In posix_fadvise(2), under "Architecture-specific variants", it says: [Due to register-alignment issues,] ... the call signature of posix_fadvise() shown in the SYNOPSIS would force a register to be wasted as padding between the fd and len arguments. Should it say "offset" instead of "len" in the last line of that quoted bit? Rationale: the SYNOPSIS shows a call signature of: int posix_fadvise(int fd, off_t offset, off_t len, int advice); in which: - fd and len aren't adjacent - it's between a shorter value followed by a longer one that I'd expect alignment padding to be needed, and the only pair where that occurs is (fd, offset) CAVEAT: I'm not familiar with either the syscall or the specific alignment issues in question, so my guess about this is just that -- a deduced guess, NOT specific knowledge. So don't make the change on my say-so; please verify it first!
(In reply to Eric Siegerman from comment #0) > In posix_fadvise(2), under "Architecture-specific variants", it says: > [Due to register-alignment issues,] > ... the call signature of posix_fadvise() shown in > the SYNOPSIS would force a register to be wasted as padding between > the fd and len arguments. > > Should it say "offset" instead of "len" in the last line of that quoted bit? > > Rationale: the SYNOPSIS shows a call signature of: > int posix_fadvise(int fd, off_t offset, off_t len, int advice); > in which: > - fd and len aren't adjacent > - it's between a shorter value followed by a longer one that I'd expect > alignment padding to be needed, and the only pair where that occurs is (fd, > offset) The change sounds right to me. I've made it. > CAVEAT: I'm not familiar with either the syscall or the specific alignment > issues in question, so my guess about this is just that -- a deduced guess, > NOT specific knowledge. So don't make the change on my say-so; please > verify it first! I'm going to trust you on this one ;-). I was the one who injected the error in the first place.