Juan Quintela wrote:
> >>>>> "hartvig" == Hartvig Ekner <email@example.com> writes:
> hartvig> I can't see that they are using wbflush in any way. Grepping
> hartvig> after wbflush through the entire 2.4 tree, it seems wbflush
> hartvig> is something only present on some dec platforms and then the
> hartvig> au1000 stuff - which would mean that any driver directly
> hartvig> calling __wbflush would be unable to compile/load on the
> hartvig> majority of kernels. Or am I missing something? (I haven't
> hartvig> been using modules under MIPS at all).
> Yes, you missed the definition of mb() :p
> quintela$ grep "mb(" drivers/net/8139* | grep -v rmb | grep -v wmb
> drivers/net/8139too.c: mb();
> hartvig> In fact, I can't find a single file including wbflush.h
> hartvig> except system.h, and it doesn't look like anybody else should
> hartvig> directly be including the wbflush.h file, but only use the
> hartvig> macros in system.h:
> hartvig> #define wmb() fast_wmb()
> hartvig> #define rmb() fast_rmb()
> hartvig> #define mb() wbflush();
> hartvig> #define iob() wbflush();
> hartvig> (which are differently defined if there is no WB configured).
> but WB is configured in :)
That's the problem! Wb does not need to be configured in, it is a mistake. My
patch was missing
the required fixes to the defconfig files, it was only described in the mail.
So I think the only
thing missing is a patch to fix all the pb/db defconfig files to look like this:
# CONFIG_CPU_ADVANCED is not set
# CONFIG_CPU_HAS_LLDSCD is not set
# CONFIG_CPU_HAS_WB is not set
... which is the setting also used by all other MIPS32 CPUs. Then everything
should be ok.
> quintela$ grep WB arch/mips/defconfig-pb1*
> Other thing is that this machine should be using wbflush at all, but
> that is a different story. I agree with (/me looks in archive),
> *your* patch removing wbflush for that boards. They are only doing
> "sync" ond wbflush, and that is the thing that __sync() already does.
> I.e. If I have to chooses wich patch to integrate, the one exporting
> __wbflush(), or the one removing it altogether from pb1*, I will
> choose removing it. It looks superflous.
> Later, Juan "who don't have that processor, and handwaving is easy".
> In theory, practice and theory are the same, but in practice they
> are different -- Larry McVoy