Re: [Qemu-devel] [PATCH 01/18] smbios: Add a function to directly add an entry

[Anthony Liguori <anthony@codemonkey.ws>]

Corey Minyard <cminyard@mvista.com> writes:

> On 08/02/2012 01:32 PM, Anthony Liguori wrote:
>> Corey Minyard <tcminyard@gmail.com> writes:
>>
>>> On 08/01/2012 09:40 PM, Anthony Liguori wrote:
>>>> Corey Minyard <cminyard@mvista.com> writes:
>>>>
>>>>> On 08/01/2012 08:15 PM, Kevin O'Connor wrote:
>>>>> Well, I should also probably add the ACPI name space definition for this
>>>>> information, too, and the SMBIOS information is not capable of passing
>>>>> all the information required for this (though the above structure can).
>>>>>
>>>>> I've been studying this, but I don't see an obvious way to dynamically
>>>>> add something to the ACPI name space.  At least an easy way.
>>>> Okay, I was actually going to ask if there was an ACPI table for this.
>>>>
>>>> Maybe this argues in favor of doing a fw_cfg interface?
>>>>
>>>> Another question--is it really necessary for all of this to be user
>>>> specified?  Can't we just use a static SMBIOS/ACPI entry?  Then SeaBIOS
>>>> only needs to be concerned with whether or not an IPMI device exists.
>>> That's a good question At least the interrupt is important for the user
>>> to be able to specify.  The specific interface type may also be
>>> important if the user is trying to accomplish some specific emulation.
>> Why is it important to specify the interrupt?  Is this important for a
>> typical user, or important for the IPMI maintainer who needs to test a
>> bunch of different scenarios? :-)
>
> I'm not too worried about the IPMI maintainer, he can hack in what he 
> likes :).
>
> I would be worried about conflicts on interrupts with other devices.  I 
> really don't know how people use qemu out in the wild, though.  If they 
> are trying to get close to some specific machine, or if nobody really 
> cares about stuff like that.

It's an LPC device?  Ther aren't going to be many of those device types
that would be user controllable (basically TPM and IPMI) so I don't
think interrupt conflicts are a real likely issue.

> I also don't know if people will be wanting this on other 
> architectures.  IPMI is certainly available on ia64.

I doubt QEMU will ever support ia64 since noone seems to care about it anymore.

> In fact it's quite 
> common there.  I've seen it on practically everything else, though it's 
> not so common.  It's in the PPC device trees for sure, and in some uboot 
> device trees on other arches.

Right, but this is specifically about SMBIOS/ACPI support which won't be
on other architectures.

>
>> If it's the later, we can probably express the interrupt number as a
>> #define in SeaBIOS, but still make it configurable in QEMU.  Then you
>> could build multiple copies of SeaBIOS and then just point QEMU at the
>> right version.
>
> That philosophy sounds like a recipe for version overload.  I'd prefer 
> to avoid that.
>
>>
>>> Two other standard emulations exist, too, one in memory and one over
>>> I2C.  I'd eventually like to add those, if for nothing else my ability
>>> to test the interfaces.
>> Right, see above.  It may be easier to just build multiple copies of the
>> BIOS then to try and make this all dynamic.
> In my experience, if you need the flexibility and don't make it dynamic, 
> you make things harder in the long run.  But adding unnecessary 
> flexibility is extra work without value.

Exactly.

> IMHO, we should either have a single IPMI interface type at a fixed 
> location with a fixed interrupt, or we should make it flexible.

I think fixed interrupt is what makes the most sense now.  If there's a
 pressing need in the future to do otherwise, we can revisit.

Regards,

Anthony Liguori

> Even if 
> we make it fixed, the BIOS will have to be told if the device is present 
> and will have to dynamically chose to add the SMBIOS table and ACPI name 
> space entries.
>
> Thanks,
>
> -corey
>
>> Regards,
>>
>> Anthony Liguori
>>
>>> If the user is trying to emulate some specific machine, setting the
>>> address is also important, and I need to add the ability to specify
>>> register spacing and the address space.  This will become more important
>>> for non-x86 machines.
>>>
>>> -corey
>>>
>>>> Regards,
>>>>
>>>> Anthony Liguori
>>>>
>>>>> -corey