Re: [PATCH v2] netlink: Replace rhash_portid with bound

Linus Torvalds [off-list ref] writes:

The PPRO_FENCE_BUG thing is rather questionable. I'm not sure it's
even documented, but I can't find the official PPro errata now. I
think I discussed it with Andy Glew back when Intel was starting to
document the memory ordering rules.

I'd be willing to get rid of it. It was apparently just an early
stepping or two.

Anybody with better google-fu than me who can find the official Intel
PPro errata?

http://download.intel.com/design/archives/processors/pro/docs/24268935.pdf

Says "NoFix" for erratas 66 and 92.


Bjørn

On Tue, Sep 22, 2015 at 1:36 PM, Bjørn Mork [off-list ref] wrote:

http://download.intel.com/design/archives/processors/pro/docs/24268935.pdf

Says "NoFix" for erratas 66 and 92.

Yeah, 66 and 92 do look like they could cause the apparent ordering of
accesses to be violated. That said, both of them <i>seem</i> to be
"processor had exclusive access to line A, and gave it away but ended
up still reading now-stale data".

And that's not what we use "smp_wmb()" or "smp_rmb()" to protect
against. If we did a write and then wanted to do an ordered read, we'd
use smp_mb(), which always does that barrier.

So I don't know whether either of those really merit our PPRO
workaround. Cache coherency is hard.

There's also errata 41, which looks like it would be a bad situation.

                 Linus