drm/amd/display: Compensate for XGMI SS downspread on dprefclk

[Why]
When XGMI is enabled, we need to adjust the dprefclk according to the
WAFL link's spread spectrum info. This is for VG20 (DCE121) only.

[How]
dce_clk_mgr already stores SS info, currently being used by audio clock.
Therefore, patch the clk_mgr's SS info with the xGMI SS info, if xGMI
is enabled. For display clock, adjust it during dce12_update_clocks()
before calling set_clock().

Since we rely on a mmhub register to reliably determine if xGMI is
enabled, the patching step needs to happen after resource_construct()
has initialized the hardware sequencer.
Signed-off-by: Leo Li <sunpeng.li@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_clk_mgr.c

@@ -450,6 +450,42 @@ void dce_clock_read_ss_info(struct dce_clk_mgr *clk_mgr_dce)
 	}
 }
 
+/**
+ * dce121_clock_patch_xgmi_ss_info() - Save XGMI spread spectrum info
+ * @clk_mgr: clock manager base structure
+ *
+ * Reads from VBIOS the XGMI spread spectrum info and saves it within
+ * the dce clock manager. This operation will overwrite the existing dprefclk
+ * SS values if the vBIOS query succeeds. Otherwise, it does nothing. It also
+ * sets the ->xgmi_enabled flag.
+ */
+void dce121_clock_patch_xgmi_ss_info(struct clk_mgr *clk_mgr)
+{
+	struct dce_clk_mgr *clk_mgr_dce = TO_DCE_CLK_MGR(clk_mgr);
+	enum bp_result result;
+	struct spread_spectrum_info info = { { 0 } };
+	struct dc_bios *bp = clk_mgr_dce->base.ctx->dc_bios;
+
+	clk_mgr_dce->xgmi_enabled = false;
+
+	result = bp->funcs->get_spread_spectrum_info(bp, AS_SIGNAL_TYPE_XGMI,
+						     0, &info);
+	if (result == BP_RESULT_OK && info.spread_spectrum_percentage != 0) {
+		clk_mgr_dce->xgmi_enabled = true;
+		clk_mgr_dce->ss_on_dprefclk = true;
+		clk_mgr_dce->dprefclk_ss_divider =
+				info.spread_percentage_divider;
+
+		if (info.type.CENTER_MODE == 0) {
+			/* Currently for DP Reference clock we
+			 * need only SS percentage for
+			 * downspread */
+			clk_mgr_dce->dprefclk_ss_percentage =
+					info.spread_spectrum_percentage;
+		}
+	}
+}
+
 void dce110_fill_display_configs(
 	const struct dc_state *context,
 	struct dm_pp_display_configuration *pp_display_cfg)
@@ -710,6 +746,13 @@ static void dce12_update_clocks(struct clk_mgr *clk_mgr,
 
 	if (should_set_clock(safe_to_lower, patched_disp_clk, clk_mgr->clks.dispclk_khz)) {
 		clock_voltage_req.clk_type = DM_PP_CLOCK_TYPE_DISPLAY_CLK;
+		/*
+		 * When xGMI is enabled, the display clk needs to be adjusted
+		 * with the WAFL link's SS percentage.
+		 */
+		if (clk_mgr_dce->xgmi_enabled)
+			patched_disp_clk = clk_mgr_adjust_dp_ref_freq_for_ss(
+					clk_mgr_dce, patched_disp_clk);
 		clock_voltage_req.clocks_in_khz = patched_disp_clk;
 		clk_mgr->clks.dispclk_khz = dce112_set_clock(clk_mgr, patched_disp_clk);
 
@@ -875,6 +918,27 @@ struct clk_mgr *dce120_clk_mgr_create(struct dc_context *ctx)
 	return &clk_mgr_dce->base;
 }
 
+struct clk_mgr *dce121_clk_mgr_create(struct dc_context *ctx)
+{
+	struct dce_clk_mgr *clk_mgr_dce = kzalloc(sizeof(*clk_mgr_dce),
+						  GFP_KERNEL);
+
+	if (clk_mgr_dce == NULL) {
+		BREAK_TO_DEBUGGER();
+		return NULL;
+	}
+
+	memcpy(clk_mgr_dce->max_clks_by_state, dce120_max_clks_by_state,
+	       sizeof(dce120_max_clks_by_state));
+
+	dce_clk_mgr_construct(clk_mgr_dce, ctx, NULL, NULL, NULL);
+
+	clk_mgr_dce->dprefclk_khz = 625000;
+	clk_mgr_dce->base.funcs = &dce120_funcs;
+
+	return &clk_mgr_dce->base;
+}
+
 void dce_clk_mgr_destroy(struct clk_mgr **clk_mgr)
 {
 	struct dce_clk_mgr *clk_mgr_dce = TO_DCE_CLK_MGR(*clk_mgr);

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

@@ -94,11 +94,37 @@ struct dce_clk_mgr {
 	 * This is basically "Crystal Frequency In KHz" (XTALIN) frequency */
 	int dfs_bypass_disp_clk;
 
-	/* Flag for Enabled SS on DPREFCLK */
+	/**
+	 * @ss_on_dprefclk:
+	 *
+	 * True if spread spectrum is enabled on the DP ref clock.
+	 */
 	bool ss_on_dprefclk;
-	/* DPREFCLK SS percentage (if down-spread enabled) */
+
+	/**
+	 * @xgmi_enabled:
+	 *
+	 * True if xGMI is enabled. On VG20, both audio and display clocks need
+	 * to be adjusted with the WAFL link's SS info if xGMI is enabled.
+	 */
+	bool xgmi_enabled;
+
+	/**
+	 * @dprefclk_ss_percentage:
+	 *
+	 * DPREFCLK SS percentage (if down-spread enabled).
+	 *
+	 * Note that if XGMI is enabled, the SS info (percentage and divider)
+	 * from the WAFL link is used instead. This is decided during
+	 * dce_clk_mgr initialization.
+	 */
 	int dprefclk_ss_percentage;
-	/* DPREFCLK SS percentage Divider (100 or 1000) */
+
+	/**
+	 * @dprefclk_ss_divider:
+	 *
+	 * DPREFCLK SS percentage Divider (100 or 1000).
+	 */
 	int dprefclk_ss_divider;
 	int dprefclk_khz;
 
@@ -163,6 +189,9 @@ struct clk_mgr *dce112_clk_mgr_create(
 
 struct clk_mgr *dce120_clk_mgr_create(struct dc_context *ctx);
 
+struct clk_mgr *dce121_clk_mgr_create(struct dc_context *ctx);
+void dce121_clock_patch_xgmi_ss_info(struct clk_mgr *clk_mgr);
+
 void dce_clk_mgr_destroy(struct clk_mgr **clk_mgr);
 
 int dentist_get_divider_from_did(int did);

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

@@ -133,6 +133,10 @@
 	SR(DCHUB_AGP_TOP), \
 	BL_REG_LIST()
 
+#define HWSEQ_VG20_REG_LIST() \
+	HWSEQ_DCE120_REG_LIST(),\
+	MMHUB_SR(MC_VM_XGMI_LFB_CNTL)
+
 #define HWSEQ_DCE112_REG_LIST() \
 	HWSEQ_DCE10_REG_LIST(), \
 	HWSEQ_PIXEL_RATE_REG_LIST(CRTC), \
@@ -298,6 +302,7 @@ struct dce_hwseq_registers {
 	uint32_t MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_LSB;
 	uint32_t MC_VM_SYSTEM_APERTURE_LOW_ADDR;
 	uint32_t MC_VM_SYSTEM_APERTURE_HIGH_ADDR;
+	uint32_t MC_VM_XGMI_LFB_CNTL;
 	uint32_t AZALIA_AUDIO_DTO;
 	uint32_t AZALIA_CONTROLLER_CLOCK_GATING;
 };
@@ -382,6 +387,11 @@ struct dce_hwseq_registers {
 	HWS_SF(, LVTMA_PWRSEQ_CNTL, LVTMA_BLON, mask_sh), \
 	HWS_SF(, LVTMA_PWRSEQ_STATE, LVTMA_PWRSEQ_TARGET_STATE_R, mask_sh)
 
+#define HWSEQ_VG20_MASK_SH_LIST(mask_sh)\
+	HWSEQ_DCE12_MASK_SH_LIST(mask_sh),\
+	HWS_SF(, MC_VM_XGMI_LFB_CNTL, PF_LFB_REGION, mask_sh),\
+	HWS_SF(, MC_VM_XGMI_LFB_CNTL, PF_MAX_REGION, mask_sh)
+
 #define HWSEQ_DCN_MASK_SH_LIST(mask_sh)\
 	HWSEQ_PIXEL_RATE_MASK_SH_LIST(mask_sh, OTG0_),\
 	HWS_SF1(OTG0_, PHYPLL_PIXEL_RATE_CNTL, PHYPLL_PIXEL_RATE_SOURCE, mask_sh), \
@@ -470,6 +480,8 @@ struct dce_hwseq_registers {
 	type PHYSICAL_PAGE_NUMBER_MSB;\
 	type PHYSICAL_PAGE_NUMBER_LSB;\
 	type LOGICAL_ADDR; \
+	type PF_LFB_REGION;\
+	type PF_MAX_REGION;\
 	type ENABLE_L1_TLB;\
 	type SYSTEM_ACCESS_MODE;\
 	type LVTMA_BLON;\

drivers/gpu/drm/amd/display/dc/dce120/dce120_hw_sequencer.c

@@ -244,6 +244,21 @@ static void dce120_update_dchub(
 	dh_data->dchub_info_valid = false;
 }
 
+/**
+ * dce121_xgmi_enabled() - Check if xGMI is enabled
+ * @hws: DCE hardware sequencer object
+ *
+ * Return true if xGMI is enabled. False otherwise.
+ */
+bool dce121_xgmi_enabled(struct dce_hwseq *hws)
+{
+	uint32_t pf_max_region;
+
+	REG_GET(MC_VM_XGMI_LFB_CNTL, PF_MAX_REGION, &pf_max_region);
+	/* PF_MAX_REGION == 0 means xgmi is disabled */
+	return !!pf_max_region;
+}
+
 void dce120_hw_sequencer_construct(struct dc *dc)
 {
 	/* All registers used by dce11.2 match those in dce11 in offset and

drivers/gpu/drm/amd/display/dc/dce120/dce120_hw_sequencer.h

@@ -30,6 +30,7 @@
 
 struct dc;
 
+bool dce121_xgmi_enabled(struct dce_hwseq *hws);
 void dce120_hw_sequencer_construct(struct dc *dc);
 
 #endif /* __DC_HWSS_DCE112_H__ */

drivers/gpu/drm/amd/display/dc/dce120/dce120_resource.c

@@ -62,6 +62,8 @@
 #include "soc15_hw_ip.h"
 #include "vega10_ip_offset.h"
 #include "nbio/nbio_6_1_offset.h"
+#include "mmhub/mmhub_9_4_0_offset.h"
+#include "mmhub/mmhub_9_4_0_sh_mask.h"
 #include "reg_helper.h"
 
 #include "dce100/dce100_resource.h"
@@ -139,6 +141,17 @@ static const struct dce110_timing_generator_offsets dce120_tg_offsets[] = {
 	.reg_name = BASE(mm ## block ## id ## _ ## reg_name ## _BASE_IDX) + \
 					mm ## block ## id ## _ ## reg_name
 
+/* MMHUB */
+#define MMHUB_BASE_INNER(seg) \
+	MMHUB_BASE__INST0_SEG ## seg
+
+#define MMHUB_BASE(seg) \
+	MMHUB_BASE_INNER(seg)
+
+#define MMHUB_SR(reg_name)\
+		.reg_name = MMHUB_BASE(mm ## reg_name ## _BASE_IDX) +  \
+					mm ## reg_name
+
 /* macros to expend register list macro defined in HW object header file
  * end *********************/
 
@@ -681,6 +694,19 @@ static const struct dce_hwseq_mask hwseq_mask = {
 		HWSEQ_DCE12_MASK_SH_LIST(_MASK)
 };
 
+/* HWSEQ regs for VG20 */
+static const struct dce_hwseq_registers dce121_hwseq_reg = {
+		HWSEQ_VG20_REG_LIST()
+};
+
+static const struct dce_hwseq_shift dce121_hwseq_shift = {
+		HWSEQ_VG20_MASK_SH_LIST(__SHIFT)
+};
+
+static const struct dce_hwseq_mask dce121_hwseq_mask = {
+		HWSEQ_VG20_MASK_SH_LIST(_MASK)
+};
+
 static struct dce_hwseq *dce120_hwseq_create(
 	struct dc_context *ctx)
 {
@@ -695,6 +721,20 @@ static struct dce_hwseq *dce120_hwseq_create(
 	return hws;
 }
 
+static struct dce_hwseq *dce121_hwseq_create(
+	struct dc_context *ctx)
+{
+	struct dce_hwseq *hws = kzalloc(sizeof(struct dce_hwseq), GFP_KERNEL);
+
+	if (hws) {
+		hws->ctx = ctx;
+		hws->regs = &dce121_hwseq_reg;
+		hws->shifts = &dce121_hwseq_shift;
+		hws->masks = &dce121_hwseq_mask;
+	}
+	return hws;
+}
+
 static const struct resource_create_funcs res_create_funcs = {
 	.read_dce_straps = read_dce_straps,
 	.create_audio = create_audio,
@@ -702,6 +742,14 @@ static const struct resource_create_funcs res_create_funcs = {
 	.create_hwseq = dce120_hwseq_create,
 };
 
+static const struct resource_create_funcs dce121_res_create_funcs = {
+	.read_dce_straps = read_dce_straps,
+	.create_audio = create_audio,
+	.create_stream_encoder = dce120_stream_encoder_create,
+	.create_hwseq = dce121_hwseq_create,
+};
+
+
 #define mi_inst_regs(id) { MI_DCE12_REG_LIST(id) }
 static const struct dce_mem_input_registers mi_regs[] = {
 		mi_inst_regs(0),
@@ -911,7 +959,8 @@ static bool construct(
 	int j;
 	struct dc_context *ctx = dc->ctx;
 	struct irq_service_init_data irq_init_data;
-	bool harvest_enabled = ASICREV_IS_VEGA20_P(ctx->asic_id.hw_internal_rev);
+	static const struct resource_create_funcs *res_funcs;
+	bool is_vg20 = ASICREV_IS_VEGA20_P(ctx->asic_id.hw_internal_rev);
 	uint32_t pipe_fuses;
 
 	ctx->dc_bios->regs = &bios_regs;
@@ -975,7 +1024,11 @@ static bool construct(
 		}
 	}
 
-	pool->base.clk_mgr = dce120_clk_mgr_create(ctx);
+	if (is_vg20)
+		pool->base.clk_mgr = dce121_clk_mgr_create(ctx);
+	else
+		pool->base.clk_mgr = dce120_clk_mgr_create(ctx);
+
 	if (pool->base.clk_mgr == NULL) {
 		dm_error("DC: failed to create display clock!\n");
 		BREAK_TO_DEBUGGER();
@@ -1008,14 +1061,14 @@ static bool construct(
 	if (!pool->base.irqs)
 		goto irqs_create_fail;
 
-	/* retrieve valid pipe fuses */
-	if (harvest_enabled)
+	/* VG20: Pipe harvesting enabled, retrieve valid pipe fuses */
+	if (is_vg20)
 		pipe_fuses = read_pipe_fuses(ctx);
 
 	/* index to valid pipe resource */
 	j = 0;
 	for (i = 0; i < pool->base.pipe_count; i++) {
-		if (harvest_enabled) {
+		if (is_vg20) {
 			if ((pipe_fuses & (1 << i)) != 0) {
 				dm_error("DC: skip invalid pipe %d!\n", i);
 				continue;
@@ -1093,10 +1146,24 @@ static bool construct(
 	pool->base.pipe_count = j;
 	pool->base.timing_generator_count = j;
 
-	if (!resource_construct(num_virtual_links, dc, &pool->base,
-			 &res_create_funcs))
+	if (is_vg20)
+		res_funcs = &dce121_res_create_funcs;
+	else
+		res_funcs = &res_create_funcs;
+
+	if (!resource_construct(num_virtual_links, dc, &pool->base, res_funcs))
 		goto res_create_fail;
 
+	/*
+	 * This is a bit of a hack. The xGMI enabled info is used to determine
+	 * if audio and display clocks need to be adjusted with the WAFL link's
+	 * SS info. This is a responsiblity of the clk_mgr. But since MMHUB is
+	 * under hwseq, and the relevant register is in MMHUB, we have to do it
+	 * here.
+	 */
+	if (is_vg20 && dce121_xgmi_enabled(dc->hwseq))
+		dce121_clock_patch_xgmi_ss_info(pool->base.clk_mgr);
+
 	/* Create hardware sequencer */
 	if (!dce120_hw_sequencer_create(dc))
 		goto controller_create_fail;