On Sun, Jul 22, 2018 at 02:18:32 +0300, Konstantin Belousov wrote: > On Sat, Jul 21, 2018 at 02:16:17AM +0000, jenkins-admin_at_FreeBSD.org wrote: > > --- lib/ofed/libmlx5__L --- > > In file included from /workspace/src/contrib/ofed/libmlx5/mlx5.h:44:0, > > from /workspace/src/contrib/ofed/libmlx5/buf.c:42: > > /workspace/obj/workspace/src/riscv.riscv64/tmp/usr/include/infiniband/udma_barrier.h:108:2: error: #error No architecture specific memory barrier defines found! > > #error No architecture specific memory barrier defines found! > > ^~~~~ > > The patch below should be enough to fix the build, modulo the syntax > errors which I cannot check. The reason is that it seems riscv is not > cross-buildable from stable/11 host: > --- includes_subdir_include/rpcsvc --- > RPCGEN_CPP=cpp\ -target\ riscv64-unknown-freebsd12.0\ --sysroot=/usr/home/konstantinb/build/bsd/DEV/obj/usr/home/konstantinb/build/bsd/DEV/src/riscv.riscv64/tmp\ -B/usr/local rpcgen -C -h -DWANT_NFS3 /usr/home/konstantinb/build/bsd/DEV/src/include/rpcsvc/klm_prot.x -o klm_prot.h > --- key_prot.h --- > error: unknown target triple 'riscv64-unknown-freebsd12.0', please use -triple or -arch > *** [key_prot.h] Error code 1 > > make[4]: stopped in /usr/home/konstantinb/build/bsd/DEV/src/include/rpcsvc > > So if somebody who care about riscv and can build the world could > confirm the fix, I will commit it. > > diff --git a/contrib/ofed/include/udma_barrier.h b/contrib/ofed/include/udma_barrier.h > index 71d264f8dce..6709d4433a7 100644 > --- a/contrib/ofed/include/udma_barrier.h > +++ b/contrib/ofed/include/udma_barrier.h > _at__at_ -104,6 +104,10 _at__at_ > #include <sys/types.h> > #include <machine/atomic.h> > #define udma_to_device_barrier() dmb() > +#elif defined(__riscv__) > +#include <sys/types.h> > +#include <machine/atomic.h> > +#define udma_to_device_barrier() fence() > #else > #error No architecture specific memory barrier defines found! > #endif Thanks for looking into this. Attached patch is slightly modified. RISC-V is using __riscv and there are few more atomic macros need to be defined. However this seems not enough, there are some c++ errors which may take longer to fix: In file included from /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/cma.h:43:0, from /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/acm.c:42: /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/cma.h: In function 'fastlock_init': /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/cma.h:60:2: error: invalid initializer atomic_store(&lock->cnt, 0); ^ In file included from /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/acm.c:42:0: /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/cma.h: In function 'fastlock_acquire': /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/cma.h:68:2: error: operand type 'struct <anonymous> *' is incompatible with argument 1 of '__atomic_fetch_add' if (atomic_fetch_add(&lock->cnt, 1) > 0) ^~ /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/cma.h: In function 'fastlock_release': /usr/home/lwhsu/freebsd-src/contrib/ofed/librdmacm/cma.h:73:2: error: operand type 'struct <anonymous> *' is incompatible with argument 1 of '__atomic_fetch_sub' if (atomic_fetch_sub(&lock->cnt, 1) > 1) ^~ *** [acm.o] Error code 1 make[6]: stopped in /usr/home/lwhsu/freebsd-src/lib/ofed/librdmacm BTW, stable/11 should be fine to cross-build riscv64, this job uses 11.2-RELEASE and riscv64-xtoolchain-gcc package, with CROSS_TOOLCHAIN=riscv64-gcc defined. Best, Li-Wen -- Li-Wen Hsu <lwhsu_at_FreeBSD.org> https://lwhsu.org
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:41:17 UTC