More useful generic-x86 machine

More useful generic-x86 machine

[Ross Burton]

Hi,

My current test hardware for Yocto work is a Celeron-class Sandy Bridge-based micro PC. As is to be expected for a machine like this, it's got modern Intel graphics and wireless. Neither of which are supported by atom-pc, as it's nominally a "generic netbook" image.

Let's face it -- it's a really bad generic netbook image, it's really a Asus eeePC701-and-similar image. Specifically, only one wifi driver, only i915 GPU driver, and so on.

I'm not arguing for a true generic kernel such as Fedora maintains which boots on almost everything, just a new machine with more flexibility. Including both i915 and i965 GPU drivers covers everything Intel-driven from the earliest netbook to the latest Ivy Bridge[1].  Including the iwl wifi drivers at least covers a good proportion of devices out there.  There are probably a few more drivers that are common and give big gains in support.  Not exactly boot on everything, but certainly boot on many.

Thoughts?

Ross

[1] Ignoring Cedar Trail, but that's already in meta-intel

Re: More useful generic-x86 machine

[Darren Hart]

On 06/13/2012 01:44 PM, Ross Burton wrote:
> Hi,
> 
> My current test hardware for Yocto work is a Celeron-class Sandy
> Bridge-based micro PC. As is to be expected for a machine like this,
> it's got modern Intel graphics and wireless. Neither of which are
> supported by atom-pc, as it's nominally a "generic netbook" image.
> 
> Let's face it -- it's a really bad generic netbook image, it's really
> a Asus eeePC701-and-similar image. Specifically, only one wifi
> driver, only i915 GPU driver, and so on.
> 
> I'm not arguing for a true generic kernel such as Fedora maintains
> which boots on almost everything, just a new machine with more
> flexibility. Including both i915 and i965 GPU drivers covers
> everything Intel-driven from the earliest netbook to the latest Ivy
> Bridge[1].  Including the iwl wifi drivers at least covers a good
> proportion of devices out there.  There are probably a few more
> drivers that are common and give big gains in support.  Not exactly
> boot on everything, but certainly boot on many.
> 
> Thoughts?

Seems reasonable to me. We should probably have 32b and 64b of this
machine as well.


-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

Re: More useful generic-x86 machine

[Bruce Ashfield]

On 12-06-13 04:44 PM, Ross Burton wrote:
> Hi,
>
> My current test hardware for Yocto work is a Celeron-class Sandy Bridge-based micro PC. As is to be expected for a machine like this, it's got modern Intel graphics and wireless. Neither of which are supported by atom-pc, as it's nominally a "generic netbook" image.
>
> Let's face it -- it's a really bad generic netbook image, it's really a Asus eeePC701-and-similar image. Specifically, only one wifi driver, only i915 GPU driver, and so on.
>
> I'm not arguing for a true generic kernel such as Fedora maintains which boots on almost everything, just a new machine with more flexibility. Including both i915 and i965 GPU drivers covers everything Intel-driven from the earliest netbook to the latest Ivy Bridge[1].  Including the iwl wifi drivers at least covers a good proportion of devices out there.  There are probably a few more drivers that are common and give big gains in support.  Not exactly boot on everything, but certainly boot on many.
>
> Thoughts?

Send patches, and define the machine. But it has to be something
that we can actually *test* and maintain. We don't define 'what if'
and 'in case' configurations. We define embedded/tight configurations
that should be extended for particular machines.

The common-pc is a bit old, so I'd suggest sending patches to that
base and we can consider them on a 1:1 basis. As long as it isn't
overly generic, and hits a defined/closed set of machines and drivers,
patches won't be denied.

Even better, just branch from one of the existing configs and add more,
that way the extensions are separate from the base.

Cheers,

Bruce

>
> Ross
>
> [1] Ignoring Cedar Trail, but that's already in meta-intel
> _______________________________________________
> yocto mailing list
> yocto@yoctoproject.org
> https://lists.yoctoproject.org/listinfo/yocto

Re: More useful generic-x86 machine

[Ross Burton]

On Wednesday, 13 June 2012 at 21:47, Darren Hart wrote:
> Seems reasonable to me. We should probably have 32b and 64b of this
> machine as well.

And x32… :)

Ross  

Re: More useful generic-x86 machine

[Bruce Ashfield]

On 12-06-13 04:52 PM, Ross Burton wrote:
> On Wednesday, 13 June 2012 at 21:47, Darren Hart wrote:
>> Seems reasonable to me. We should probably have 32b and 64b of this
>> machine as well.
>
> And x32… :)

 From the kernel point of view, these are just configuration extensions
to a base, which is where this discussion started (the kernel, I'm
excluding userspace on purpose). So this should be one machine with
these as overlays, not three different machines.

Cheers,

Bruce

>
> Ross
>
>
> _______________________________________________
> yocto mailing list
> yocto@yoctoproject.org
> https://lists.yoctoproject.org/listinfo/yocto

Re: More useful generic-x86 machine

[Darren Hart]

On 06/13/2012 01:55 PM, Bruce Ashfield wrote:
> On 12-06-13 04:52 PM, Ross Burton wrote:
>> On Wednesday, 13 June 2012 at 21:47, Darren Hart wrote:
>>> Seems reasonable to me. We should probably have 32b and 64b of this
>>> machine as well.
>>
>> And x32… :)
> 
>  From the kernel point of view, these are just configuration extensions
> to a base, which is where this discussion started (the kernel, I'm
> excluding userspace on purpose). So this should be one machine with
> these as overlays, not three different machines.

I would have thought the three different architectures would have called
for three different machines. How would this work from the KMACHINE
meta-data perspective?

--
Darren

> 
> Cheers,
> 
> Bruce
> 
>>
>> Ross
>>
>>
>> _______________________________________________
>> yocto mailing list
>> yocto@yoctoproject.org
>> https://lists.yoctoproject.org/listinfo/yocto
> 

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

Re: More useful generic-x86 machine

[Bruce Ashfield]

On 12-06-13 05:00 PM, Darren Hart wrote:
>
>
> On 06/13/2012 01:55 PM, Bruce Ashfield wrote:
>> On 12-06-13 04:52 PM, Ross Burton wrote:
>>> On Wednesday, 13 June 2012 at 21:47, Darren Hart wrote:
>>>> Seems reasonable to me. We should probably have 32b and 64b of this
>>>> machine as well.
>>>
>>> And x32… :)
>>
>>    From the kernel point of view, these are just configuration extensions
>> to a base, which is where this discussion started (the kernel, I'm
>> excluding userspace on purpose). So this should be one machine with
>> these as overlays, not three different machines.
>
> I would have thought the three different architectures would have called
> for three different machines. How would this work from the KMACHINE
> meta-data perspective?

I've had dual endian machines for ages for MIPS. This is no
different. You use a common machine, and then just trigger fragments
that change the few options that are different like endianess, etc.
So there's a single KMACHINE definition with additions.

Granted, this was more important when there was a strict 1:1 branch ->
machine mapping. But it still makes sense to keep things as small
as possible.  We can do the same thing with three KMACHINE definitions
that include a common base, and that's nominally three machines, but
the slippery slope is that they start to diverge .. since they are
described by three different top level options.

I'm splitting a hair, I just wanted to point out that I wouldn't
call word size, endianess or other ABI differences big differences
in a machine definition.

Cheers,

Bruce

>
> --
> Darren
>
>>
>> Cheers,
>>
>> Bruce
>>
>>>
>>> Ross
>>>
>>>
>>> _______________________________________________
>>> yocto mailing list
>>> yocto@yoctoproject.org
>>> https://lists.yoctoproject.org/listinfo/yocto
>>
>

Re: More useful generic-x86 machine

[Stewart, David C]

> From: yocto-bounces@yoctoproject.org [mailto:yocto-
> bounces@yoctoproject.org] On Behalf Of Ross Burton
> Sent: Wednesday, June 13, 2012 1:45 PM
>
> Hi,
> 
> My current test hardware for Yocto work is a Celeron-class Sandy Bridge-
> based micro PC. As is to be expected for a machine like this, it's got modern
> Intel graphics and wireless. Neither of which are supported by atom-pc, as
> it's nominally a "generic netbook" image.
> 
> Let's face it -- it's a really bad generic netbook image, it's really a Asus
> eeePC701-and-similar image. Specifically, only one wifi driver, only i915 GPU
> driver, and so on.

I thought we had some Sandy Bridge BSP already?

> I'm not arguing for a true generic kernel such as Fedora maintains which
> boots on almost everything, just a new machine with more flexibility.
> Including both i915 and i965 GPU drivers covers everything Intel-driven from
> the earliest netbook to the latest Ivy Bridge[1].  Including the iwl wifi drivers
> at least covers a good proportion of devices out there.  There are probably a
> few more drivers that are common and give big gains in support.  Not exactly
> boot on everything, but certainly boot on many.
> 
> Thoughts?
> 
> Ross
> 
> [1] Ignoring Cedar Trail, but that's already in meta-intel
> _______________________________________________
> yocto mailing list
> yocto@yoctoproject.org
> https://lists.yoctoproject.org/listinfo/yocto

Re: More useful generic-x86 machine

[Darren Hart]

On 06/13/2012 03:50 PM, Stewart, David C wrote:
>> From: yocto-bounces@yoctoproject.org [mailto:yocto-
>> bounces@yoctoproject.org] On Behalf Of Ross Burton
>> Sent: Wednesday, June 13, 2012 1:45 PM
>>
>> Hi,
>>
>> My current test hardware for Yocto work is a Celeron-class Sandy Bridge-
>> based micro PC. As is to be expected for a machine like this, it's got modern
>> Intel graphics and wireless. Neither of which are supported by atom-pc, as
>> it's nominally a "generic netbook" image.
>>
>> Let's face it -- it's a really bad generic netbook image, it's really a Asus
>> eeePC701-and-similar image. Specifically, only one wifi driver, only i915 GPU
>> driver, and so on.
> 
> I thought we had some Sandy Bridge BSP already?

We do. The problem is that doesn't define things like which wifi device
is used. There seems to be a reasonable amount of interest in a much
more generally useful x86 BSP. This seems to fit the goals of the
atom-pc BSP, which has become a bit stale.

> 
>> I'm not arguing for a true generic kernel such as Fedora maintains which
>> boots on almost everything, just a new machine with more flexibility.
>> Including both i915 and i965 GPU drivers covers everything Intel-driven from
>> the earliest netbook to the latest Ivy Bridge[1].  Including the iwl wifi drivers
>> at least covers a good proportion of devices out there.  There are probably a
>> few more drivers that are common and give big gains in support.  Not exactly
>> boot on everything, but certainly boot on many.
>>
>> Thoughts?
>>
>> Ross
>>
>> [1] Ignoring Cedar Trail, but that's already in meta-intel
>> _______________________________________________
>> yocto mailing list
>> yocto@yoctoproject.org
>> https://lists.yoctoproject.org/listinfo/yocto
> _______________________________________________
> yocto mailing list
> yocto@yoctoproject.org
> https://lists.yoctoproject.org/listinfo/yocto
> 

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel