drm/i915/selftest: test aligned offsets for 64K
authorMatthew Auld <matthew.auld@intel.com>
Mon, 29 Oct 2018 20:37:34 +0000 (20:37 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Wed, 31 Oct 2018 08:48:46 +0000 (08:48 +0000)
When using softpin it's not enough to just pad the vma size, we also
need to ensure the vma offset is at the start of the pt boundary, if we
plan to utilize 64K pages. Therefore to improve test coverage we should
use both aligned and unaligned gtt offsets in igt_write_huge.

Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20181029203734.21936-1-matthew.auld@intel.com
drivers/gpu/drm/i915/selftests/huge_pages.c

index 256001b..26c065c 100644 (file)
@@ -1135,7 +1135,8 @@ static int igt_write_huge(struct i915_gem_context *ctx,
        n = 0;
        for_each_engine(engine, i915, id) {
                if (!intel_engine_can_store_dword(engine)) {
-                       pr_info("store-dword-imm not supported on engine=%u\n", id);
+                       pr_info("store-dword-imm not supported on engine=%u\n",
+                               id);
                        continue;
                }
                engines[n++] = engine;
@@ -1167,17 +1168,30 @@ static int igt_write_huge(struct i915_gem_context *ctx,
                engine = engines[order[i] % n];
                i = (i + 1) % (n * I915_NUM_ENGINES);
 
-               err = __igt_write_huge(ctx, engine, obj, size, offset_low, dword, num + 1);
+               /*
+                * In order to utilize 64K pages we need to both pad the vma
+                * size and ensure the vma offset is at the start of the pt
+                * boundary, however to improve coverage we opt for testing both
+                * aligned and unaligned offsets.
+                */
+               if (obj->mm.page_sizes.sg & I915_GTT_PAGE_SIZE_64K)
+                       offset_low = round_down(offset_low,
+                                               I915_GTT_PAGE_SIZE_2M);
+
+               err = __igt_write_huge(ctx, engine, obj, size, offset_low,
+                                      dword, num + 1);
                if (err)
                        break;
 
-               err = __igt_write_huge(ctx, engine, obj, size, offset_high, dword, num + 1);
+               err = __igt_write_huge(ctx, engine, obj, size, offset_high,
+                                      dword, num + 1);
                if (err)
                        break;
 
                if (igt_timeout(end_time,
                                "%s timed out on engine=%u, offset_low=%llx offset_high=%llx, max_page_size=%x\n",
-                               __func__, engine->id, offset_low, offset_high, max_page_size))
+                               __func__, engine->id, offset_low, offset_high,
+                               max_page_size))
                        break;
        }