Re: [coreboot] Plugin support

[Myles Watson]

On Mon, Feb 23, 2009 at 10:43 AM, Myles Watson  wrote:
> On Mon, Feb 23, 2009 at 10:22 AM, Stefan Reinauer  wrote:
> > On 23.02.2009 18:04 Uhr, Myles Watson wrote:
> > > I implemented the *_PLUGIN_SUPPORT config options for v3.  I thought
> > > it was simple since I was just enabling the defines already in the
> > > code.  In retrospect I may have implemented them backward.
> > >
> > > The space in the ROM consumed by PLUGIN_SUPPORT is very small if you
> > > already have support for that bus.  If you don't have support for the
> > > bus it could be significant.  I wanted to make it so that e.g.
> > > HYPERTRANSPORT_PLUGIN_SUPPORT was the default if you had
> > > HYPERTRANSPORT_SUPPORT.  I'd also like people to be able to select
> > > CARDBUS_PLUGIN_SUPPORT or NO_CARDBUS_PLUGIN_SUPPORT.
> >
> > How much code do all those "plugins" together consume, given you have a
> > system with none of that buses in the dts?
>
> Coreboot ROM Image (QEMU) with no plugin support:
>  normal/option_table (932 bytes @0x50);loadaddress 0x0 entry 0x0
>  normal/initram/segment0 (440 bytes @0x450);loadaddress 0x0 entry 0x0x20
>  normal/stage2/segment0 (194780 bytes, zeroes compressed to 1 bytes
> @0x660);loadaddress 0x0xd130 entry 0x0x2000
>  normal/stage2/segment1 (35188 bytes, lzma compressed to 18410 bytes
> @0x6c0);loadaddress 0x0x2000 entry 0x0x2000
>  normal/stage2/segment2 (6076 bytes, lzma compressed to 360 bytes
> @0x4f00);loadaddress 0x0xb974 entry 0x0x2000
>  bootblock (20480 bytes @0x3b000)
> Total size = 41071B 40KB (0xa06f)
>
> Coreboot ROM Image with all plugins:
>  normal/option_table (932 bytes @0x50);loadaddress 0x0 entry 0x0
>  normal/initram/segment0 (440 bytes @0x450);loadaddress 0x0 entry 0x0x20
>  normal/stage2/segment0 (194784 bytes, zeroes compressed to 1 bytes
> @0x660);loadaddress 0x0xe1c0 entry 0x0x2000
>  normal/stage2/segment1 (39340 bytes, lzma compressed to 20416 bytes
> @0x6c0);loadaddress 0x0x2000 entry 0x0x2000
>  normal/stage2/segment2 (6164 bytes, lzma compressed to 376 bytes
> @0x56d0);loadaddress 0x0xc9ac entry 0x0x2000
>  bootblock (20480 bytes @0x3b000)
> Total size = 43093B 42KB (0xa855)
>
> It looks like about 2K.  In order for it to build I had to define a
> few HT-specific things and put in  an empty mainboard.h, but it won't
> be hard if we want to make it the default to always include all buses.

Coreboot ROM Image with everything but HyperTransport:
  normal/option_table (932 bytes @0x50);loadaddress 0x0 entry 0x0
  normal/initram/segment0 (440 bytes @0x450);loadaddress 0x0 entry 0x0x20
  normal/stage2/segment0 (194784 bytes, zeroes compressed to 1 bytes
@0x660);loadaddress 0x0xda70 entry 0x0x2000
  normal/stage2/segment1 (37468 bytes, lzma compressed to 19388 bytes
@0x6c0);loadaddress 0x0x2000 entry 0x0x2000
  normal/stage2/segment2 (6164 bytes, lzma compressed to 379 bytes
@0x52d0);loadaddress 0x0xc25c entry 0x0x2000
  bootblock (20480 bytes @0x3b000)
Total size = 42068B 41KB (0xa454)

Maybe this is a good middle ground.  I haven't seen a PCI->HT bridge yet :)

Thanks,
Myles
> > > Maybe we could get rid of the *PLUGIN_SUPPORT options all together and
> > > just allow the developer to select CARDBUS_SUPPORT or
> > > REMOVE_CARDBUS_SUPPORT directly?  We don't want people to deselect
> > > HYPERTRANSPORT_SUPPORT when they have an Opteron, though.
> >
> > If we're talking about a significant amount of code that is disabled, I
> > tend to agree. On the other hand, Ron brought up the very valuable point
> > that we want to try remove as many conditional compilations as possible
> > to keep the code simple (and: testable)
>
> the fewer #ifdefs the better for me too.
> > > Could I get some feedback on what people want?  It's very low on my
> > > priority list, but I'm willing to take ownership since I put it there.
> >
> > Sure.. :-)
>
> Thanks,
> Myles