[Re: First RSPRO deployed!] flash utility mkfwimage and RSPRO boot question

[Milan Obuch]

On Monday 29 November 2010 18:13:16 Warner Losh wrote:
> On 11/26/2010 11:35, Milan Obuch wrote:
> > On Friday 26 November 2010 17:59:05 Warner Losh wrote:
> >> On 11/26/2010 08:54, Milan Obuch wrote:
> >>> On Thursday 25 November 2010 12:12:06 Luiz Otavio O Souza wrote:
> > [ snip]
> > 
> >>>> Building world works with CPUTYPE=mips32 and for ports, add the
> >>>> following line to /etc/make.conf:
> >>>> 
> >>>> CFLAGS=-O2 -pipe -march=mips32
> >>> 
> >>> Hmm, I put just TARGET_ARCH=mipseb into /etc/make.conf and compiled
> >>> almost all world (accidentally broken just before end, but that's a
> >>> pilot error in this case) and sysutils/screen port compiled and run
> >>> with no CPUTYPE nor CFLAGS definition.
> >>> 
> >>> That should not be read as better from any side, but surely simpler :)
> >>> Maybe even TARGET_ARCH should be known from running kernel, this would
> >>> be similar how native build works on both i386 and amd64 platforms.
> >>> Now, with screen in place I am going to test buildworld again (with
> >>> some WITHOUT_XXX=YES in /etc/make.conf) to verify native builds...
> >>> they are slowish here, which is expected, but seems to work, which is
> >>> great.
> >> 
> >> what does uname -a tell you?
> > 
> > FreeBSD argo.dino.sk 9.0-CURRENT FreeBSD 9.0-CURRENT #3: Wed Nov 24
> > 10:16:14 CET 2010    
> > root at wind.dino.sk:/tmp/rspro/obj/mips.mipseb/data/src9/sys/RSPRO mips
> > 
> > Currently running world and kernel is from cross build on i386 host.
> > 
> > Native buildworld just ended, I am going to try native buildkernel... for
> > that I must apply some patches (already published on this list).
> 
> Juli just committed some patches to make this better.  I was going to do
> it, but she got impatient waiting for my overbooked schedule to free up...
> 
> Warner

Well, over weekend it did not work, now it compiles.

There is a problem, however.

With no /etc/make.conf (TARGET nor TARGET_ARCH defined) both buildworld and 
buildkernel operation fails complaining about 'unknown target mips:mips'. 
After defining TARGET_ARCH=mipseb make buildworld succeeds, but this is not 
actually native buildworld - nake installworld complains about unknown target 
mips:mips as well, when TARGET_ARCH is defined, make errors with 'ERROR: 
Please set DESTDIR!' - so actually this is considered cross build.

MAKEOBJDIRPREFIX is set to /usr/obj/mips.mipseb and this again indicates it is 
a cross-build. So I think for mips architecture we need to set TARGET_ARCH 
internally, somehow - after all, maybe endianness should be visible in output 
of uname -someswitch...
Maybe uname -m or uname -p should be altered to output mipsel/mipseb... I 
think I saw it once, but now they are both mips...

Regards,
Milan