gpu: nvgpu: Pre-Population of zbc entries

The default zbc entries were never populated in zbc HW table
because the conditional flag "gr->sw_ready" was always set thus
avoided the zbc default loading function call. Now zbc default
loading would happen only during boot time in sw structure.Hw
zbc regs would be loaded from that structure every time a
railgate exit happens.

Bug 1580210

Change-Id: Ie3e40738cbc84cf724c3f3871f15b17a5c84025a
Signed-off-by: Sujeet Baranwal <sbaranwal@nvidia.com>
Reviewed-on: http://git-master/r/662306
Reviewed-by: Sami Kiminki <skiminki@nvidia.com>
Tested-by: Lauri Peltonen <lpeltonen@nvidia.com>
Reviewed-by: Arto Merilainen <amerilainen@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>

1 files changed, 20 insertions, 22 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
index 4d101845..4b48b838 100644
--- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
@@ -3805,7 +3805,6 @@ static int gr_gk20a_load_zbc_table(struct gk20a *g, struct gr_gk20a *gr)
 {
         int i, ret;
 
-        mutex_init(&gr->zbc_lock);
         for (i = 0; i < gr->max_used_color_index; i++) {
                 struct zbc_color_table *c_tbl = &gr->zbc_col_tbl[i];
                 struct zbc_entry zbc_val;
@@ -3842,39 +3841,39 @@ int gr_gk20a_load_zbc_default_table(struct gk20a *g, struct gr_gk20a *gr)
         struct zbc_entry zbc_val;
         u32 i, err;
 
+        mutex_init(&gr->zbc_lock);
+
         /* load default color table */
         zbc_val.type = GK20A_ZBC_TYPE_COLOR;
 
-        zbc_val.format = gr_ds_zbc_color_fmt_val_zero_v();
+        /* Opaque black (i.e. solid black, fmt 0x28 = A8B8G8R8) */
+        zbc_val.format = gr_ds_zbc_color_fmt_val_a8_b8_g8_r8_v();
         for (i = 0; i < GK20A_ZBC_COLOR_VALUE_SIZE; i++) {
                 zbc_val.color_ds[i] = 0;
                 zbc_val.color_l2[i] = 0;
         }
+        zbc_val.color_l2[0] = 0xff000000;
+        zbc_val.color_ds[3] = 0x3f800000;
         err = gr_gk20a_add_zbc(g, gr, &zbc_val);
 
-        zbc_val.format = gr_ds_zbc_color_fmt_val_unorm_one_v();
-        for (i = 0; i < GK20A_ZBC_COLOR_VALUE_SIZE; i++) {
-                zbc_val.color_ds[i] = 0xffffffff;
-                zbc_val.color_l2[i] = 0x3f800000;
-        }
-        err |= gr_gk20a_add_zbc(g, gr, &zbc_val);
-
-        zbc_val.format = gr_ds_zbc_color_fmt_val_rf32_gf32_bf32_af32_v();
+        /* Transparent black = (fmt 1 = zero) */
+        zbc_val.format = gr_ds_zbc_color_fmt_val_zero_v();
         for (i = 0; i < GK20A_ZBC_COLOR_VALUE_SIZE; i++) {
                 zbc_val.color_ds[i] = 0;
                 zbc_val.color_l2[i] = 0;
         }
-        err |= gr_gk20a_add_zbc(g, gr, &zbc_val);
+        err = gr_gk20a_add_zbc(g, gr, &zbc_val);
 
-        zbc_val.format = gr_ds_zbc_color_fmt_val_rf32_gf32_bf32_af32_v();
+        /* Opaque white (i.e. solid white) = (fmt 2 = uniform 1) */
+        zbc_val.format = gr_ds_zbc_color_fmt_val_unorm_one_v();
         for (i = 0; i < GK20A_ZBC_COLOR_VALUE_SIZE; i++) {
                 zbc_val.color_ds[i] = 0x3f800000;
-                zbc_val.color_l2[i] = 0x3f800000;
+                zbc_val.color_l2[i] = 0xffffffff;
         }
         err |= gr_gk20a_add_zbc(g, gr, &zbc_val);
 
         if (!err)
-                gr->max_default_color_index = 4;
+                gr->max_default_color_index = 3;
         else {
                 gk20a_err(dev_from_gk20a(g),
                            "fail to load default zbc color table\n");
@@ -3885,13 +3884,13 @@ int gr_gk20a_load_zbc_default_table(struct gk20a *g, struct gr_gk20a *gr)
         zbc_val.type = GK20A_ZBC_TYPE_DEPTH;
 
         zbc_val.format = gr_ds_zbc_z_fmt_val_fp32_v();
-        zbc_val.depth = 0;
-        err = gr_gk20a_add_zbc(g, gr, &zbc_val);
-
-        zbc_val.format = gr_ds_zbc_z_fmt_val_fp32_v();
         zbc_val.depth = 0x3f800000;
         err |= gr_gk20a_add_zbc(g, gr, &zbc_val);
 
+        zbc_val.format = gr_ds_zbc_z_fmt_val_fp32_v();
+        zbc_val.depth = 0;
+        err = gr_gk20a_add_zbc(g, gr, &zbc_val);
+
         if (!err)
                 gr->max_default_depth_index = 2;
         else {
@@ -4311,10 +4310,7 @@ static int gk20a_init_gr_setup_hw(struct gk20a *g)
         data = gk20a_readl(g, gr_status_mask_r());
         gk20a_writel(g, gr_status_mask_r(), data & gr->status_disable_mask);
 
-        if (gr->sw_ready)
-                gr_gk20a_load_zbc_table(g, gr);
-        else
-                gr_gk20a_load_zbc_default_table(g, gr);
+        gr_gk20a_load_zbc_table(g, gr);
 
         g->ops.ltc.init_cbc(g, gr);
 
@@ -4625,6 +4621,8 @@ static int gk20a_init_gr_setup_sw(struct gk20a *g)
         if (err)
                 goto clean_up;
 
+        gr_gk20a_load_zbc_default_table(g, gr);
+
         mutex_init(&gr->ctx_mutex);
         spin_lock_init(&gr->ch_tlb_lock);