Add comments about syscall wrapper problems.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@484 a5019735-40e9-0310-863c-91ae7b9d1cf9
diff --git a/README_PACKAGERS b/README_PACKAGERS
index a739dee..c5301b9 100644
--- a/README_PACKAGERS
+++ b/README_PACKAGERS
@@ -54,6 +54,28 @@
from valgrind.
+-- Try and ensure that the /usr/include/asm/unistd.h file on the
+ build machine contains an entry for all the system calls that
+ the kernels on the target machines can actually support. On my
+ Red Hat 7.2 (kernel 2.4.9) box the highest-numbered entry is
+ #define __NR_fcntl64 221
+ but I have heard of 2.2 boxes where it stops at 179 or so.
+
+ Reason for this is that at build time, support for syscalls
+ is compiled in -- or not -- depending on which of these __NR_*
+ symbols is defined. Problems arise when /usr/include/asm/unistd.h
+ fails to give an entry for a system call which is actually
+ available in the target kernel. In that case, valgrind will
+ abort if asked to handle such a syscall. This is despite the
+ fact that (usually) valgrind's sources actually contain the
+ code to deal with the syscall.
+
+ Several people have reported having this problem. So, please
+ be aware of it. If it's useful, the syscall wrappers are
+ all done in vg_syscall_mem.c; you might want to have a little
+ look in there.
+
+
-- Please test the final installation works by running it on
something huge. I suggest checking that it can start and
exit successfully both Mozilla-1.0 and OpenOffice.org 1.0.