x86_64 BSP as GSOC2016 project

[Saket Sinha]

Hi,

Congratulations to the coreboot team for selection in GSOC 2016.

I am interested in working on x86_64 BSP as GSOC2016 project with
RTEMS and with initial guidance from Joel as to how to get started, I
am working on the same.

Looking forward to working with the community for the same.
Regards,
Saket Sinha


On Wed, Feb 10, 2016 at 7:08 PM, Joel Sherrill <joel at rtems.org> wrote:
>
> On Feb 9, 2016 10:37 PM, "Saket Sinha" <saket.sinha89 at gmail.com> wrote:
>>
>> Hi Joel,
>>
>> Thanks for your inputs.
>>
>>
>>
>> On Wed, Feb 10, 2016 at 1:06 AM, Joel Sherrill <joel at rtems.org> wrote:
>> >
>> >
>> > On Tue, Feb 9, 2016 at 12:08 PM, Saket Sinha <saket.sinha89 at gmail.com>
>> > wrote:
>> >>
>> >> Hi,
>> >>
>> >> I think this is early for this but I need to enquire if x86_64_BSP (
>> >> https://devel.rtems.org/wiki/Developer/Projects/Open/x86_64_BSP )
>> >> could be taken as a GSOC project this year.
>> >>
>> >
>> > Yes. Assuming you mean the x86_64 bit port and accompanying BSP.
>> >
>> > If you noticed my post yesterday, we have tripped across a new embedded
>> > PC
>> > without legacy PCI BIOS support.
>> >
>>
>> Could you provide the link. I am not able to locate it on mailing list
>> or your blog.
>
> I only posted a question this week had anyone attempted to use RTEMS on a PC
> without legacy BIOS. I think the answer is no.
>
> I have done a little searching for how to use the newer UEFI services to
> replace those we use the video and PCI but so far have only found high level
> information.
>
> Sorry. This one is a known need with only early research. I was going to dig
> into the FreeBSD source next.
>
>> > Beyond the obvious port, for the purposes of your effort, there are
>> > a couple of bugs and some modernization needed by the pc386 BSP which
>> > you will likely have to address.
>> >
>> > + PCI BIOS uses legacy support. Needs to support new and old for 32-bit
>> > and new only for 64 bit.
>> > + Better APIC support. pc386 uses legacy PIC and some LPIC for SMP.
>> > Probably Ok for both 32 and 64 bit to support APIC only.
>> > + i386 does not have Thread Local Support. Both should have it.
>> > + i386 has a ticket for SMP synchronization during context switch.
>> > The code does not use atomics. Should be fixed and x86_64 follow the
>> > correct pattern.
>> >
>>
>> So now my question is that could these be solved on an emulator like
>> qemu-x86, atleast for an initial POC ?
>> I mean is a x86_64 hardware required for it( though I have a Minnowmax
>> board.)
>>
>
> I think qemu can be used for all of this. You can certainly do the APIC
> support without addressing the legacy video and PCI BIOS.
>
> Similar for the TLS and SMP issue.
>
> I believe the entire thing could be brought up and debugged on qemu with
> checkouts on real HW. Either the Minnow or even a desktop PC with the right
> BIOS settings. None of this is particularly new. APIC dates back at least a
> decade.
>
>>
>> > Basically x86_64 should assume a modern (non-legacy) PC and pc386
>> > needs updating to support newer systems. Common hardware platform
>> > so hopefully the software is standard.
>> >
>> > I was going to write this up as a "PC386 Modernization" Project but some
>> > of it makes sense as a side-effect of your project. That project idea
>> > also
>> > included killing AT bus NICs which you wouldn't have any reason to get
>> > near.
>> >
>> > You need a subtask list (e.g. todo list) and we need to help you flesh
>> > it
>> > out. I only hit a few non-obvious highlights.
>> >
>>
>> Let me know how to dig deeper on this. Looking at the code, right  but
>> than exactly what sections ?
>>
>
> libbsp/shared/i386/pci is the current legacy PCI code.
>
> The IRQ code should also be under shared.
>
> I will have to feel to see where the video probe that is failing is located.
>
> The nice thing is that each of these three can be completed in the context
> of the current BSP.
>
>> Regards,
>> Saket Sinha