On Mon, Sep 18, 2023 at 10:00:18AM +0900, Akira Yokosawa wrote:
> There remain obsolete word choices of "data dependencies" in section
> Alpha.
>
> It is load-to-load address dependencies that other archs respect.
>
> Signed-off-by: Akira Yokosawa <akiyks@xxxxxxxxx>
> ---
> Hi Paul,
>
> I caught them while attempting to add index markers to
> address/data/control dependency.
Good catch, queued and pushed, thank you!!!
Thanx, Paul
> Thanks, Akira
> --
> memorder/memorder.tex | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/memorder/memorder.tex b/memorder/memorder.tex
> index 1a4edf8bbb6d..86bb8f41b5f8 100644
> --- a/memorder/memorder.tex
> +++ b/memorder/memorder.tex
> @@ -5403,7 +5403,7 @@ One could place an \co{smp_rmb()} primitive
> between the pointer fetch and dereference in order to force Alpha
> to order the pointer fetch with the later dependent load.
> However, this imposes unneeded overhead on systems (such as \ARM,
> -Itanium, PPC, and SPARC) that respect data dependencies on the read side.
> +Itanium, PPC, and SPARC) that respect address dependencies on the read side.
> A \co{smp_read_barrier_depends()} primitive was therefore added to the
> Linux kernel to eliminate overhead on these systems, but was removed
> in v5.9 of the Linux kernel in favor of augmenting Alpha's definition
> @@ -5428,7 +5428,7 @@ Upon receipt of such an IPI, a CPU would execute a memory-barrier
> instruction, implementing a system-wide memory barrier similar to that
> provided by the Linux kernel's \co{sys_membarrier()} system call.
> Additional logic is required to avoid deadlocks.
> -Of course, CPUs that respect data dependencies would define such a barrier
> +Of course, CPUs that respect address dependencies would define such a barrier
> to simply be \co{smp_store_release()}.
> However, this approach was deemed by the Linux community
> to impose excessive overhead~\cite{McKenney01f}, and to their point would
>
> base-commit: 8224a12917dd115ead6fce0f785b19a4c7d50f18
> --
> 2.25.1
>
