On 8/6/2015 11:00 PM, Daniel Vetter wrote:
On Thu, Aug 6, 2015 at 10:17 PM, kbuild test robot
<fengguang.wu@xxxxxxxxx> wrote:
1070 if (IS_ENABLED(CONFIG_X86_32))
1071 /* While we have a proliferation of size_t variables
1072 * we cannot represent the full ppgtt size on 32bit,
1073 * so limit it to the same size as the GGTT (currently
1074 * 2GiB).
1075 */
1076 ppgtt->base.total = to_i915(ppgtt->base.dev)->gtt.base.total;
1077 ppgtt->base.cleanup = gen8_ppgtt_cleanup;
1078 ppgtt->base.allocate_va_range = gen8_alloc_va_range;
1079 ppgtt->base.insert_entries = gen8_ppgtt_insert_entries;
1080 ppgtt->base.clear_range = gen8_ppgtt_clear_range;
1081 ppgtt->base.unbind_vma = ppgtt_unbind_vma;
1082 ppgtt->base.bind_vma = ppgtt_bind_vma;
1083
1084 ppgtt->switch_mm = gen8_mm_switch;
1085
1086 ret = __pdp_init(false, &ppgtt->pdp);
So the first argument of pdp_init ist struct drm_device *dev and yes
the first thing it does is deref it.
*dev is used only for I915_PDPES_PER_PDP/USES_FULL_48BIT_PPGTT, which in
this path is always false. I didn't expect kbuild to complain. I'll
change it with the other modifications I'm about to send.
How exactly was this tested again?
Oh and the hunk right below with the CONFIG_X86_32 needs to got too -
if we're 48b safe then we should be 32b safe too ;-)
Yes, after the offset change it'll be completely safe. I left it in this
patch as that was only moving code around.
I'll include a proper "revert-me" patch, saying why it is safe now.
-Michel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
