Why do flush page cache twice when change TT's cache attribute

[Scott Fang]

yes, flush page twice cause system slow in some CPU.
Meanwhile, set_memory_uc/wc doesn't only do flush page cache, but also
change page table attribute, so your solution seems to have some problem.

2012/3/20 Konrad Rzeszutek Wilk <konrad.wilk at oracle.com>

> On Tue, Mar 20, 2012 at 10:15:02AM +0800, Scott Fang wrote:
> > Can I do the optimization like:
> >
> >     if (ttm->caching_state == tt_cached)
> >  -        drm_clflush_pages(ttm->pages, ttm->num_pages);
> > +        for (i = 0; i < ttm->num_pages; ++i)
> > +           if (PageHighMem(ttm->pages[i]))
> > +                drm_clflush_pages(&ttm->pages[i], 1);
> >
> > only do flush cache when high memory and leave the linear memory flush in
> > function set_memory_uc/wc?
>
> So what are you trying to solve? I mean one way to fix this
> is to do:
> >
> > 2012/3/20 Jerome Glisse <j.glisse at gmail.com>
> >
> > > On Mon, 2012-03-19 at 23:11 +0800, Scott Fang wrote:
> > > > In function ttm_tt_set_caching
> > > > ,,,,,,,
> > > >
> > > >     if (ttm->caching_state == tt_cached)
> > > >         drm_clflush_pages(ttm->pages, ttm->num_pages);
>               goto out;
> > > >
> > > >     for (i = 0; i < ttm->num_pages; ++i) {
> > > >         cur_page = ttm->pages[i];
> > > >         if (likely(cur_page != NULL)) {
> > > >             ret = ttm_tt_set_page_caching(cur_page,
> > > >                               ttm->caching_state,
> > > >                               c_state);
> > > >             if (unlikely(ret != 0))
> > > >                 goto out_err;
> > > >         }
> > > >     }
> > >
> out:
> > > >     ttm->caching_state = c_state;
> > > >
> > > >     return 0;
>
> Is the problem with calling page change twice making the machine slow?
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/dri-devel/attachments/20120321/9f991ca4/attachment.htm>