Opened 12 years ago

Closed 12 years ago

#1285 closed bug (invalid)

missing symbol __ucmpdi2 in kernel_x86

Reported by: mmu_man Owned by: axeld
Priority: normal Milestone: R1
Component: System/Kernel Version: R1/pre-alpha1
Keywords: Cc:
Blocked By: Blocking:
Has a Patch: no Platform: All


When trying to load a driver build in zeta (not sure R5 has it though, but I suppose it did):

KERN: "/boot/home/config/add-ons/kernel/drivers/dev/oss": could not resolve symbol '__ucmpdi2'

Seems it's missing from the libc runtime linked to the kernel:

[revol@Zeta /Data/haiku/trunk]# nm /haiku/beos/system/kernel_x86 |grep cmpdi
800936e0 T __cmpdi2
[revol@Zeta /Data/haiku/trunk]# nm /boot/beos/system/kernel_intel |grep cmpdi
00076fc0 T __cmpdi2
00077000 T __ucmpdi2

Seems it's present in R5-bone as well, so...

Change History (6)

comment:1 Changed 12 years ago by mmu_man

it seems it is present in libgcc.a, at least in my gcc... I think libgcc.a should be linked in fully, with the --whole-archive ld arg, else it results in only needed symbols being linked in. Not sure how to do it with jam... probably have to change KernelLd rule.. or maybe the ldscript ? I suppose SEARCH_DIR(libgcc) tells to pick needed symbols in it, maybe changing it to INPUT() ?

comment:2 Changed 12 years ago by mmu_man

Hmm, didn't work. Changing KernelLd to do this:

LINKLIBS on $(1) = $(TARGET_STATIC_LIBSUPC++) --whole-archive $(TARGET_GCC_LIBGCC) ;

did the thing, but linking fails as it requires other symbols like fread, fwrite, which are useland stuff... I suppose we need to force the symbol in but not the whole archive.

comment:3 Changed 12 years ago by mmu_man

adding an


in the kernel ld script seems to do the job... not sure it's the best way though. comments ?

comment:4 Changed 12 years ago by bonefish

ATM, it's as easy as that: Kernel modules requiring libgcc features just link against libgcc directly.

comment:5 Changed 12 years ago by mmu_man

Except this module was compiled in BeOS. This means bin compatibility is broken.

comment:6 Changed 12 years ago by mmu_man

Resolution: invalid
Status: newclosed

Hmm ok, it seems more logical to link libgcc in, as it might change with gcc versions anyway. I suppose the exported symbol on R5 was probably just a side effect of something needing it in the kernel itself.

Note: See TracTickets for help on using tickets.