[PATCH 18/24] drm/i915/lmem: support CPU relocations

[Matthew Auld]

Add LMEM support for the CPU reloc path. When doing relocations we have
both a GPU and CPU reloc path, as well as some debugging options to force a
particular path. The GPU reloc path is preferred when the object
is not currently idle, otherwise we use the CPU reloc path. Since we
can't kmap the object, and the mappable aperture might not be available,
add support for mapping it through LMEMBAR.

Signed-off-by: Matthew Auld <matthew.auld at intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Cc: Abdiel Janulgue <abdiel.janulgue at linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi at intel.com>
---
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 55 +++++++++++++++++--
 1 file changed, 51 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index e96901888323..eafdf26d6862 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -15,6 +15,7 @@
 #include "display/intel_frontbuffer.h"
 
 #include "gem/i915_gem_ioctls.h"
+#include "gem/i915_gem_lmem.h"
 #include "gt/intel_context.h"
 #include "gt/intel_engine_pool.h"
 #include "gt/intel_gt.h"
@@ -251,6 +252,7 @@ struct i915_execbuffer {
 		bool has_llc : 1;
 		bool has_fence : 1;
 		bool needs_unfenced : 1;
+		bool is_lmem : 1;
 
 		struct intel_context *ce;
 		struct i915_request *rq;
@@ -909,6 +911,7 @@ static void reloc_cache_init(struct reloc_cache *cache,
 	cache->has_fence = cache->gen < 4;
 	cache->needs_unfenced = INTEL_INFO(i915)->unfenced_needs_alignment;
 	cache->node.flags = 0;
+	cache->is_lmem = false;
 	cache->ce = NULL;
 	cache->rq = NULL;
 	cache->rq_size = 0;
@@ -967,10 +970,14 @@ static void reloc_cache_reset(struct reloc_cache *cache)
 	} else {
 		struct i915_ggtt *ggtt = cache_to_ggtt(cache);
 
-		intel_gt_flush_ggtt_writes(ggtt->vm.gt);
+		if (!cache->is_lmem)
+			intel_gt_flush_ggtt_writes(ggtt->vm.gt);
 		io_mapping_unmap_atomic((void __iomem *)vaddr);
 
-		if (drm_mm_node_allocated(&cache->node)) {
+		if (cache->is_lmem) {
+			i915_gem_object_unpin_pages((struct drm_i915_gem_object *)cache->node.mm);
+			cache->is_lmem = false;
+		} else if (drm_mm_node_allocated(&cache->node)) {
 			ggtt->vm.clear_range(&ggtt->vm,
 					     cache->node.start,
 					     cache->node.size);
@@ -1018,6 +1025,42 @@ static void *reloc_kmap(struct drm_i915_gem_object *obj,
 	return vaddr;
 }
 
+static void *reloc_lmem(struct drm_i915_gem_object *obj,
+			struct reloc_cache *cache,
+			unsigned long page)
+{
+	void *vaddr;
+	int err;
+
+	GEM_BUG_ON(use_cpu_reloc(cache, obj));
+
+	if (cache->vaddr) {
+		io_mapping_unmap_atomic((void __force __iomem *) unmask_page(cache->vaddr));
+	} else {
+		err = i915_gem_object_pin_pages(obj);
+		if (err)
+			return ERR_PTR(err);
+
+		i915_gem_object_lock(obj);
+		err = i915_gem_object_set_to_wc_domain(obj, true);
+		i915_gem_object_unlock(obj);
+		if (err) {
+			i915_gem_object_unpin_pages(obj);
+			return ERR_PTR(err);
+		}
+
+		cache->node.mm = (void *)obj;
+		cache->is_lmem = true;
+	}
+
+	vaddr = i915_gem_object_lmem_io_map_page_atomic(obj, page);
+
+	cache->vaddr = (unsigned long)vaddr;
+	cache->page = page;
+
+	return vaddr;
+}
+
 static void *reloc_iomap(struct drm_i915_gem_object *obj,
 			 struct reloc_cache *cache,
 			 unsigned long page)
@@ -1093,8 +1136,12 @@ static void *reloc_vaddr(struct drm_i915_gem_object *obj,
 		vaddr = unmask_page(cache->vaddr);
 	} else {
 		vaddr = NULL;
-		if ((cache->vaddr & KMAP) == 0)
-			vaddr = reloc_iomap(obj, cache, page);
+		if ((cache->vaddr & KMAP) == 0) {
+			if (i915_gem_object_is_lmem(obj))
+				vaddr = reloc_lmem(obj, cache, page);
+			else
+				vaddr = reloc_iomap(obj, cache, page);
+		}
 		if (!vaddr)
 			vaddr = reloc_kmap(obj, cache, page);
 	}
-- 
2.20.1