gpu: nvgpu: Implement ELPG flush for gm20b

ELPG flush is initiated from a common broadcast register, but must be
waited on via per-L2 registers. Split gk20a and gm20b versions of
the flush.

Change-Id: I75c2d65e8da311b50d35bee70308b60464ec2d4d
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/401545
Reviewed-by: Automatic_Commit_Validation_User

1 files changed, 34 insertions, 0 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/ltc_gk20a.c b/drivers/gpu/nvgpu/gk20a/ltc_gk20a.c
index c1ba2aee..9f5317fc 100644
--- a/drivers/gpu/nvgpu/gk20a/ltc_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/ltc_gk20a.c
@@ -212,6 +212,40 @@ void gk20a_ltc_isr(struct gk20a *g)
         gk20a_writel(g, ltc_ltc0_ltss_intr_r(), intr);
 }
 
+/* Flushes the compression bit cache as well as "data".
+ * Note: the name here is a bit of a misnomer.  ELPG uses this
+ * internally... but ELPG doesn't have to be on to do it manually.
+ */
+static void gk20a_mm_g_elpg_flush_locked(struct gk20a *g)
+{
+        u32 data;
+        s32 retry = 100;
+
+        gk20a_dbg_fn("");
+
+        /* Make sure all previous writes are committed to the L2. There's no
+           guarantee that writes are to DRAM. This will be a sysmembar internal
+           to the L2. */
+        gk20a_writel(g, ltc_ltcs_ltss_g_elpg_r(),
+                     ltc_ltcs_ltss_g_elpg_flush_pending_f());
+        do {
+                data = gk20a_readl(g, ltc_ltc0_ltss_g_elpg_r());
+
+                if (ltc_ltc0_ltss_g_elpg_flush_v(data) ==
+                    ltc_ltc0_ltss_g_elpg_flush_pending_v()) {
+                        gk20a_dbg_info("g_elpg_flush 0x%x", data);
+                        retry--;
+                        usleep_range(20, 40);
+                } else
+                        break;
+        } while (retry >= 0 || !tegra_platform_is_silicon());
+
+        if (retry < 0)
+                gk20a_warn(dev_from_gk20a(g),
+                            "g_elpg_flush too many retries");
+
+}
+
 void gk20a_init_ltc(struct gpu_ops *gops)
 {
         gops->ltc.determine_L2_size_bytes = gk20a_determine_L2_size_bytes;