Re: [RFC][PATCH v3 2/6] mm, directio: fix fork vs direct-io race (read(2) side IOW gup(write) side)

[Andrea Arcangeli <aarcange@redhat.com>]

On Tue, Apr 14, 2009 at 12:45:41PM -0400, Jeff Moyer wrote:
> So, if you're continuously submitting async read I/O, you will starve
> out the fork() call indefinitely.  I agree that you want to allow

IIRC rwsem good enough to stop the down_read when a down_write is
blocked. Otherwise page fault flood in threads would also starve any
mmap or similar call. Still with this approach fork will start to hang
indefinitely waiting for I/O, making it an I/O bound call, and not a
CPU call anymore, which may severely impact interactive-ness of
applications.

As long as fork is useful in the first place to provide memory
protection of different code with different
memory-corruption-trust-levels (otherwise nobody should use fork at
all, and vfork [or better spawn] should become the only option), then
fork from a thread pool is also reasonable. Either fork is totally
useless as a whole (which I wouldn't argue too much about), or if you
agree fork makes any sense, it can also make sense if intermixed with
clone(CLONE_VM) and hopefully it should behave CPU bound like CLONE_VM.