On Fri, 2010-06-18 at 15:30 +0200, Arnd Bergmann wrote: > > I have been using the asm-generic/system.h header to implement my > > version of arch/xx/include/asm/system.h: it appears that the version of > > cmpxchg defined in this generic header does not handle correctly > > non-long arguments on an LP64 arch. > > That looks like a correct observation. It's also true on ILP32 > architectures. > > I am not sure though if it is actually supposed to operate on other > types, I think there was some disagreement on this in the past. > What code specifically did you find needs to do cmpxchg on non-long > data? asm-generic/atomic.h defines atomic_add_unless to use atomic_cmpxchg which uses cmpxchg and all of that is supposed to work on struct atomic_t * which contains an int defined in include/linux/types.h atomic_add_unless is used through atomic_inc_not_zero to do some refcounting of sockets in the network stack so it did not take me long to notice that it appeared to not work. > > - > > -#define cmpxchg(ptr, o, n) \ > > - ((__typeof__(*(ptr))) __cmpxchg((unsigned long *)(ptr), \ > > - (unsigned long)(o), \ > > - (unsigned long)(n))) > > +#define cmpxchg(ptr, o, n) cmpxchg_local(ptr, o, n) > > This seems to match what we have in include/asm-generic/cmpxchg.h. > Maybe the best option would be to include that in asm-generic/system.h. This appears to work, indeed. > Nobody so far is using asm-generic/system.h, not even tile, which > uses most of the generic headers. This makes it quite likely that > the file is not correct right now. If you think you can improve it, > go wild. > > BTW, what architecture are you working on? Is this something you > plan to submit for inclusion soon? I am working on a version of the linux kernel network stack which runs in userspace as a shared library in a network simulator. My 'arch' port (named sim) provides glue code to replace all the code used by net/ in kernel/, lib/, crypto/, fs/ et al. The code is pretty rough and just started to run udp and netlink sockets yesterday but I could post it if others are interested. Mathieu -- Mathieu Lacage <mathieu.lacage@xxxxxxxxxxxxxxx> Tel: +33 4 9238 5056 -- To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html