Re: [RESEND][PATCH 1/3] meta: add new qemuarma9 machine definition

[Bruce Ashfield <bruce.ashfield@windriver.com>]

On 2015-05-14 09:46 AM, Richard Purdie wrote:
> On Wed, 2015-05-13 at 18:17 -0700, Andre McCurdy wrote:
>> On Tue, May 12, 2015 at 7:47 AM, Martin Jansa <martin.jansa@gmail.com> wrote:
>>> On Tue, May 12, 2015 at 03:25:43PM +0100, Burton, Ross wrote:
>>>> On 11 May 2015 at 20:52, Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
>>>> wrote:
>>>>
>>>>> Currently qemuarm is limited to 256 Mb of RAM. Sometimes this is too
>>>>> little to run necessary applications. Add a new arm configuration based
>>>>> on Versatile Express board, Cortex-A9 CPU, allowing up to 1Gb of RAM.
>>>>>
>>>>
>>>> Not sure I'm keen on oe-core having two almost-identical qemuarm machines.
>>>> Why not just change the qemuarm machine to use the A9?
>>>
>>> Then we should officially drop thumb1 support, because current qemuarm
>>> builds are quite broken when thumb is enabled and dropping current
>>> qemuarm or replacing it with A9 variant will prevent oe-core to be
>>> testable on autobuilder. See
>>> https://bugzilla.yoctoproject.org/show_bug.cgi?id=7717
>>
>> +1 for updating qemuarm to an ARMv7 CPU.
>
> One thing I did notice about the new proposed arm machine was the lack
> of graphics support. We really do need a machine with graphics. If we
> could get a machine which had graphics and more memory that would be
> much more attractive to switch to.
>
> This also has implications on the kernel support (cc Bruce).

I've been using the qemuarma9 machine in some different contexts for
a while now, and in fact, there's a BSP definition in linux-yocto
already for it.

So from that point of view, the kernel impacts are understood.

But not only does the qemuarma9 lack graphics, it also has issues
with disk and USB, so generally it isn't as usable as the arm926
qemu variant.

There are other options that have newer CPUs, or just changing the
cpu .. but a wholesale switch to the "qemuarma9" machine tends to
bring some new challenges.

Bruce

>
>> As for dropping thumb1 support that's probably fine too - although
>> technically (if someone really did want to keep thumb1 support alive)
>> I guess nothing prevents testing thumb1 binaries on an ARMv7 CPU?
>
> Just guessing but they might work in some cases an a v7 CPU but fail on
> older ones due to alignment constraints?
>
> Cheers,
>
> Richard
>
>