drm/i915: Apply batch location restrictions before pinning

We special case the position of the batch within the GTT to prevent
negative self-relocation deltas from underflowing. However, that
restriction is being applied after a trial pin of the batch in its
current position. Thus we are not rejecting an invalid location if the
batch has been used before, leading to an assertion if we happen to need
to rearrange the entire payload. In the worst case, this may cause a GPU
hang on gen7 or perhaps missing state.

References: https://bugs.freedesktop.org/show_bug.cgi?id=105720
Fixes: 2889caa92321 ("drm/i915: Eliminate lots of iterations over the execobjects array")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Martin Peres <martin.peres@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180610194325.13467-2-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
(cherry picked from commit 746c8f143afad7aaa66c484485fc39888d437a3f)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>

1 files changed, 27 insertions, 22 deletions

diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index f627a8c47c58..22df17c8ca9b 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -489,7 +489,9 @@ eb_validate_vma(struct i915_execbuffer *eb,
 }
 
 static int
-eb_add_vma(struct i915_execbuffer *eb, unsigned int i, struct i915_vma *vma)
+eb_add_vma(struct i915_execbuffer *eb,
+	   unsigned int i, unsigned batch_idx,
+	   struct i915_vma *vma)
 {
 	struct drm_i915_gem_exec_object2 *entry = &eb->exec[i];
 	int err;
@@ -522,6 +524,24 @@ eb_add_vma(struct i915_execbuffer *eb, unsigned int i, struct i915_vma *vma)
 	eb->flags[i] = entry->flags;
 	vma->exec_flags = &eb->flags[i];
 
+	/*
+	 * SNA is doing fancy tricks with compressing batch buffers, which leads
+	 * to negative relocation deltas. Usually that works out ok since the
+	 * relocate address is still positive, except when the batch is placed
+	 * very low in the GTT. Ensure this doesn't happen.
+	 *
+	 * Note that actual hangs have only been observed on gen7, but for
+	 * paranoia do it everywhere.
+	 */
+	if (i == batch_idx) {
+		if (!(eb->flags[i] & EXEC_OBJECT_PINNED))
+			eb->flags[i] |= __EXEC_OBJECT_NEEDS_BIAS;
+		if (eb->reloc_cache.has_fence)
+			eb->flags[i] |= EXEC_OBJECT_NEEDS_FENCE;
+
+		eb->batch = vma;
+	}
+
 	err = 0;
 	if (eb_pin_vma(eb, entry, vma)) {
 		if (entry->offset != vma->node.start) {
@@ -716,7 +736,7 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 {
 	struct radix_tree_root *handles_vma = &eb->ctx->handles_vma;
 	struct drm_i915_gem_object *obj;
-	unsigned int i;
+	unsigned int i, batch;
 	int err;
 
 	if (unlikely(i915_gem_context_is_closed(eb->ctx)))
@@ -728,6 +748,8 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 	INIT_LIST_HEAD(&eb->relocs);
 	INIT_LIST_HEAD(&eb->unbound);
 
+	batch = eb_batch_index(eb);
+
 	for (i = 0; i < eb->buffer_count; i++) {
 		u32 handle = eb->exec[i].handle;
 		struct i915_lut_handle *lut;
@@ -770,33 +792,16 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 		lut->handle = handle;
 
 add_vma:
-		err = eb_add_vma(eb, i, vma);
+		err = eb_add_vma(eb, i, batch, vma);
 		if (unlikely(err))
 			goto err_vma;
 
 		GEM_BUG_ON(vma != eb->vma[i]);
 		GEM_BUG_ON(vma->exec_flags != &eb->flags[i]);
+		GEM_BUG_ON(drm_mm_node_allocated(&vma->node) &&
+			   eb_vma_misplaced(&eb->exec[i], vma, eb->flags[i]));
 	}
 
-	/* take note of the batch buffer before we might reorder the lists */
-	i = eb_batch_index(eb);
-	eb->batch = eb->vma[i];
-	GEM_BUG_ON(eb->batch->exec_flags != &eb->flags[i]);
-
-	/*
-	 * SNA is doing fancy tricks with compressing batch buffers, which leads
-	 * to negative relocation deltas. Usually that works out ok since the
-	 * relocate address is still positive, except when the batch is placed
-	 * very low in the GTT. Ensure this doesn't happen.
-	 *
-	 * Note that actual hangs have only been observed on gen7, but for
-	 * paranoia do it everywhere.
-	 */
-	if (!(eb->flags[i] & EXEC_OBJECT_PINNED))
-		eb->flags[i] |= __EXEC_OBJECT_NEEDS_BIAS;
-	if (eb->reloc_cache.has_fence)
-		eb->flags[i] |= EXEC_OBJECT_NEEDS_FENCE;
-
 	eb->args->flags |= __EXEC_VALIDATED;
 	return eb_reserve(eb);