Re: ioremap_uc() followed by set_memory_wc() - burrying MTRR

[Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 2938 bytes --]

On Wed, 2015-04-22 at 10:17 -0600, Jason Gunthorpe wrote:
> On Wed, Apr 22, 2015 at 05:23:28PM +0200, Luis R. Rodriguez wrote:
> > On Tue, Apr 21, 2015 at 11:39:39PM -0600, Jason Gunthorpe wrote:
> > > On Wed, Apr 22, 2015 at 01:39:07AM +0200, Luis R. Rodriguez wrote:
> > > > > Mike, do you think the time is right to just remove the iPath driver?
> > > > 
> > > > With PAT now being default the driver effectively won't work
> > > > with write-combining on modern kernels. Even if systems are old
> > > > they likely had PAT support, when upgrading kernels PAT will work
> > > > but write-combing won't on ipath.
> > > 
> > > Sorry, do you mean the driver already doesn't get WC? Or do you mean
> > > after some more pending patches are applied?
> > 
> > No, you have to consider the system used and the effects of calls used
> > on the driver in light of this table:
> 
> So, just to be clear:
> 
> At some point Linux started setting the PAT bits during
> ioremap_nocache, which overrides MTRR, and at that point the driver
> became broken on all PAT capable systems?
> 
> Not only that, but we've only just noticed it now, and no user ever
> complained?
> 
> So that means either no users exist, or all users are on non-PAT
> systems?
> 
> This driver only works on x86-64 systems. Are there any x86-64 systems
> that are not PAT capable? IIRC even the first Opteron had PAT, but my
> memory is fuzzy from back then :|
> 
> > Another option in order to enable this type of checks at run time
> > and still be able to build the driver on standard distributions and
> > just prevent if from loading on PAT systems is to have some code in
> > place which would prevent the driver from loading if PAT was
> > enabled, this would enable folks to disable PAT via a kernel command
> > line option, and if that was used then the driver probe would
> > complete.
> 
> This seems like a reasonble option to me. At the very least we might
> learn if anyone is still using these cards.
> 
> I'd also love to remove the driver if it turns out there are actually
> no users. qib substantially replaces it except for a few very old
> cards.

To be precise, the split is that ipath powers the old HTX bus cards that
only work in AMD systems, qib is all PCI-e cards.  I still have a few
HTX cards, but I no longer have any systems with HTX slots, so we
haven't even used this driver in testing for 3 or 4 years now.  And
these are all old SDR cards, where the performance numbers were 800MB/s
with WC enabled, 50MB/s without it.

> Mike?
> 
> Jason
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
              GPG KeyID: 0E572FDD



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]