On Thursday, December 16, 2010 03:03:21 pm Linus Torvalds wrote:
> On Thu, Dec 16, 2010 at 9:38 AM, Bjorn Helgaas <bjorn.helgaas@xxxxxx> wrote:
> >
> > This adds arch_remove_reservations(), which an arch can implement if it
> > needs to protect part of the address space from allocation.
>
> Oh, I just realized - we've had gcc (or maybe "as") bugs when the weak
> symbol is in the same compilation unit as the caller, and some part of
> the toolchain just ends up short-circuiting it (either gcc inlines it,
> or maybe it was that as resolves it early). Making the "weak" part not
> work, because it binds strongly.
>
> It may be that we don't support those gcc versions any more, but I
> thought I'd bring the issue up. It's safer to put the weak functions
> in some other file if possible.
I'd be glad to repost and do that if necessary. Does anybody
know for sure if it is?
I did find these cases where we have a weak symbol definition in
the same file as the caller, as well as another version elsewhere,
so maybe those toolchains are indeed no longer supported?
arch_disable_smp_support()
defined and called in init/main.c (used for "nosmp" arg)
also defined in arch/x86/kernel/apic/io_apic.c
smp_setup_processor_id()
defined and called in init/main.c
also defined in arch/s390/kernel/smp.c
also defined in arch/sparc/kernel/smp_64.c
arch_enable_nonboot_cpus_begin()
defined and called in kernel/cpu.c
also defined in arch/x86/kernel/smpboot.c
arch_irq_work_raise()
defined and called in kernel/irq_work.c
also defined in arch/sparc/kernel/pcr.c
also defined in arch/x86/kernel/irq_work.c
arch_jump_label_text_poke_early()
defined and called in kernel/jump_label.c
also defined in arch/sparc/kernel/jump_label.c
also defined in arch/x86/kernel/jump_label.c
arch_deref_entry_point()
defined and called in kernel/kprobes.c
also defined in arch/ia64/kernel/kprobes.c
also defined in arch/powerpc/kernel/kprobes.c
arch_scale_smt_power()
defined and called in kernel/sched_fair.c
also defined in arch/x86/kernel/cpu/sched.c
memblock_nid_range()
defined and called in mm/memblock.c
also defined in arch/sparc/mm/init_64.c
If this toolchain bug is still a problem, I would think all of the
above would be defects, so my guess is that we don't have to worry
about it any more.
Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
