[Buildroot] Build jobbing

[Buildroot] Build jobbing

[Michael S. Zick]

When Buildroot is tested, do you use anything other than: -j1?

I have been using -j5 to keep my quad-core out of the idle loop;
but am beginning to suspect there is something not right about that.

Mike

[Buildroot] Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

 Michael> When Buildroot is tested, do you use anything other than: -j1?

Sure.

 Michael> I have been using -j5 to keep my quad-core out of the idle loop;
 Michael> but am beginning to suspect there is something not right about that.

Notice that you have to set BR2_JLEVEL under build options - The
toplevel make must NOT be run with -jN.

-- 
Bye, Peter Korsgaard

[Buildroot] Build jobbing

[Michael S. Zick]

On Fri November 27 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> 
>  Michael> When Buildroot is tested, do you use anything other than: -j1?
> 
> Sure.
> 
>  Michael> I have been using -j5 to keep my quad-core out of the idle loop;
>  Michael> but am beginning to suspect there is something not right about that.
> 
> Notice that you have to set BR2_JLEVEL under build options - The
> toplevel make must NOT be run with -jN.
> 

Yup, got that right.

Now using the repository head -

The cross-uClibc will not build with "gnu-hash" option unless the host binutils
already support "gnu-hash" (??) - Easy, I just disable that and pretended I 
didn't get the message.

ltrace does not build - just disabled it for now.
mii-diag-2.11 does not build - just disable it for now.
We already know about sstrip - I bug reported that one (I think).

nand_{erase,lock,unlock) in BB-1.15.2 does not build.

continuing...
will go back and investigate those things *later*.

Mike

[Buildroot] Build jobbing

[Michael S. Zick]

On Fri November 27 2009, Michael S. Zick wrote:
> On Fri November 27 2009, Peter Korsgaard wrote:
> > >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> > 
> >  Michael> When Buildroot is tested, do you use anything other than: -j1?
> > 
> > Sure.
> > 
> >  Michael> I have been using -j5 to keep my quad-core out of the idle loop;
> >  Michael> but am beginning to suspect there is something not right about that.
> > 
> > Notice that you have to set BR2_JLEVEL under build options - The
> > toplevel make must NOT be run with -jN.
> > 
> 
> Yup, got that right.
> 
> Now using the repository head -
> 
> The cross-uClibc will not build with "gnu-hash" option unless the host binutils
> already support "gnu-hash" (??) - Easy, I just disable that and pretended I 
> didn't get the message.
> 
> ltrace does not build - just disabled it for now.
> mii-diag-2.11 does not build - just disable it for now.
> We already know about sstrip - I bug reported that one (I think).
> 
> nand_{erase,lock,unlock) in BB-1.15.2 does not build.
> 

nbd-server.c:267: Incompatible implicit declaration of built-in function "index"
Disabled, continuing.

Mike
> continuing...
> will go back and investigate those things *later*.
> 
> Mike
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
> 
> 

[Buildroot] Build jobbing

[Michael S. Zick]

On Fri November 27 2009, Michael S. Zick wrote:
> On Fri November 27 2009, Michael S. Zick wrote:
> > On Fri November 27 2009, Peter Korsgaard wrote:
> > > >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> > > 
> > >  Michael> When Buildroot is tested, do you use anything other than: -j1?
> > > 
> > > Sure.
> > > 
> > >  Michael> I have been using -j5 to keep my quad-core out of the idle loop;
> > >  Michael> but am beginning to suspect there is something not right about that.
> > > 
> > > Notice that you have to set BR2_JLEVEL under build options - The
> > > toplevel make must NOT be run with -jN.
> > > 
> > 
> > Yup, got that right.
> > 
> > Now using the repository head -
> > 
> > The cross-uClibc will not build with "gnu-hash" option unless the host binutils
> > already support "gnu-hash" (??) - Easy, I just disable that and pretended I 
> > didn't get the message.
> > 
> > ltrace does not build - just disabled it for now.
> > mii-diag-2.11 does not build - just disable it for now.
> > We already know about sstrip - I bug reported that one (I think).
> > 
> > nand_{erase,lock,unlock) in BB-1.15.2 does not build.
> > 
> 
> nbd-server.c:267: Incompatible implicit declaration of built-in function "index"
> Disabled, continuing.
> 

startup-notification-0.9 does not build, no X11 development libraries found.
(No libraries is correct, in that I did not intend to build anything X11-ish,
it was the selection of startup-notification that should have been blocked.)

Oh,
kexec will not build, forget why (I think that one was the "no such instruction" 
in a *.S for MIPS) - will look into that one later.

Both disabled, continuing.
Mike
> Mike
> > continuing...
> > will go back and investigate those things *later*.
> > 
> > Mike
> > _______________________________________________
> > buildroot mailing list
> > buildroot at busybox.net
> > http://lists.busybox.net/mailman/listinfo/buildroot
> > 
> > 
> 
> 
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
> 
> 

[Buildroot] Build jobbing

[Michael S. Zick]

On Fri November 27 2009, Michael S. Zick wrote:
> On Fri November 27 2009, Michael S. Zick wrote:
> > On Fri November 27 2009, Michael S. Zick wrote:
> > > On Fri November 27 2009, Peter Korsgaard wrote:
> > > > >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> > > > 
> > > >  Michael> When Buildroot is tested, do you use anything other than: -j1?
> > > > 
> > > > Sure.
> > > > 
> > > >  Michael> I have been using -j5 to keep my quad-core out of the idle loop;
> > > >  Michael> but am beginning to suspect there is something not right about that.
> > > > 
> > > > Notice that you have to set BR2_JLEVEL under build options - The
> > > > toplevel make must NOT be run with -jN.
> > > > 
> > > 
> > > Yup, got that right.
> > > 
> > > Now using the repository head -
> > > 
> > > The cross-uClibc will not build with "gnu-hash" option unless the host binutils
> > > already support "gnu-hash" (??) - Easy, I just disable that and pretended I 
> > > didn't get the message.
> > > 
> > > ltrace does not build - just disabled it for now.
> > > mii-diag-2.11 does not build - just disable it for now.
> > > We already know about sstrip - I bug reported that one (I think).
> > > 
> > > nand_{erase,lock,unlock) in BB-1.15.2 does not build.
> > > 
> > 
> > nbd-server.c:267: Incompatible implicit declaration of built-in function "index"
> > Disabled, continuing.
> > 
> 
> startup-notification-0.9 does not build, no X11 development libraries found.
> (No libraries is correct, in that I did not intend to build anything X11-ish,
> it was the selection of startup-notification that should have been blocked.)
> 
> Oh,
> kexec will not build, forget why (I think that one was the "no such instruction" 
> in a *.S for MIPS) - will look into that one later.
> 
> Both disabled, continuing.
>

sylpheed-2.4.8 - fails to build due to no GTK+ 
Not having GTK+ is correct, but should have disabled sylpheed selection.

sysstat-9.0.5 - fails to build due to "undefined ref to libintl_gettext"
but both libintl and gettext where auto-selected to be built. (???)

Both disabled, continuing.

Mike
Mike
> Mike
> > Mike
> > > continuing...
> > > will go back and investigate those things *later*.
> > > 
> > > Mike
> > > _______________________________________________
> > > buildroot mailing list
> > > buildroot at busybox.net
> > > http://lists.busybox.net/mailman/listinfo/buildroot
> > > 
> > > 
> > 
> > 
> > _______________________________________________
> > buildroot mailing list
> > buildroot at busybox.net
> > http://lists.busybox.net/mailman/listinfo/buildroot
> > 
> > 
> 
> 
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
> 
> 

[Buildroot] Build jobbing

[Michael S. Zick]

Well, the build completed - -
Now will see any of this stuff runs.  ;)

Will be doing another complete (as possible) build RSN,
even if this stuff runs - I didn't notice that I had
dropped the tune=24kc option during the switch from
2009.08 to "head" - shame on me.

You haven't seen that patch yet - because there are a 
slew of 24kc cores and also a slew of gcc options
to match - need time (and MTI documentation) to sort
it all out.
This processor core **should** be a 24kc-pro (hardware fp)
but that only narrows it down to three or four fp implementations.


Mike

[Buildroot] Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

Hi,

 Michael> nbd-server.c:267: Incompatible implicit declaration of built-in function "index"
 Michael> Disabled, continuing.

It sounds like the patches somehow didn't get applied - What does
output/build/nbd-2.9.11/.applied_patches_list contain?

It should contain:

cat output/build/nbd-2.9.11/.applied_patches_list
nbd-fix-build-with-disable-lfs.patch
nbd-server-fix-read-eagain.patch
nbd-server-susv3-legacy.patch

Can you rm output/build/nbd-2.9.11 and try again, and this time check if
nbd-server-susv3-legacy.patch gets applied?

-- 
Bye, Peter Korsgaard

[Buildroot] Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

Hi,

 Michael> ltrace does not build - just disabled it for now.

It builds here:

build/ltrace-0.5/ltrace: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked (uses shared libs), not stripped

What exactly goes wrong?

 Michael> mii-diag-2.11 does not build - just disable it for now.

It builds here:

build/mii-diag-2.11/mii-diag: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked (uses shared libs), not stripped

What exactly goes wrong?

 Michael> nand_{erase,lock,unlock) in BB-1.15.2 does not build.

There's a busybox patch pending for building the flash stuff with
2.6.31.x kernel headers. I'll do a bit more testing and commit.

-- 
Bye, Peter Korsgaard

[Buildroot] Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
>

I'll go back through all of those today, one at a time, and post
either patches or a better description of each problem.

I am gaining confidence that I have most (all?) of the cockpit
errors out of my setup here.

Some of those things I know how to fix,
Others, I needed the additional set of eyes and suggestions
just like you provided.

Mike 
> Hi,
> 
>  Michael> ltrace does not build - just disabled it for now.
> 
> It builds here:
> 
> build/ltrace-0.5/ltrace: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked (uses shared libs), not stripped
> 
> What exactly goes wrong?
> 
>  Michael> mii-diag-2.11 does not build - just disable it for now.
> 
> It builds here:
> 
> build/mii-diag-2.11/mii-diag: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked (uses shared libs), not stripped
> 
> What exactly goes wrong?
> 
>  Michael> nand_{erase,lock,unlock) in BB-1.15.2 does not build.
> 
> There's a busybox patch pending for building the flash stuff with
> 2.6.31.x kernel headers. I'll do a bit more testing and commit.
> 

[Buildroot] Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Michael S. Zick wrote:
> On Sat November 28 2009, Peter Korsgaard wrote:
> > >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> >
> 
> I'll go back through all of those today, one at a time, and post
> either patches or a better description of each problem.
> 

First off, what worked - -

My configuration menu patch to expose the "build a static busybox"
worked just as intended while doing a full, dynamic system build.

During the build of uClibc, both shared and static libraries need
to always be built -
That is because the library loader utilities need to be linked as
static executables so that the dynamic library system can be fixed
if (when) it breaks.

Perhaps we need a menu selection to drop the static libraries from the 
target fs image but they are required inside of the build system.

On the other hand, that might be considered a "site specific detail",
and just like splitting the fs image, should be left to post-buildroot
scripting by the buildroot user.

The "mips32-r2, tune=24kc, hard-float" without specifying *which*
hard-float type to gcc seems to be working (BB uses float in e2fsck).
Or perhaps I just got lucky with that one on this SoC (a Realtec RTD-1073).
Realtec licensed the 24kc-pro (hardware float) core back in 2004/2005.

After running the uClibc test suite, I'll try the Lua numerical test suite.
Lua only has one internal number type: Double. So if that works....

Mike
> I am gaining confidence that I have most (all?) of the cockpit
> errors out of my setup here.
> 
> Some of those things I know how to fix,
> Others, I needed the additional set of eyes and suggestions
> just like you provided.
> 
> Mike 
> > Hi,
> > 
> >  Michael> ltrace does not build - just disabled it for now.
> > 
> > It builds here:
> > 
> > build/ltrace-0.5/ltrace: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked (uses shared libs), not stripped
> > 
> > What exactly goes wrong?
> > 
> >  Michael> mii-diag-2.11 does not build - just disable it for now.
> > 
> > It builds here:
> > 
> > build/mii-diag-2.11/mii-diag: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked (uses shared libs), not stripped
> > 
> > What exactly goes wrong?
> > 
> >  Michael> nand_{erase,lock,unlock) in BB-1.15.2 does not build.
> > 
> > There's a busybox patch pending for building the flash stuff with
> > 2.6.31.x kernel headers. I'll do a bit more testing and commit.
> > 
> 
> 
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
> 
> 

[Buildroot] Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Michael S. Zick wrote:
> 
> After running the uClibc test suite, I'll try the Lua numerical test suite.
> Lua only has one internal number type: Double. So if that works....
> 
> Mike
> > I am gaining confidence that I have most (all?) of the cockpit
> > errors out of my setup here.
> > 

Well, not quite all of the cockpit error:

/ # chroot /tmp/work
/ # lua
Lua 5.1.4  Copyright (C) 1994-2008 Lua.org, PUC-Rio
> print(math.sin(12))
table index is NaN
> os.exit()
/ #

Takes deep breath, starts over with those compiler and build options.  ;)

Mike

[Buildroot] Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

Hi,

 Michael> nand_{erase,lock,unlock) in BB-1.15.2 does not build.

I take it that you mean flash_eraseall / flash_unlock / flash_lock?

It works for me after 4e385ee7 - Could you update and try again?

-- 
Bye, Peter Korsgaard

[Buildroot] Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> 
> Hi,
> 
>  Michael> nand_{erase,lock,unlock) in BB-1.15.2 does not build.
> 
> I take it that you mean flash_eraseall / flash_unlock / flash_lock?
> 
> It works for me after 4e385ee7 - Could you update and try again?
> 

Yes, it was a long day.

Indeed, these are not "options", without them the firmware will
never get off the first cylinder of the media storage device.

Mike

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Michael S. Zick wrote:
> On Sat November 28 2009, Peter Korsgaard wrote:
> > >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> > 
> > Hi,
> > 
> >  Michael> nand_{erase,lock,unlock) in BB-1.15.2 does not build.
> > 
> > I take it that you mean flash_eraseall / flash_unlock / flash_lock?
> > 
> > It works for me after 4e385ee7 - Could you update and try again?
> > 

The build didn't get far enough to find out, it looks like it broke
on the recent scsi commit:

In file included from miscutils/eject.c:20:
/Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h:145: error: expected specifier-qualifier-list before ?u8?
/Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h: In function ?scsi_varlen_cdb_length?:
/Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h:156: error: ?struct scsi_varlen_cdb_hdr? has no member named ?additional_cdb_length?
make[2]: *** [miscutils/eject.o] Error 1
make[1]: *** [miscutils] Error 2
make[1]: *** Waiting for unfinished jobs....

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Michael S. Zick wrote:
> On Sat November 28 2009, Michael S. Zick wrote:
> > On Sat November 28 2009, Peter Korsgaard wrote:
> > > >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> > > 
> > > Hi,
> > > 
> > >  Michael> nand_{erase,lock,unlock) in BB-1.15.2 does not build.
> > > 
> > > I take it that you mean flash_eraseall / flash_unlock / flash_lock?
> > > 
> > > It works for me after 4e385ee7 - Could you update and try again?
> > > 
>

After disabling eject/eject-scsi support. . .
Looks like it has a dependency on jffs2:

  CC      miscutils/flash_lock_unlock.o
miscutils/flash_eraseall.c:15:28: error: mtd/jffs2-user.h: No such file or directory
miscutils/flash_eraseall.c: In function ?flash_eraseall_main?:
miscutils/flash_eraseall.c:52: error: storage size of ?cleanmarker? isn?t known
miscutils/flash_eraseall.c:83: warning: implicit declaration of function ?cpu_to_je16?
miscutils/flash_eraseall.c:83: error: ?JFFS2_MAGIC_BITMASK? undeclared (first use in this function)
miscutils/flash_eraseall.c:83: error: (Each undeclared identifier is reported only once
miscutils/flash_eraseall.c:83: error: for each function it appears in.)
miscutils/flash_eraseall.c:84: error: ?JFFS2_NODETYPE_CLEANMARKER? undeclared (first use in this function)
miscutils/flash_eraseall.c:86: warning: implicit declaration of function ?cpu_to_je32?
miscutils/flash_eraseall.c:86: error: invalid application of ?sizeof? to incomplete type ?struct jffs2_unknown_node?
miscutils/flash_eraseall.c:121: error: invalid application of ?sizeof? to incomplete type ?struct jffs2_unknown_node?
miscutils/flash_eraseall.c:52: warning: unused variable ?cleanmarker?
make[2]: *** [miscutils/flash_eraseall.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [miscutils] Error 2
make[1]: *** Waiting for unfinished jobs....


Mike 
> The build didn't get far enough to find out, it looks like it broke
> on the recent scsi commit:
> 
> In file included from miscutils/eject.c:20:
> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h:145: error: expected specifier-qualifier-list before ?u8?
> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h: In function ?scsi_varlen_cdb_length?:
> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h:156: error: ?struct scsi_varlen_cdb_hdr? has no member named ?additional_cdb_length?
> make[2]: *** [miscutils/eject.o] Error 1
> make[1]: *** [miscutils] Error 2
> make[1]: *** Waiting for unfinished jobs....
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot

[Buildroot] BB update Was: Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

 Michael> After disabling eject/eject-scsi support. . .
 Michael> Looks like it has a dependency on jffs2:

 Michael>   CC      miscutils/flash_lock_unlock.o
 Michael> miscutils/flash_eraseall.c:15:28: error: mtd/jffs2-user.h: No such file or directory

That's fixed by 4e385ee7 - Are you sure your tree is uptodate?

-- 
Bye, Peter Korsgaard

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> 
>  Michael> After disabling eject/eject-scsi support. . .
>  Michael> Looks like it has a dependency on jffs2:
> 
>  Michael>   CC      miscutils/flash_lock_unlock.o
>  Michael> miscutils/flash_eraseall.c:15:28: error: mtd/jffs2-user.h: No such file or directory
> 
> That's fixed by 4e385ee7 - Are you sure your tree is uptodate?
> 

I am no git expert, but it looks like it to me:

commit b4b561df9d484757845deaf1df91c418f5e68b78
Merge: 8c366a6... 4e385ee...
Author: Mike Z <minimod@morethan.org>
Date:   Sat Nov 28 09:22:25 2009 -0600

    Merge branch 'master' of git://git.buildroot.net/buildroot

commit 4e385ee794978e04701171f06cbfad8b3796e468
Author: Peter Korsgaard <jacmet@sunsite.dk>
Date:   Sat Nov 28 15:05:52 2009 +0100

    busybox: 1.15.2 flash_eraseall fix for modern kernel headers

    86cfb70ca upstream.

    Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>

commit 57922bfbf324745fb486775a53725ab573bfbf12
Author: Peter Korsgaard <jacmet@sunsite.dk>
Date:   Sat Nov 28 14:25:13 2009 +0100

    busybox: look at kernel headers for include files as well

    E.G. for flash_* applets.

    Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>

The git://git.buildroot.net/buildroot is a clone of the local
repository clone, I pulled them both before this build.

Did something move in the Buildroot tree?  I did not start over
with a distclean tree.  Whatever distclean does now.  ;)

Mike

[Buildroot] BB update Was: Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

Hi,

 >> That's fixed by 4e385ee7 - Are you sure your tree is uptodate?
 >> 

 Michael> I am no git expert, but it looks like it to me:

To me as well.

 Michael> Did something move in the Buildroot tree?  I did not start over
 Michael> with a distclean tree.  Whatever distclean does now.  ;)

No, but you DO have to run make clean for new patches to take effect -
Did you do that?

-- 
Bye, Peter Korsgaard

[Buildroot] BB update Was: Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

Hi,

 Michael> The build didn't get far enough to find out, it looks like it broke
 Michael> on the recent scsi commit:

 Michael> In file included from miscutils/eject.c:20:
 Michael> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h:145: error: expected specifier-qualifier-list before ?u8?
 Michael> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h: In function ?scsi_varlen_cdb_length?:
 Michael> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h:156: error: ?struct scsi_varlen_cdb_hdr? has no member named ?additional_cdb_length?

What kernel headers version are you using? I checked in a fix for
2.6.31.6 kernel headers today (c589012eecd)

-- 
Bye, Peter Korsgaard

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> 
> Hi,
> 
>  Michael> The build didn't get far enough to find out, it looks like it broke
>  Michael> on the recent scsi commit:
> 
>  Michael> In file included from miscutils/eject.c:20:
>  Michael> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h:145: error: expected specifier-qualifier-list before ?u8?
>  Michael> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h: In function ?scsi_varlen_cdb_length?:
>  Michael> /Builds/BLDRT/br-head/output/toolchain/linux/include/scsi/scsi.h:156: error: ?struct scsi_varlen_cdb_hdr? has no member named ?additional_cdb_length?
> 
> What kernel headers version are you using? I checked in a fix for
> 2.6.31.6 kernel headers today (c589012eecd)
> 

It looks like this:

  ? ?             Toolchain type (Buildroot toolchain)  --->                                 ? ?
  ? ?             *** Kernel Header Options ***                                              ? ?
  ? ?             Kernel Headers (Linux 2.6.31.x kernel headers)  --->                       ? ?
  ? ?             *** uClibc Options ***                                                     ? ?
  ? ?             uClibc C library Version (uClibc 0.9.30.1)  --->                           ? ?
  ? ?         (toolchain/uClibc/uClibc-0.9.30.config) uClibc configuration file to use?      ? ?
  ? ?         [ ] Thread library debugging                                                   ? ?
  ? ?         [*] Compile and install uClibc tests                                           ? ?
  ? ?             *** Binutils Options ***                                                   ? ?
  ? ?             Binutils Version (binutils 2.19.1)  --->                                   ? ?
  ? ?         ()  Additional binutils options                                                ? ?
  ? ?             *** GCC Options ***                                                        ? ?
  ? ?             GCC compiler Version (gcc 4.3.4)  --->                                     ? ?
  ? ?         [*] Enable toolchain with --sysroot support                                    ? ?
  ? ?         [*] Enable setjmp/longjmp exceptions?                                          ? ?
  ?

Mike

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Michael S. Zick wrote:
>   ? ?             GCC compiler Version (gcc 4.3.4)  --->                                     ? ?
>   ? ?         [*] Enable toolchain with --sysroot support                                    ? ?
>   ? ?         [*] Enable setjmp/longjmp exceptions?                                          ? ?
>

One more nit - 

Enable setjmp/longjmp should default to yes if Lua is selected.

Lua uses 'pure' C code, I.E: setjmp/longjmp because that is the
only type of exception handling in ansi-c standard.

Mike

[Buildroot] BB update Was: Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

 Michael> One more nit - 

 Michael> Enable setjmp/longjmp should default to yes if Lua is selected.

 Michael> Lua uses 'pure' C code, I.E: setjmp/longjmp because that is the
 Michael> only type of exception handling in ansi-c standard.

setjmp exceptions is afaik something else. I atleast just built lua
without it being enabled.

-- 
Bye, Peter Korsgaard

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sat November 28 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> 
>  Michael> One more nit - 
> 
>  Michael> Enable setjmp/longjmp should default to yes if Lua is selected.
> 
>  Michael> Lua uses 'pure' C code, I.E: setjmp/longjmp because that is the
>  Michael> only type of exception handling in ansi-c standard.
> 
> setjmp exceptions is afaik something else. I atleast just built lua
> without it being enabled.
> 

Look at my print(math.sin(12)) example -
bet you don't see any backtrace.

Mike

[Buildroot] BB update Was: Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

Hi,

 >> setjmp exceptions is afaik something else. I atleast just built lua
 >> without it being enabled.

 Michael> Look at my print(math.sin(12)) example -
 Michael> bet you don't see any backtrace.

Backtrace? Where should I see a backtrace?

Lua on my host (Debian):
% lua
Lua 5.1.4  Copyright (C) 1994-2008 Lua.org, PUC-Rio
> print(math.sin(12))
-0.53657291800043
> os.exit()

Lua built from BR:

% sudo chroot /tmp/blah /bin/sh
/ # lua
Lua 5.1.4  Copyright (C) 1994-2008 Lua.org, PUC-Rio
> print(math.sin(12))
-0.53657291800043
> os.exit()
/ # exit

Looks the same to me.

And this is without the setjmp exceptions:

% grep SJL .config
# BR2_GCC_USE_SJLJ_EXCEPTIONS is not set

-- 
Bye, Peter Korsgaard

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sun November 29 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> 
> Hi,
> 
>  Michael> Try:
> 
>  Michael> Lua 5.1.4  Copyright (C) 1994-2008 Lua.org, PUC-Rio
>  >> print(math.sin(garbage))
>  Michael> stdin:1: bad argument #1 to 'sin' (number expected, got nil)
>  Michael> stack traceback:
>  Michael>         [C]: in function 'sin'
>  Michael>         stdin:1: in main chunk
>  Michael>         [C]: ?
> 
> 
> I get exactly the same on BR:
> 
> sudo chroot /tmp/blah /bin/sh
> / # lua
> Lua 5.1.4  Copyright (C) 1994-2008 Lua.org, PUC-Rio
> > print(math.sin(garbage))
> stdin:1: bad argument #1 to 'sin' (number expected, got nil)
> stack traceback:
>         [C]: in function 'sin'
>         stdin:1: in main chunk
>         [C]: ?
> 

Found what might lead to the references I need:
http://www.mips.com/products/processors/32-64-bit-cores/mips32-24k/index.cfm#specifications

Which immediately shows that I made an error - about mid-page;
Just because the core has "core extend" does not mean it has 
hardware floating point.
duh...
It would seem that if you have enough gates for a pair of DSP
cores, you would have enough to "waste" a few thousand on H.F.P.

I'll dig into the problem and try to sort it all out today.
Will also try to figure out why I didn't get a backtrace from
a NaN error in Lua.  The "no backtrace" is a separate problem
from "no double" number type.

At first glance, it looks like a case for a Config.in.mips sub-menu.
Just to get all of the gcc options and core options to match up.

Mike

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sun November 29 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> 
> Hi,
> 
>  Michael> Well, then the text in the menu must be mis-leading. . .
>  Michael> it must be refering to generation/use of setjmp/longjmp
>  Michael> somewhere other than in the target applications compiled.
> 
> config BR2_GCC_USE_SJLJ_EXCEPTIONS
>         bool "Enable setjmp/longjmp exceptions?"
>         help
>           For some platforms, the 

            internal 

>           stack unwinding 

            that gcc generates for stack traces and debugging info

>           works perfectly, 
>           while other platforms must use setjmp/longjmp exceptions for
>           proper stack unwinding during exception handling.  

            This does not effect the use of setjmp/longjmp in applications
            compiled by gcc.

>           Most people can leave this set to n.
> 
> That's imho fairly clearly not about basic setjmp/longjmp support.
> 

Only when you are thinking about building a compiler, but if instead you
happen to be an applications programmer, who doesn't know or care how
the compiler works internally...

I only now recognize why it seems clear to you - because I did live
on the gcc list once upon a time - not all of the buildroot users
have that background.

>  Michael> The most recent build has the option enabled, will see if
>  Michael> it makes a difference.
> 
>  Michael> I did not test yesterday's build because of local work rules:
>  Michael> 1) No shift longer than 18 hours;
>  Michael> 2) A minimum of 4 hours off between shifts.
> 
> ;)
> 

Mike

[Buildroot] BB update Was: Build jobbing

[Peter Korsgaard]

>>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:

 >> That's imho fairly clearly not about basic setjmp/longjmp support.

 Michael> Only when you are thinking about building a compiler, but if
 Michael> instead you happen to be an applications programmer, who
 Michael> doesn't know or care how the compiler works internally...

Then he probably shouldn't be changing stuff under toolchain options ;)

 Michael> I only now recognize why it seems clear to you - because I did live
 Michael> on the gcc list once upon a time - not all of the buildroot users
 Michael> have that background.

Actually I haven't done much gcc development. The
BR2_GCC_USE_SJLJ_EXCEPTIONS option got added back in 2004 by Erik really
in beginning of BR development. I don't know of anything needing it
today.

In general, we try to use sensible defaults - If you don't understand an
option, just leave it at default.

-- 
Bye, Peter Korsgaard

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sun November 29 2009, Peter Korsgaard wrote:
> >>>>> "Michael" == Michael S Zick <minimod@morethan.org> writes:
> 
>  >> That's imho fairly clearly not about basic setjmp/longjmp support.
> 
>  Michael> Only when you are thinking about building a compiler, but if
>  Michael> instead you happen to be an applications programmer, who
>  Michael> doesn't know or care how the compiler works internally...
> 
> Then he probably shouldn't be changing stuff under toolchain options ;)
> 
>  Michael> I only now recognize why it seems clear to you - because I did live
>  Michael> on the gcc list once upon a time - not all of the buildroot users
>  Michael> have that background.
> 
> Actually I haven't done much gcc development. The
> BR2_GCC_USE_SJLJ_EXCEPTIONS option got added back in 2004 by Erik really
> in beginning of BR development. I don't know of anything needing it
> today.
> 

2004?  That explains that -
Back in 2004 there where three possibilities for stack unwind code in the
generated objects, the "third" - setjmp/longjmp, was a user specified
option for use on arch's that did not have either of the other two complete.

If Erik had put in all three, or
If Erik had put in a "use compiler default model" / "use setjmp-longjmp"
then that help message would have been comprehensible then *and* now.

> In general, we try to use sensible defaults - If you don't understand an
> option, just leave it at default.
> 

Ah, but I did - - -
Until I encountered an application that failed to generate a backtrace -
And then I re-read the help message with the "application code" viewpoint
rather than the "compiler generated objects" viewpoint.

*and* we should hold off on refining this help message until I get a chance
to see if it does effect the application's use of setjmp/longjmp - - -

I haven't had a chance to run yesterday's build yet.
Instead, I spent this morning's time registering with MTI and sucking their
technical reference library dry.  At least on the subject of the 24K family.
Everything short of the FPGA code to load onto my Xilinx development board
and "grow" (programically) my own core.

I am also now the licensed owner of an MTI routine that *will* ID this core.

My plan of attack:
Use a spreadsheet to produce a feature matrix -
Then populate that with the appropriate GCC options -
And I expect to end up with something that will need a mips-submenu.

That is just for the 24K family - -
Users of Buildroot, using other MIPS families can do their own feature 
matrix and edit the mips-submenu following the model I set.

Between the configuration menu/matrix and the runtime cpuid thingy -
I expect I can get this system to turn out something better than 1984 MIPS code.

Mike

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sun November 29 2009, Michael S. Zick wrote:
> 
> Between the configuration menu/matrix and the runtime cpuid thingy -
> I expect I can get this system to turn out something better than 1984 MIPS code.
> 

Dream on. . .

First footnote of interest found:

"Floating-point operations on very small numbers (denormalized numbers) and illegal
numbers (NaNs) must be handled by software.  Also some IEEE exceptions and exceptional
results will cause an FPU exception.  The FPU emulator is thus mandatory for full IEEE
compliance, ___even when the system has a hardware FPU___."

Emphasis added.

So we can't configure with a simple:
soft-float / hard-float configuration variable - -
the choices are: soft-float / sort-of-hard-float + a bit of help.

That will also impact that "help message" about not loading the FPU emulator into
a kernel more recent than 1997....
If your running MIPS, and your core has hardware floating point and you choose:
hard-float - - you still need the FPU emulator loaded for IEEE compliance.

Translation: the FPU emulator is not optional, you need it regardless of the 
hard-float / soft-float selection.

I guess we have to go all the way back to the uClibc question:
"does your machine have an FPU?" yes/no
to
"does your machine have an FPU?" yes/sort-of/no

Just in case I needed a reminder of why I retired from this *&^&*% twenty years ago.
As if I need another one.  ;)

Mike

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sun November 29 2009, Michael S. Zick wrote:
> 
> My plan of attack:
> Use a spreadsheet to produce a feature matrix -
> Then populate that with the appropriate GCC options -
>

Not required, MTI already did that, see:

"MIPS Toolchain Specifics", Document: MD00624, Revision 01.52, April 8, 2009.

At least something about this got simpler, finally.  ;)

Mike
> And I expect to end up with something that will need a mips-submenu.
> 

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sun November 29 2009, Michael S. Zick wrote:
> If Erik had put in a "use compiler default model" / "use setjmp-longjmp"
> 

I thought that we had agreed the above was the meaning of that option -
only effects the compiler's model of stack unwind information.
**NOT** user space. (unless you count gdb - which has to understand it).

> > In general, we try to use sensible defaults - If you don't understand an
> > option, just leave it at default.
> > 

guess what, the "sensible defaults" results are in:

/ # uname -a
Linux Venus 2.6.12.6-VENUS #13 Wed Jul 15 17:57:07 CST 2009 mips unknown
/ # chroot /tmp/work
/ # lua
Lua 5.1.4  Copyright (C) 1994-2008 Lua.org, PUC-Rio
> print(math.sin(12))
-0.53657291800043
> print(math.sin(garbage))
stdin:1: bad argument #1 to 'sin' (number expected, got nil)
stack traceback:
        [C]: in function 'sin'
        stdin:1: in main chunk
        [C]: ?
> os.exit()
/ #

1) The computation of math.sin(12) (as a double) is incorrect - -
That is because I didn't know about the "sort-of-Hard-Float" yesterday.
I should be able to fix that once I get the math options corrected.

2) It didn't trap, just produced an erroneous value (?wrong fp options?).

3) I now have a stack traceback (implemented in Lua as setjmp/longjmp)!

Number three is the one specific to this discussion.

That 2004-Eric option is affecting the use of setjmp/longjmp *in application
code*.  
I think we are agreed that was not the intent, 
regardless of the wording of the help message.

So I will continue with translating this 0.5Gbytes of MIPS documentation into
a menuconfig file - - - (number 1 & 2 above).

Somebody familiar with the build system take a look at how this option is being
used please.  Assign it a bug-tracking number if no "quick fix" is seen.

*) it shouldn't be needed by gcc-4.x for MIPS; "compiler default" should work.
*) if it leads to a "--disable-setjmp" option being passed to applications;
that needs to be sorted out.

Mike

[Buildroot] BB update Was: Build jobbing

[Michael S. Zick]

On Sun November 29 2009, Michael S. Zick wrote:
> On Sun November 29 2009, Michael S. Zick wrote:
> / # uname -a
> Linux Venus 2.6.12.6-VENUS #13 Wed Jul 15 17:57:07 CST 2009 mips unknown
> / # chroot /tmp/work
> / # lua
> Lua 5.1.4  Copyright (C) 1994-2008 Lua.org, PUC-Rio
> > print(math.sin(12))
> -0.53657291800043
>

Geesh! I must need some sleep, that is the correct sign of 12 *radians*!
duh...
goes to take a long nap.

Mike