Xen devel - Feb 2009 - [RFC] Disable xen platform device when it is not needed.

I plan to create the patch to disable xen platform device when it
is not needed. Please let me know if anyone has a comment.

Windows guest shows xen platform device as unknown
device with "!" mark in the device manager, when PV drivers are not
loaded. So I''d like to disable xen platform device.

If following conditions are all true, I will disable xen platform
device. Otherwise, xen platform device will be enabled.

    - All entries in "disk" parameter have "ioemu:" prefix
in DEV field.
        Example:
        disk = [ ''/var/images/guest.img,ioemu:hda,w'',
'',ioemu:hdc:cdrom,r'' ]

    - All entries in "vif" parameter have "type=ioemu".
        Example:
        vif = [ ''type=ioemu, bridge=eth0'',
''type=ioemu, bridge=eth1'' ]

    - "vscsi" parameter is not specified.

Xen platform device has functions for PV drivers and guest
firmware (hvmloader and rombios). The functions for PV drivers will be
disabled when xen platform device is disabled. This is no problem. But
the function for guest firmware should not be disabled.

I will move the function for guest firmware from xen platform device
to host bridge. The interface between host bridge and guest firmware
will be in PCI configuration space, because host bridge does not have
io resource and memory resource.

The function which I will move is switching RW/RO state of ROM memory
area. Please refer
qemu-xen/hw/xen_platform.c:xen_platform_ioport_writeb. I will not move
the other functions.

Thanks,
--
Yuji Shimada


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> 
> I plan to create the patch to disable xen platform device when it
> is not needed. Please let me know if anyone has a comment.
> 
> Windows guest shows xen platform device as unknown
> device with "!" mark in the device manager, when PV drivers are
not
> loaded. So I''d like to disable xen platform device.
> 
> If following conditions are all true, I will disable xen platform
> device. Otherwise, xen platform device will be enabled.
> 
>     - All entries in "disk" parameter have "ioemu:"
prefix in DEV
field.
>         Example:
>         disk = [ ''/var/images/guest.img,ioemu:hda,w'',
'',ioemu:hdc:cdrom,r''
> ]
> 
>     - All entries in "vif" parameter have "type=ioemu".
>         Example:
>         vif = [ ''type=ioemu, bridge=eth0'',
''type=ioemu, bridge=eth1'' ]
> 
>     - "vscsi" parameter is not specified.
I still leave my disks and network as ioemu even when using PV drivers,
as I simply need to use the /NOGPLPV boot option to switch back to them.

I recommend you add a ''no_platform_pci=1'' option or something
like that,
rather than try and guess that the user doesn''t want it.

I''m also not sure that xen will boot without the disks being in ioemu
mode, so using a non-ioemu disk for the situation where you want to use
PV drivers will not give you a bootable system.

James

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

James Harper writes:
> > 
> > I plan to create the patch to disable xen platform device when it
> > is not needed. Please let me know if anyone has a comment.
> > 
> > Windows guest shows xen platform device as unknown
> > device with "!" mark in the device manager, when PV drivers
are not
> > loaded. So I''d like to disable xen platform device.
> > 
> > If following conditions are all true, I will disable xen platform
> > device. Otherwise, xen platform device will be enabled.
> > 
> >     - All entries in "disk" parameter have
"ioemu:" prefix in DEV
> field.
> >         Example:
> >         disk = [
''/var/images/guest.img,ioemu:hda,w'',
> '',ioemu:hdc:cdrom,r''
> > ]
> > 
> >     - All entries in "vif" parameter have
"type=ioemu".
> >         Example:
> >         vif = [ ''type=ioemu, bridge=eth0'',
''type=ioemu, bridge=eth1'' ]
> > 
> >     - "vscsi" parameter is not specified.
> 
> I still leave my disks and network as ioemu even when using PV drivers,
> as I simply need to use the /NOGPLPV boot option to switch back to them.
> 
> I recommend you add a ''no_platform_pci=1'' option or
something like that,
> rather than try and guess that the user doesn''t want it.
Besides, we can add a PV disk by ''xm block-attach'' command
after boot.
My vote would be a ''xen_platform_pci=0'' option (default=1).
> I''m also not sure that xen will boot without the disks being in
ioemu
> mode, so using a non-ioemu disk for the situation where you want to use
> PV drivers will not give you a bootable system.
The PCI passthrough enables it.

Thanks,
Kouya

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Thank you for your comments.

On Wed, 18 Feb 2009 11:29:54 +1100
"James Harper" <james.harper@bendigoit.com.au> wrote:
> > 
> > I plan to create the patch to disable xen platform device when it
> > is not needed. Please let me know if anyone has a comment.
> > 
> > Windows guest shows xen platform device as unknown
> > device with "!" mark in the device manager, when PV drivers
are not
> > loaded. So I''d like to disable xen platform device.
> > 
> > If following conditions are all true, I will disable xen platform
> > device. Otherwise, xen platform device will be enabled.
> > 
> >     - All entries in "disk" parameter have
"ioemu:" prefix in DEV
> field.
> >         Example:
> >         disk = [
''/var/images/guest.img,ioemu:hda,w'',
> '',ioemu:hdc:cdrom,r''
> > ]
> > 
> >     - All entries in "vif" parameter have
"type=ioemu".
> >         Example:
> >         vif = [ ''type=ioemu, bridge=eth0'',
''type=ioemu, bridge=eth1'' ]
> > 
> >     - "vscsi" parameter is not specified.
> 
> I still leave my disks and network as ioemu even when using PV drivers,
> as I simply need to use the /NOGPLPV boot option to switch back to them.
When "ioemu:" or "type=ioemu" are not specified, both
emulated device
and PV device are enabled. This is current xen''s behavior.

So my idea does not break your usage.
> I recommend you add a ''no_platform_pci=1'' option or
something like that,
> rather than try and guess that the user doesn''t want it.
I think xen''s users are not interested in whether platform device is
enabled or disabled. When they find that the device manager shows "!",
they will complain. On the other hand, they are interested in type of
disk and vif, because this affects the performance of disk and vif. So
I think it is better to use "ioemu:" and "type=ioemu".

What do you think about this?

Thanks,
--
Yuji Shimada


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> 
> Thank you for your comments.
> 
> On Wed, 18 Feb 2009 11:29:54 +1100
> "James Harper" <james.harper@bendigoit.com.au> wrote:
> 
> > >
> > > I plan to create the patch to disable xen platform device when it
> > > is not needed. Please let me know if anyone has a comment.
> > >
> > > Windows guest shows xen platform device as unknown
> > > device with "!" mark in the device manager, when PV
drivers are
not
> > > loaded. So I''d like to disable xen platform device.
> > >
> > > If following conditions are all true, I will disable xen platform
> > > device. Otherwise, xen platform device will be enabled.
> > >
> > >     - All entries in "disk" parameter have
"ioemu:" prefix in DEV
> > field.
> > >         Example:
> > >         disk = [
''/var/images/guest.img,ioemu:hda,w'',
> > '',ioemu:hdc:cdrom,r''
> > > ]
> > >
> > >     - All entries in "vif" parameter have
"type=ioemu".
> > >         Example:
> > >         vif = [ ''type=ioemu, bridge=eth0'',
''type=ioemu,
bridge=eth1'' ]
> > >
> > >     - "vscsi" parameter is not specified.
> >
> > I still leave my disks and network as ioemu even when using PV
drivers,
> > as I simply need to use the /NOGPLPV boot option to switch back to
them.
> 
> When "ioemu:" or "type=ioemu" are not specified, both
emulated device
> and PV device are enabled. This is current xen''s behavior.
> 
> So my idea does not break your usage.
Maybe I don''t understand your idea, but I think you are saying that if
ioemu is specified then the PV drivers won''t work because you will
disable the platform pci device.

James


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>From: James Harper
>Sent: Wednesday, February 18, 2009 8:30 AM
>
>I''m also not sure that xen will boot without the disks being in
ioemu
>mode, so using a non-ioemu disk for the situation where you want to use
>PV drivers will not give you a bootable system.
>
Anthony Liguori posted a pxeboot feature to allow booting from pv disk:
http://www.mailinglistarchive.com/xen-devel@lists.xensource.com/msg47004.html

Thanks,
Kevin
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> > Thank you for your comments.
> > 
> > On Wed, 18 Feb 2009 11:29:54 +1100
> > "James Harper" <james.harper@bendigoit.com.au> wrote:
> > 
> > > >
> > > > I plan to create the patch to disable xen platform device
when it
> > > > is not needed. Please let me know if anyone has a comment.
> > > >
> > > > Windows guest shows xen platform device as unknown
> > > > device with "!" mark in the device manager, when
PV drivers are
> not
> > > > loaded. So I''d like to disable xen platform device.
> > > >
> > > > If following conditions are all true, I will disable xen
platform
> > > > device. Otherwise, xen platform device will be enabled.
> > > >
> > > >     - All entries in "disk" parameter have
"ioemu:" prefix in DEV
> > > field.
> > > >         Example:
> > > >         disk = [
''/var/images/guest.img,ioemu:hda,w'',
> > > '',ioemu:hdc:cdrom,r''
> > > > ]
> > > >
> > > >     - All entries in "vif" parameter have
"type=ioemu".
> > > >         Example:
> > > >         vif = [ ''type=ioemu, bridge=eth0'',
''type=ioemu,
> bridge=eth1'' ]
> > > >
> > > >     - "vscsi" parameter is not specified.
> > >
> > > I still leave my disks and network as ioemu even when using PV
> drivers,
> > > as I simply need to use the /NOGPLPV boot option to switch back
to
> them.
> > 
> > When "ioemu:" or "type=ioemu" are not specified,
both emulated device
> > and PV device are enabled. This is current xen''s behavior.
> > 
> > So my idea does not break your usage.
> 
> Maybe I don''t understand your idea, but I think you are saying
that if
> ioemu is specified then the PV drivers won''t work because you will
> disable the platform pci device.
I suppose your understanding is right. 

BTW, I have one question.
If you want to use the PV drivers, I don''t think xen requires
"ioemu" option.
Without "ioemu" option, xen will enable the emulated devices.
Why do you need "ioemu" option?

Thanks,
--
Shohei Fujiwara

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> > > So my idea does not break your usage.
> >
> > Maybe I don''t understand your idea, but I think you are
saying that
if
> > ioemu is specified then the PV drivers won''t work because you
will
> > disable the platform pci device.
> 
> I suppose your understanding is right.
> 
> BTW, I have one question.
> If you want to use the PV drivers, I don''t think xen requires
"ioemu"
> option.
> Without "ioemu" option, xen will enable the emulated devices.
> Why do you need "ioemu" option?
> 
For booting. I don''t think PV boot in HVM is supported properly yet is
it? That post I was directed to appears to just be an idea.

Also, the instructions for setting up GPLPV specify to add a separate
entry into boot.ini with a ''/NOGPLPV'' option in it which will
disable
the PV device drivers and use the ioemu drivers, which is useful for
testing to see if an error is being caused by the GPLPV drivers or is
unrelated. Also safe mode will disable the PV drivers (I''m not sure if
that will stay that way though). It would be a nuisance to have to edit
the domu config file each time you wanted to flip between the two modes.

Do you have an objection to something like an explicit
''xen_platform_pci=0'' option in the config to disable the PCI
device?
That seems like a far more sensible way of doing things IMHO if users
are really bothered by the extra PCI device.

James


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 18/02/2009 00:15, "Yuji Shimada"
<shimada-yxb@necst.nec.co.jp> wrote:
> Windows guest shows xen platform device as unknown
> device with "!" mark in the device manager, when PV drivers are
not
> loaded. So I''d like to disable xen platform device.
Is it really worth the hassle just to get rid of that? A number of
responders have said they don''t want the device auto disabled, and
noone is
going to bother to manually disable the device just to get rid of a warning
triangle in device manager.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 18/02/2009 03:20, "James Harper"
<james.harper@bendigoit.com.au> wrote:
>> When "ioemu:" or "type=ioemu" are not specified,
both emulated device
>> and PV device are enabled. This is current xen''s behavior.
>> 
>> So my idea does not break your usage.
> 
> Maybe I don''t understand your idea, but I think you are saying
that if
> ioemu is specified then the PV drivers won''t work because you will
> disable the platform pci device.
That''s my understanding. I''m not keen.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Thank all for the comments.

I will create the patch to add "xen_platform_pci" option in guest
config file to disable xen platform device.

    xen_platform_pci=0: Disable xen platform device.
    xen_platform_pci=1: Enable xen platform device. (default)

Thanks,
--
Yuji Shimada

On Wed, 18 Feb 2009 09:15:43 +0900
Yuji Shimada <shimada-yxb@necst.nec.co.jp> wrote:
> I plan to create the patch to disable xen platform device when it
> is not needed. Please let me know if anyone has a comment.
> 
> Windows guest shows xen platform device as unknown
> device with "!" mark in the device manager, when PV drivers are
not
> loaded. So I''d like to disable xen platform device.
> 
> If following conditions are all true, I will disable xen platform
> device. Otherwise, xen platform device will be enabled.
> 
>     - All entries in "disk" parameter have "ioemu:"
prefix in DEV field.
>         Example:
>         disk = [ ''/var/images/guest.img,ioemu:hda,w'',
'',ioemu:hdc:cdrom,r'' ]
> 
>     - All entries in "vif" parameter have "type=ioemu".
>         Example:
>         vif = [ ''type=ioemu, bridge=eth0'',
''type=ioemu, bridge=eth1'' ]
> 
>     - "vscsi" parameter is not specified.
> 
> Xen platform device has functions for PV drivers and guest
> firmware (hvmloader and rombios). The functions for PV drivers will be
> disabled when xen platform device is disabled. This is no problem. But
> the function for guest firmware should not be disabled.
> 
> I will move the function for guest firmware from xen platform device
> to host bridge. The interface between host bridge and guest firmware
> will be in PCI configuration space, because host bridge does not have
> io resource and memory resource.
> 
> The function which I will move is switching RW/RO state of ROM memory
> area. Please refer
> qemu-xen/hw/xen_platform.c:xen_platform_ioport_writeb. I will not move
> the other functions.
> 
> Thanks,
> --
> Yuji Shimada
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 19/02/2009 06:44, "Yuji Shimada"
<shimada-yxb@necst.nec.co.jp> wrote:
> Thank all for the comments.
> 
> I will create the patch to add "xen_platform_pci" option in guest
> config file to disable xen platform device.
> 
>     xen_platform_pci=0: Disable xen platform device.
>     xen_platform_pci=1: Enable xen platform device. (default)
I still might veto this. I don''t really want the platform device I/O
ports
scattered elsewhere for this rather frivolous reason.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 19/02/2009 09:39, "Keir Fraser" <keir.fraser@eu.citrix.com>
wrote:
>> I will create the patch to add "xen_platform_pci" option in
guest
>> config file to disable xen platform device.
>> 
>>     xen_platform_pci=0: Disable xen platform device.
>>     xen_platform_pci=1: Enable xen platform device. (default)
> 
> I still might veto this. I don''t really want the platform device
I/O ports
> scattered elsewhere for this rather frivolous reason.
Actually it''s okay, you can alias the
''platform_flags'' byte of the ioport
BAR to fixed byte port 0x10. There''s a handler for that already in
xen_platform.c (platform_fixed_ioport_{read,write}1). That''s the only
port
that hvmloader/rombios cares about.

Hvmloader/rombios can then always use fixed single-byte ioport 0x10. This is
actually better since the code is simpler and also rombios accesses that
port on S3 suspend/resume, at which point an OS could have relocated the
xen_platform ioport BAR! Using a fixed port is therefore safer.

Notice that I said you must *alias* this ioport byte. You cannot remove it
from the xen_platform BAR as old saved guests will rely on it being there,
and removing it will break their S3. We have to be backward compatible with
the Xen-3.3 virtual hardware interface for saved guests.

So, no veto after all!

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Thu, 19 Feb 2009 09:53:54 +0000
Keir Fraser <keir.fraser@eu.citrix.com> wrote:
> On 19/02/2009 09:39, "Keir Fraser"
<keir.fraser@eu.citrix.com> wrote:
> 
> >> I will create the patch to add "xen_platform_pci" option
in guest
> >> config file to disable xen platform device.
> >> 
> >>     xen_platform_pci=0: Disable xen platform device.
> >>     xen_platform_pci=1: Enable xen platform device. (default)
> > 
> > I still might veto this. I don''t really want the platform
device I/O ports
> > scattered elsewhere for this rather frivolous reason.
> 
> Actually it''s okay, you can alias the
''platform_flags'' byte of the ioport
> BAR to fixed byte port 0x10. There''s a handler for that already in
> xen_platform.c (platform_fixed_ioport_{read,write}1). That''s the
only port
> that hvmloader/rombios cares about.
> 
> Hvmloader/rombios can then always use fixed single-byte ioport 0x10. This
is
> actually better since the code is simpler and also rombios accesses that
> port on S3 suspend/resume, at which point an OS could have relocated the
> xen_platform ioport BAR! Using a fixed port is therefore safer.
> 
> Notice that I said you must *alias* this ioport byte. You cannot remove it
> from the xen_platform BAR as old saved guests will rely on it being there,
> and removing it will break their S3. We have to be backward compatible with
> the Xen-3.3 virtual hardware interface for saved guests.
> 
> So, no veto after all!
I will create the patch as follows.

- Alias the ''platform_flags'' byte of the ioport BAR to fixed
byte port
  0x10, and don''t remove existing ''platform_flags''
byte.

- Change guest firmware to use new fixed byte port 0x10.

- Add "xen_platform_pci" option in guest config file.

    xen_platform_pci=0: Disable xen platform device.
    xen_platform_pci=1: Enable xen platform device. (default)

Thanks,
--
Yuji Shimada


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 20/02/2009 07:52, "Yuji Shimada"
<shimada-yxb@necst.nec.co.jp> wrote:
> I will create the patch as follows.
> 
> - Alias the ''platform_flags'' byte of the ioport BAR to
fixed byte port
>   0x10, and don''t remove existing
''platform_flags'' byte.
> 
> - Change guest firmware to use new fixed byte port 0x10.
> 
> - Add "xen_platform_pci" option in guest config file.
> 
>     xen_platform_pci=0: Disable xen platform device.
>     xen_platform_pci=1: Enable xen platform device. (default)
Agreed.

 K.



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Yuji Shimada writes ("Re: [Xen-devel] [RFC] Disable xen platform device
when it is not needed."):
> - Alias the ''platform_flags'' byte of the ioport BAR to
fixed byte port
>   0x10, and don''t remove existing
''platform_flags'' byte.
How will this interact with the `disable qemu PCI devices in HVM
domains'' feature, which also uses ioport 0x10 ?

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 23/02/2009 08:43, "Ian Jackson" <Ian.Jackson@eu.citrix.com>
wrote:
> Yuji Shimada writes ("Re: [Xen-devel] [RFC] Disable xen platform
device when
> it is not needed."):
>> - Alias the ''platform_flags'' byte of the ioport BAR
to fixed byte port
>>   0x10, and don''t remove existing
''platform_flags'' byte.
> 
> How will this interact with the `disable qemu PCI devices in HVM
> domains'' feature, which also uses ioport 0x10 ?
That feature implements WORD and DWORD ports at address 0x10, but leaves the
BYTE port at that address available for other uses.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Keir Fraser writes ("Re: [Xen-devel] [RFC] Disable xen platform device when
it is not needed."):
> On 23/02/2009 08:43, "Ian Jackson"
<Ian.Jackson@eu.citrix.com> wrote:
> > How will this interact with the `disable qemu PCI devices in HVM
> > domains'' feature, which also uses ioport 0x10 ?
> 
> That feature implements WORD and DWORD ports at address 0x10, but leaves
the
> BYTE port at that address available for other uses.
Ah, err, yes.  Ugh.  Maybe a table of uses of ioports 0x10 (and 0x12!)
would be a sensible addition to xen-unstable.hg/doc/ somewhere.

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 23/02/2009 09:35, "Ian Jackson" <Ian.Jackson@eu.citrix.com>
wrote:
> Keir Fraser writes ("Re: [Xen-devel] [RFC] Disable xen platform device
when it
> is not needed."):
>> On 23/02/2009 08:43, "Ian Jackson"
<Ian.Jackson@eu.citrix.com> wrote:
>>> How will this interact with the `disable qemu PCI devices in HVM
>>> domains'' feature, which also uses ioport 0x10 ?
>> 
>> That feature implements WORD and DWORD ports at address 0x10, but
leaves the
>> BYTE port at that address available for other uses.
> 
> Ah, err, yes.  Ugh.  Maybe a table of uses of ioports 0x10 (and 0x12!)
> would be a sensible addition to xen-unstable.hg/doc/ somewhere.
Well, byte 0x10 is available and is the first one, so it seemed to make
sense to me. ;-) The protocol on the fixed ports is already kind of ugly and
makes use of different meanings for different access sizes, so it''s in
the
existing spirit also. More documentation is almost always nice, so I
can''t
disagree with that.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel