drm/i915: Release shortlived maps of longlived objects

Some objects we map once during their construction, and then never
access their mappings again, even if they are kept around for the
duration of the driver. Keeping those pages mapped, often vmapped, is
therefore wasteful and we should release the maps as soon as we no
longer need them.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200708173748.32734-3-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gem/i915_gem_object.h

@@ -394,6 +394,8 @@ static inline void i915_gem_object_unpin_map(struct drm_i915_gem_object *obj)
 	i915_gem_object_unpin_pages(obj);
 }
 
+void __i915_gem_object_release_map(struct drm_i915_gem_object *obj);
+
 void
 i915_gem_object_flush_write_domain(struct drm_i915_gem_object *obj,
 				   unsigned int flush_domains);

drivers/gpu/drm/i915/gem/i915_gem_pages.c

@@ -408,6 +408,21 @@ void __i915_gem_object_flush_map(struct drm_i915_gem_object *obj,
 	}
 }
 
+void __i915_gem_object_release_map(struct drm_i915_gem_object *obj)
+{
+	GEM_BUG_ON(!obj->mm.mapping);
+
+	/*
+	 * We allow removing the mapping from underneath pinned pages!
+	 *
+	 * Furthermore, since this is an unsafe operation reserved only
+	 * for construction time manipulation, we ignore locking prudence.
+	 */
+	unmap_object(obj, page_mask_bits(fetch_and_zero(&obj->mm.mapping)));
+
+	i915_gem_object_unpin_map(obj);
+}
+
 struct scatterlist *
 i915_gem_object_get_sg(struct drm_i915_gem_object *obj,
 		       unsigned int n,

drivers/gpu/drm/i915/gt/gen7_renderclear.c

@@ -396,7 +396,7 @@ int gen7_setup_clear_gpr_bb(struct intel_engine_cs * const engine,
 	emit_batch(vma, memset(batch, 0, bv.max_size), &bv);
 
 	i915_gem_object_flush_map(vma->obj);
-	i915_gem_object_unpin_map(vma->obj);
+	__i915_gem_object_release_map(vma->obj);
 
 	return 0;
 }

drivers/gpu/drm/i915/gt/intel_lrc.c

@@ -3937,7 +3937,7 @@ static int intel_init_workaround_bb(struct intel_engine_cs *engine)
 	GEM_BUG_ON(batch_ptr - batch > CTX_WA_BB_OBJ_SIZE);
 
 	__i915_gem_object_flush_map(wa_ctx->vma->obj, 0, batch_ptr - batch);
-	i915_gem_object_unpin_map(wa_ctx->vma->obj);
+	__i915_gem_object_release_map(wa_ctx->vma->obj);
 	if (ret)
 		lrc_destroy_wa_ctx(engine);
 

drivers/gpu/drm/i915/gt/intel_renderstate.c

@@ -150,7 +150,7 @@ static int render_state_setup(struct intel_renderstate *so,
 	ret = 0;
 out:
 	__i915_gem_object_flush_map(so->vma->obj, 0, i * sizeof(u32));
-	i915_gem_object_unpin_map(so->vma->obj);
+	__i915_gem_object_release_map(so->vma->obj);
 	return ret;
 }
 

drivers/gpu/drm/i915/gt/intel_ring_submission.c

@@ -543,7 +543,7 @@ alloc_context_vma(struct intel_engine_cs *engine)
 			   vaddr, engine->context_size);
 
 		i915_gem_object_flush_map(obj);
-		i915_gem_object_unpin_map(obj);
+		__i915_gem_object_release_map(obj);
 	}
 
 	vma = i915_vma_instance(obj, &engine->gt->ggtt->vm, NULL);

drivers/gpu/drm/i915/i915_perf.c

@@ -1772,7 +1772,7 @@ static int alloc_noa_wait(struct i915_perf_stream *stream)
 	GEM_BUG_ON(cs - batch > PAGE_SIZE / sizeof(*batch));
 
 	i915_gem_object_flush_map(bo);
-	i915_gem_object_unpin_map(bo);
+	__i915_gem_object_release_map(bo);
 
 	stream->noa_wait = vma;
 	return 0;
@@ -1867,7 +1867,7 @@ alloc_oa_config_buffer(struct i915_perf_stream *stream,
 	*cs++ = 0;
 
 	i915_gem_object_flush_map(obj);
-	i915_gem_object_unpin_map(obj);
+	__i915_gem_object_release_map(obj);
 
 	oa_bo->vma = i915_vma_instance(obj,
 				       &stream->engine->gt->ggtt->vm,