* you have coherent shared 
storage.

Regards,

Anthony Liguori

>
>>
>> Regards,
>>
>> Anthony Liguori
>>
>>>
>>> Kevin
>>>
>>
>>
>
>

Re: [Qemu-devel] KVM call agenda for October 25

[Andreas Färber]

Am 25.10.2011 15:18, schrieb Dor Laor:
> [...] it would be nice to mark a VM un-migratable [snip]

Speaking of which, I'm working on the missing migration support for AHCI
but fear I won't quite make it for the Nov 1 deadline.

Andreas

-- 
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg

Re: [Qemu-devel] KVM call agenda for October 25

[Kevin Wolf]

Am 25.10.2011 15:05, schrieb Anthony Liguori:
> On 10/25/2011 07:35 AM, Kevin Wolf wrote:
>> Am 24.10.2011 13:35, schrieb Paolo Bonzini:
>>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>>>
>>>> Hi
>>>>
>>>> Please send in any agenda items you are interested in covering.
>>>
>>> - What's left to merge for 1.0.
>>
>> I would still like to cache the default cache mode (probably to
>> cache=writeback). We don't allow guests to toggle WCE yet which Anthony
>> would have liked to see before doing the change. Is it a strict requirement?
> 
> I don't see a way around it.  If the default mode is cache=writeback, then we're 
> open to data corruption in any guest where barrier=0.  With guest togglable WCE, 
> it ends up being a guest configuration issue so we can more or less defer 
> responsibility.

So do you think that offering a WCE inside the guest would be a real
solution or just a way to have an excuse?

Christoph said that OSes don't usually change this by themselves, it
would need an administrator manually changing the setting. But if we
require that, we can just as well require that the administrator set
cache=writethrough on the qemu command line.

> Do you think it's a good idea to change the default mode w/o guest WCE toggle 
> support?  What's your view about older guests if we change the default mode? 
> What's your main motivation for wanting to change the default mode?

Because people are constantly complaining about the awful
(cache=writethrough) performance they get before they are told they
should use a different cache option. And they are right. The
out-of-the-box experience with qemu's block performance really sucks.

> I'd be much more open to changing the default mode to cache=none FWIW since the 
> risk of data loss there is much, much lower.

I think people said that they'd rather not have cache=none as default
because O_DIRECT doesn't work everywhere.

Kevin

Re: [Qemu-devel] KVM call agenda for October 25

[Anthony Liguori]

On 10/25/2011 08:56 AM, Kevin Wolf wrote:
> Am 25.10.2011 15:05, schrieb Anthony Liguori:
>> On 10/25/2011 07:35 AM, Kevin Wolf wrote:
>>> Am 24.10.2011 13:35, schrieb Paolo Bonzini:
>>>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>>>>
>>>>> Hi
>>>>>
>>>>> Please send in any agenda items you are interested in covering.
>>>>
>>>> - What's left to merge for 1.0.
>>>
>>> I would still like to cache the default cache mode (probably to
>>> cache=writeback). We don't allow guests to toggle WCE yet which Anthony
>>> would have liked to see before doing the change. Is it a strict requirement?
>>
>> I don't see a way around it.  If the default mode is cache=writeback, then we're
>> open to data corruption in any guest where barrier=0.  With guest togglable WCE,
>> it ends up being a guest configuration issue so we can more or less defer
>> responsibility.
>
> So do you think that offering a WCE inside the guest would be a real
> solution or just a way to have an excuse?

No, it offers a mechanism to "fix mistakes" at run-time verses at start up time. 
  It also means that you can make template images that understand that they 
don't support barriers and change the WCE setting appropriately.

> Christoph said that OSes don't usually change this by themselves, it
> would need an administrator manually changing the setting. But if we
> require that, we can just as well require that the administrator set
> cache=writethrough on the qemu command line.

The administrator of the guest != the administrator of the host.

>> Do you think it's a good idea to change the default mode w/o guest WCE toggle
>> support?  What's your view about older guests if we change the default mode?
>> What's your main motivation for wanting to change the default mode?
>
> Because people are constantly complaining about the awful
> (cache=writethrough) performance they get before they are told they
> should use a different cache option. And they are right. The
> out-of-the-box experience with qemu's block performance really sucks.

With qcow2 you mean, right?

>> I'd be much more open to changing the default mode to cache=none FWIW since the
>> risk of data loss there is much, much lower.
>
> I think people said that they'd rather not have cache=none as default
> because O_DIRECT doesn't work everywhere.

Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
works on ext[234], btrfs, nfs.

Regards,

Anthony Liguori

> Kevin
>

Re: [Qemu-devel] KVM call agenda for October 25

[Kevin Wolf]

Am 25.10.2011 16:06, schrieb Anthony Liguori:
> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
>>> On 10/25/2011 07:35 AM, Kevin Wolf wrote:
>>>> Am 24.10.2011 13:35, schrieb Paolo Bonzini:
>>>>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>>>>>
>>>>>> Hi
>>>>>>
>>>>>> Please send in any agenda items you are interested in covering.
>>>>>
>>>>> - What's left to merge for 1.0.
>>>>
>>>> I would still like to cache the default cache mode (probably to
>>>> cache=writeback). We don't allow guests to toggle WCE yet which Anthony
>>>> would have liked to see before doing the change. Is it a strict requirement?
>>>
>>> I don't see a way around it.  If the default mode is cache=writeback, then we're
>>> open to data corruption in any guest where barrier=0.  With guest togglable WCE,
>>> it ends up being a guest configuration issue so we can more or less defer
>>> responsibility.
>>
>> So do you think that offering a WCE inside the guest would be a real
>> solution or just a way to have an excuse?
> 
> No, it offers a mechanism to "fix mistakes" at run-time verses at start up time. 

This is true (in both directions). But I think it's independent from the
right default.

>   It also means that you can make template images that understand that they 
> don't support barriers and change the WCE setting appropriately.

Isn't that really a job for management tools?

>> Christoph said that OSes don't usually change this by themselves, it
>> would need an administrator manually changing the setting. But if we
>> require that, we can just as well require that the administrator set
>> cache=writethrough on the qemu command line.
> 
> The administrator of the guest != the administrator of the host.

But the administrator of the guest == the owner of the qemu instance,
no? He should be the one to use the management tools and configure his VMs.

>>> Do you think it's a good idea to change the default mode w/o guest WCE toggle
>>> support?  What's your view about older guests if we change the default mode?
>>> What's your main motivation for wanting to change the default mode?
>>
>> Because people are constantly complaining about the awful
>> (cache=writethrough) performance they get before they are told they
>> should use a different cache option. And they are right. The
>> out-of-the-box experience with qemu's block performance really sucks.
> 
> With qcow2 you mean, right?

No, with any format, including raw. Which isn't surprising at all,
O_SYNC makes writes very expensive.

>>> I'd be much more open to changing the default mode to cache=none FWIW since the
>>> risk of data loss there is much, much lower.
>>
>> I think people said that they'd rather not have cache=none as default
>> because O_DIRECT doesn't work everywhere.
> 
> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
> works on ext[234], btrfs, nfs.

I think tmpfs was named (and failing to start with default settings on
tmpfs would be nasty enough), but iirc Alex had another one.

Kevin

Re: [Qemu-devel] KVM call agenda for October 25

[Alexander Graf]

On 25.10.2011, at 17:32, Kevin Wolf <kwolf@redhat.com> wrote:

> Am 25.10.2011 16:06, schrieb Anthony Liguori:
>> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
>>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
>>>> On 10/25/2011 07:35 AM, Kevin Wolf wrote:
>>>>> Am 24.10.2011 13:35, schrieb Paolo Bonzini:
>>>>>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>>>>>> 
>>>>>>> Hi
>>>>>>> 
>>>>>>> Please send in any agenda items you are interested in covering.
>>>>>> 
>>>>>> - What's left to merge for 1.0.
>>>>> 
>>>>> I would still like to cache the default cache mode (probably to
>>>>> cache=writeback). We don't allow guests to toggle WCE yet which Anthony
>>>>> would have liked to see before doing the change. Is it a strict requirement?
>>>> 
>>>> I don't see a way around it.  If the default mode is cache=writeback, then we're
>>>> open to data corruption in any guest where barrier=0.  With guest togglable WCE,
>>>> it ends up being a guest configuration issue so we can more or less defer
>>>> responsibility.
>>> 
>>> So do you think that offering a WCE inside the guest would be a real
>>> solution or just a way to have an excuse?
>> 
>> No, it offers a mechanism to "fix mistakes" at run-time verses at start up time. 
> 
> This is true (in both directions). But I think it's independent from the
> right default.
> 
>>  It also means that you can make template images that understand that they 
>> don't support barriers and change the WCE setting appropriately.
> 
> Isn't that really a job for management tools?
> 
>>> Christoph said that OSes don't usually change this by themselves, it
>>> would need an administrator manually changing the setting. But if we
>>> require that, we can just as well require that the administrator set
>>> cache=writethrough on the qemu command line.
>> 
>> The administrator of the guest != the administrator of the host.
> 
> But the administrator of the guest == the owner of the qemu instance,
> no? He should be the one to use the management tools and configure his VMs.
> 
>>>> Do you think it's a good idea to change the default mode w/o guest WCE toggle
>>>> support?  What's your view about older guests if we change the default mode?
>>>> What's your main motivation for wanting to change the default mode?
>>> 
>>> Because people are constantly complaining about the awful
>>> (cache=writethrough) performance they get before they are told they
>>> should use a different cache option. And they are right. The
>>> out-of-the-box experience with qemu's block performance really sucks.
>> 
>> With qcow2 you mean, right?
> 
> No, with any format, including raw. Which isn't surprising at all,
> O_SYNC makes writes very expensive.
> 
>>>> I'd be much more open to changing the default mode to cache=none FWIW since the
>>>> risk of data loss there is much, much lower.
>>> 
>>> I think people said that they'd rather not have cache=none as default
>>> because O_DIRECT doesn't work everywhere.
>> 
>> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
>> works on ext[234], btrfs, nfs.
> 
> I think tmpfs was named (and failing to start with default settings on
> tmpfs would be nasty enough), but iirc Alex had another one.

Yeah, IIRC NFS also failed on me :)

Alex

> 
> Kevin

Re: [Qemu-devel] KVM call agenda for October 25

[Kevin Wolf]

Am 25.10.2011 16:06, schrieb Anthony Liguori:
> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
>>> I'd be much more open to changing the default mode to cache=none FWIW since the
>>> risk of data loss there is much, much lower.
>>
>> I think people said that they'd rather not have cache=none as default
>> because O_DIRECT doesn't work everywhere.
> 
> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
> works on ext[234], btrfs, nfs.

Besides file systems (and probably OSes) that don't support O_DIRECT,
there's another case: Our defaults don't work on 4k sector disks today.
You need to explicitly specify the logical_block_size qdev property for
cache=none to work on them.

And changing this default isn't trivial as the right value doesn't only
depend on the host disk, but it's also guest visible. The only way out
would be bounce buffers, but I'm not sure that doing that silently is a
good idea...

Kevin

Re: [Qemu-devel] KVM call agenda for October 25

[Markus Armbruster]

Kevin Wolf <kwolf@redhat.com> writes:

> Am 25.10.2011 16:06, schrieb Anthony Liguori:
>> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
>>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
>>>> I'd be much more open to changing the default mode to cache=none FWIW since the
>>>> risk of data loss there is much, much lower.
>>>
>>> I think people said that they'd rather not have cache=none as default
>>> because O_DIRECT doesn't work everywhere.
>> 
>> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
>> works on ext[234], btrfs, nfs.
>
> Besides file systems (and probably OSes) that don't support O_DIRECT,
> there's another case: Our defaults don't work on 4k sector disks today.
> You need to explicitly specify the logical_block_size qdev property for
> cache=none to work on them.
>
> And changing this default isn't trivial as the right value doesn't only
> depend on the host disk, but it's also guest visible. The only way out
> would be bounce buffers, but I'm not sure that doing that silently is a
> good idea...

Sector size is a device property.

If the user asks for a 4K sector disk, and the backend can't support
that, we need to reject the configuration.  Just like we reject
read-only backends for read/write disks.

If the backend can only support it by using bounce buffers, I'd say
reject it unless the user explicitly permits bounce buffers.  But that's
debatable.

It's okay to default device properties to some backend-dependent value,
if that improves usability.

Re: [Qemu-devel] KVM call agenda for October 25

[Paolo Bonzini]

On 10/26/2011 10:48 AM, Markus Armbruster wrote:
> Sector size is a device property.
>
> If the user asks for a 4K sector disk, and the backend can't support
> that, we need to reject the configuration.  Just like we reject
> read-only backends for read/write disks.

Isn't it the other way round, i.e. the user asks for a 512-byte sector 
disk (i.e. the default) with cache=none but the disk has 4k sectors? 
We're basically saying "choose between NFS and migration if you have 4k 
sector disks but your guest doesn't support them".  Understandable 
perhaps, but not exactly kind, and virtualization is also about 
shielding from this kind of hardware dependency even at the cost of 
performance.  QEMU should just warn about performance degradations, 
erroring out would be a policy decision that should be up to management.

> It's okay to default device properties to some backend-dependent value,
> if that improves usability.

On the other hand, not all guests support 4k-sectors properly.

Paolo

Re: [Qemu-devel] KVM call agenda for October 25

[Daniel P. Berrange]

On Wed, Oct 26, 2011 at 10:48:12AM +0200, Markus Armbruster wrote:
> Kevin Wolf <kwolf@redhat.com> writes:
> 
> > Am 25.10.2011 16:06, schrieb Anthony Liguori:
> >> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
> >>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
> >>>> I'd be much more open to changing the default mode to cache=none FWIW since the
> >>>> risk of data loss there is much, much lower.
> >>>
> >>> I think people said that they'd rather not have cache=none as default
> >>> because O_DIRECT doesn't work everywhere.
> >> 
> >> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
> >> works on ext[234], btrfs, nfs.
> >
> > Besides file systems (and probably OSes) that don't support O_DIRECT,
> > there's another case: Our defaults don't work on 4k sector disks today.
> > You need to explicitly specify the logical_block_size qdev property for
> > cache=none to work on them.
> >
> > And changing this default isn't trivial as the right value doesn't only
> > depend on the host disk, but it's also guest visible. The only way out
> > would be bounce buffers, but I'm not sure that doing that silently is a
> > good idea...
> 
> Sector size is a device property.
> 
> If the user asks for a 4K sector disk, and the backend can't support
> that, we need to reject the configuration.  Just like we reject
> read-only backends for read/write disks.

I don't see why we need to reject a guest disk with 4k sectors,
just because the host disk only has 512 byte sectors. A guest
sector size that's a larger multiple of host sector size should
work just fine. It just means any guest sector write will update
8 host sectors at a time. We only have problems if guest sector
size is not a multiple of host sector size, in which case bounce
buffers are the only option (other than rejecting the config
which is not too nice).

IIUC, current QEMU behaviour is

           Guest 512    Guest 4k
 Host 512   * OK          OK
 Host 4k    * I/O Err     OK

'*' marks defaults

IMHO, QEMU needs to work withot I/O errors in all of these
combinations, even if this means having to use bounce buffers
in some of them. That said, IMHO the default should be for
QEMU to avoid bounce buffers, which implies it should either
chose guest sector size to match host sector size, or it
should unconditionally use 4k guest. IMHO we need the former

           Guest 512  Guest 4k
 Host 512   *OK         OK
 Host 4k     OK        *OK


Yes, I know there are other wierd sector sizes besides 512
and 4k, but the same general principals apply of either one
being a multiple of the other, or needing to use bounce
buffers.

> If the backend can only support it by using bounce buffers, I'd say
> reject it unless the user explicitly permits bounce buffers.  But that's
> debatable.

I don't think it really adds value for QEMU to force the user to specify
some extra magic flag in order to make the user's requested config
actually be honoured. If a config needs bounce buffers, QEMU should just
do it, without needing 'use-bounce-buffers=1'. A higher level mgmt app
is in a better position to inform users about the consequences.


Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|

Re: [Qemu-devel] KVM call agenda for October 25

[Markus Armbruster]

Paolo Bonzini <pbonzini@redhat.com> writes:

> On 10/26/2011 10:48 AM, Markus Armbruster wrote:
>> Sector size is a device property.
>>
>> If the user asks for a 4K sector disk, and the backend can't support
>> that, we need to reject the configuration.  Just like we reject
>> read-only backends for read/write disks.
>
> Isn't it the other way round, i.e. the user asks for a 512-byte sector
> disk (i.e. the default) with cache=none but the disk has 4k sectors?

Let me rephrase: If the user asks for a FOO disk, and the backend can't
support that, we need to reject the configuration.  Just like we reject
read-only backends for read/write disks.

> We're basically saying "choose between NFS and migration if you have
> 4k sector disks but your guest doesn't support them".  Understandable
> perhaps, but not exactly kind, and virtualization is also about
> shielding from this kind of hardware dependency even at the cost of
> performance.  QEMU should just warn about performance degradations,
> erroring out would be a policy decision that should be up to
> management.

I don't have strong opinions on that.

>> It's okay to default device properties to some backend-dependent value,
>> if that improves usability.
>
> On the other hand, not all guests support 4k-sectors properly.

You can't pick perfect defaults for all conceivable guests.  Life's
tough.

Re: [Qemu-devel] KVM call agenda for October 25

[Kevin Wolf]

Am 26.10.2011 11:57, schrieb Daniel P. Berrange:
> On Wed, Oct 26, 2011 at 10:48:12AM +0200, Markus Armbruster wrote:
>> Kevin Wolf <kwolf@redhat.com> writes:
>>
>>> Am 25.10.2011 16:06, schrieb Anthony Liguori:
>>>> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
>>>>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
>>>>>> I'd be much more open to changing the default mode to cache=none FWIW since the
>>>>>> risk of data loss there is much, much lower.
>>>>>
>>>>> I think people said that they'd rather not have cache=none as default
>>>>> because O_DIRECT doesn't work everywhere.
>>>>
>>>> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
>>>> works on ext[234], btrfs, nfs.
>>>
>>> Besides file systems (and probably OSes) that don't support O_DIRECT,
>>> there's another case: Our defaults don't work on 4k sector disks today.
>>> You need to explicitly specify the logical_block_size qdev property for
>>> cache=none to work on them.
>>>
>>> And changing this default isn't trivial as the right value doesn't only
>>> depend on the host disk, but it's also guest visible. The only way out
>>> would be bounce buffers, but I'm not sure that doing that silently is a
>>> good idea...
>>
>> Sector size is a device property.
>>
>> If the user asks for a 4K sector disk, and the backend can't support
>> that, we need to reject the configuration.  Just like we reject
>> read-only backends for read/write disks.
> 
> I don't see why we need to reject a guest disk with 4k sectors,
> just because the host disk only has 512 byte sectors. A guest
> sector size that's a larger multiple of host sector size should
> work just fine. It just means any guest sector write will update
> 8 host sectors at a time. We only have problems if guest sector
> size is not a multiple of host sector size, in which case bounce
> buffers are the only option (other than rejecting the config
> which is not too nice).
> 
> IIUC, current QEMU behaviour is
> 
>            Guest 512    Guest 4k
>  Host 512   * OK          OK
>  Host 4k    * I/O Err     OK
> 
> '*' marks defaults
> 
> IMHO, QEMU needs to work withot I/O errors in all of these
> combinations, even if this means having to use bounce buffers
> in some of them. That said, IMHO the default should be for
> QEMU to avoid bounce buffers, which implies it should either
> chose guest sector size to match host sector size, or it
> should unconditionally use 4k guest. IMHO we need the former
> 
>            Guest 512  Guest 4k
>  Host 512   *OK         OK
>  Host 4k     OK        *OK

I'm not sure if a 4k host should imply a 4k guest by default. This means
that some guests wouldn't be able to run on a 4k host. On the other
hand, for those guests that can do 4k, it would be the much better option.

So I think this decision is the hard thing about it.

> Yes, I know there are other wierd sector sizes besides 512
> and 4k, but the same general principals apply of either one
> being a multiple of the other, or needing to use bounce
> buffers.
> 
>> If the backend can only support it by using bounce buffers, I'd say
>> reject it unless the user explicitly permits bounce buffers.  But that's
>> debatable.
> 
> I don't think it really adds value for QEMU to force the user to specify
> some extra magic flag in order to make the user's requested config
> actually be honoured. 

The user's requested config is often enough something like "-hda
foo.img". Give me a working disk, I don't care how do you it. (And of
course I don't tell you what sector sizes my guest can cope with)

> If a config needs bounce buffers, QEMU should just
> do it, without needing 'use-bounce-buffers=1'. A higher level mgmt app
> is in a better position to inform users about the consequences.

A higher level management app doesn't exist in the general case.

Kevin

Re: [Qemu-devel] KVM call agenda for October 25

[Daniel P. Berrange]

On Wed, Oct 26, 2011 at 01:23:05PM +0200, Kevin Wolf wrote:
> Am 26.10.2011 11:57, schrieb Daniel P. Berrange:
> > On Wed, Oct 26, 2011 at 10:48:12AM +0200, Markus Armbruster wrote:
> >> Kevin Wolf <kwolf@redhat.com> writes:
> >>
> >>> Am 25.10.2011 16:06, schrieb Anthony Liguori:
> >>>> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
> >>>>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
> >>>>>> I'd be much more open to changing the default mode to cache=none FWIW since the
> >>>>>> risk of data loss there is much, much lower.
> >>>>>
> >>>>> I think people said that they'd rather not have cache=none as default
> >>>>> because O_DIRECT doesn't work everywhere.
> >>>>
> >>>> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
> >>>> works on ext[234], btrfs, nfs.
> >>>
> >>> Besides file systems (and probably OSes) that don't support O_DIRECT,
> >>> there's another case: Our defaults don't work on 4k sector disks today.
> >>> You need to explicitly specify the logical_block_size qdev property for
> >>> cache=none to work on them.
> >>>
> >>> And changing this default isn't trivial as the right value doesn't only
> >>> depend on the host disk, but it's also guest visible. The only way out
> >>> would be bounce buffers, but I'm not sure that doing that silently is a
> >>> good idea...
> >>
> >> Sector size is a device property.
> >>
> >> If the user asks for a 4K sector disk, and the backend can't support
> >> that, we need to reject the configuration.  Just like we reject
> >> read-only backends for read/write disks.
> > 
> > I don't see why we need to reject a guest disk with 4k sectors,
> > just because the host disk only has 512 byte sectors. A guest
> > sector size that's a larger multiple of host sector size should
> > work just fine. It just means any guest sector write will update
> > 8 host sectors at a time. We only have problems if guest sector
> > size is not a multiple of host sector size, in which case bounce
> > buffers are the only option (other than rejecting the config
> > which is not too nice).
> > 
> > IIUC, current QEMU behaviour is
> > 
> >            Guest 512    Guest 4k
> >  Host 512   * OK          OK
> >  Host 4k    * I/O Err     OK
> > 
> > '*' marks defaults
> > 
> > IMHO, QEMU needs to work withot I/O errors in all of these
> > combinations, even if this means having to use bounce buffers
> > in some of them. That said, IMHO the default should be for
> > QEMU to avoid bounce buffers, which implies it should either
> > chose guest sector size to match host sector size, or it
> > should unconditionally use 4k guest. IMHO we need the former
> > 
> >            Guest 512  Guest 4k
> >  Host 512   *OK         OK
> >  Host 4k     OK        *OK
> 
> I'm not sure if a 4k host should imply a 4k guest by default. This means
> that some guests wouldn't be able to run on a 4k host. On the other
> hand, for those guests that can do 4k, it would be the much better option.
> 
> So I think this decision is the hard thing about it.

I guess it somewhat depends whether we want to strive for

 1. Give the user the fastest working config by default
 2. Give the user a working config by default
 3. Give the user the fastest (possibly broken) config by default

IMHO 3 is not a serious option, but I could see 2 as a reasonable
tradeoff to avoid complexity in chosing QEMU defaults. The user
would have a working config with 512 sectors, but sub-optimal perf
on 4k hosts due to bounce buffering. Ideally libvirt or other
higher app would be setting the best block size that a guest
can support by default, so bounce buffers would rarely be needed.
So only people using QEMU directly without setting a block size
would ordinarily suffer the bounce buffer perf hit on a 4k host
host

Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|

Re: [Qemu-devel] KVM call agenda for October 25

[Kevin Wolf]

Am 26.10.2011 13:39, schrieb Daniel P. Berrange:
> On Wed, Oct 26, 2011 at 01:23:05PM +0200, Kevin Wolf wrote:
>> Am 26.10.2011 11:57, schrieb Daniel P. Berrange:
>>> On Wed, Oct 26, 2011 at 10:48:12AM +0200, Markus Armbruster wrote:
>>>> Kevin Wolf <kwolf@redhat.com> writes:
>>>>
>>>>> Am 25.10.2011 16:06, schrieb Anthony Liguori:
>>>>>> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
>>>>>>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
>>>>>>>> I'd be much more open to changing the default mode to cache=none FWIW since the
>>>>>>>> risk of data loss there is much, much lower.
>>>>>>>
>>>>>>> I think people said that they'd rather not have cache=none as default
>>>>>>> because O_DIRECT doesn't work everywhere.
>>>>>>
>>>>>> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it 
>>>>>> works on ext[234], btrfs, nfs.
>>>>>
>>>>> Besides file systems (and probably OSes) that don't support O_DIRECT,
>>>>> there's another case: Our defaults don't work on 4k sector disks today.
>>>>> You need to explicitly specify the logical_block_size qdev property for
>>>>> cache=none to work on them.
>>>>>
>>>>> And changing this default isn't trivial as the right value doesn't only
>>>>> depend on the host disk, but it's also guest visible. The only way out
>>>>> would be bounce buffers, but I'm not sure that doing that silently is a
>>>>> good idea...
>>>>
>>>> Sector size is a device property.
>>>>
>>>> If the user asks for a 4K sector disk, and the backend can't support
>>>> that, we need to reject the configuration.  Just like we reject
>>>> read-only backends for read/write disks.
>>>
>>> I don't see why we need to reject a guest disk with 4k sectors,
>>> just because the host disk only has 512 byte sectors. A guest
>>> sector size that's a larger multiple of host sector size should
>>> work just fine. It just means any guest sector write will update
>>> 8 host sectors at a time. We only have problems if guest sector
>>> size is not a multiple of host sector size, in which case bounce
>>> buffers are the only option (other than rejecting the config
>>> which is not too nice).
>>>
>>> IIUC, current QEMU behaviour is
>>>
>>>            Guest 512    Guest 4k
>>>  Host 512   * OK          OK
>>>  Host 4k    * I/O Err     OK
>>>
>>> '*' marks defaults
>>>
>>> IMHO, QEMU needs to work withot I/O errors in all of these
>>> combinations, even if this means having to use bounce buffers
>>> in some of them. That said, IMHO the default should be for
>>> QEMU to avoid bounce buffers, which implies it should either
>>> chose guest sector size to match host sector size, or it
>>> should unconditionally use 4k guest. IMHO we need the former
>>>
>>>            Guest 512  Guest 4k
>>>  Host 512   *OK         OK
>>>  Host 4k     OK        *OK
>>
>> I'm not sure if a 4k host should imply a 4k guest by default. This means
>> that some guests wouldn't be able to run on a 4k host. On the other
>> hand, for those guests that can do 4k, it would be the much better option.
>>
>> So I think this decision is the hard thing about it.
> 
> I guess it somewhat depends whether we want to strive for
> 
>  1. Give the user the fastest working config by default
>  2. Give the user a working config by default
>  3. Give the user the fastest (possibly broken) config by default
> 
> IMHO 3 is not a serious option, but I could see 2 as a reasonable
> tradeoff to avoid complexity in chosing QEMU defaults. The user
> would have a working config with 512 sectors, but sub-optimal perf
> on 4k hosts due to bounce buffering. Ideally libvirt or other
> higher app would be setting the best block size that a guest
> can support by default, so bounce buffers would rarely be needed.
> So only people using QEMU directly without setting a block size
> would ordinarily suffer the bounce buffer perf hit on a 4k host
> host

Yes, I'm currently tending towards this plus a warning on stderr if
bounce buffering is used.

Or, coming back to the original subject of this discussion, we can
default to cache=writeback and forget about alignment. If you specify
cache=none, you have to take care to explicitly specify a block size >
512 bytes, too.

Maybe the best is actually to do both: Default to cache=writeback,
completely avoiding bounce buffers. If the user specifies cache=none,
but doesn't change the sector size of the virtual disk, print a warning
and enable bounce buffers.

Kevin

Re: [Qemu-devel] KVM call agenda for October 25

[Anthony Liguori]

On 10/25/2011 10:32 AM, Kevin Wolf wrote:
> Am 25.10.2011 16:06, schrieb Anthony Liguori:
>> On 10/25/2011 08:56 AM, Kevin Wolf wrote:
>>> Am 25.10.2011 15:05, schrieb Anthony Liguori:
>>>> On 10/25/2011 07:35 AM, Kevin Wolf wrote:
>>>>> Am 24.10.2011 13:35, schrieb Paolo Bonzini:
>>>>>> On 10/24/2011 01:04 PM, Juan Quintela wrote:
>>>>>>>
>>>>>>> Hi
>>>>>>>
>>>>>>> Please send in any agenda items you are interested in covering.
>>>>>>
>>>>>> - What's left to merge for 1.0.
>>>>>
>>>>> I would still like to cache the default cache mode (probably to
>>>>> cache=writeback). We don't allow guests to toggle WCE yet which Anthony
>>>>> would have liked to see before doing the change. Is it a strict requirement?
>>>>
>>>> I don't see a way around it.  If the default mode is cache=writeback, then we're
>>>> open to data corruption in any guest where barrier=0.  With guest togglable WCE,
>>>> it ends up being a guest configuration issue so we can more or less defer
>>>> responsibility.
>>>
>>> So do you think that offering a WCE inside the guest would be a real
>>> solution or just a way to have an excuse?
>>
>> No, it offers a mechanism to "fix mistakes" at run-time verses at start up time.
>
> This is true (in both directions). But I think it's independent from the
> right default.
>
>>    It also means that you can make template images that understand that they
>> don't support barriers and change the WCE setting appropriately.
>
> Isn't that really a job for management tools?
>
>>> Christoph said that OSes don't usually change this by themselves, it
>>> would need an administrator manually changing the setting. But if we
>>> require that, we can just as well require that the administrator set
>>> cache=writethrough on the qemu command line.
>>
>> The administrator of the guest != the administrator of the host.
>
> But the administrator of the guest == the owner of the qemu instance,
> no? He should be the one to use the management tools and configure his VMs.

You're really talking about a multi-tenancy virtualization management solution. 
  There really aren't a lot of these today.  The most common variant is a IaaS 
platform where the end-user API is mostly just create a VM, destroy a VM. 
There's not a lot of dynamic configurability (just look at EC2s API).

>>>> Do you think it's a good idea to change the default mode w/o guest WCE toggle
>>>> support?  What's your view about older guests if we change the default mode?
>>>> What's your main motivation for wanting to change the default mode?
>>>
>>> Because people are constantly complaining about the awful
>>> (cache=writethrough) performance they get before they are told they
>>> should use a different cache option. And they are right. The
>>> out-of-the-box experience with qemu's block performance really sucks.
>>
>> With qcow2 you mean, right?
>
> No, with any format, including raw. Which isn't surprising at all,
> O_SYNC makes writes very expensive.
>
>>>> I'd be much more open to changing the default mode to cache=none FWIW since the
>>>> risk of data loss there is much, much lower.
>>>
>>> I think people said that they'd rather not have cache=none as default
>>> because O_DIRECT doesn't work everywhere.
>>
>> Where doesn't it work these days?  I know it doesn't work on tmpfs.  I know it
>> works on ext[234], btrfs, nfs.
>
> I think tmpfs was named (and failing to start with default settings on
> tmpfs would be nasty enough), but iirc Alex had another one.

Alex?  We can detect tmpfs with fsstat and do the right thing.

Regards,

Anthony Liguori


>
> Kevin
>