tegrakernel/kernel/nvidia/drivers/video/tegra/dc/t21x/bandwidth.c

/*
 * bandwidth.c: Functions required for dc bandwidth calculations.
 *
 * Copyright (c) 2010-2019, NVIDIA CORPORATION, All rights reserved.
 *
 * Author: Jon Mayo <jmayo@nvidia.com>
 *
 * This software is licensed under the terms of the GNU General Public
 * License version 2, as published by the Free Software Foundation, and
 * may be copied, distributed, and modified under those terms.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 */

#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/clk.h>
#include <linux/clk/tegra.h>
#include <linux/math64.h>

#include <linux/nvhost.h>
#include <trace/events/display.h>

#include <linux/platform/tegra/latency_allowance.h>
#include <linux/platform/tegra/emc_bwmgr.h>
#include <linux/platform/tegra/tegra_emc.h>
#include <linux/platform/tegra/mc.h>
#include <linux/platform/tegra/bwmgr_mc.h>

#include <uapi/video/tegra_dc_ext.h>
#include "dc.h"
#include "dc_reg.h"
#include "dc_config.h"
#include "dc_priv.h"

static int use_dynamic_emc = 1;

module_param_named(use_dynamic_emc, use_dynamic_emc, int, 0644);

DEFINE_MUTEX(tegra_dcs_total_bw_lock);

/* windows A, B, C for first and second display */
static const enum tegra_la_id la_id_tab[2][DC_N_WINDOWS] = {
	/* first display */
	{
		TEGRA_LA_DISPLAY_0A,
		TEGRA_LA_DISPLAY_0B,
		TEGRA_LA_DISPLAY_0C,
		TEGRA_LA_DISPLAYD,
		TEGRA_LA_DISPLAY_T,
	},
	/* second display */
	{
		TEGRA_LA_DISPLAY_0AB,
		TEGRA_LA_DISPLAY_0BB,
		TEGRA_LA_DISPLAY_0CB,
	},
};

static bool is_internal_win(enum tegra_la_id id)
{
	return ((id == TEGRA_LA_DISPLAY_0A) || (id == TEGRA_LA_DISPLAY_0B) ||
		(id == TEGRA_LA_DISPLAY_0C) || (id == TEGRA_LA_DISPLAYD) ||
		(id == TEGRA_LA_DISPLAY_HC) || (id == TEGRA_LA_DISPLAY_T));
}

static unsigned int num_active_internal_wins(struct tegra_dc *dc)
{
	unsigned int num_active_internal_wins = 0;
	int i = 0;

	for_each_set_bit(i, &dc->valid_windows,
			tegra_dc_get_numof_dispwindows()) {
		struct tegra_dc_win *curr_win = tegra_dc_get_window(dc, i);
		enum tegra_la_id curr_win_la_id =
				la_id_tab[dc->ctrl_num][curr_win->idx];

		if (!is_internal_win(curr_win_la_id))
			continue;

		if (WIN_IS_ENABLED(curr_win))
			num_active_internal_wins++;
	}

	return num_active_internal_wins;
}

static unsigned int num_active_external_wins(struct tegra_dc *dc)
{
	unsigned int num_active_external_wins = 0;
	int i = 0;

	for_each_set_bit(i, &dc->valid_windows,
			tegra_dc_get_numof_dispwindows()) {
		struct tegra_dc_win *curr_win = tegra_dc_get_window(dc, i);
		enum tegra_la_id curr_win_la_id =
				la_id_tab[dc->ctrl_num][curr_win->idx];

		if (is_internal_win(curr_win_la_id))
			continue;

		if (WIN_IS_ENABLED(curr_win))
			num_active_external_wins++;
	}

	return num_active_external_wins;
}

/*
 * Note about fixed point arithmetic:
 * ----------------------------------
 * calc_disp_params(...) contains fixed point values and arithmetic due to the
 * need to use floating point values. All fixed point values have the "_fp" or
 * "_FP" suffix in their name. Functions/values used to convert between real and
 * fixed point values are listed below:
 *    - la_params.fp_factor
 *    - la_params.la_real_to_fp(real_val)
 *    - la_params.la_fp_to_real(fp_val)
 */

#define T12X_LA_BW_DISRUPTION_TIME_EMCCLK_FP			2362000
#define T12X_LA_STATIC_LA_SNAP_ARB_TO_ROW_SRT_EMCCLK_FP	54000
#define T12X_LA_CONS_MEM_EFFICIENCY_FP				500
#define T12X_LA_ROW_SRT_SZ_BYTES	(64 * (T12X_LA_MC_EMEM_NUM_SLOTS + 1))
#define T12X_LA_MC_EMEM_NUM_SLOTS				63
#define T12X_LA_MAX_DRAIN_TIME_USEC				10

/*
 * Function outputs:
 *    - disp_params->thresh_lwm_bytes
 *    - disp_params->spool_up_buffering_adj_bytes
 *    - disp_params->total_dc0_bw
 *    - disp_params->total_dc1_bw
 */
static void calc_disp_params(struct tegra_dc *dc,
				struct tegra_dc_win *w,
				enum tegra_la_id la_id,
				unsigned long emc_freq_hz,
				unsigned int bw_mbps,
				struct dc_to_la_params *disp_params) {
	const struct disp_client *disp_clients_info =
						tegra_la_disp_clients_info;
	struct la_to_dc_params la_params = tegra_get_la_to_dc_params();
	unsigned int bw_mbps_fp = la_params.la_real_to_fp(bw_mbps);
	bool active = WIN_IS_ENABLED(w);
	bool win_rotated = false;
	unsigned int surface_width = 0;
	bool vertical_scaling_enabled = false;
	bool pitch = !WIN_IS_BLOCKLINEAR(w) && !WIN_IS_TILED(w);
	bool planar = tegra_dc_is_yuv_planar(w->fmt);
	bool packed_yuv422 =
			((tegra_dc_fmt(w->fmt) ==
				TEGRA_DC_EXT_FMT_T_U8_Y8__V8_Y8) ||
			(tegra_dc_fmt(w->fmt) ==
				TEGRA_DC_EXT_FMT_T_U8_Y8__V8_Y8_TRUE) ||
			(tegra_dc_fmt(w->fmt) ==
				TEGRA_DC_EXT_FMT_T_Y8_U8__Y8_V8));
	/* all of tegra's YUV formats(420 and 422) fetch 2 bytes per pixel,
	 * but the size reported by tegra_dc_fmt_bpp for the planar version
	 * is of the luma plane's size only. */
	unsigned int bytes_per_pixel = tegra_dc_is_yuv_planar(w->fmt) ?
				2 * tegra_dc_fmt_bpp(w->fmt) / 8 :
				tegra_dc_fmt_bpp(w->fmt) / 8;
	struct tegra_dc_mode mode = dc->mode;
	unsigned int total_h = mode.h_active +
				mode.h_front_porch +
				mode.h_back_porch +
				mode.h_sync_width;
	unsigned int total_v = mode.v_active +
				mode.v_front_porch +
				mode.v_back_porch +
				mode.v_sync_width;
	unsigned int total_screen_area = total_h * total_v;
	unsigned int total_active_area = mode.h_active * mode.v_active;
	unsigned int total_blank_area = total_screen_area - total_active_area;
	unsigned int c1_fp = 0;
	unsigned int c2 = 0;
	unsigned int c3 = 0;
	unsigned int bpp_for_line_buffer_storage_fp = 0;
	unsigned int reqd_buffering_thresh_disp_bytes_fp = 0;
	unsigned int latency_buffering_available_in_reqd_buffering_fp = 0;
	unsigned long emc_freq_khz = emc_freq_hz / 1000;
	unsigned long emc_freq_mhz = emc_freq_khz / 1000;
	unsigned int bw_disruption_time_usec_fp =
					T12X_LA_BW_DISRUPTION_TIME_EMCCLK_FP /
					emc_freq_mhz;
	unsigned int effective_row_srt_sz_bytes_fp =
		min((unsigned long)la_params.la_real_to_fp(min(
			(unsigned long)T12X_LA_ROW_SRT_SZ_BYTES *
			(la_params.dram_width_bits /
			(T12X_LA_MC_EMEM_NUM_SLOTS + 1)),
			(2 * la_params.dram_width_bits / 8) *
			min(emc_freq_mhz + 50, 400ul))),
			(T12X_LA_MAX_DRAIN_TIME_USEC * emc_freq_mhz -
			la_params.la_fp_to_real(
			T12X_LA_STATIC_LA_SNAP_ARB_TO_ROW_SRT_EMCCLK_FP)) *
			2 *
			la_params.dram_width_bits /
			8 *
			T12X_LA_CONS_MEM_EFFICIENCY_FP);
	unsigned long helper_1st =
		(unsigned long)effective_row_srt_sz_bytes_fp *
		la_params.fp_factor;
	unsigned long helper_2nd =
		(unsigned long)(emc_freq_mhz * la_params.dram_width_bits / 4 *
		 T12X_LA_CONS_MEM_EFFICIENCY_FP);
	unsigned long helper_3rd =
		(unsigned long)T12X_LA_STATIC_LA_SNAP_ARB_TO_ROW_SRT_EMCCLK_FP /
		emc_freq_mhz;
	unsigned int drain_time_usec_fp =
		(unsigned int) (helper_1st / helper_2nd + helper_3rd);
	unsigned int total_latency_usec_fp =
		drain_time_usec_fp +
		la_params.static_la_minus_snap_arb_to_row_srt_emcclks_fp /
		emc_freq_mhz;
	unsigned int bw_disruption_buffering_bytes_fp =
					bw_mbps *
					max(bw_disruption_time_usec_fp,
						total_latency_usec_fp) +
					la_params.la_real_to_fp(1)/2;
	unsigned int reqd_lines = 0;
	unsigned int lines_of_latency = 0;
	unsigned int thresh_lwm_bytes = 0;
	unsigned int total_buf_sz_bytes =
	disp_clients_info[DISP_CLIENT_LA_ID(la_id)].line_buf_sz_bytes +
	disp_clients_info[DISP_CLIENT_LA_ID(la_id)].mccif_size_bytes;
	unsigned int num_active_wins_to_use = is_internal_win(la_id) ?
						num_active_internal_wins(dc) :
						num_active_external_wins(dc);
	unsigned int total_active_space_bw = 0;
	unsigned int total_vblank_bw = 0;
	unsigned int bw_other_wins = 0;
	unsigned int bw_display_fp = 0;
	unsigned int bw_delta_fp = 0;
	unsigned int fill_rate_other_wins_fp = 0;
	unsigned int dvfs_time_nsec =
			tegra_get_dvfs_clk_change_latency_nsec(emc_freq_khz);
	unsigned int data_shortfall_other_wins_fp = 0;
	unsigned int duration_usec_fp = 0;
	unsigned int spool_up_buffering_adj_bytes = 0;
	unsigned int curr_dc_head_bw = 0;

	disp_params->drain_time_usec_fp = drain_time_usec_fp;

	if (w->flags & TEGRA_WIN_FLAG_SCAN_COLUMN) {
		win_rotated = true;
		surface_width = dfixed_trunc(w->h);
		vertical_scaling_enabled = (dfixed_trunc(w->w) == w->out_w) ?
					false : true;
	} else {
		win_rotated = false;
		surface_width = dfixed_trunc(w->w);
		vertical_scaling_enabled = (dfixed_trunc(w->h) == w->out_h) ?
					false : true;
	}

	if ((disp_clients_info[DISP_CLIENT_LA_ID(la_id)].line_buf_sz_bytes
									== 0) ||
		(pitch == true)) {
		reqd_lines = 0;
	} else if (win_rotated && planar) {
		if (vertical_scaling_enabled)
			reqd_lines = 17;
		else
			reqd_lines = 16;
	} else {
		if (win_rotated) {
			if (vertical_scaling_enabled)
				reqd_lines = 16 / bytes_per_pixel + 1;
			else
				reqd_lines = 16 / bytes_per_pixel;
		} else {
			if (vertical_scaling_enabled)
				reqd_lines = 3;
			else
				reqd_lines = 1;
		}
	}

	if (reqd_lines > 0 && !vertical_scaling_enabled && win_rotated)
		lines_of_latency = 1;
	else
		lines_of_latency = 0;


	if (((tegra_dc_fmt(w->fmt) ==
		TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N422R) ||
		(tegra_dc_fmt(w->fmt) ==
			TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N422R_TRUE) ||
		(tegra_dc_fmt(w->fmt) ==
			TEGRA_DC_EXT_FMT_T_V8_Y8__U8_Y8) ||
		(tegra_dc_fmt(w->fmt) ==
			TEGRA_DC_EXT_FMT_T_V8_Y8__U8_Y8_TRUE)) &&
		!win_rotated) {
		c1_fp = la_params.la_real_to_fp(5) / 2;
	} else {
		c1_fp = la_params.la_real_to_fp(1);
	}

	if ((((tegra_dc_fmt(w->fmt) ==
			TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N420) ||
		(tegra_dc_fmt(w->fmt) ==
				TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N420_TRUE) ||
		(tegra_dc_fmt(w->fmt) ==
			TEGRA_DC_EXT_FMT_T_Y8___U8V8_N420) ||
		(tegra_dc_fmt(w->fmt) ==
			TEGRA_DC_EXT_FMT_T_Y8___V8U8_N420) ||
		(tegra_dc_fmt(w->fmt) ==
			TEGRA_DC_EXT_FMT_T_Y8___U8V8_N420_TRUE) ||
		(tegra_dc_fmt(w->fmt) ==
			TEGRA_DC_EXT_FMT_T_Y8___V8U8_N420_TRUE)) &&
		!win_rotated) ||
		(tegra_dc_is_yuv(w->fmt) && win_rotated)) {
		c2 = 3;
	} else {
		c2 = bytes_per_pixel;
	}

	c3 = (packed_yuv422 && win_rotated) ? 2 : 1;
	latency_buffering_available_in_reqd_buffering_fp = active *
							surface_width *
							lines_of_latency *
							c1_fp *
							c2 *
							c3;

	switch (tegra_dc_fmt(w->fmt)) {
	/* YUV 420 case*/
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N420:
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N420_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___U8V8_N420:
	case TEGRA_DC_EXT_FMT_T_Y8___V8U8_N420:
	case TEGRA_DC_EXT_FMT_T_Y8___U8V8_N420_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___V8U8_N420_TRUE:
		c1_fp = (win_rotated) ?
			la_params.la_real_to_fp(2) :
			la_params.la_real_to_fp(3);
		break;

	/* YUV 422 case */
	case TEGRA_DC_EXT_FMT_T_U8_Y8__V8_Y8:
	case TEGRA_DC_EXT_FMT_T_Y8_U8__Y8_V8:
	case TEGRA_DC_EXT_FMT_T_U8_Y8__V8_Y8_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N422:
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N422_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___U8V8_N422:
	case TEGRA_DC_EXT_FMT_T_Y8___V8U8_N422:
	case TEGRA_DC_EXT_FMT_T_Y8___U8V8_N422_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___V8U8_N422_TRUE:
		c1_fp = (win_rotated) ?
			la_params.la_real_to_fp(3) :
			la_params.la_real_to_fp(2);
		break;

	/* YUV 422R case */
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N422R:
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N422R_TRUE:
	case TEGRA_DC_EXT_FMT_T_V8_Y8__U8_Y8:
	case TEGRA_DC_EXT_FMT_T_V8_Y8__U8_Y8_TRUE:
		c1_fp = (win_rotated) ?
			la_params.la_real_to_fp(2) :
			la_params.la_real_to_fp(5);
		break;

	/* YUV 444 case */
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N444:
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N444_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___U8V8_N444_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___V8U8_N444_TRUE:
		c1_fp = la_params.la_real_to_fp(3);
		break;

	default:
		c1_fp = la_params.la_real_to_fp(bytes_per_pixel);
		break;
	}

	c2 = (packed_yuv422 && win_rotated) ? 2 : 1;
	bpp_for_line_buffer_storage_fp = c1_fp * c2;
	reqd_buffering_thresh_disp_bytes_fp = active *
						surface_width *
						reqd_lines *
						bpp_for_line_buffer_storage_fp;
	thresh_lwm_bytes =
		la_params.la_fp_to_real(reqd_buffering_thresh_disp_bytes_fp);
	thresh_lwm_bytes +=
		la_params.la_fp_to_real(
			(bw_disruption_buffering_bytes_fp >=
			latency_buffering_available_in_reqd_buffering_fp) ?
			(bw_disruption_buffering_bytes_fp -
			latency_buffering_available_in_reqd_buffering_fp) : 0);
	disp_params->thresh_lwm_bytes = thresh_lwm_bytes;


	if (is_internal_win(la_id)) {
		int i = 0;

		for_each_set_bit(i, &dc->valid_windows,
				tegra_dc_get_numof_dispwindows()) {
			struct tegra_dc_win *curr_win =
				tegra_dc_get_window(dc, i);
			enum tegra_la_id curr_win_la_id =
					la_id_tab[dc->ctrl_num][curr_win->idx];
			unsigned int curr_win_bw = 0;

			if (!is_internal_win(curr_win_la_id))
				continue;

			curr_win_bw = max(curr_win->bandwidth,
						curr_win->new_bandwidth);
			/* our bandwidth is in kbytes/sec, but LA takes MBps.
			 * round up bandwidth to next 1MBps */
			if (curr_win_bw != UINT_MAX)
				curr_win_bw = curr_win_bw / 1000 + 1;

			total_active_space_bw += curr_win_bw;
		}
	} else {
		int i = 0;

		for_each_set_bit(i, &dc->valid_windows,
				tegra_dc_get_numof_dispwindows()) {
			struct tegra_dc_win *curr_win =
				tegra_dc_get_window(dc, i);
			enum tegra_la_id curr_win_la_id =
					la_id_tab[dc->ctrl_num][curr_win->idx];
			unsigned int curr_win_bw = 0;

			if (is_internal_win(curr_win_la_id))
				continue;

			curr_win_bw = max(curr_win->bandwidth,
						curr_win->new_bandwidth);
			/* our bandwidth is in kbytes/sec, but LA takes MBps.
			 * round up bandwidth to next 1MBps */
			if (curr_win_bw != UINT_MAX)
				curr_win_bw = curr_win_bw / 1000 + 1;

			total_active_space_bw += curr_win_bw;
		}
	}


	if ((disp_clients_info[DISP_CLIENT_LA_ID(la_id)].win_type  ==
						TEGRA_LA_DISP_WIN_TYPE_FULL) ||
		(disp_clients_info[DISP_CLIENT_LA_ID(la_id)].win_type  ==
						TEGRA_LA_DISP_WIN_TYPE_FULLA) ||
		(disp_clients_info[DISP_CLIENT_LA_ID(la_id)].win_type  ==
						TEGRA_LA_DISP_WIN_TYPE_FULLB)) {
		total_vblank_bw = total_buf_sz_bytes / total_blank_area;
	} else {
		total_vblank_bw = 0;
	}


	bw_display_fp = la_params.disp_catchup_factor_fp *
			max(total_active_space_bw,
				total_vblank_bw);
	if (active)
		bw_delta_fp = bw_mbps_fp -
				(bw_display_fp /
				num_active_wins_to_use);


	bw_other_wins = total_active_space_bw - bw_mbps;

	if (num_active_wins_to_use > 0) {
		fill_rate_other_wins_fp =
				bw_display_fp *
				(num_active_wins_to_use - active) /
				num_active_wins_to_use -
				la_params.la_real_to_fp(bw_other_wins);
	} else {
		fill_rate_other_wins_fp = 0;
	}

	data_shortfall_other_wins_fp = dvfs_time_nsec *
					bw_other_wins *
					la_params.fp_factor /
					1000;

	duration_usec_fp = (fill_rate_other_wins_fp == 0) ? 0 :
				data_shortfall_other_wins_fp *
				la_params.fp_factor /
				fill_rate_other_wins_fp;


	spool_up_buffering_adj_bytes = (bw_delta_fp > 0) ?
					(bw_delta_fp *
					duration_usec_fp /
					(la_params.fp_factor *
					la_params.fp_factor)) :
					0;
	disp_params->spool_up_buffering_adj_bytes =
						spool_up_buffering_adj_bytes;

	mutex_lock(&tegra_dcs_total_bw_lock);
	curr_dc_head_bw = max(dc->new_bw_kbps, dc->bw_kbps);
	/* our bandwidth is in kbytes/sec, but LA takes MBps.
	 * round up bandwidth to next 1MBps */
	if (curr_dc_head_bw != ULONG_MAX)
		curr_dc_head_bw = curr_dc_head_bw / 1000 + 1;

	if (dc->ctrl_num)
		disp_params->total_dc1_bw = curr_dc_head_bw;
	else
		disp_params->total_dc0_bw = curr_dc_head_bw;

	mutex_unlock(&tegra_dcs_total_bw_lock);
}

/*
 * tegra_dc_process_bandwidth_renegotiate() is only called in code
 * sections wrapped by CONFIG_TEGRA_ISOMGR.  Thus it is also wrapped
 * to avoid "defined but not used" compiler error.
 */
#ifdef CONFIG_TEGRA_ISOMGR
static void tegra_dc_process_bandwidth_renegotiate(struct tegra_dc *dc,
						struct tegra_dc_bw_data *bw)
{
	tegra_dc_ext_process_bandwidth_renegotiate(dc->ctrl_num, bw);
}
#endif

/* uses the larger of w->bandwidth or w->new_bandwidth */
static int tegra_dc_handle_latency_allowance(struct tegra_dc *dc,
	struct tegra_dc_win *w, int set_la)
{
	int ret = 0;
	unsigned long bw;
	struct dc_to_la_params disp_params;
	struct clk *dram_clk = NULL;
	unsigned long emc_freq_hz = 0;
	unsigned long dram_freq_hz = 0;

	BUG_ON(dc->ctrl_num >= ARRAY_SIZE(la_id_tab));
	BUG_ON(w->idx >= ARRAY_SIZE(*la_id_tab));
	BUG_ON(w->dc->ndev->id >= ARRAY_SIZE(la_id_tab));

	bw = max(w->bandwidth, w->new_bandwidth);

	/* our bandwidth is in kbytes/sec, but LA takes MBps.
	 * round up bandwidth to next 1MBps */
	if (bw != ULONG_MAX)
		bw = bw / 1000 + 1;

	/* use clk_round_rate on root dram clock instead to get correct rate */
	dram_clk = clk_get_sys("tegra_emc", "emc");
	dram_freq_hz = set_la ?
		tegra_emc_bw_to_freq_req(bw * 1000000) : UINT_MAX;
	/* clk_round_rate returns the next dram frequency */
	dram_freq_hz = clk_round_rate(dram_clk, dram_freq_hz);
	emc_freq_hz = dram_freq_hz / bwmgr_get_emc_to_dram_freq_factor();

	while (1) {
		int err;
		unsigned long next_emc_freq_hz = 0;
		calc_disp_params(dc, w,
				la_id_tab[dc->ctrl_num][w->idx],
				emc_freq_hz, bw, &disp_params);

		if (!set_la) {
			err = tegra_check_disp_latency_allowance(
				la_id_tab[dc->ndev->id][w->idx],
				emc_freq_hz, bw, disp_params);
			ret = err;
			break;
		}

		err = tegra_set_disp_latency_allowance(
			la_id_tab[dc->ndev->id][w->idx],
			emc_freq_hz, bw, disp_params);

		if (!err) {
			/* tegra_bwmgr_set_emc sets dram frequency */
			tegra_bwmgr_set_emc(dc->emc_la_handle,
				dram_freq_hz, TEGRA_BWMGR_SET_EMC_FLOOR);
			break;
		}

		dram_freq_hz = clk_round_rate(dram_clk, dram_freq_hz + 1000000);

		next_emc_freq_hz = dram_freq_hz /
			bwmgr_get_emc_to_dram_freq_factor();

		if (emc_freq_hz == next_emc_freq_hz) {
			ret = err;
			break;
		}
		emc_freq_hz = next_emc_freq_hz;

	}

	clk_put(dram_clk);

	return ret;
}

static int tegra_dc_set_latency_allowance(struct tegra_dc *dc,
	struct tegra_dc_win *w)
{
	return tegra_dc_handle_latency_allowance(dc, w, 1);
}

#ifdef CONFIG_TEGRA_ISOMGR
static int tegra_dc_check_latency_allowance(struct tegra_dc *dc,
	struct tegra_dc_win *w)
{
	return tegra_dc_handle_latency_allowance(dc, w, 0);
}
#endif

/* determine if a row is within a window */
static int tegra_dc_line_in_window(int line, struct tegra_dc_win *win)
{
	int win_first = win->out_y;
	int win_last = win_first + win->out_h - 1;

	return (line >= win_first && line <= win_last);
}

/* check overlapping window combinations to find the max bandwidth. */
static unsigned long tegra_dc_find_max_bandwidth(struct tegra_dc_win *wins[],
						 unsigned n)
{
	unsigned i;
	unsigned j;
	long max = 0;

	for (i = 0; i < n; i++) {
		struct tegra_dc_win *a = wins[i];
		long bw = 0;
		int a_first = a->out_y;

		if (!WIN_IS_ENABLED(a))
			continue;
		for (j = 0; j < n; j++) {
			struct tegra_dc_win *b = wins[j];

			if (!WIN_IS_ENABLED(b))
				continue;
			if (tegra_dc_line_in_window(a_first, b))
				bw += b->new_bandwidth;
		}
		if (max < bw)
			max = bw;
	}
	return max;
}

static inline int tegra_dc_is_yuv420(int fmt)
{
	switch (fmt) {
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N420:
	case TEGRA_DC_EXT_FMT_T_Y8___U8___V8_N420_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___U8V8_N420:
	case TEGRA_DC_EXT_FMT_T_Y8___V8U8_N420:
	case TEGRA_DC_EXT_FMT_T_Y8___U8V8_N420_TRUE:
	case TEGRA_DC_EXT_FMT_T_Y8___V8U8_N420_TRUE:
		return 1;
	default:
		return 0;
	}
}

/*
 * Calculate peak EMC bandwidth for each enabled window =
 * pixel_clock * win_bpp * (use_v_filter ? 2 : 1)) * H_scale_factor *
 * (windows_tiling ? 2 : 1)
 *
 * note:
 * (*) We use 2 tap V filter on T2x/T3x, so need double BW if use V filter
 * (*) Tiling mode on T30 and DDR3 requires double BW
 *
 * return:
 * bandwidth in kBps
 */
static unsigned long tegra_dc_calc_win_bandwidth(struct tegra_dc *dc,
	struct tegra_dc_win *w)
{
	u64 ret;
	int tiled_windows_bw_multiplier;
	unsigned long bpp;
	unsigned in_w;
	unsigned out_w;
	unsigned in_h;
	unsigned out_h;

	if (!dc || !WIN_IS_ENABLED(w))
		return 0;

	if (dfixed_trunc(w->w) == 0 || dfixed_trunc(w->h) == 0 ||
	    w->out_w == 0 || w->out_h == 0)
		return 0;
	if (w->flags & TEGRA_WIN_FLAG_SCAN_COLUMN) {
		/* rotated: PRESCALE_SIZE swapped, but WIN_SIZE is unchanged */
		in_w = dfixed_trunc(w->h);
		out_w = w->out_h;
		in_h = dfixed_trunc(w->w);
		out_h = w->out_w;
	} else {
		in_w = dfixed_trunc(w->w); /* normal output, not rotated */
		out_w = w->out_w;
		in_h = dfixed_trunc(w->h);
		out_h = w->out_h;
	}

	tiled_windows_bw_multiplier =
		tegra_mc_get_tiled_memory_bandwidth_multiplier();

	/* all of tegra's YUV formats(420 and 422) fetch 2 bytes per pixel,
	 * but the size reported by tegra_dc_fmt_bpp for the planar version
	 * is of the luma plane's size only. */
	bpp = tegra_dc_is_yuv_planar(w->fmt) ?
		2 * tegra_dc_fmt_bpp(w->fmt) : tegra_dc_fmt_bpp(w->fmt);
	if (tegra_dc_is_yuv420(w->fmt))
		bpp = 16;

	ret = (dc->mode.pclk / 1000UL) * (bpp / 8);
	ret *= in_w;
	ret = div_u64(ret, out_w * (WIN_IS_TILED(w) ?
		      tiled_windows_bw_multiplier : 1));

	if (in_h > out_h) {
		/* vertical downscaling enabled  */
		ret *= in_h;
		ret = div_u64(ret, out_h);
	}

	return ret;
}

unsigned long tegra_dc_get_bandwidth(
	struct tegra_dc_win *windows[], int n)
{
	int i;

	if (tegra_dc_is_nvdisplay())
		return -EINVAL;

	BUG_ON(n > tegra_dc_get_numof_dispwindows());

	/* emc rate and latency allowance both need to know per window
	 * bandwidths */
	for (i = 0; i < n; i++) {
		struct tegra_dc_win *w = windows[i];

		if (w)
			w->new_bandwidth =
				tegra_dc_calc_win_bandwidth(w->dc, w);
	}

	return tegra_dc_find_max_bandwidth(windows, n);
}
EXPORT_SYMBOL(tegra_dc_get_bandwidth);

#ifdef CONFIG_TEGRA_ISOMGR
/* to save power, call when display memory clients would be idle */
void tegra_dc_clear_bandwidth(struct tegra_dc *dc)
{
	int latency;

	if (tegra_dc_is_nvdisplay())
		return;

	trace_clear_bandwidth(dc);
	latency = tegra_isomgr_reserve(dc->isomgr_handle, 0, 1000);
	if (latency) {
		dc->reserved_bw = 0;
		latency = tegra_isomgr_realize(dc->isomgr_handle);
		WARN_ONCE(!latency, "tegra_isomgr_realize failed\n");
	} else {
		dev_dbg(&dc->ndev->dev, "Failed to clear bw.\n");
		tegra_dc_process_bandwidth_renegotiate(dc, NULL);
	}
	dc->bw_kbps = 0;
}
#else
/* to save power, call when display memory clients would be idle */
void tegra_dc_clear_bandwidth(struct tegra_dc *dc)
{
	if (tegra_dc_is_nvdisplay())
		return;

	trace_clear_bandwidth(dc);
	if (tegra_dc_is_clk_enabled(dc->emc_clk))
		tegra_disp_clk_disable_unprepare(dc->emc_clk);
	dc->bw_kbps = 0;
}

/* bw in kByte/second. returns Hz for EMC frequency */
static inline unsigned long tegra_dc_kbps_to_emc(unsigned long bw)
{
	unsigned long freq;

	if (bw == ULONG_MAX)
		return ULONG_MAX;

	freq = tegra_emc_bw_to_freq_req(bw);
	if (freq >= (ULONG_MAX / 1000))
		return ULONG_MAX; /* freq too big - clamp at max */

	if (WARN_ONCE((freq * 1000) < freq, "Bandwidth Overflow"))
		return ULONG_MAX; /* should never occur because of above. */
	return freq * 1000;
}
#endif

/* use the larger of dc->bw_kbps or dc->new_bw_kbps, and copies
 * dc->new_bw_kbps into dc->bw_kbps.
 * calling this function both before and after a flip is sufficient to select
 * the best possible frequency and latency allowance.
 * set use_new to true to force dc->new_bw_kbps programming.
 */
void tegra_dc_program_bandwidth(struct tegra_dc *dc, bool use_new)
{
	unsigned i;

	if (tegra_dc_is_nvdisplay())
		return;

	if (!dc->enabled)
		return;

	if (!tegra_platform_is_silicon())
		return;

	if (use_new || dc->bw_kbps != dc->new_bw_kbps) {
		long bw = max(dc->bw_kbps, dc->new_bw_kbps);

#ifdef CONFIG_TEGRA_ISOMGR
		int latency;

		if (!dc->isomgr_handle)
			return;

		/* reserve atleast the minimum bandwidth. */
		bw = max(bw, tegra_calc_min_bandwidth(dc));
		latency = tegra_isomgr_reserve(dc->isomgr_handle, bw, 1000);
		if (latency) {
			dc->reserved_bw = bw;
			latency = tegra_isomgr_realize(dc->isomgr_handle);
			WARN_ONCE(!latency, "tegra_isomgr_realize failed\n");
		} else {
			dev_dbg(&dc->ndev->dev, "Failed to reserve bw %ld.\n",
									bw);
			tegra_dc_process_bandwidth_renegotiate(dc, NULL);
		}
#else /* EMC version */
		int emc_freq;

		/* going from 0 to non-zero */
		if (!dc->bw_kbps && dc->new_bw_kbps &&
			!tegra_dc_is_clk_enabled(dc->emc_clk))
			tegra_disp_clk_prepare_enable(dc->emc_clk);

		emc_freq = tegra_dc_kbps_to_emc(bw);
		clk_set_rate(dc->emc_clk, emc_freq);

		/* going from non-zero to 0 */
		if (dc->bw_kbps && !dc->new_bw_kbps &&
			tegra_dc_is_clk_enabled(dc->emc_clk))
			tegra_disp_clk_disable_unprepare(dc->emc_clk);
#endif
		dc->bw_kbps = dc->new_bw_kbps;
	}

	for_each_set_bit(i, &dc->valid_windows,
			tegra_dc_get_numof_dispwindows()) {
		struct tegra_dc_win *w = tegra_dc_get_window(dc, i);

		if ((use_new || w->bandwidth != w->new_bandwidth) &&
			w->new_bandwidth != 0)
			tegra_dc_set_latency_allowance(dc, w);
		trace_program_bandwidth(dc);
		w->bandwidth = w->new_bandwidth;
	}
}

int tegra_dc_set_dynamic_emc(struct tegra_dc *dc)
{
	unsigned long new_rate;
	int nwins = tegra_dc_get_numof_dispwindows();
	struct tegra_dc_win *windows[nwins];
	unsigned i;
	unsigned len;

	if (!use_dynamic_emc)
		return 0;

	for (i = 0, len = 0; i < tegra_dc_get_numof_dispwindows(); i++) {
		struct tegra_dc_win *win = tegra_dc_get_window(dc, i);
		if (win)
			windows[len++] = win;
	}
#ifdef CONFIG_TEGRA_ISOMGR
	new_rate = tegra_dc_get_bandwidth(windows, len);
#else
	if (tegra_dc_has_multiple_dc())
		new_rate = ULONG_MAX;
	else
		new_rate = tegra_dc_get_bandwidth(windows, len);
#endif

	dc->new_bw_kbps = new_rate;
	trace_set_dynamic_emc(dc);

	return 0;
}

/* return the minimum bandwidth in kbps for display to function */
long tegra_calc_min_bandwidth(struct tegra_dc *dc)
{
	unsigned  pclk;

	if (WARN_ONCE(!dc, "dc is NULL") ||
		WARN_ONCE(!dc->out, "dc->out is NULL!"))
		return 0;

	pclk = tegra_dc_get_out_max_pixclock(dc);
	if (!pclk) {
		 if (dc->out->type == TEGRA_DC_OUT_HDMI) {
			if (tegra_dc_is_t21x())
				pclk = KHZ2PICOS(600000); /* 600MHz max */
			else
				pclk = KHZ2PICOS(150000); /* 150MHz max */
		} else if ((dc->out->type == TEGRA_DC_OUT_DP) ||
			(dc->out->type == TEGRA_DC_OUT_FAKE_DP) ||
			(dc->out->type == TEGRA_DC_OUT_NULL)) {
			if (dc->mode.pclk)
				pclk = KHZ2PICOS(dc->mode.pclk / 1000);
			else
				pclk = KHZ2PICOS(25200); /* vga */
		} else {
			if (!WARN_ONCE(!dc->mode.pclk,
				"pclk is not set, bandwidth calc cannot work"))
				pclk = KHZ2PICOS(dc->mode.pclk / 1000);
			else
				return 0;
		}
	}

	return PICOS2KHZ(pclk) * 4; /* support a single 32bpp window */
}

#ifdef CONFIG_TEGRA_ISOMGR
int tegra_dc_bandwidth_negotiate_bw(struct tegra_dc *dc,
			struct tegra_dc_win *windows[], int n)
{
	int i;
	u32 bw;
	int err;
	int latency;

	if (tegra_dc_is_nvdisplay())
		return 0;

	mutex_lock(&dc->lock);
	/*
	 * isomgr will update available bandwidth through a callback.
	 * If available bandwidth is less than proposed bw fail the ioctl.
	 * If proposed bw is larger than reserved bw, make it in effect
	 * immediately. Otherwise, bandwidth will be adjusted in flips.
	 */
	bw = tegra_dc_get_bandwidth(windows, n);
	if (bw > dc->available_bw) {
		mutex_unlock(&dc->lock);
		return -1;
	} else if (bw <= dc->reserved_bw) {
		mutex_unlock(&dc->lock);
		return 0;
	}

	latency = tegra_isomgr_reserve(dc->isomgr_handle, bw, 1000);
	if (!latency) {
		dev_dbg(&dc->ndev->dev, "Failed to reserve proposed bw %d.\n",
									bw);
		mutex_unlock(&dc->lock);
		return -1;
	}

	/*
	 * Only realize if bw required > reserved bw
	 * If realized, update dc->bw_kbps
	 */
	if (dc->reserved_bw < bw) {
		dc->reserved_bw = bw;
		latency = tegra_isomgr_realize(dc->isomgr_handle);
		if (!latency) {
			WARN_ONCE(!latency, "tegra_isomgr_realize failed\n");
			mutex_unlock(&dc->lock);
			return -1;
		}
		dc->bw_kbps = bw;
	}

	for (i = 0; i < n; i++) {
		struct tegra_dc_win *w = windows[i];

		if ((w->bandwidth != w->new_bandwidth) &&
			w->new_bandwidth != 0) {
			err = tegra_dc_check_latency_allowance(dc, w);
			if (err) {
				WARN_ONCE(!latency, "tegra_dc_check_latency_allowance failed\n");
				mutex_unlock(&dc->lock);
				return -1;
			}
		}
	}

	mutex_unlock(&dc->lock);

	return 0;
}
EXPORT_SYMBOL(tegra_dc_bandwidth_negotiate_bw);

void tegra_dc_bandwidth_renegotiate(void *p, u32 avail_bw)
{
	struct tegra_dc_bw_data data;
	struct tegra_dc *dc = p;

	if (WARN_ONCE(!dc, "dc is NULL!"))
		return;

	if (dc->available_bw == avail_bw)
		return;

	/* tegra_isomgr_get_total_iso_bw() uses the iso_client_id passed only
	 * for t19x. For pre-t19x the function argument passed is ignored.
	 * bandwidth.c is only used on T21x. On T186/T194, we always use
	 * TEGRA_ISO_CLIENT_DISP_0 since there's only one display isoclient,
	 * but this isn't the case for T21x. For T21x, there are per-DC
	 * isoclients, and each one has its own corresponding id.
	 */
	data.total_bw = tegra_isomgr_get_total_iso_bw(TEGRA_ISO_CLIENT_DISP_0);
	data.avail_bw = avail_bw;
	data.resvd_bw = dc->reserved_bw;

	tegra_dc_process_bandwidth_renegotiate(dc, &data);

	mutex_lock(&dc->lock);
	dc->available_bw = avail_bw;
	mutex_unlock(&dc->lock);
}
#endif