drm/amd/display: Change dmcu init sequence for dmcub loading dmcu FW.

[Why]
DMCU isn't intiliazed properly by dmcub loading due to dmcub initialize
sequence.

[How]
Change dmcu init sequece to meet dmcub initilize.
Signed-off-by: Yongqiang Sun <yongqiang.sun@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Reviewed-by: Hersen Wu <hersenxs.wu@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/display/dc/dce/dce_dmcu.c

@@ -59,6 +59,12 @@
 #define MCP_BL_SET_PWM_FRAC 0x6A  /* Enable or disable Fractional PWM */
 #define MASTER_COMM_CNTL_REG__MASTER_COMM_INTERRUPT_MASK   0x00000001L
 
+// PSP FW version
+#define mmMP0_SMN_C2PMSG_58				0x1607A
+
+//Register access policy version
+#define mmMP0_SMN_C2PMSG_91				0x1609B
+
 static bool dce_dmcu_init(struct dmcu *dmcu)
 {
 	// Do nothing
@@ -373,6 +379,7 @@ static bool dcn10_dmcu_init(struct dmcu *dmcu)
 	const struct dc_config *config = &dmcu->ctx->dc->config;
 	bool status = false;
 
+	PERF_TRACE();
 	/*  Definition of DC_DMCU_SCRATCH
 	 *  0 : firmare not loaded
 	 *  1 : PSP load DMCU FW but not initialized
@@ -429,9 +436,23 @@ static bool dcn10_dmcu_init(struct dmcu *dmcu)
 		break;
 	}
 
+	PERF_TRACE();
 	return status;
 }
 
+#if defined(CONFIG_DRM_AMD_DC_DCN2_1)
+static bool dcn21_dmcu_init(struct dmcu *dmcu)
+{
+	struct dce_dmcu *dmcu_dce = TO_DCE_DMCU(dmcu);
+	uint32_t dmcub_psp_version = REG_READ(DMCUB_SCRATCH15);
+
+	if (dmcu->auto_load_dmcu && dmcub_psp_version == 0) {
+		return false;
+	}
+
+	return dcn10_dmcu_init(dmcu);
+}
+#endif
 
 static bool dcn10_dmcu_load_iram(struct dmcu *dmcu,
 		unsigned int start_offset,
@@ -818,6 +839,21 @@ static const struct dmcu_funcs dcn20_funcs = {
 };
 #endif
 
+#if defined(CONFIG_DRM_AMD_DC_DCN2_1)
+static const struct dmcu_funcs dcn21_funcs = {
+	.dmcu_init = dcn21_dmcu_init,
+	.load_iram = dcn10_dmcu_load_iram,
+	.set_psr_enable = dcn10_dmcu_set_psr_enable,
+	.setup_psr = dcn10_dmcu_setup_psr,
+	.get_psr_state = dcn10_get_dmcu_psr_state,
+	.set_psr_wait_loop = dcn10_psr_wait_loop,
+	.get_psr_wait_loop = dcn10_get_psr_wait_loop,
+	.is_dmcu_initialized = dcn10_is_dmcu_initialized,
+	.lock_phy = dcn20_lock_phy,
+	.unlock_phy = dcn20_unlock_phy
+};
+#endif
+
 static void dce_dmcu_construct(
 	struct dce_dmcu *dmcu_dce,
 	struct dc_context *ctx,
@@ -836,6 +872,26 @@ static void dce_dmcu_construct(
 	dmcu_dce->dmcu_mask = dmcu_mask;
 }
 
+#if defined(CONFIG_DRM_AMD_DC_DCN2_1)
+static void dcn21_dmcu_construct(
+		struct dce_dmcu *dmcu_dce,
+		struct dc_context *ctx,
+		const struct dce_dmcu_registers *regs,
+		const struct dce_dmcu_shift *dmcu_shift,
+		const struct dce_dmcu_mask *dmcu_mask)
+{
+	uint32_t psp_version = 0;
+
+	dce_dmcu_construct(dmcu_dce, ctx, regs, dmcu_shift, dmcu_mask);
+
+	if (!IS_FPGA_MAXIMUS_DC(ctx->dce_environment)) {
+		psp_version = dm_read_reg(ctx, mmMP0_SMN_C2PMSG_58);
+		dmcu_dce->base.auto_load_dmcu = (psp_version > 0x00110029);
+		dmcu_dce->base.psp_version = psp_version;
+	}
+}
+#endif
+
 struct dmcu *dce_dmcu_create(
 	struct dc_context *ctx,
 	const struct dce_dmcu_registers *regs,
@@ -903,6 +959,29 @@ struct dmcu *dcn20_dmcu_create(
 }
 #endif
 
+#if defined(CONFIG_DRM_AMD_DC_DCN2_1)
+struct dmcu *dcn21_dmcu_create(
+	struct dc_context *ctx,
+	const struct dce_dmcu_registers *regs,
+	const struct dce_dmcu_shift *dmcu_shift,
+	const struct dce_dmcu_mask *dmcu_mask)
+{
+	struct dce_dmcu *dmcu_dce = kzalloc(sizeof(*dmcu_dce), GFP_KERNEL);
+
+	if (dmcu_dce == NULL) {
+		BREAK_TO_DEBUGGER();
+		return NULL;
+	}
+
+	dcn21_dmcu_construct(
+		dmcu_dce, ctx, regs, dmcu_shift, dmcu_mask);
+
+	dmcu_dce->base.funcs = &dcn21_funcs;
+
+	return &dmcu_dce->base;
+}
+#endif
+
 void dce_dmcu_destroy(struct dmcu **dmcu)
 {
 	struct dce_dmcu *dmcu_dce = TO_DCE_DMCU(*dmcu);

drivers/gpu/drm/amd/display/dc/dce/dce_dmcu.h

@@ -71,6 +71,10 @@
 	DMCU_COMMON_REG_LIST_DCE_BASE(), \
 	SR(DMU_MEM_PWR_CNTL)
 
+#define DMCU_DCN20_REG_LIST()\
+	DMCU_DCN10_REG_LIST(), \
+	SR(DMCUB_SCRATCH15)
+
 #define DMCU_SF(reg_name, field_name, post_fix)\
 	.field_name = reg_name ## __ ## field_name ## post_fix
 
@@ -175,6 +179,7 @@ struct dce_dmcu_registers {
 	uint32_t DMCU_INTERRUPT_TO_UC_EN_MASK;
 	uint32_t SMU_INTERRUPT_CONTROL;
 	uint32_t DC_DMCU_SCRATCH;
+	uint32_t DMCUB_SCRATCH15;
 };
 
 struct dce_dmcu {
@@ -269,6 +274,14 @@ struct dmcu *dcn20_dmcu_create(
 	const struct dce_dmcu_mask *dmcu_mask);
 #endif
 
+#if defined(CONFIG_DRM_AMD_DC_DCN2_1)
+struct dmcu *dcn21_dmcu_create(
+	struct dc_context *ctx,
+	const struct dce_dmcu_registers *regs,
+	const struct dce_dmcu_shift *dmcu_shift,
+	const struct dce_dmcu_mask *dmcu_mask);
+#endif
+
 void dce_dmcu_destroy(struct dmcu **dmcu);
 
 static const uint32_t abm_gain_stepsize = 0x0060;

drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c

@@ -1285,7 +1285,7 @@ static void dcn10_init_hw(struct dc *dc)
 		abm->funcs->abm_init(abm);
 	}
 
-	if (dmcu != NULL)
+	if (dmcu != NULL && !dmcu->auto_load_dmcu)
 		dmcu->funcs->dmcu_init(dmcu);
 
 	if (abm != NULL && dmcu != NULL)

drivers/gpu/drm/amd/display/dc/dcn21/dcn21_resource.c

@@ -350,7 +350,7 @@ static const struct bios_registers bios_regs = {
 };
 
 static const struct dce_dmcu_registers dmcu_regs = {
-		DMCU_DCN10_REG_LIST()
+		DMCU_DCN20_REG_LIST()
 };
 
 static const struct dce_dmcu_shift dmcu_shift = {
@@ -1727,7 +1727,7 @@ static bool construct(
 		goto create_fail;
 	}
 
-	pool->base.dmcu = dcn20_dmcu_create(ctx,
+	pool->base.dmcu = dcn21_dmcu_create(ctx,
 			&dmcu_regs,
 			&dmcu_shift,
 			&dmcu_mask);

drivers/gpu/drm/amd/display/dc/inc/hw/dmcu.h

@@ -52,6 +52,8 @@ struct dmcu {
 	enum dmcu_state dmcu_state;
 	struct dmcu_version dmcu_version;
 	unsigned int cached_wait_loop_number;
+	uint32_t psp_version;
+	bool auto_load_dmcu;
 };
 
 struct dmcu_funcs {