libplayload and coreinfo

[Jordan Crouse]

Greetings.  I've been talking about it for a while, and I've gotten
a few people excited, so I figured it was time to actually release
something useful.

I believe that there is a lot of power in coreboot for AMD customers,
especially in the embedded market which my team supports.  Part of
that power is in the low cost and high visibility of the open source
code, but also in the flexibility of the payload model.  Now, for a very
low cost a customer can get not only the hardware initialization, but 
also an blank page where they can add their own special, be it a fully
fledged payload, or just an intermediary before loading an operating
system.

This is where libpayload comes in - it abstracts away the icky bits
of payload initialization, and gives the user a semblance of a normal
C library that they can use to develop a payload without having to
worry about re-implementing malloc() and printf() and all the other
usual friends.

I am happy to release the first proof of concept for libpayload:

http://coreboot.org/~jcrouse/libpayload_0.1.tar.gz

This is a conglomeration of code from three projects - the 
payload initialization code from FILO, superb curses code from
our good friend Uwe Hermann, and various libc and driver bits
from coreboot-v3.  I combined them together into a library,
mixed kconfig into the mix and baked at 300F for 2 hours.  This is
proof of concept code for the most part, and I wouldn't go off
and put this in your airplane autopilot quite yet, but its a good
start.

And to celebrate the good start, how about something practical to
go with it?  Here is something I call coreinfo:

http://coreboot.org/~jcrouse/coreinfo_0.1.tar.gz

Coreinfo is a curses based payload that outputs lots of information
about the CPU and system.  Or rather, it will - for now it just 
shows the cpuid of the processor, and has a rather slick PCI device
browser.  The most important thing is that it shows that libpayload
can be used for a relatively complex payload without really breathing
hard.  Here's a screenshot:

http://coreboot.org/~jcrouse/coreinfo_screenshot.png

So, please, enjoy.  Flame me as needed, and once everybody has
digested this for a bit, we'll meet back and discuss what the
next step is.  Obviously, I think more libc functions are needed,
filesystem support is useful, and maybe even graphics are in our future.

Jordan

[ron minnich]

That is really cool :-)

ron

[Jordan Crouse]

On 04/03/08 14:39 -0800, ron minnich wrote:
> That is really cool :-)

I would like to point out that all the eye-candy is really thanks to Uwe
and his awesome tinycurses implmentation, without which none of this
would have been possible.  printf() is good and all, but I prefer 
pretty colors.

Thanks, Uwe!
> ron

[Peter Stuge]

On Tue, Mar 04, 2008 at 03:45:35PM -0700, Jordan Crouse wrote:
> http://coreboot.org/~jcrouse/coreinfo_screenshot.png

On Tue, Mar 04, 2008 at 03:52:40PM -0700, Jordan Crouse wrote:
> I would like to point out that all the eye-candy is really thanks
> to Uwe and his awesome tinycurses implmentation

Vielen Dank and Thank You!

This made my day which was otherwise not so good.

It is beautiful!

//Peter

[Martin Trautmann]

Jordan Crouse wrote:
> And to celebrate the good start, how about something practical to
> go with it?  Here is something I call coreinfo:
>
> http://coreboot.org/~jcrouse/coreinfo_0.1.tar.gz

Thanks!

I'll have to find out about the proper stdio.h include file
or sprintf declaration.

When I use my /usr/include/stdio.h I get redefinitions of
  typedef ‘FILE’
and
  struct _IO_FILE

Otherwise without including stdio.h I see
incompatible implicit declaration of built-in function ‘sprintf’

Martin

[Jordan Crouse]

On 06/03/08 09:27 +0100, Martin Trautmann wrote:
> Jordan Crouse wrote:
> > And to celebrate the good start, how about something practical to
> > go with it?  Here is something I call coreinfo:
> >
> > http://coreboot.org/~jcrouse/coreinfo_0.1.tar.gz
>
> Thanks!
>
> I'll have to find out about the proper stdio.h include file
> or sprintf declaration.
>
> When I use my /usr/include/stdio.h I get redefinitions of
>   typedef ‘FILE’
> and
>   struct _IO_FILE
>
> Otherwise without including stdio.h I see
> incompatible implicit declaration of built-in function ‘sprintf’

Oops - yep.  sprintf is missing from the header file.

Add:

int sprintf(char * buf, const char *fmt, ...);

to include/libpayload.h

[Uwe Hermann]

On Tue, Mar 04, 2008 at 03:45:35PM -0700, Jordan Crouse wrote:
> I am happy to release the first proof of concept for libpayload:
>
> http://coreboot.org/~jcrouse/libpayload_0.1.tar.gz

Great stuff, thanks!
> This is a conglomeration of code from three projects - the 
> payload initialization code from FILO, superb curses code from
> our good friend Uwe Hermann, and various libc and driver bits
> from coreboot-v3.  I combined them together into a library,
> mixed kconfig into the mix and baked at 300F for 2 hours.  This is
> proof of concept code for the most part, and I wouldn't go off
> and put this in your airplane autopilot quite yet, but its a good
> start.
>
> And to celebrate the good start, how about something practical to
> go with it?  Here is something I call coreinfo:
>
> http://coreboot.org/~jcrouse/coreinfo_0.1.tar.gz

I think Jordan tested this on SimNow, but I'm happy to confirm that
it runs very nicely on real hardware, in my case on the ASUS A8V-E Deluxe.

FYI: We (or rather: mostly Jordan) are working on a fixed-up libpayload
version which will be BSD-licensed, so commercial users can benefit too.

For that to work, I'm hereby relicensing all my code which is part of
libpayload (tinycurses mostly) under the 3-clause BSD license. The plan
is that the same will happen with code written by Jordan, and the rest
of the currently non-BSD code which is not authored by us will be
shamelessly replaced by code stolen from FreeBSD/OpenBSD/NetBSD
and/or rewritten by us from scratch.

The goal is to have a small, fully-BSD libpayload soon, with each file
having a clear license header. This should then probably go into svn, maybe
under a new 'payloads' directory or so (where other libpayload-based
payloads will also live, e.g. coreinfo, lbmenu, etc).

Also, I'll rework my (long-overdue) GSoC project from last year (lbmenu)
to use libpayload after that is done, and beat it with a stick until it
becomes reasonably suited for real-life usage.

Uwe.

[Jordan Crouse]

On 07/03/08 00:45 +0100, Uwe Hermann wrote:
> FYI: We (or rather: mostly Jordan) are working on a fixed-up libpayload
> version which will be BSD-licensed, so commercial users can benefit too.
>
> For that to work, I'm hereby relicensing all my code which is part of
> libpayload (tinycurses mostly) under the 3-clause BSD license. The plan
> is that the same will happen with code written by Jordan, and the rest
> of the currently non-BSD code which is not authored by us will be
> shamelessly replaced by code stolen from FreeBSD/OpenBSD/NetBSD
> and/or rewritten by us from scratch.

Thanks Uwe!

To that end, here's our current status.  If anybody out there is 
bored, pick a feature and start rewriting (or borrow from *BSD).

These are the files of questionable license status from FILO - these need
to be replaced:

./libc/malloc.c
./include/elf_boot.h
./include/arch/io.h
./i386/context.c
./i386/context.h
./i386/segment.c
./i386/segment.h
./i386/switch.S

These are files of known status (all GPLv2) that have been borrowed from
various projects.  These either need to be replaced with BSD versions,
or permission acquired to relicense them:

./libc/memory.c (from coreboot-v3)
./libc/string.c (from coreboot-v3)
./libc/printf.c (from the kernel)
./libc/ipchecksum.c (from coreboot-v3)
./include/checksum.h (from coreboot-v3)
./drivers/serial.c (from coreboot-v3)

Jordan