Shorter version: -m elf32ppc_fbsd (and elf_i386_fbsd ?) vs. -Wl, -m, elf32ppc_fbsd problems (11.0-CURRENT and 10.1-STABLE)

Fri Apr 10 21:21:20 UTC 2015

I had written:

> So I’ve restarted the powerpc64 FreeBSD hosted tests, now based on the notation
> 
> -Wl,-m -Wl,elf32ppc_fbsd
> 
> which sys.mk’s _LDFLAGS assignment should handle okay.

Based on the system/normal gcc 4.2.1 context I’ve successfully updated/rebuilt/booted:

A) powerpc64 10.1-STABLE -r281235 (self updated from my prior snapshot)
B) powerpc64 11.0-CURRENT -r281236
C) powerpc 11.0-CURRENT -r281236 (with -r281243 applied to fix register save/restore)

(B) and (C) were built from (A)’s context (using a separate 11.0-CURRENT svn-based source tree) and DESTDIR used to installkernel and installworld to other SSD’s that were being updated.

The powerpc64-xtoolchain-gcc experimental update on a different PowerMac G5 G5 did okay as far as it got based on the notation. But it did not complete. It will be a while before I research the following references to compiler support routines for restoring various general purpose registers.

> /usr/local/bin/powerpc64-portbld-freebsd11.0-gcc -ffreestanding -msoft-float -Os  -I/usr/src/sys/boot/powerpc/boot1.chrp/../../common -I/usr/src/sys/boot/powerpc/boot1.chrp/../../../  -D_STANDALONE -m
> 32 -mcpu=powerpc -m32 -mcpu=powerpc -std=gnu99     -nostdlib -static -Wl,-N -Wl,-m -Wl,elf32ppc_fbsd -Wl,-m -Wl,elf32ppc_fbsd -o boot1.elf boot1.o ashldi3.o syncicache.o 

got:

> boot1.o: In function `__puts':
> boot1.c:(.text+0xf4): undefined reference to `_restgpr_28_x'
> boot1.o: In function `__printf':
> boot1.c:(.text+0x4fc): undefined reference to `_restgpr_24_x'
> boot1.o: In function `ofw_getprop':
> boot1.c:(.text+0x628): undefined reference to `_restgpr_31_x'
> boot1.o: In function `ofw_close':
> boot1.c:(.text+0x694): undefined reference to `_restgpr_31_x'
> boot1.o: In function `dskread':
> boot1.c:(.text+0x79c): undefined reference to `_restgpr_25_x'
> boot1.o: In function `fsread':
> boot1.c:(.text+0xce4): undefined reference to `_restgpr_14_x'
> boot1.o: In function `domount':
> boot1.c:(.text+0xdb8): undefined reference to `_restgpr_28_x'
> boot1.o: In function `ofw_write.constprop.2':
> boot1.c:(.text+0xe40): undefined reference to `_restgpr_30_x'
> boot1.o: In function `putchar':
> boot1.c:(.text+0xe90): undefined reference to `_restgpr_30_x'
> boot1.o: In function `main':
> boot1.c:(.text.startup+0x528): undefined reference to `_restgpr_18_x'
> collect2: error: ld returned 1 exit status
> *** [boot1.elf] Error code 1

Before the -Wl, notation changes I did not get this far so -WL,-m -Wl,elf32ppc_fbsd did help.

===
Mark Millard
markmi at dsl-only.net

On 2015-Apr-9, at 08:46 PM, Mark Millard <markmi at dsl-only.net> wrote:

I had written:

> I now see one place where "-Wl,-m,elf32ppc_fbsd" type of notation in LDFLAGS would not be handled if it ended up involved:
> 
> share/mk/sys.mk:_LDFLAGS        =       ${LDFLAGS:S/-Wl,//g}    # strip -Wl, for LD
> 
> This notation does not deal with turning the extra comma back into a space.

So I’ve restarted the powerpc64 FreeBSD hosted tests, now based on the notation

-Wl,-m -Wl,elf32ppc_fbsd

which sys.mk’s _LDFLAGS assignment should handle okay.

One test is building 11.0-CURRENT -r281236 via powerpc64-xtoolchain-gcc and the other is building 10.1-STABLE -r281235 via the system/normal gcc 4.2.1.

As my builds take considerable time if they actually complete, it will be a while before I can try any other variations, such as trying a gcc 4.2.1 cross-build for powerpc.

Unfortunately for the purpose at hand: I’m not set up to test any tier 1 FreeBSD environments at this time. So, for example, I can not report observations for amd64 building elf_i386_fbsd materials. My tests may be useful but are not sufficient of themselves to justify edits that anyone worries might damage tier 1 build-ability.

The places I found with notation to adjust if in general the updated notation works are:

> LDFLAGS+=       -m elf32ppc_fbsd
> /usr/src/sys/boot/ofw/Makefile.inc

> LDFLAGS+=       -m elf32ppc_fbsd
> /usr/src/sys/boot/uboot/Makefile.inc

> LDFLAGS+=       -m elf32ppc_fbsd
> /usr/src/sys/boot/powerpc/Makefile.inc

and the tier 1 case using elf_i386_fbsd:

> LD_FLAGS+=      -m elf_i386_fbsd
> /usr/src/sys/boot/i386/Makefile.inc

===
Mark Millard
markmi at dsl-only.net

Just for reference…

On 2015-Apr-9, at 07:35 PM, Warner Losh <imp at bsdimp.com> wrote:

> 
>> On Apr 9, 2015, at 7:56 PM, Mark Millard <markmi at dsl-only.net> wrote:
>> 
>> From share/mk/bsd.README :
>> 
>> LDFLAGS         Additional loader flags. Passed to the loader via CC,
>>             since that's used to link programs as well, so loader
>>             specific flags need to be prefixed with -Wl, to work.
>> 
>> But the following 3 powerpc (non-64) examples do not use the -Wl, notation:
>> 
>>> LDFLAGS+=       -m elf32ppc_fbsd
>>> /usr/src/sys/boot/ofw/Makefile.inc
>> 
>> 
>>> LDFLAGS+=       -m elf32ppc_fbsd
>>> /usr/src/sys/boot/uboot/Makefile.inc
>> 
>> 
>>> LDFLAGS+=       -m elf32ppc_fbsd
>>> /usr/src/sys/boot/powerpc/Makefile.inc
>> 
>> In fact I get errors such as (for that last one when using powerpc64-gcc via powerpc64-xtoolchain-gcc, executed on a powerpc64):
>> 
>>> powerpc64-portbld-freebsd11.0-gcc: error: elf32ppc_fbsd: No such file or directory
>>> powerpc64-portbld-freebsd11.0-gcc: error: elf32ppc_fbsd: No such file or directory
>>> powerpc64-portbld-freebsd11.0-gcc: error: unrecognized command line option '-m'
>>> powerpc64-portbld-freebsd11.0-gcc: error: unrecognized command line option '-m'
>>> 
>>> *** [boot1.elf] Error code 1
>>> 
>>> make[6]: stopped in /usr/srcC/sys/boot/powerpc/boot1.chrp
>>> 1 error
>> 
>> I do not know if the space between -m and elf... creates a problem for -Wl, use or not. I would guess that
>> 
>> -Wl,-m,elf32pcc_fbsd
>> 
>> is the proper notation for putting the space through to the ld variant used. But I’m not to the point of testing the behavior of that yet.
>> 
>> 
>> 
>> i386 seems to have a similar example, although I’m not using such a FreeBSD environment.
>> 
>>> LD_FLAGS+=      -m elf_i386_fbsd
>>> /usr/src/sys/boot/i386/Makefile.inc
>> 
>> 
>> 
>> (This note is shorter in part because figured out more context than I had last time.)
> 
> I think much of this is historical accident where the boot Makefiles used to call ld directly, then were converted to call gcc, and gcc allowed the -m notation like this as a historical compatibility.
> 
> Do thinks still work if you use -Wl, notation?
> 
> Warner
> 

I have a powerpc64-xtoolchain-gcc build going now but that will take a while. I’ll also need to try a normal one from a gcc 4.2.1 environment and I’ve not started that yet. And I’m only set up to test powerpc64 and powerpc. The tier 1 consequences (i386 and amd64) are outside my environment.

Also I sent out another note after discovering a potential problem...

> I now see one place where "-Wl,-m,elf32ppc_fbsd" type of notation in LDFLAGS would not be handled if it ended up involved:
> 
> share/mk/sys.mk:_LDFLAGS        =       ${LDFLAGS:S/-Wl,//g}    # strip -Wl, for LD
> 
> This notation does not deal with turning the extra comma back into a space.

===
Mark Millard
markmi at dsl-only.net