gpu: nvgpu: Wait for BAR1 bind

Wait for BAR1 bind to complete before continuing. The register to
wait exists Maxwell onwards.

Change-Id: Ie3736033fdb748c5da8d7a6085ad6d63acaf41f5
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/1123941

drivers/gpu/nvgpu/gk20a/gk20a.h

@@ -475,6 +475,7 @@ struct gpu_ops {
 		void (*init_pdb)(struct gk20a *g, void *inst_ptr, u64 pdb_addr);
 		u64 (*get_iova_addr)(struct gk20a *g, struct scatterlist *sgl,
 					 u32 flags);
+		int (*bar1_bind)(struct gk20a *g, u64 bar1_iova);
 	} mm;
 	struct {
 		int (*init_therm_setup_hw)(struct gk20a *g);

drivers/gpu/nvgpu/gk20a/mm_gk20a.c

@@ -493,15 +493,7 @@ int gk20a_init_mm_setup_hw(struct gk20a *g)
 			g->ops.fb.set_use_full_comp_tag_line(g);
 
 
-	inst_pa = (u32)(inst_pa >> bar1_instance_block_shift_gk20a());
+	g->ops.mm.bar1_bind(g, inst_pa);
-	gk20a_dbg_info("bar1 inst block ptr: 0x%08x",  (u32)inst_pa);
-
-	gk20a_writel(g, bus_bar1_block_r(),
-		     (g->mm.vidmem_is_vidmem ?
-		       bus_bar1_block_target_sys_mem_ncoh_f() :
-		       bus_bar1_block_target_vid_mem_f()) |
-		     bus_bar1_block_mode_virtual_f() |
-		     bus_bar1_block_ptr_f(inst_pa));
 
 	if (g->ops.mm.init_bar2_mm_hw_setup) {
 		err = g->ops.mm.init_bar2_mm_hw_setup(g);
@@ -516,6 +508,21 @@ int gk20a_init_mm_setup_hw(struct gk20a *g)
 	return 0;
 }
 
+static int gk20a_mm_bar1_bind(struct gk20a *g, u64 bar1_iova)
+{
+	u64 inst_pa = (u32)(bar1_iova >> bar1_instance_block_shift_gk20a());
+	gk20a_dbg_info("bar1 inst block ptr: 0x%08x",  (u32)inst_pa);
+
+	gk20a_writel(g, bus_bar1_block_r(),
+		     (g->mm.vidmem_is_vidmem ?
+		       bus_bar1_block_target_sys_mem_ncoh_f() :
+		       bus_bar1_block_target_vid_mem_f()) |
+		     bus_bar1_block_mode_virtual_f() |
+		     bus_bar1_block_ptr_f(inst_pa));
+
+	return 0;
+}
+
 int gk20a_init_mm_support(struct gk20a *g)
 {
 	u32 err;
@@ -3919,4 +3926,5 @@ void gk20a_init_mm(struct gpu_ops *gops)
 	gops->mm.get_mmu_levels = gk20a_mm_get_mmu_levels;
 	gops->mm.init_pdb = gk20a_mm_init_pdb;
 	gops->mm.init_mm_setup_hw = gk20a_init_mm_setup_hw;
+	gops->mm.bar1_bind = gk20a_mm_bar1_bind;
 }

drivers/gpu/nvgpu/gm20b/hw_bus_gm20b.h

@@ -106,6 +106,58 @@ static inline u32 bus_bar2_block_ptr_shift_v(void)
 {
 	return 0x0000000c;
 }
+static inline u32 bus_bind_status_r(void)
+{
+	return 0x00001710;
+}
+static inline u32 bus_bind_status_bar1_pending_v(u32 r)
+{
+	return (r >> 0) & 0x1;
+}
+static inline u32 bus_bind_status_bar1_pending_empty_f(void)
+{
+	return 0x0;
+}
+static inline u32 bus_bind_status_bar1_pending_busy_f(void)
+{
+	return 0x1;
+}
+static inline u32 bus_bind_status_bar1_outstanding_v(u32 r)
+{
+	return (r >> 1) & 0x1;
+}
+static inline u32 bus_bind_status_bar1_outstanding_false_f(void)
+{
+	return 0x0;
+}
+static inline u32 bus_bind_status_bar1_outstanding_true_f(void)
+{
+	return 0x2;
+}
+static inline u32 bus_bind_status_bar2_pending_v(u32 r)
+{
+	return (r >> 2) & 0x1;
+}
+static inline u32 bus_bind_status_bar2_pending_empty_f(void)
+{
+	return 0x0;
+}
+static inline u32 bus_bind_status_bar2_pending_busy_f(void)
+{
+	return 0x4;
+}
+static inline u32 bus_bind_status_bar2_outstanding_v(u32 r)
+{
+	return (r >> 3) & 0x1;
+}
+static inline u32 bus_bind_status_bar2_outstanding_false_f(void)
+{
+	return 0x0;
+}
+static inline u32 bus_bind_status_bar2_outstanding_true_f(void)
+{
+	return 0x8;
+}
 static inline u32 bus_intr_0_r(void)
 {
 	return 0x00001100;

drivers/gpu/nvgpu/gm20b/mm_gm20b.c

@@ -14,12 +14,14 @@
  */
 
 #include <linux/pm_runtime.h>
+#include <linux/delay.h>
 #include "gk20a/gk20a.h"
 #include "mm_gm20b.h"
 #include "hw_gmmu_gm20b.h"
 #include "hw_fb_gm20b.h"
 #include "hw_gr_gm20b.h"
 #include "hw_ram_gm20b.h"
+#include "hw_bus_gm20b.h"
 
 static int gm20b_mm_mmu_vpr_info_fetch_wait(struct gk20a *g,
 		const unsigned int msec)
@@ -133,6 +135,32 @@ static bool gm20b_mm_support_sparse(struct gk20a *g)
 	return true;
 }
 
+static int gm20b_mm_bar1_bind(struct gk20a *g, u64 bar1_iova)
+{
+	int retry = 1000;
+	u64 inst_pa = (u32)(bar1_iova >> bar1_instance_block_shift_gk20a());
+	gk20a_dbg_info("bar1 inst block ptr: 0x%08x",  (u32)inst_pa);
+
+	gk20a_writel(g, bus_bar1_block_r(),
+		     (g->mm.vidmem_is_vidmem ?
+		       bus_bar1_block_target_sys_mem_ncoh_f() :
+		       bus_bar1_block_target_vid_mem_f()) |
+		     bus_bar1_block_mode_virtual_f() |
+		     bus_bar1_block_ptr_f(inst_pa));
+	do {
+		u32 val = gk20a_readl(g, bus_bind_status_r());
+		u32 pending = bus_bind_status_bar1_pending_v(val);
+		u32 outstanding = bus_bind_status_bar1_outstanding_v(val);
+		if (!pending && !outstanding)
+			break;
+
+		udelay(5);
+		retry--;
+	} while (retry >= 0 || !tegra_platform_is_silicon());
+
+	return retry ? -EINVAL : 0;
+}
+
 void gm20b_init_mm(struct gpu_ops *gops)
 {
 	gops->mm.support_sparse = gm20b_mm_support_sparse;
@@ -155,4 +183,5 @@ void gm20b_init_mm(struct gpu_ops *gops)
 	gops->mm.get_mmu_levels = gk20a_mm_get_mmu_levels;
 	gops->mm.init_pdb = gk20a_mm_init_pdb;
 	gops->mm.init_mm_setup_hw = gk20a_init_mm_setup_hw;
+	gops->mm.bar1_bind = gm20b_mm_bar1_bind;
 }