[virt-tools-list] [virt-manager PATCH] Allow installation of i686 guests on x86_64 machines

[Cole Robinson]

On 03/03/2018 04:03 PM, Martin Kletzander wrote:
> On Sat, Mar 03, 2018 at 03:53:59PM -0500, Cole Robinson wrote:
>> On 03/01/2018 03:52 AM, Martin Kletzander wrote:
>>> Do that by changing the code that was disabling it into code that
>>> just warns in
>>> such case.
>>>
>>> Signed-off-by: Martin Kletzander <mkletzan at redhat.com>
>>> ---
>>>  virtManager/create.py | 10 ++++------
>>>  1 file changed, 4 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/virtManager/create.py b/virtManager/create.py
>>> index 0a73309372d9..9a40aec4a068 100644
>>> --- a/virtManager/create.py
>>> +++ b/virtManager/create.py
>>> @@ -495,6 +495,10 @@ class vmmCreate(vmmGObjectUI):
>>>                  msg = _("Failed to setup UEFI for AArch64: %s\n"
>>>                          "Install options are limited.") % e
>>>                  self._show_arch_warning(msg)
>>> +        elif (self._capsinfo.arch == "i686" and
>>> +              self.conn.caps.host.cpu.arch == "x86_64"):
>>> +            msg = _("You are installing 32bit guest on 64bit host")
>>> +            self._show_arch_warning(msg)
>>>
>>>          # Install Options
>>>          method_tree = self.widget("method-tree")
>>> @@ -824,12 +828,6 @@ class vmmCreate(vmmGObjectUI):
>>>              if guest.os_type == self._capsinfo.os_type:
>>>                  archs.append(guest.arch)
>>>
>>> -        # Combine x86/i686 to avoid confusion
>>> -        if (self.conn.caps.host.cpu.arch == "x86_64" and
>>> -            "x86_64" in archs and "i686" in archs):
>>> -            archs.remove("i686")
>>> -        archs.sort()
>>> -
>>>          prios = ["x86_64", "i686", "aarch64", "armv7l", "ppc64",
>>> "ppc64le",
>>>              "s390x"]
>>>          if self.conn.caps.host.cpu.arch not in prios:
>>>
>>
>> The idea behind hiding the option is 1) it should be rare that someone
>> actually wants their VM to present a 32bit cpu on 64bit host, 2) hiding
>> it means when only x86_64 qemu is installed we can entirely hide the
>> 'advanced options' expander.
>>
>> Example: on rhel7 x86_64, libvirt advertises arch=i686 and arch=x86_64
>> for /usr/libexec/qemu-kvm. If we hide i686, there's only one arch option
>> available, and virt-manager will hide the entire 'advanced options'
>> expander. This is ideal IMO, otherwise users might go clicking, see the
>> i686 option, misinterpret it to mean OS arch (which in the context of
>> virt I've seen people mistake many times), set things to i686 needlessly.
>>
> 
> Sure I get that, that's why I kept it as a warning.
> 
>> Are there benefits I'm missng of doing arch=i686 on an x86_64 host
>> exactly? Doesn't it just map to using qemu32 as the default CPU?
>>
> 
> To be honest, I'm not sure if qemu32 CPU type means also 32bit CPU or
> just limited instruction set and how it looks from the guest OS POV.
> The reason for this patch emerged simply when we needed to test libvirt
> build failure on 32bit machine and it was just easier to install 32bit
> VM than setting up a cross-build.  Or rather installing all 32bit
> dependencies properly.  I'd love to hear how to do this properly, I'm
> not saying this patch is needed =)

I think this is an example of that arch confusion actually. You can
install a 32 bit OS into a 64 bit machine, doesn't matter if the CPU is
x86_64, and building in that should reproduce the 32bit build error

So in this case if i686 was a selectable option it would have just given
you worse performance since we would have used model=qemu32 implicitly
rather than virt-manager's default of a more modern CPU (the model we
get out of host capabilities output)

- Cole