video: tegra: nvmap: implement full L2 clean/flush

When allocation is bigger than L2 size it is more optimal to
flush or writeback whole L2 instead of doing maintenance
for each allocated page.

bug 983964

Signed-off-by: Kirill Artamonov <kartamonov@nvidia.com>
Change-Id: Ieaa70875b92920567ad7cd75eca6eac8197f46de
Reviewed-on: http://git-master/r/108511
Reviewed-by: Simone Willett <swillett@nvidia.com>
Tested-by: Simone Willett <swillett@nvidia.com>

4 files changed, 45 insertions, 12 deletions

diff --git a/drivers/video/tegra/Kconfig b/drivers/video/tegra/Kconfig
index 650c418ccb52..68cf54264403 100644
--- a/drivers/video/tegra/Kconfig
+++ b/drivers/video/tegra/Kconfig
@@ -119,7 +119,7 @@ config NVMAP_PAGE_POOL_SIZE
 	default 0x0
 
 config NVMAP_CACHE_MAINT_BY_SET_WAYS
-	bool "Enalbe cache maintenance by set/ways"
+	bool "Enable cache maintenance by set/ways"
 	depends on TEGRA_NVMAP
 	help
 	 Say Y here to reduce cache maintenance overhead by MVA.
@@ -127,6 +127,14 @@ config NVMAP_CACHE_MAINT_BY_SET_WAYS
 	 where inner cache includes only L1. For the systems, where inner cache
 	 includes L1 and L2, keep this option disabled.
 
+config NVMAP_OUTER_CACHE_MAINT_BY_SET_WAYS
+	bool "Enable outer cache maintenance by set/ways"
+	depends on TEGRA_NVMAP
+	help
+	  Say Y here if you want to optimize cache maintenance for ranges
+	  bigger than size of outer cache. This option has no effect on
+	  system without outer cache.
+
 config NVMAP_VPR
 	bool "Enable VPR Heap."
 	depends on TEGRA_NVMAP
diff --git a/drivers/video/tegra/nvmap/nvmap_common.h b/drivers/video/tegra/nvmap/nvmap_common.h
index 6da010720bb2..2a1e2b4b3c4b 100644
--- a/drivers/video/tegra/nvmap/nvmap_common.h
+++ b/drivers/video/tegra/nvmap/nvmap_common.h
@@ -23,7 +23,8 @@
 extern void v7_flush_kern_cache_all(void *);
 extern void v7_clean_kern_cache_all(void *);
 
-#define FLUSH_CLEAN_BY_SET_WAY_THRESHOLD (8 * PAGE_SIZE)
+#define FLUSH_CLEAN_BY_SET_WAY_THRESHOLD_INNER (8 * PAGE_SIZE)
+#define FLUSH_CLEAN_BY_SET_WAY_THRESHOLD_OUTER (1024 * 1024)
 
 static inline void inner_flush_cache_all(void)
 {
diff --git a/drivers/video/tegra/nvmap/nvmap_dev.c b/drivers/video/tegra/nvmap/nvmap_dev.c
index 02ccaeec4965..0c12348db88e 100644
--- a/drivers/video/tegra/nvmap/nvmap_dev.c
+++ b/drivers/video/tegra/nvmap/nvmap_dev.c
@@ -290,7 +290,7 @@ int nvmap_flush_heap_block(struct nvmap_client *client,
 	if (prot == NVMAP_HANDLE_UNCACHEABLE || prot == NVMAP_HANDLE_WRITE_COMBINE)
 		goto out;
 
-	if (len >= FLUSH_CLEAN_BY_SET_WAY_THRESHOLD) {
+	if (len >= FLUSH_CLEAN_BY_SET_WAY_THRESHOLD_INNER) {
 		inner_flush_cache_all();
 		if (prot != NVMAP_HANDLE_INNER_CACHEABLE)
 			outer_flush_range(block->base, block->base + len);
diff --git a/drivers/video/tegra/nvmap/nvmap_ioctl.c b/drivers/video/tegra/nvmap/nvmap_ioctl.c
index e3adc0abc174..da974b2c6846 100644
--- a/drivers/video/tegra/nvmap/nvmap_ioctl.c
+++ b/drivers/video/tegra/nvmap/nvmap_ioctl.c
@@ -550,14 +550,32 @@ static void heap_page_cache_maint(struct nvmap_client *client,
 	}
 }
 
+static bool fast_cache_maint_outer(unsigned long start,
+		unsigned long end, unsigned int op)
+{
+	bool result = false;
+#if defined(CONFIG_NVMAP_OUTER_CACHE_MAINT_BY_SET_WAYS)
+	if (end - start >= FLUSH_CLEAN_BY_SET_WAY_THRESHOLD_OUTER) {
+		if (op == NVMAP_CACHE_OP_WB_INV) {
+			outer_flush_all();
+			result = true;
+		}
+		if (op == NVMAP_CACHE_OP_WB) {
+			outer_clean_all();
+			result = true;
+		}
+	}
+#endif
+	return result;
+}
+
 static bool fast_cache_maint(struct nvmap_client *client, struct nvmap_handle *h,
 	unsigned long start, unsigned long end, unsigned int op)
 {
 	int ret = false;
-
 #if defined(CONFIG_NVMAP_CACHE_MAINT_BY_SET_WAYS)
 	if ((op == NVMAP_CACHE_OP_INV) ||
-		((end - start) < FLUSH_CLEAN_BY_SET_WAY_THRESHOLD))
+		((end - start) < FLUSH_CLEAN_BY_SET_WAY_THRESHOLD_INNER))
 		goto out;
 
 	if (op == NVMAP_CACHE_OP_WB_INV)
@@ -565,13 +583,19 @@ static bool fast_cache_maint(struct nvmap_client *client, struct nvmap_handle *h
 	else if (op == NVMAP_CACHE_OP_WB)
 		inner_clean_cache_all();
 
-	if (h->heap_pgalloc && (h->flags != NVMAP_HANDLE_INNER_CACHEABLE)) {
-		heap_page_cache_maint(client, h, start, end, op,
-				false, true, NULL, 0, 0);
-	} else if (h->flags != NVMAP_HANDLE_INNER_CACHEABLE) {
-		start += h->carveout->base;
-		end += h->carveout->base;
-		outer_cache_maint(op, start, end - start);
+	/* outer maintenance */
+	if (h->flags != NVMAP_HANDLE_INNER_CACHEABLE ) {
+		if(!fast_cache_maint_outer(start, end, op))
+		{
+			if (h->heap_pgalloc) {
+				heap_page_cache_maint(client, h, start,
+					end, op, false, true, NULL, 0, 0);
+			} else  {
+				start += h->carveout->base;
+				end += h->carveout->base;
+				outer_cache_maint(op, start, end - start);
+			}
+		}
 	}
 	ret = true;
 out: