The ioremap for stolen RAM is about 8MB - we do actually need that mapped for the console framebuffer. The GTT tables are pretty small (64K or so) and the rest of the GTT space if ever used doesn't get an ioremap.
It's a bit different to the i915 world because the CPU cannot indirect via the GTT but must go directly.
Afaik this comment is bogus. And the previous one sounds like it's no clear whether this holds (it does, at least for the i915.ko driver). Maybe kill them both?
Possibly.
+static void psb_pages_clflush(struct psb_mmu_driver *driver,
struct page *page[], unsigned long num_pages)+{
- int i;
- if (!driver->has_clflush)
return ;- for (i = 0; i < num_pages; i++)
psb_page_clflush(driver, *page++);+}
drm_clflush_pages does that for you.
Ok. Handy to know.
I my next patch series for i915 drm_cache.c will grow a drm_clflush_virt_range, which would perfectly fit here. I'll try not to forget gma500 ...
Thanks