Re: [GIT PULL] device-dax for 5.1: PMEM as RAM

[Dan Williams <dan.j.williams@intel.com>]

[ add Tony, who has wrestled with how to detect rep; movs recover-ability ]

On Sun, Mar 10, 2019 at 1:02 PM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> On Sun, Mar 10, 2019 at 12:54 PM Dan Williams <dan.j.williams@intel.com> wrote:
> >
> > Hi Linus, please pull from:
> >
> >   git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm
> > tags/devdax-for-5.1
> >
> > ...to receive new device-dax infrastructure to allow persistent memory
> > and other "reserved" / performance differentiated memories, to be
> > assigned to the core-mm as "System RAM".
>
> I'm not pulling this until I get official Intel clarification on the
> whole "pmem vs rep movs vs machine check" behavior.
>
> Last I saw it was deadly and didn't work, and we have a whole "mc-safe
> memory copy" thing for it in the kernel because repeat string
> instructions didn't work correctly on nvmem.
>
> No way am I exposing any users to something like that.
>
> We need a way to know when it works and when it doesn't, and only do
> it when it's safe.

Unfortunately this particular b0rkage is not constrained to nvmem.
I.e. there's nothing specific about nvmem requiring mc-safe memory
copy, it's a cpu problem consuming any poison regardless of
source-media-type with "rep; movs".