[Intel-gfx] [PATCH] drm/i915: Handle unmappable buffers during error state capture

[Chris Wilson]

As the buffer is not necessarily accessible through the GTT at the time
of a GPU hang, and capturing some of its contents is far more valuable
than skipping it, provide a clflushed fallback read path. We still
prefer to read through the GTT as that is more consistent with the GPU
access of the same buffer. So example it will demonstrate any errorneous
tiling or swizzling of the command buffer as seen by the GPU.

This becomes necessary with use of CPU relocations and lazy GTT binding,
but could potentially happen anyway as a result of a pathological error.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
---

We continued discussing how best to handle this in the light of the
desirability of unmappable command buffers and decided that a CPU
fallback path was necessary to maintain the utility of the
i915_error_state and a prerequisite for LLC relocations.
-Chris

---
 drivers/gpu/drm/i915/i915_irq.c |   28 +++++++++++++++++++++++-----
 1 files changed, 23 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index b40004b..08877a7 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -720,7 +720,6 @@ i915_error_object_create(struct drm_i915_private *dev_priv,
 	reloc_offset = src->gtt_offset;
 	for (page = 0; page < page_count; page++) {
 		unsigned long flags;
-		void __iomem *s;
 		void *d;
 
 		d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
@@ -728,10 +727,29 @@ i915_error_object_create(struct drm_i915_private *dev_priv,
 			goto unwind;
 
 		local_irq_save(flags);
-		s = io_mapping_map_atomic_wc(dev_priv->mm.gtt_mapping,
-					     reloc_offset);
-		memcpy_fromio(d, s, PAGE_SIZE);
-		io_mapping_unmap_atomic(s);
+		if (reloc_offset < dev_priv->mm.gtt_mappable_end) {
+			void __iomem *s;
+
+			/* Simply ignore tiling or any overlapping fence.
+			 * It's part of the error state, and this hopefully
+			 * captures what the GPU read.
+			 */
+
+			s = io_mapping_map_atomic_wc(dev_priv->mm.gtt_mapping,
+						     reloc_offset);
+			memcpy_fromio(d, s, PAGE_SIZE);
+			io_mapping_unmap_atomic(s);
+		} else {
+			void *s;
+
+			drm_clflush_pages(&src->pages[page], 1);
+
+			s = kmap_atomic(src->pages[page]);
+			memcpy(d, s, PAGE_SIZE);
+			kunmap_atomic(s);
+
+			drm_clflush_pages(&src->pages[page], 1);
+		}
 		local_irq_restore(flags);
 
 		dst->pages[page] = d;
-- 
1.7.7.3