Hi
I have he follow piece of assembler code powerpc.s running on powerpc6
redhat4.5
with GCC4.1.1
when i do a nm on the the object the symbol is undefined
Can you see whats wrong with the code i have below that the symbols
os_atomic_lock
os_atomic_unlock, etc are showing up as undefined ...
Please HELP .......
Thanks
nm powerpc.o |grep os_atomic_lock
0000000000000018 n os_atomic_lock
U os_atomic_lock
====
.set r0,0
.set SP,1
.set RTOC,2
.set r3,3
.set r4,4
.set r5,5
.set r6,6
.set r7,7
.set r8,8
.set r9,9
.extern os_atomic_add
.extern os_atomic_lock
.extern os_atomic_unlock
.extern os_atomic_set
.extern os_atomic_get
.extern os_atomic_init
.file "powerpc.s"
.sect H.10.NO_SYMBOL{PR} # I have no idea what this does
os_atomic_add:
lwarx r5, 0, r3
add r0, r4, r5
stwcx. r0, 0, r3
bne- os_atomic_add
mr r3, r0
blr
os_atomic_lock:
li r0, 1
loop:
# loop cheaply while not zero (this might speed things up)
# lwz r5, 0(r3)
# cmpwi r5, 0
# bne- loop
lwarx r5, 0, r3
cmpw r0, r5
beq- loop
stwcx. r0, 0, r3
bne- loop
isync
blr
os_atomic_unlock:
sync
os_atomic_init:
li r0, 0
stw r0, 0(r3)
blr
os_atomic_set:
#sync ?
stw r4, 0(r3)
blr
os_atomic_get:
#sync ?
lwz r3, 0(r3)
blr
/usr/bin/as -mppc64 -a64 -as -o powerpc.o
Nano2 Nano wrote: > Hi > I have he follow piece of assembler code powerpc.s running on powerpc6 > redhat4.5 > with GCC4.1.1 I wonder then why you are asking this question on an ARM development forum? Nano2 Nano wrote: > > when i do a nm on the the object the symbol is undefined > Can you see whats wrong with the code i have below that the symbols > os_atomic_lock > os_atomic_unlock, etc are showing up as undefined ... I am not sure but some versions of GCC add a _ prefix to external symbols, wheras teh assembler does not. So the high level language code may be looking for _os_atomic_lock. To accommodate this possibility I usually double define symbols thus: .extern os_atomic_lock .extern _os_atomic_lock ... os_atomic_lock: _os_atomic_lock: li r0, 1 Take a look at the extern symbols in the code that references this to see if it has silently prefixed the symbols. Also if you are using C++ you would have to specify that the symbol uses C linkage to avoid name mangling. If you still have problems you really should post teh build log. The actual output from teh build will be far more useful that your description. Clifford
Clifford Slocombe wrote: > Nano2 Nano wrote: >> Hi >> I have he follow piece of assembler code powerpc.s running on powerpc6 >> redhat4.5 >> with GCC4.1.1 > I wonder then why you are asking this question on an ARM development > forum? > > Nano2 Nano wrote: >> >> when i do a nm on the the object the symbol is undefined >> Can you see whats wrong with the code i have below that the symbols >> os_atomic_lock >> os_atomic_unlock, etc are showing up as undefined ... > I am not sure but some versions of GCC add a _ prefix to external > symbols, wheras teh assembler does not. So the high level language code > may be looking for _os_atomic_lock. To accommodate this possibility I > usually double define symbols thus: > > .extern os_atomic_lock > .extern _os_atomic_lock > > ... Clifford , I had the symbols defined as global like so .globl .os_atomic_add .globl .os_atomic_lock .globl .os_atomic_unlock .globl .os_atomic_set .globl .os_atomic_get .globl .os_atomic_init .os_atomic_add: But still got the undefined symbols so i declared them as extern .... The version of GCC is 4.1.1 Any ideas why this was so with defining them .globl ? Thanks help much appreciated .. > os_atomic_lock: > _os_atomic_lock: > li r0, 1 > > Take a look at the extern symbols in the code that references this to > see if it has silently prefixed the symbols. Also if you are using C++ > you would have to specify that the symbol uses C linkage to avoid name > mangling. > > If you still have problems you really should post teh build log. The > actual output from teh build will be far more useful that your > description. > > Clifford
Its going to be far easier if you post the actual build log. All that information is output to aid fault finding so you need to provide that benefit to us too.
Nano2 Nano wrote: > Clifford Slocombe wrote: >> Nano2 Nano wrote: >>> Hi >>> I have he follow piece of assembler code powerpc.s running on powerpc6 >>> redhat4.5 >>> with GCC4.1.1 >> I wonder then why you are asking this question on an ARM development >> forum? >> >> Nano2 Nano wrote: >>> >>> when i do a nm on the the object the symbol is undefined >>> Can you see whats wrong with the code i have below that the symbols >>> os_atomic_lock >>> os_atomic_unlock, etc are showing up as undefined ... >> I am not sure but some versions of GCC add a _ prefix to external >> symbols, wheras teh assembler does not. So the high level language code >> may be looking for _os_atomic_lock. To accommodate this possibility I >> usually double define symbols thus: >> >> .extern os_atomic_lock >> .extern _os_atomic_lock >> >> ... > Clifford , > I had the symbols defined as global like so > > .globl .os_atomic_add > .globl .os_atomic_lock > .globl .os_atomic_unlock > .globl .os_atomic_set > .globl .os_atomic_get > .globl .os_atomic_init > > .os_atomic_add: > > But still got the undefined symbols so i declared them as extern .... > The version of GCC is 4.1.1 > > Any ideas why this was so with defining them .globl ? > > Thanks help much appreciated .. > >> os_atomic_lock: >> _os_atomic_lock: >> li r0, 1 >> >> Take a look at the extern symbols in the code that references this to >> see if it has silently prefixed the symbols. Also if you are using C++ >> you would have to specify that the symbol uses C linkage to avoid name >> mangling. >> >> If you still have problems you really should post teh build log. The >> actual output from teh build will be far more useful that your >> description. >> >> Clifford /home/t/cvs/4.3.7B1023/src/dsd/obj.ppc64/libdsd.so: undefined reference to `os_atomic_get' Clifford , This is what i have g++ -L/home/t/cvs/F4.3.7B1023/src/lib/obj.ppc64 -L/usr/X11R6/lib64 -L /home/t/cvs/F4.3.7B1023/src/lib/obj.ppc64 -L /home/t/cvs/F4.3.7B1023/src/dsd/obj.ppc64 -L /usr/X11R6/lib64 -m64 -L /usr/local64/lib -L /usr/local64/lib -L /usr/local64/lib -O -fPIC -shared -o libon.so powerpc.o svr4.o subr.o cllist.o md5.o buffer.o ftp.o network.o poller.o tp.o gettime.o base64.o oosubr.o rfile.o qsort.o -lnsl -ldl -lpthread -lCosNaming -lOB -lJTC powerpc.o is the assembler code above and builds fine when i do the nm on the new shared library nm libon.so |grep os_atomic 0000000000000000 n _os_atomic_add 0000000000000000 n os_atomic_add U os_atomic_add 0000000000000050 n _os_atomic_get 0000000000000050 n os_atomic_get U os_atomic_get 000000000000003c n _os_atomic_init 000000000000003c n os_atomic_init U os_atomic_init 0000000000000018 n _os_atomic_lock 0000000000000018 n os_atomic_lock 0000000000000048 n _os_atomic_set 0000000000000048 n os_atomic_set U os_atomic_set 0000000000000038 n _os_atomic_unlock 0000000000000038 n os_atomic_unlock I would expect the symbold to be declared different.. ======================================================================== ================= So when I link the new shared library in with the following c++ -L/home/t/cvs/F4.3.7B1023/src/lib/obj.ppc64 -L/usr/X11R6/lib64 -L /home/t/cvs/F4.3.7B1023/src/lib/obj.ppc64 -L /home/t/cvs/F4.3.7B1023/src/dsd/obj.ppc64 -L /usr/X11R6/lib64 -m64 -L /usr/local64/lib -L /usr/local64/lib -L /usr/local64/lib -o enc enc.o -ldsd -lnsl -ldl -lpthread -lCosNaming -lOB -lJTC -lon /home/t/cvs/F4.3.7B1023/src/dsd/obj.ppc64/libdsd.so: undefined reference to `os_atomic_lock' /home/t/cvs/F4.3.7B1023/src/dsd/obj.ppc64/libdsd.so: undefined reference to `os_atomic_add' /home/t/cvs/FW_4.3.7B1023/src/dsd/obj.ppc64/libdsd.so: undefined reference to `os_atomic_get' /home/t/cvs/FW_4.3.7B1023/src/dsd/obj.ppc64/libdsd.so: undefined reference to `os_atomic_init' /home/t/cvs/FW_4.3.7B1023/src/dsd/obj.ppc64/libdsd.so: undefined reference to `os_read_tick' /home/t/cvs/FW_4.3.7B1023/src/dsd/obj.ppc64/libdsd.so: undefined reference to `os_atomic_set' /home/t/cvs/FW_4.3.7B1023/src/dsd/obj.ppc64/libdsd.so: undefined reference to `os_atomic_unlock' collect2: ld returned 1 exit status Any ideas .... Thanks
You don't have to re-quote the entire thread every time you know!
U means undefined, but you also have entries marked with 'n'. I am not
sure what 'n' means, it is not listed in the nm documentation. However
these 'n' symbols clearly are defined because they have addresses.
Moreover the addresses are at what look like likely locations.
The log shows that you are using C++ compilation. Are you accessing
these symbols from C++ code? If so what do the declarations look like?
They need to have extern "C" linkage; for example:
extern "C" void os_atomic_add( void ) ;
or
extern "C"
{
void os_atomic_add( void ) ;
void os_atomic_lock( void ) ;
...
}
It seems to me that you are merely showing the result of your
investigation to some problem that you are not revealing. It would be
far more productive to explain what the problem is, since I suspect that
your investigation is going in the wrong direction and you are turning
up "red herrings". We need to know rather where you are coming from, not
where you have arrived at.
I see you are blissfully ignoring the fact that this is not a PowerPC
development forum!
Clifford
Watch this topic
Disable multi-page viewPlease log in before posting. Registration is free and takes only a minute.
Existing account
Do you have a Google/GoogleMail account? No registration required!
Log in with Google account
Log in with Google account
No account? Register here.