/*
 * Copyright © 2012 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 * Authors:
 *    Eugeni Dodonov <eugeni.dodonov@intel.com>
 *
 */

#include <linux/module.h>
#include <linux/pm_runtime.h>

#include <drm/drm_atomic_helper.h>
#include <drm/drm_fourcc.h>
#include <drm/drm_plane_helper.h>

#include "display/intel_atomic.h"
#include "display/intel_display_types.h"
#include "display/intel_fbc.h"
#include "display/intel_sprite.h"

#include "gt/intel_llc.h"

#include "i915_drv.h"
#include "i915_irq.h"
#include "i915_trace.h"
#include "intel_pm.h"
#include "intel_sideband.h"
#include "../../../platform/x86/intel_ips.h"

static void gen9_init_clock_gating(struct drm_i915_private *dev_priv)
{
	if (HAS_LLC(dev_priv)) {
		/*
		 * WaCompressedResourceDisplayNewHashMode:skl,kbl
		 * Display WA #0390: skl,kbl
		 *
		 * Must match Sampler, Pixel Back End, and Media. See
		 * WaCompressedResourceSamplerPbeMediaNewHashMode.
		 */
		I915_WRITE(CHICKEN_PAR1_1,
			   I915_READ(CHICKEN_PAR1_1) |
			   SKL_DE_COMPRESSED_HASH_MODE);
	}

	/* See Bspec note for PSR2_CTL bit 31, Wa#828:skl,bxt,kbl,cfl */
	I915_WRITE(CHICKEN_PAR1_1,
		   I915_READ(CHICKEN_PAR1_1) | SKL_EDP_PSR_FIX_RDWRAP);

	/* WaEnableChickenDCPR:skl,bxt,kbl,glk,cfl */
	I915_WRITE(GEN8_CHICKEN_DCPR_1,
		   I915_READ(GEN8_CHICKEN_DCPR_1) | MASK_WAKEMEM);

	/* WaFbcTurnOffFbcWatermark:skl,bxt,kbl,cfl */
	/* WaFbcWakeMemOn:skl,bxt,kbl,glk,cfl */
	I915_WRITE(DISP_ARB_CTL, I915_READ(DISP_ARB_CTL) |
		   DISP_FBC_WM_DIS |
		   DISP_FBC_MEMORY_WAKE);

	/* WaFbcHighMemBwCorruptionAvoidance:skl,bxt,kbl,cfl */
	I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) |
		   ILK_DPFC_DISABLE_DUMMY0);

	if (IS_SKYLAKE(dev_priv)) {
		/* WaDisableDopClockGating */
		I915_WRITE(GEN7_MISCCPCTL, I915_READ(GEN7_MISCCPCTL)
			   & ~GEN7_DOP_CLOCK_GATE_ENABLE);
	}
}

static void bxt_init_clock_gating(struct drm_i915_private *dev_priv)
{
	gen9_init_clock_gating(dev_priv);

	/* WaDisableSDEUnitClockGating:bxt */
	I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
		   GEN8_SDEUNIT_CLOCK_GATE_DISABLE);

	/*
	 * FIXME:
	 * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only.
	 */
	I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
		   GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ);

	/*
	 * Wa: Backlight PWM may stop in the asserted state, causing backlight
	 * to stay fully on.
	 */
	I915_WRITE(GEN9_CLKGATE_DIS_0, I915_READ(GEN9_CLKGATE_DIS_0) |
		   PWM1_GATING_DIS | PWM2_GATING_DIS);

	/*
	 * Lower the display internal timeout.
	 * This is needed to avoid any hard hangs when DSI port PLL
	 * is off and a MMIO access is attempted by any privilege
	 * application, using batch buffers or any other means.
	 */
	I915_WRITE(RM_TIMEOUT, MMIO_TIMEOUT_US(950));
}

static void glk_init_clock_gating(struct drm_i915_private *dev_priv)
{
	gen9_init_clock_gating(dev_priv);

	/*
	 * WaDisablePWMClockGating:glk
	 * Backlight PWM may stop in the asserted state, causing backlight
	 * to stay fully on.
	 */
	I915_WRITE(GEN9_CLKGATE_DIS_0, I915_READ(GEN9_CLKGATE_DIS_0) |
		   PWM1_GATING_DIS | PWM2_GATING_DIS);

	/* WaDDIIOTimeout:glk */
	if (IS_GLK_REVID(dev_priv, 0, GLK_REVID_A1)) {
		u32 val = I915_READ(CHICKEN_MISC_2);
		val &= ~(GLK_CL0_PWR_DOWN |
			 GLK_CL1_PWR_DOWN |
			 GLK_CL2_PWR_DOWN);
		I915_WRITE(CHICKEN_MISC_2, val);
	}

}

static void i915_pineview_get_mem_freq(struct drm_i915_private *dev_priv)
{
	u32 tmp;

	tmp = I915_READ(CLKCFG);

	switch (tmp & CLKCFG_FSB_MASK) {
	case CLKCFG_FSB_533:
		dev_priv->fsb_freq = 533; /* 133*4 */
		break;
	case CLKCFG_FSB_800:
		dev_priv->fsb_freq = 800; /* 200*4 */
		break;
	case CLKCFG_FSB_667:
		dev_priv->fsb_freq =  667; /* 167*4 */
		break;
	case CLKCFG_FSB_400:
		dev_priv->fsb_freq = 400; /* 100*4 */
		break;
	}

	switch (tmp & CLKCFG_MEM_MASK) {
	case CLKCFG_MEM_533:
		dev_priv->mem_freq = 533;
		break;
	case CLKCFG_MEM_667:
		dev_priv->mem_freq = 667;
		break;
	case CLKCFG_MEM_800:
		dev_priv->mem_freq = 800;
		break;
	}

	/* detect pineview DDR3 setting */
	tmp = I915_READ(CSHRDDR3CTL);
	dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
}

static void i915_ironlake_get_mem_freq(struct drm_i915_private *dev_priv)
{
	u16 ddrpll, csipll;

	ddrpll = intel_uncore_read16(&dev_priv->uncore, DDRMPLL1);
	csipll = intel_uncore_read16(&dev_priv->uncore, CSIPLL0);

	switch (ddrpll & 0xff) {
	case 0xc:
		dev_priv->mem_freq = 800;
		break;
	case 0x10:
		dev_priv->mem_freq = 1066;
		break;
	case 0x14:
		dev_priv->mem_freq = 1333;
		break;
	case 0x18:
		dev_priv->mem_freq = 1600;
		break;
	default:
		DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
				 ddrpll & 0xff);
		dev_priv->mem_freq = 0;
		break;
	}

	switch (csipll & 0x3ff) {
	case 0x00c:
		dev_priv->fsb_freq = 3200;
		break;
	case 0x00e:
		dev_priv->fsb_freq = 3733;
		break;
	case 0x010:
		dev_priv->fsb_freq = 4266;
		break;
	case 0x012:
		dev_priv->fsb_freq = 4800;
		break;
	case 0x014:
		dev_priv->fsb_freq = 5333;
		break;
	case 0x016:
		dev_priv->fsb_freq = 5866;
		break;
	case 0x018:
		dev_priv->fsb_freq = 6400;
		break;
	default:
		DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
				 csipll & 0x3ff);
		dev_priv->fsb_freq = 0;
		break;
	}
}

static const struct cxsr_latency cxsr_latency_table[] = {
	{1, 0, 800, 400, 3382, 33382, 3983, 33983},    /* DDR2-400 SC */
	{1, 0, 800, 667, 3354, 33354, 3807, 33807},    /* DDR2-667 SC */
	{1, 0, 800, 800, 3347, 33347, 3763, 33763},    /* DDR2-800 SC */
	{1, 1, 800, 667, 6420, 36420, 6873, 36873},    /* DDR3-667 SC */
	{1, 1, 800, 800, 5902, 35902, 6318, 36318},    /* DDR3-800 SC */

	{1, 0, 667, 400, 3400, 33400, 4021, 34021},    /* DDR2-400 SC */
	{1, 0, 667, 667, 3372, 33372, 3845, 33845},    /* DDR2-667 SC */
	{1, 0, 667, 800, 3386, 33386, 3822, 33822},    /* DDR2-800 SC */
	{1, 1, 667, 667, 6438, 36438, 6911, 36911},    /* DDR3-667 SC */
	{1, 1, 667, 800, 5941, 35941, 6377, 36377},    /* DDR3-800 SC */

	{1, 0, 400, 400, 3472, 33472, 4173, 34173},    /* DDR2-400 SC */
	{1, 0, 400, 667, 3443, 33443, 3996, 33996},    /* DDR2-667 SC */
	{1, 0, 400, 800, 3430, 33430, 3946, 33946},    /* DDR2-800 SC */
	{1, 1, 400, 667, 6509, 36509, 7062, 37062},    /* DDR3-667 SC */
	{1, 1, 400, 800, 5985, 35985, 6501, 36501},    /* DDR3-800 SC */

	{0, 0, 800, 400, 3438, 33438, 4065, 34065},    /* DDR2-400 SC */
	{0, 0, 800, 667, 3410, 33410, 3889, 33889},    /* DDR2-667 SC */
	{0, 0, 800, 800, 3403, 33403, 3845, 33845},    /* DDR2-800 SC */
	{0, 1, 800, 667, 6476, 36476, 6955, 36955},    /* DDR3-667 SC */
	{0, 1, 800, 800, 5958, 35958, 6400, 36400},    /* DDR3-800 SC */

	{0, 0, 667, 400, 3456, 33456, 4103, 34106},    /* DDR2-400 SC */
	{0, 0, 667, 667, 3428, 33428, 3927, 33927},    /* DDR2-667 SC */
	{0, 0, 667, 800, 3443, 33443, 3905, 33905},    /* DDR2-800 SC */
	{0, 1, 667, 667, 6494, 36494, 6993, 36993},    /* DDR3-667 SC */
	{0, 1, 667, 800, 5998, 35998, 6460, 36460},    /* DDR3-800 SC */

	{0, 0, 400, 400, 3528, 33528, 4255, 34255},    /* DDR2-400 SC */
	{0, 0, 400, 667, 3500, 33500, 4079, 34079},    /* DDR2-667 SC */
	{0, 0, 400, 800, 3487, 33487, 4029, 34029},    /* DDR2-800 SC */
	{0, 1, 400, 667, 6566, 36566, 7145, 37145},    /* DDR3-667 SC */
	{0, 1, 400, 800, 6042, 36042, 6584, 36584},    /* DDR3-800 SC */
};

static const struct cxsr_latency *intel_get_cxsr_latency(bool is_desktop,
							 bool is_ddr3,
							 int fsb,
							 int mem)
{
	const struct cxsr_latency *latency;
	int i;

	if (fsb == 0 || mem == 0)
		return NULL;

	for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
		latency = &cxsr_latency_table[i];
		if (is_desktop == latency->is_desktop &&
		    is_ddr3 == latency->is_ddr3 &&
		    fsb == latency->fsb_freq && mem == latency->mem_freq)
			return latency;
	}

	DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");

	return NULL;
}

static void chv_set_memory_dvfs(struct drm_i915_private *dev_priv, bool enable)
{
	u32 val;

	vlv_punit_get(dev_priv);

	val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
	if (enable)
		val &= ~FORCE_DDR_HIGH_FREQ;
	else
		val |= FORCE_DDR_HIGH_FREQ;
	val &= ~FORCE_DDR_LOW_FREQ;
	val |= FORCE_DDR_FREQ_REQ_ACK;
	vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val);

	if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) &
		      FORCE_DDR_FREQ_REQ_ACK) == 0, 3))
		DRM_ERROR("timed out waiting for Punit DDR DVFS request\n");

	vlv_punit_put(dev_priv);
}

static void chv_set_memory_pm5(struct drm_i915_private *dev_priv, bool enable)
{
	u32 val;

	vlv_punit_get(dev_priv);

	val = vlv_punit_read(dev_priv, PUNIT_REG_DSPSSPM);
	if (enable)
		val |= DSP_MAXFIFO_PM5_ENABLE;
	else
		val &= ~DSP_MAXFIFO_PM5_ENABLE;
	vlv_punit_write(dev_priv, PUNIT_REG_DSPSSPM, val);

	vlv_punit_put(dev_priv);
}

#define FW_WM(value, plane) \
	(((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK)

static bool _intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable)
{
	bool was_enabled;
	u32 val;

	if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
		was_enabled = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN;
		I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0);
		POSTING_READ(FW_BLC_SELF_VLV);
	} else if (IS_G4X(dev_priv) || IS_I965GM(dev_priv)) {
		was_enabled = I915_READ(FW_BLC_SELF) & FW_BLC_SELF_EN;
		I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0);
		POSTING_READ(FW_BLC_SELF);
	} else if (IS_PINEVIEW(dev_priv)) {
		val = I915_READ(DSPFW3);
		was_enabled = val & PINEVIEW_SELF_REFRESH_EN;
		if (enable)
			val |= PINEVIEW_SELF_REFRESH_EN;
		else
			val &= ~PINEVIEW_SELF_REFRESH_EN;
		I915_WRITE(DSPFW3, val);
		POSTING_READ(DSPFW3);
	} else if (IS_I945G(dev_priv) || IS_I945GM(dev_priv)) {
		was_enabled = I915_READ(FW_BLC_SELF) & FW_BLC_SELF_EN;
		val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) :
			       _MASKED_BIT_DISABLE(FW_BLC_SELF_EN);
		I915_WRITE(FW_BLC_SELF, val);
		POSTING_READ(FW_BLC_SELF);
	} else if (IS_I915GM(dev_priv)) {
		/*
		 * FIXME can't find a bit like this for 915G, and
		 * and yet it does have the related watermark in
		 * FW_BLC_SELF. What's going on?
		 */
		was_enabled = I915_READ(INSTPM) & INSTPM_SELF_EN;
		val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) :
			       _MASKED_BIT_DISABLE(INSTPM_SELF_EN);
		I915_WRITE(INSTPM, val);
		POSTING_READ(INSTPM);
	} else {
		return false;
	}

	trace_intel_memory_cxsr(dev_priv, was_enabled, enable);

	DRM_DEBUG_KMS("memory self-refresh is %s (was %s)\n",
		      enableddisabled(enable),
		      enableddisabled(was_enabled));

	return was_enabled;
}

/**
 * intel_set_memory_cxsr - Configure CxSR state
 * @dev_priv: i915 device
 * @enable: Allow vs. disallow CxSR
 *
 * Allow or disallow the system to enter a special CxSR
 * (C-state self refresh) state. What typically happens in CxSR mode
 * is that several display FIFOs may get combined into a single larger
 * FIFO for a particular plane (so called max FIFO mode) to allow the
 * system to defer memory fetches longer, and the memory will enter
 * self refresh.
 *
 * Note that enabling CxSR does not guarantee that the system enter
 * this special mode, nor does it guarantee that the system stays
 * in that mode once entered. So this just allows/disallows the system
 * to autonomously utilize the CxSR mode. Other factors such as core
 * C-states will affect when/if the system actually enters/exits the
 * CxSR mode.
 *
 * Note that on VLV/CHV this actually only controls the max FIFO mode,
 * and the system is free to enter/exit memory self refresh at any time
 * even when the use of CxSR has been disallowed.
 *
 * While the system is actually in the CxSR/max FIFO mode, some plane
 * control registers will not get latched on vblank. Thus in order to
 * guarantee the system will respond to changes in the plane registers
 * we must always disallow CxSR prior to making changes to those registers.
 * Unfortunately the system will re-evaluate the CxSR conditions at
 * frame start which happens after vblank start (which is when the plane
 * registers would get latched), so we can't proceed with the plane update
 * during the same frame where we disallowed CxSR.
 *
 * Certain platforms also have a deeper HPLL SR mode. Fortunately the
 * HPLL SR mode depends on CxSR itself, so we don't have to hand hold
 * the hardware w.r.t. HPLL SR when writing to plane registers.
 * Disallowing just CxSR is sufficient.
 */
bool intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable)
{
	bool ret;

	mutex_lock(&dev_priv->wm.wm_mutex);
	ret = _intel_set_memory_cxsr(dev_priv, enable);
	if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv))
		dev_priv->wm.vlv.cxsr = enable;
	else if (IS_G4X(dev_priv))
		dev_priv->wm.g4x.cxsr = enable;
	mutex_unlock(&dev_priv->wm.wm_mutex);

	return ret;
}

/*
 * Latency for FIFO fetches is dependent on several factors:
 *   - memory configuration (speed, channels)
 *   - chipset
 *   - current MCH state
 * It can be fairly high in some situations, so here we assume a fairly
 * pessimal value.  It's a tradeoff between extra memory fetches (if we
 * set this value too high, the FIFO will fetch frequently to stay full)
 * and power consumption (set it too low to save power and we might see
 * FIFO underruns and display "flicker").
 *
 * A value of 5us seems to be a good balance; safe for very low end
 * platforms but not overly aggressive on lower latency configs.
 */
static const int pessimal_latency_ns = 5000;

#define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \
	((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8))

static void vlv_get_fifo_size(struct intel_crtc_state *crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	struct vlv_fifo_state *fifo_state = &crtc_state->wm.vlv.fifo_state;
	enum pipe pipe = crtc->pipe;
	int sprite0_start, sprite1_start;

	switch (pipe) {
		u32 dsparb, dsparb2, dsparb3;
	case PIPE_A:
		dsparb = I915_READ(DSPARB);
		dsparb2 = I915_READ(DSPARB2);
		sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 0, 0);
		sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 8, 4);
		break;
	case PIPE_B:
		dsparb = I915_READ(DSPARB);
		dsparb2 = I915_READ(DSPARB2);
		sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 16, 8);
		sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 24, 12);
		break;
	case PIPE_C:
		dsparb2 = I915_READ(DSPARB2);
		dsparb3 = I915_READ(DSPARB3);
		sprite0_start = VLV_FIFO_START(dsparb3, dsparb2, 0, 16);
		sprite1_start = VLV_FIFO_START(dsparb3, dsparb2, 8, 20);
		break;
	default:
		MISSING_CASE(pipe);
		return;
	}

	fifo_state->plane[PLANE_PRIMARY] = sprite0_start;
	fifo_state->plane[PLANE_SPRITE0] = sprite1_start - sprite0_start;
	fifo_state->plane[PLANE_SPRITE1] = 511 - sprite1_start;
	fifo_state->plane[PLANE_CURSOR] = 63;
}

static int i9xx_get_fifo_size(struct drm_i915_private *dev_priv,
			      enum i9xx_plane_id i9xx_plane)
{
	u32 dsparb = I915_READ(DSPARB);
	int size;

	size = dsparb & 0x7f;
	if (i9xx_plane == PLANE_B)
		size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;

	DRM_DEBUG_KMS("FIFO size - (0x%08x) %c: %d\n",
		      dsparb, plane_name(i9xx_plane), size);

	return size;
}

static int i830_get_fifo_size(struct drm_i915_private *dev_priv,
			      enum i9xx_plane_id i9xx_plane)
{
	u32 dsparb = I915_READ(DSPARB);
	int size;

	size = dsparb & 0x1ff;
	if (i9xx_plane == PLANE_B)
		size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
	size >>= 1; /* Convert to cachelines */

	DRM_DEBUG_KMS("FIFO size - (0x%08x) %c: %d\n",
		      dsparb, plane_name(i9xx_plane), size);

	return size;
}

static int i845_get_fifo_size(struct drm_i915_private *dev_priv,
			      enum i9xx_plane_id i9xx_plane)
{
	u32 dsparb = I915_READ(DSPARB);
	int size;

	size = dsparb & 0x7f;
	size >>= 2; /* Convert to cachelines */

	DRM_DEBUG_KMS("FIFO size - (0x%08x) %c: %d\n",
		      dsparb, plane_name(i9xx_plane), size);

	return size;
}

/* Pineview has different values for various configs */
static const struct intel_watermark_params pineview_display_wm = {
	.fifo_size = PINEVIEW_DISPLAY_FIFO,
	.max_wm = PINEVIEW_MAX_WM,
	.default_wm = PINEVIEW_DFT_WM,
	.guard_size = PINEVIEW_GUARD_WM,
	.cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params pineview_display_hplloff_wm = {
	.fifo_size = PINEVIEW_DISPLAY_FIFO,
	.max_wm = PINEVIEW_MAX_WM,
	.default_wm = PINEVIEW_DFT_HPLLOFF_WM,
	.guard_size = PINEVIEW_GUARD_WM,
	.cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params pineview_cursor_wm = {
	.fifo_size = PINEVIEW_CURSOR_FIFO,
	.max_wm = PINEVIEW_CURSOR_MAX_WM,
	.default_wm = PINEVIEW_CURSOR_DFT_WM,
	.guard_size = PINEVIEW_CURSOR_GUARD_WM,
	.cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
	.fifo_size = PINEVIEW_CURSOR_FIFO,
	.max_wm = PINEVIEW_CURSOR_MAX_WM,
	.default_wm = PINEVIEW_CURSOR_DFT_WM,
	.guard_size = PINEVIEW_CURSOR_GUARD_WM,
	.cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params i965_cursor_wm_info = {
	.fifo_size = I965_CURSOR_FIFO,
	.max_wm = I965_CURSOR_MAX_WM,
	.default_wm = I965_CURSOR_DFT_WM,
	.guard_size = 2,
	.cacheline_size = I915_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params i945_wm_info = {
	.fifo_size = I945_FIFO_SIZE,
	.max_wm = I915_MAX_WM,
	.default_wm = 1,
	.guard_size = 2,
	.cacheline_size = I915_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params i915_wm_info = {
	.fifo_size = I915_FIFO_SIZE,
	.max_wm = I915_MAX_WM,
	.default_wm = 1,
	.guard_size = 2,
	.cacheline_size = I915_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params i830_a_wm_info = {
	.fifo_size = I855GM_FIFO_SIZE,
	.max_wm = I915_MAX_WM,
	.default_wm = 1,
	.guard_size = 2,
	.cacheline_size = I830_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params i830_bc_wm_info = {
	.fifo_size = I855GM_FIFO_SIZE,
	.max_wm = I915_MAX_WM/2,
	.default_wm = 1,
	.guard_size = 2,
	.cacheline_size = I830_FIFO_LINE_SIZE,
};
static const struct intel_watermark_params i845_wm_info = {
	.fifo_size = I830_FIFO_SIZE,
	.max_wm = I915_MAX_WM,
	.default_wm = 1,
	.guard_size = 2,
	.cacheline_size = I830_FIFO_LINE_SIZE,
};

/**
 * intel_wm_method1 - Method 1 / "small buffer" watermark formula
 * @pixel_rate: Pipe pixel rate in kHz
 * @cpp: Plane bytes per pixel
 * @latency: Memory wakeup latency in 0.1us units
 *
 * Compute the watermark using the method 1 or "small buffer"
 * formula. The caller may additonally add extra cachelines
 * to account for TLB misses and clock crossings.
 *
 * This method is concerned with the short term drain rate
 * of the FIFO, ie. it does not account for blanking periods
 * which would effectively reduce the average drain rate across
 * a longer period. The name "small" refers to the fact the
 * FIFO is relatively small compared to the amount of data
 * fetched.
 *
 * The FIFO level vs. time graph might look something like:
 *
 *   |\   |\
 *   | \  | \
 * __---__---__ (- plane active, _ blanking)
 * -> time
 *
 * or perhaps like this:
 *
 *   |\|\  |\|\
 * __----__----__ (- plane active, _ blanking)
 * -> time
 *
 * Returns:
 * The watermark in bytes
 */
static unsigned int intel_wm_method1(unsigned int pixel_rate,
				     unsigned int cpp,
				     unsigned int latency)
{
	u64 ret;

	ret = mul_u32_u32(pixel_rate, cpp * latency);
	ret = DIV_ROUND_UP_ULL(ret, 10000);

	return ret;
}

/**
 * intel_wm_method2 - Method 2 / "large buffer" watermark formula
 * @pixel_rate: Pipe pixel rate in kHz
 * @htotal: Pipe horizontal total
 * @width: Plane width in pixels
 * @cpp: Plane bytes per pixel
 * @latency: Memory wakeup latency in 0.1us units
 *
 * Compute the watermark using the method 2 or "large buffer"
 * formula. The caller may additonally add extra cachelines
 * to account for TLB misses and clock crossings.
 *
 * This method is concerned with the long term drain rate
 * of the FIFO, ie. it does account for blanking periods
 * which effectively reduce the average drain rate across
 * a longer period. The name "large" refers to the fact the
 * FIFO is relatively large compared to the amount of data
 * fetched.
 *
 * The FIFO level vs. time graph might look something like:
 *
 *    |\___       |\___
 *    |    \___   |    \___
 *    |        \  |        \
 * __ --__--__--__--__--__--__ (- plane active, _ blanking)
 * -> time
 *
 * Returns:
 * The watermark in bytes
 */
static unsigned int intel_wm_method2(unsigned int pixel_rate,
				     unsigned int htotal,
				     unsigned int width,
				     unsigned int cpp,
				     unsigned int latency)
{
	unsigned int ret;

	/*
	 * FIXME remove once all users are computing
	 * watermarks in the correct place.
	 */
	if (WARN_ON_ONCE(htotal == 0))
		htotal = 1;

	ret = (latency * pixel_rate) / (htotal * 10000);
	ret = (ret + 1) * width * cpp;

	return ret;
}

/**
 * intel_calculate_wm - calculate watermark level
 * @pixel_rate: pixel clock
 * @wm: chip FIFO params
 * @fifo_size: size of the FIFO buffer
 * @cpp: bytes per pixel
 * @latency_ns: memory latency for the platform
 *
 * Calculate the watermark level (the level at which the display plane will
 * start fetching from memory again).  Each chip has a different display
 * FIFO size and allocation, so the caller needs to figure that out and pass
 * in the correct intel_watermark_params structure.
 *
 * As the pixel clock runs, the FIFO will be drained at a rate that depends
 * on the pixel size.  When it reaches the watermark level, it'll start
 * fetching FIFO line sized based chunks from memory until the FIFO fills
 * past the watermark point.  If the FIFO drains completely, a FIFO underrun
 * will occur, and a display engine hang could result.
 */
static unsigned int intel_calculate_wm(int pixel_rate,
				       const struct intel_watermark_params *wm,
				       int fifo_size, int cpp,
				       unsigned int latency_ns)
{
	int entries, wm_size;

	/*
	 * Note: we need to make sure we don't overflow for various clock &
	 * latency values.
	 * clocks go from a few thousand to several hundred thousand.
	 * latency is usually a few thousand
	 */
	entries = intel_wm_method1(pixel_rate, cpp,
				   latency_ns / 100);
	entries = DIV_ROUND_UP(entries, wm->cacheline_size) +
		wm->guard_size;
	DRM_DEBUG_KMS("FIFO entries required for mode: %d\n", entries);

	wm_size = fifo_size - entries;
	DRM_DEBUG_KMS("FIFO watermark level: %d\n", wm_size);

	/* Don't promote wm_size to unsigned... */
	if (wm_size > wm->max_wm)
		wm_size = wm->max_wm;
	if (wm_size <= 0)
		wm_size = wm->default_wm;

	/*
	 * Bspec seems to indicate that the value shouldn't be lower than
	 * 'burst size + 1'. Certainly 830 is quite unhappy with low values.
	 * Lets go for 8 which is the burst size since certain platforms
	 * already use a hardcoded 8 (which is what the spec says should be
	 * done).
	 */
	if (wm_size <= 8)
		wm_size = 8;

	return wm_size;
}

static bool is_disabling(int old, int new, int threshold)
{
	return old >= threshold && new < threshold;
}

static bool is_enabling(int old, int new, int threshold)
{
	return old < threshold && new >= threshold;
}

static int intel_wm_num_levels(struct drm_i915_private *dev_priv)
{
	return dev_priv->wm.max_level + 1;
}

static bool intel_wm_plane_visible(const struct intel_crtc_state *crtc_state,
				   const struct intel_plane_state *plane_state)
{
	struct intel_plane *plane = to_intel_plane(plane_state->base.plane);

	/* FIXME check the 'enable' instead */
	if (!crtc_state->base.active)
		return false;

	/*
	 * Treat cursor with fb as always visible since cursor updates
	 * can happen faster than the vrefresh rate, and the current
	 * watermark code doesn't handle that correctly. Cursor updates
	 * which set/clear the fb or change the cursor size are going
	 * to get throttled by intel_legacy_cursor_update() to work
	 * around this problem with the watermark code.
	 */
	if (plane->id == PLANE_CURSOR)
		return plane_state->base.fb != NULL;
	else
		return plane_state->base.visible;
}

static struct intel_crtc *single_enabled_crtc(struct drm_i915_private *dev_priv)
{
	struct intel_crtc *crtc, *enabled = NULL;

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		if (intel_crtc_active(crtc)) {
			if (enabled)
				return NULL;
			enabled = crtc;
		}
	}

	return enabled;
}

static void pineview_update_wm(struct intel_crtc *unused_crtc)
{
	struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev);
	struct intel_crtc *crtc;
	const struct cxsr_latency *latency;
	u32 reg;
	unsigned int wm;

	latency = intel_get_cxsr_latency(!IS_MOBILE(dev_priv),
					 dev_priv->is_ddr3,
					 dev_priv->fsb_freq,
					 dev_priv->mem_freq);
	if (!latency) {
		DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
		intel_set_memory_cxsr(dev_priv, false);
		return;
	}

	crtc = single_enabled_crtc(dev_priv);
	if (crtc) {
		const struct drm_display_mode *adjusted_mode =
			&crtc->config->base.adjusted_mode;
		const struct drm_framebuffer *fb =
			crtc->base.primary->state->fb;
		int cpp = fb->format->cpp[0];
		int clock = adjusted_mode->crtc_clock;

		/* Display SR */
		wm = intel_calculate_wm(clock, &pineview_display_wm,
					pineview_display_wm.fifo_size,
					cpp, latency->display_sr);
		reg = I915_READ(DSPFW1);
		reg &= ~DSPFW_SR_MASK;
		reg |= FW_WM(wm, SR);
		I915_WRITE(DSPFW1, reg);
		DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);

		/* cursor SR */
		wm = intel_calculate_wm(clock, &pineview_cursor_wm,
					pineview_display_wm.fifo_size,
					4, latency->cursor_sr);
		reg = I915_READ(DSPFW3);
		reg &= ~DSPFW_CURSOR_SR_MASK;
		reg |= FW_WM(wm, CURSOR_SR);
		I915_WRITE(DSPFW3, reg);

		/* Display HPLL off SR */
		wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
					pineview_display_hplloff_wm.fifo_size,
					cpp, latency->display_hpll_disable);
		reg = I915_READ(DSPFW3);
		reg &= ~DSPFW_HPLL_SR_MASK;
		reg |= FW_WM(wm, HPLL_SR);
		I915_WRITE(DSPFW3, reg);

		/* cursor HPLL off SR */
		wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
					pineview_display_hplloff_wm.fifo_size,
					4, latency->cursor_hpll_disable);
		reg = I915_READ(DSPFW3);
		reg &= ~DSPFW_HPLL_CURSOR_MASK;
		reg |= FW_WM(wm, HPLL_CURSOR);
		I915_WRITE(DSPFW3, reg);
		DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);

		intel_set_memory_cxsr(dev_priv, true);
	} else {
		intel_set_memory_cxsr(dev_priv, false);
	}
}

/*
 * Documentation says:
 * "If the line size is small, the TLB fetches can get in the way of the
 *  data fetches, causing some lag in the pixel data return which is not
 *  accounted for in the above formulas. The following adjustment only
 *  needs to be applied if eight whole lines fit in the buffer at once.
 *  The WM is adjusted upwards by the difference between the FIFO size
 *  and the size of 8 whole lines. This adjustment is always performed
 *  in the actual pixel depth regardless of whether FBC is enabled or not."
 */
static unsigned int g4x_tlb_miss_wa(int fifo_size, int width, int cpp)
{
	int tlb_miss = fifo_size * 64 - width * cpp * 8;

	return max(0, tlb_miss);
}

static void g4x_write_wm_values(struct drm_i915_private *dev_priv,
				const struct g4x_wm_values *wm)
{
	enum pipe pipe;

	for_each_pipe(dev_priv, pipe)
		trace_g4x_wm(intel_get_crtc_for_pipe(dev_priv, pipe), wm);

	I915_WRITE(DSPFW1,
		   FW_WM(wm->sr.plane, SR) |
		   FW_WM(wm->pipe[PIPE_B].plane[PLANE_CURSOR], CURSORB) |
		   FW_WM(wm->pipe[PIPE_B].plane[PLANE_PRIMARY], PLANEB) |
		   FW_WM(wm->pipe[PIPE_A].plane[PLANE_PRIMARY], PLANEA));
	I915_WRITE(DSPFW2,
		   (wm->fbc_en ? DSPFW_FBC_SR_EN : 0) |
		   FW_WM(wm->sr.fbc, FBC_SR) |
		   FW_WM(wm->hpll.fbc, FBC_HPLL_SR) |
		   FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE0], SPRITEB) |
		   FW_WM(wm->pipe[PIPE_A].plane[PLANE_CURSOR], CURSORA) |
		   FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE0], SPRITEA));
	I915_WRITE(DSPFW3,
		   (wm->hpll_en ? DSPFW_HPLL_SR_EN : 0) |
		   FW_WM(wm->sr.cursor, CURSOR_SR) |
		   FW_WM(wm->hpll.cursor, HPLL_CURSOR) |
		   FW_WM(wm->hpll.plane, HPLL_SR));

	POSTING_READ(DSPFW1);
}

#define FW_WM_VLV(value, plane) \
	(((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV)

static void vlv_write_wm_values(struct drm_i915_private *dev_priv,
				const struct vlv_wm_values *wm)
{
	enum pipe pipe;

	for_each_pipe(dev_priv, pipe) {
		trace_vlv_wm(intel_get_crtc_for_pipe(dev_priv, pipe), wm);

		I915_WRITE(VLV_DDL(pipe),
			   (wm->ddl[pipe].plane[PLANE_CURSOR] << DDL_CURSOR_SHIFT) |
			   (wm->ddl[pipe].plane[PLANE_SPRITE1] << DDL_SPRITE_SHIFT(1)) |
			   (wm->ddl[pipe].plane[PLANE_SPRITE0] << DDL_SPRITE_SHIFT(0)) |
			   (wm->ddl[pipe].plane[PLANE_PRIMARY] << DDL_PLANE_SHIFT));
	}

	/*
	 * Zero the (unused) WM1 watermarks, and also clear all the
	 * high order bits so that there are no out of bounds values
	 * present in the registers during the reprogramming.
	 */
	I915_WRITE(DSPHOWM, 0);
	I915_WRITE(DSPHOWM1, 0);
	I915_WRITE(DSPFW4, 0);
	I915_WRITE(DSPFW5, 0);
	I915_WRITE(DSPFW6, 0);

	I915_WRITE(DSPFW1,
		   FW_WM(wm->sr.plane, SR) |
		   FW_WM(wm->pipe[PIPE_B].plane[PLANE_CURSOR], CURSORB) |
		   FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_PRIMARY], PLANEB) |
		   FW_WM_VLV(wm->pipe[PIPE_A].plane[PLANE_PRIMARY], PLANEA));
	I915_WRITE(DSPFW2,
		   FW_WM_VLV(wm->pipe[PIPE_A].plane[PLANE_SPRITE1], SPRITEB) |
		   FW_WM(wm->pipe[PIPE_A].plane[PLANE_CURSOR], CURSORA) |
		   FW_WM_VLV(wm->pipe[PIPE_A].plane[PLANE_SPRITE0], SPRITEA));
	I915_WRITE(DSPFW3,
		   FW_WM(wm->sr.cursor, CURSOR_SR));

	if (IS_CHERRYVIEW(dev_priv)) {
		I915_WRITE(DSPFW7_CHV,
			   FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_SPRITE1], SPRITED) |
			   FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_SPRITE0], SPRITEC));
		I915_WRITE(DSPFW8_CHV,
			   FW_WM_VLV(wm->pipe[PIPE_C].plane[PLANE_SPRITE1], SPRITEF) |
			   FW_WM_VLV(wm->pipe[PIPE_C].plane[PLANE_SPRITE0], SPRITEE));
		I915_WRITE(DSPFW9_CHV,
			   FW_WM_VLV(wm->pipe[PIPE_C].plane[PLANE_PRIMARY], PLANEC) |
			   FW_WM(wm->pipe[PIPE_C].plane[PLANE_CURSOR], CURSORC));
		I915_WRITE(DSPHOWM,
			   FW_WM(wm->sr.plane >> 9, SR_HI) |
			   FW_WM(wm->pipe[PIPE_C].plane[PLANE_SPRITE1] >> 8, SPRITEF_HI) |
			   FW_WM(wm->pipe[PIPE_C].plane[PLANE_SPRITE0] >> 8, SPRITEE_HI) |
			   FW_WM(wm->pipe[PIPE_C].plane[PLANE_PRIMARY] >> 8, PLANEC_HI) |
			   FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE1] >> 8, SPRITED_HI) |
			   FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE0] >> 8, SPRITEC_HI) |
			   FW_WM(wm->pipe[PIPE_B].plane[PLANE_PRIMARY] >> 8, PLANEB_HI) |
			   FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE1] >> 8, SPRITEB_HI) |
			   FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE0] >> 8, SPRITEA_HI) |
			   FW_WM(wm->pipe[PIPE_A].plane[PLANE_PRIMARY] >> 8, PLANEA_HI));
	} else {
		I915_WRITE(DSPFW7,
			   FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_SPRITE1], SPRITED) |
			   FW_WM_VLV(wm->pipe[PIPE_B].plane[PLANE_SPRITE0], SPRITEC));
		I915_WRITE(DSPHOWM,
			   FW_WM(wm->sr.plane >> 9, SR_HI) |
			   FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE1] >> 8, SPRITED_HI) |
			   FW_WM(wm->pipe[PIPE_B].plane[PLANE_SPRITE0] >> 8, SPRITEC_HI) |
			   FW_WM(wm->pipe[PIPE_B].plane[PLANE_PRIMARY] >> 8, PLANEB_HI) |
			   FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE1] >> 8, SPRITEB_HI) |
			   FW_WM(wm->pipe[PIPE_A].plane[PLANE_SPRITE0] >> 8, SPRITEA_HI) |
			   FW_WM(wm->pipe[PIPE_A].plane[PLANE_PRIMARY] >> 8, PLANEA_HI));
	}

	POSTING_READ(DSPFW1);
}

#undef FW_WM_VLV

static void g4x_setup_wm_latency(struct drm_i915_private *dev_priv)
{
	/* all latencies in usec */
	dev_priv->wm.pri_latency[G4X_WM_LEVEL_NORMAL] = 5;
	dev_priv->wm.pri_latency[G4X_WM_LEVEL_SR] = 12;
	dev_priv->wm.pri_latency[G4X_WM_LEVEL_HPLL] = 35;

	dev_priv->wm.max_level = G4X_WM_LEVEL_HPLL;
}

static int g4x_plane_fifo_size(enum plane_id plane_id, int level)
{
	/*
	 * DSPCNTR[13] supposedly controls whether the
	 * primary plane can use the FIFO space otherwise
	 * reserved for the sprite plane. It's not 100% clear
	 * what the actual FIFO size is, but it looks like we
	 * can happily set both primary and sprite watermarks
	 * up to 127 cachelines. So that would seem to mean
	 * that either DSPCNTR[13] doesn't do anything, or that
	 * the total FIFO is >= 256 cachelines in size. Either
	 * way, we don't seem to have to worry about this
	 * repartitioning as the maximum watermark value the
	 * register can hold for each plane is lower than the
	 * minimum FIFO size.
	 */
	switch (plane_id) {
	case PLANE_CURSOR:
		return 63;
	case PLANE_PRIMARY:
		return level == G4X_WM_LEVEL_NORMAL ? 127 : 511;
	case PLANE_SPRITE0:
		return level == G4X_WM_LEVEL_NORMAL ? 127 : 0;
	default:
		MISSING_CASE(plane_id);
		return 0;
	}
}

static int g4x_fbc_fifo_size(int level)
{
	switch (level) {
	case G4X_WM_LEVEL_SR:
		return 7;
	case G4X_WM_LEVEL_HPLL:
		return 15;
	default:
		MISSING_CASE(level);
		return 0;
	}
}

static u16 g4x_compute_wm(const struct intel_crtc_state *crtc_state,
			  const struct intel_plane_state *plane_state,
			  int level)
{
	struct intel_plane *plane = to_intel_plane(plane_state->base.plane);
	struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
	const struct drm_display_mode *adjusted_mode =
		&crtc_state->base.adjusted_mode;
	unsigned int latency = dev_priv->wm.pri_latency[level] * 10;
	unsigned int clock, htotal, cpp, width, wm;

	if (latency == 0)
		return USHRT_MAX;

	if (!intel_wm_plane_visible(crtc_state, plane_state))
		return 0;

	cpp = plane_state->base.fb->format->cpp[0];

	/*
	 * Not 100% sure which way ELK should go here as the
	 * spec only says CL/CTG should assume 32bpp and BW
	 * doesn't need to. But as these things followed the
	 * mobile vs. desktop lines on gen3 as well, let's
	 * assume ELK doesn't need this.
	 *
	 * The spec also fails to list such a restriction for
	 * the HPLL watermark, which seems a little strange.
	 * Let's use 32bpp for the HPLL watermark as well.
	 */
	if (IS_GM45(dev_priv) && plane->id == PLANE_PRIMARY &&
	    level != G4X_WM_LEVEL_NORMAL)
		cpp = max(cpp, 4u);

	clock = adjusted_mode->crtc_clock;
	htotal = adjusted_mode->crtc_htotal;

	width = drm_rect_width(&plane_state->base.dst);

	if (plane->id == PLANE_CURSOR) {
		wm = intel_wm_method2(clock, htotal, width, cpp, latency);
	} else if (plane->id == PLANE_PRIMARY &&
		   level == G4X_WM_LEVEL_NORMAL) {
		wm = intel_wm_method1(clock, cpp, latency);
	} else {
		unsigned int small, large;

		small = intel_wm_method1(clock, cpp, latency);
		large = intel_wm_method2(clock, htotal, width, cpp, latency);

		wm = min(small, large);
	}

	wm += g4x_tlb_miss_wa(g4x_plane_fifo_size(plane->id, level),
			      width, cpp);

	wm = DIV_ROUND_UP(wm, 64) + 2;

	return min_t(unsigned int, wm, USHRT_MAX);
}

static bool g4x_raw_plane_wm_set(struct intel_crtc_state *crtc_state,
				 int level, enum plane_id plane_id, u16 value)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	bool dirty = false;

	for (; level < intel_wm_num_levels(dev_priv); level++) {
		struct g4x_pipe_wm *raw = &crtc_state->wm.g4x.raw[level];

		dirty |= raw->plane[plane_id] != value;
		raw->plane[plane_id] = value;
	}

	return dirty;
}

static bool g4x_raw_fbc_wm_set(struct intel_crtc_state *crtc_state,
			       int level, u16 value)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	bool dirty = false;

	/* NORMAL level doesn't have an FBC watermark */
	level = max(level, G4X_WM_LEVEL_SR);

	for (; level < intel_wm_num_levels(dev_priv); level++) {
		struct g4x_pipe_wm *raw = &crtc_state->wm.g4x.raw[level];

		dirty |= raw->fbc != value;
		raw->fbc = value;
	}

	return dirty;
}

static u32 ilk_compute_fbc_wm(const struct intel_crtc_state *crtc_state,
			      const struct intel_plane_state *plane_state,
			      u32 pri_val);

static bool g4x_raw_plane_wm_compute(struct intel_crtc_state *crtc_state,
				     const struct intel_plane_state *plane_state)
{
	struct intel_plane *plane = to_intel_plane(plane_state->base.plane);
	int num_levels = intel_wm_num_levels(to_i915(plane->base.dev));
	enum plane_id plane_id = plane->id;
	bool dirty = false;
	int level;

	if (!intel_wm_plane_visible(crtc_state, plane_state)) {
		dirty |= g4x_raw_plane_wm_set(crtc_state, 0, plane_id, 0);
		if (plane_id == PLANE_PRIMARY)
			dirty |= g4x_raw_fbc_wm_set(crtc_state, 0, 0);
		goto out;
	}

	for (level = 0; level < num_levels; level++) {
		struct g4x_pipe_wm *raw = &crtc_state->wm.g4x.raw[level];
		int wm, max_wm;

		wm = g4x_compute_wm(crtc_state, plane_state, level);
		max_wm = g4x_plane_fifo_size(plane_id, level);

		if (wm > max_wm)
			break;

		dirty |= raw->plane[plane_id] != wm;
		raw->plane[plane_id] = wm;

		if (plane_id != PLANE_PRIMARY ||
		    level == G4X_WM_LEVEL_NORMAL)
			continue;

		wm = ilk_compute_fbc_wm(crtc_state, plane_state,
					raw->plane[plane_id]);
		max_wm = g4x_fbc_fifo_size(level);

		/*
		 * FBC wm is not mandatory as we
		 * can always just disable its use.
		 */
		if (wm > max_wm)
			wm = USHRT_MAX;

		dirty |= raw->fbc != wm;
		raw->fbc = wm;
	}

	/* mark watermarks as invalid */
	dirty |= g4x_raw_plane_wm_set(crtc_state, level, plane_id, USHRT_MAX);

	if (plane_id == PLANE_PRIMARY)
		dirty |= g4x_raw_fbc_wm_set(crtc_state, level, USHRT_MAX);

 out:
	if (dirty) {
		DRM_DEBUG_KMS("%s watermarks: normal=%d, SR=%d, HPLL=%d\n",
			      plane->base.name,
			      crtc_state->wm.g4x.raw[G4X_WM_LEVEL_NORMAL].plane[plane_id],
			      crtc_state->wm.g4x.raw[G4X_WM_LEVEL_SR].plane[plane_id],
			      crtc_state->wm.g4x.raw[G4X_WM_LEVEL_HPLL].plane[plane_id]);

		if (plane_id == PLANE_PRIMARY)
			DRM_DEBUG_KMS("FBC watermarks: SR=%d, HPLL=%d\n",
				      crtc_state->wm.g4x.raw[G4X_WM_LEVEL_SR].fbc,
				      crtc_state->wm.g4x.raw[G4X_WM_LEVEL_HPLL].fbc);
	}

	return dirty;
}

static bool g4x_raw_plane_wm_is_valid(const struct intel_crtc_state *crtc_state,
				      enum plane_id plane_id, int level)
{
	const struct g4x_pipe_wm *raw = &crtc_state->wm.g4x.raw[level];

	return raw->plane[plane_id] <= g4x_plane_fifo_size(plane_id, level);
}

static bool g4x_raw_crtc_wm_is_valid(const struct intel_crtc_state *crtc_state,
				     int level)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);

	if (level > dev_priv->wm.max_level)
		return false;

	return g4x_raw_plane_wm_is_valid(crtc_state, PLANE_PRIMARY, level) &&
		g4x_raw_plane_wm_is_valid(crtc_state, PLANE_SPRITE0, level) &&
		g4x_raw_plane_wm_is_valid(crtc_state, PLANE_CURSOR, level);
}

/* mark all levels starting from 'level' as invalid */
static void g4x_invalidate_wms(struct intel_crtc *crtc,
			       struct g4x_wm_state *wm_state, int level)
{
	if (level <= G4X_WM_LEVEL_NORMAL) {
		enum plane_id plane_id;

		for_each_plane_id_on_crtc(crtc, plane_id)
			wm_state->wm.plane[plane_id] = USHRT_MAX;
	}

	if (level <= G4X_WM_LEVEL_SR) {
		wm_state->cxsr = false;
		wm_state->sr.cursor = USHRT_MAX;
		wm_state->sr.plane = USHRT_MAX;
		wm_state->sr.fbc = USHRT_MAX;
	}

	if (level <= G4X_WM_LEVEL_HPLL) {
		wm_state->hpll_en = false;
		wm_state->hpll.cursor = USHRT_MAX;
		wm_state->hpll.plane = USHRT_MAX;
		wm_state->hpll.fbc = USHRT_MAX;
	}
}

static int g4x_compute_pipe_wm(struct intel_crtc_state *crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
	struct intel_atomic_state *state =
		to_intel_atomic_state(crtc_state->base.state);
	struct g4x_wm_state *wm_state = &crtc_state->wm.g4x.optimal;
	int num_active_planes = hweight8(crtc_state->active_planes &
					 ~BIT(PLANE_CURSOR));
	const struct g4x_pipe_wm *raw;
	const struct intel_plane_state *old_plane_state;
	const struct intel_plane_state *new_plane_state;
	struct intel_plane *plane;
	enum plane_id plane_id;
	int i, level;
	unsigned int dirty = 0;

	for_each_oldnew_intel_plane_in_state(state, plane,
					     old_plane_state,
					     new_plane_state, i) {
		if (new_plane_state->base.crtc != &crtc->base &&
		    old_plane_state->base.crtc != &crtc->base)
			continue;

		if (g4x_raw_plane_wm_compute(crtc_state, new_plane_state))
			dirty |= BIT(plane->id);
	}

	if (!dirty)
		return 0;

	level = G4X_WM_LEVEL_NORMAL;
	if (!g4x_raw_crtc_wm_is_valid(crtc_state, level))
		goto out;

	raw = &crtc_state->wm.g4x.raw[level];
	for_each_plane_id_on_crtc(crtc, plane_id)
		wm_state->wm.plane[plane_id] = raw->plane[plane_id];

	level = G4X_WM_LEVEL_SR;

	if (!g4x_raw_crtc_wm_is_valid(crtc_state, level))
		goto out;

	raw = &crtc_state->wm.g4x.raw[level];
	wm_state->sr.plane = raw->plane[PLANE_PRIMARY];
	wm_state->sr.cursor = raw->plane[PLANE_CURSOR];
	wm_state->sr.fbc = raw->fbc;

	wm_state->cxsr = num_active_planes == BIT(PLANE_PRIMARY);

	level = G4X_WM_LEVEL_HPLL;

	if (!g4x_raw_crtc_wm_is_valid(crtc_state, level))
		goto out;

	raw = &crtc_state->wm.g4x.raw[level];
	wm_state->hpll.plane = raw->plane[PLANE_PRIMARY];
	wm_state->hpll.cursor = raw->plane[PLANE_CURSOR];
	wm_state->hpll.fbc = raw->fbc;

	wm_state->hpll_en = wm_state->cxsr;

	level++;

 out:
	if (level == G4X_WM_LEVEL_NORMAL)
		return -EINVAL;

	/* invalidate the higher levels */
	g4x_invalidate_wms(crtc, wm_state, level);

	/*
	 * Determine if the FBC watermark(s) can be used. IF
	 * this isn't the case we prefer to disable the FBC
	 ( watermark(s) rather than disable the SR/HPLL
	 * level(s) entirely.
	 */
	wm_state->fbc_en = level > G4X_WM_LEVEL_NORMAL;

	if (level >= G4X_WM_LEVEL_SR &&
	    wm_state->sr.fbc > g4x_fbc_fifo_size(G4X_WM_LEVEL_SR))
		wm_state->fbc_en = false;
	else if (level >= G4X_WM_LEVEL_HPLL &&
		 wm_state->hpll.fbc > g4x_fbc_fifo_size(G4X_WM_LEVEL_HPLL))
		wm_state->fbc_en = false;

	return 0;
}

static int g4x_compute_intermediate_wm(struct intel_crtc_state *new_crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(new_crtc_state->base.crtc);
	struct g4x_wm_state *intermediate = &new_crtc_state->wm.g4x.intermediate;
	const struct g4x_wm_state *optimal = &new_crtc_state->wm.g4x.optimal;
	struct intel_atomic_state *intel_state =
		to_intel_atomic_state(new_crtc_state->base.state);
	const struct intel_crtc_state *old_crtc_state =
		intel_atomic_get_old_crtc_state(intel_state, crtc);
	const struct g4x_wm_state *active = &old_crtc_state->wm.g4x.optimal;
	enum plane_id plane_id;

	if (!new_crtc_state->base.active || drm_atomic_crtc_needs_modeset(&new_crtc_state->base)) {
		*intermediate = *optimal;

		intermediate->cxsr = false;
		intermediate->hpll_en = false;
		goto out;
	}

	intermediate->cxsr = optimal->cxsr && active->cxsr &&
		!new_crtc_state->disable_cxsr;
	intermediate->hpll_en = optimal->hpll_en && active->hpll_en &&
		!new_crtc_state->disable_cxsr;
	intermediate->fbc_en = optimal->fbc_en && active->fbc_en;

	for_each_plane_id_on_crtc(crtc, plane_id) {
		intermediate->wm.plane[plane_id] =
			max(optimal->wm.plane[plane_id],
			    active->wm.plane[plane_id]);

		WARN_ON(intermediate->wm.plane[plane_id] >
			g4x_plane_fifo_size(plane_id, G4X_WM_LEVEL_NORMAL));
	}

	intermediate->sr.plane = max(optimal->sr.plane,
				     active->sr.plane);
	intermediate->sr.cursor = max(optimal->sr.cursor,
				      active->sr.cursor);
	intermediate->sr.fbc = max(optimal->sr.fbc,
				   active->sr.fbc);

	intermediate->hpll.plane = max(optimal->hpll.plane,
				       active->hpll.plane);
	intermediate->hpll.cursor = max(optimal->hpll.cursor,
					active->hpll.cursor);
	intermediate->hpll.fbc = max(optimal->hpll.fbc,
				     active->hpll.fbc);

	WARN_ON((intermediate->sr.plane >
		 g4x_plane_fifo_size(PLANE_PRIMARY, G4X_WM_LEVEL_SR) ||
		 intermediate->sr.cursor >
		 g4x_plane_fifo_size(PLANE_CURSOR, G4X_WM_LEVEL_SR)) &&
		intermediate->cxsr);
	WARN_ON((intermediate->sr.plane >
		 g4x_plane_fifo_size(PLANE_PRIMARY, G4X_WM_LEVEL_HPLL) ||
		 intermediate->sr.cursor >
		 g4x_plane_fifo_size(PLANE_CURSOR, G4X_WM_LEVEL_HPLL)) &&
		intermediate->hpll_en);

	WARN_ON(intermediate->sr.fbc > g4x_fbc_fifo_size(1) &&
		intermediate->fbc_en && intermediate->cxsr);
	WARN_ON(intermediate->hpll.fbc > g4x_fbc_fifo_size(2) &&
		intermediate->fbc_en && intermediate->hpll_en);

out:
	/*
	 * If our intermediate WM are identical to the final WM, then we can
	 * omit the post-vblank programming; only update if it's different.
	 */
	if (memcmp(intermediate, optimal, sizeof(*intermediate)) != 0)
		new_crtc_state->wm.need_postvbl_update = true;

	return 0;
}

static void g4x_merge_wm(struct drm_i915_private *dev_priv,
			 struct g4x_wm_values *wm)
{
	struct intel_crtc *crtc;
	int num_active_pipes = 0;

	wm->cxsr = true;
	wm->hpll_en = true;
	wm->fbc_en = true;

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		const struct g4x_wm_state *wm_state = &crtc->wm.active.g4x;

		if (!crtc->active)
			continue;

		if (!wm_state->cxsr)
			wm->cxsr = false;
		if (!wm_state->hpll_en)
			wm->hpll_en = false;
		if (!wm_state->fbc_en)
			wm->fbc_en = false;

		num_active_pipes++;
	}

	if (num_active_pipes != 1) {
		wm->cxsr = false;
		wm->hpll_en = false;
		wm->fbc_en = false;
	}

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		const struct g4x_wm_state *wm_state = &crtc->wm.active.g4x;
		enum pipe pipe = crtc->pipe;

		wm->pipe[pipe] = wm_state->wm;
		if (crtc->active && wm->cxsr)
			wm->sr = wm_state->sr;
		if (crtc->active && wm->hpll_en)
			wm->hpll = wm_state->hpll;
	}
}

static void g4x_program_watermarks(struct drm_i915_private *dev_priv)
{
	struct g4x_wm_values *old_wm = &dev_priv->wm.g4x;
	struct g4x_wm_values new_wm = {};

	g4x_merge_wm(dev_priv, &new_wm);

	if (memcmp(old_wm, &new_wm, sizeof(new_wm)) == 0)
		return;

	if (is_disabling(old_wm->cxsr, new_wm.cxsr, true))
		_intel_set_memory_cxsr(dev_priv, false);

	g4x_write_wm_values(dev_priv, &new_wm);

	if (is_enabling(old_wm->cxsr, new_wm.cxsr, true))
		_intel_set_memory_cxsr(dev_priv, true);

	*old_wm = new_wm;
}

static void g4x_initial_watermarks(struct intel_atomic_state *state,
				   struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);

	mutex_lock(&dev_priv->wm.wm_mutex);
	crtc->wm.active.g4x = crtc_state->wm.g4x.intermediate;
	g4x_program_watermarks(dev_priv);
	mutex_unlock(&dev_priv->wm.wm_mutex);
}

static void g4x_optimize_watermarks(struct intel_atomic_state *state,
				    struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);

	if (!crtc_state->wm.need_postvbl_update)
		return;

	mutex_lock(&dev_priv->wm.wm_mutex);
	crtc->wm.active.g4x = crtc_state->wm.g4x.optimal;
	g4x_program_watermarks(dev_priv);
	mutex_unlock(&dev_priv->wm.wm_mutex);
}

/* latency must be in 0.1us units. */
static unsigned int vlv_wm_method2(unsigned int pixel_rate,
				   unsigned int htotal,
				   unsigned int width,
				   unsigned int cpp,
				   unsigned int latency)
{
	unsigned int ret;

	ret = intel_wm_method2(pixel_rate, htotal,
			       width, cpp, latency);
	ret = DIV_ROUND_UP(ret, 64);

	return ret;
}

static void vlv_setup_wm_latency(struct drm_i915_private *dev_priv)
{
	/* all latencies in usec */
	dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM2] = 3;

	dev_priv->wm.max_level = VLV_WM_LEVEL_PM2;

	if (IS_CHERRYVIEW(dev_priv)) {
		dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM5] = 12;
		dev_priv->wm.pri_latency[VLV_WM_LEVEL_DDR_DVFS] = 33;

		dev_priv->wm.max_level = VLV_WM_LEVEL_DDR_DVFS;
	}
}

static u16 vlv_compute_wm_level(const struct intel_crtc_state *crtc_state,
				const struct intel_plane_state *plane_state,
				int level)
{
	struct intel_plane *plane = to_intel_plane(plane_state->base.plane);
	struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
	const struct drm_display_mode *adjusted_mode =
		&crtc_state->base.adjusted_mode;
	unsigned int clock, htotal, cpp, width, wm;

	if (dev_priv->wm.pri_latency[level] == 0)
		return USHRT_MAX;

	if (!intel_wm_plane_visible(crtc_state, plane_state))
		return 0;

	cpp = plane_state->base.fb->format->cpp[0];
	clock = adjusted_mode->crtc_clock;
	htotal = adjusted_mode->crtc_htotal;
	width = crtc_state->pipe_src_w;

	if (plane->id == PLANE_CURSOR) {
		/*
		 * FIXME the formula gives values that are
		 * too big for the cursor FIFO, and hence we
		 * would never be able to use cursors. For
		 * now just hardcode the watermark.
		 */
		wm = 63;
	} else {
		wm = vlv_wm_method2(clock, htotal, width, cpp,
				    dev_priv->wm.pri_latency[level] * 10);
	}

	return min_t(unsigned int, wm, USHRT_MAX);
}

static bool vlv_need_sprite0_fifo_workaround(unsigned int active_planes)
{
	return (active_planes & (BIT(PLANE_SPRITE0) |
				 BIT(PLANE_SPRITE1))) == BIT(PLANE_SPRITE1);
}

static int vlv_compute_fifo(struct intel_crtc_state *crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
	const struct g4x_pipe_wm *raw =
		&crtc_state->wm.vlv.raw[VLV_WM_LEVEL_PM2];
	struct vlv_fifo_state *fifo_state = &crtc_state->wm.vlv.fifo_state;
	unsigned int active_planes = crtc_state->active_planes & ~BIT(PLANE_CURSOR);
	int num_active_planes = hweight8(active_planes);
	const int fifo_size = 511;
	int fifo_extra, fifo_left = fifo_size;
	int sprite0_fifo_extra = 0;
	unsigned int total_rate;
	enum plane_id plane_id;

	/*
	 * When enabling sprite0 after sprite1 has already been enabled
	 * we tend to get an underrun unless sprite0 already has some
	 * FIFO space allcoated. Hence we always allocate at least one
	 * cacheline for sprite0 whenever sprite1 is enabled.
	 *
	 * All other plane enable sequences appear immune to this problem.
	 */
	if (vlv_need_sprite0_fifo_workaround(active_planes))
		sprite0_fifo_extra = 1;

	total_rate = raw->plane[PLANE_PRIMARY] +
		raw->plane[PLANE_SPRITE0] +
		raw->plane[PLANE_SPRITE1] +
		sprite0_fifo_extra;

	if (total_rate > fifo_size)
		return -EINVAL;

	if (total_rate == 0)
		total_rate = 1;

	for_each_plane_id_on_crtc(crtc, plane_id) {
		unsigned int rate;

		if ((active_planes & BIT(plane_id)) == 0) {
			fifo_state->plane[plane_id] = 0;
			continue;
		}

		rate = raw->plane[plane_id];
		fifo_state->plane[plane_id] = fifo_size * rate / total_rate;
		fifo_left -= fifo_state->plane[plane_id];
	}

	fifo_state->plane[PLANE_SPRITE0] += sprite0_fifo_extra;
	fifo_left -= sprite0_fifo_extra;

	fifo_state->plane[PLANE_CURSOR] = 63;

	fifo_extra = DIV_ROUND_UP(fifo_left, num_active_planes ?: 1);

	/* spread the remainder evenly */
	for_each_plane_id_on_crtc(crtc, plane_id) {
		int plane_extra;

		if (fifo_left == 0)
			break;

		if ((active_planes & BIT(plane_id)) == 0)
			continue;

		plane_extra = min(fifo_extra, fifo_left);
		fifo_state->plane[plane_id] += plane_extra;
		fifo_left -= plane_extra;
	}

	WARN_ON(active_planes != 0 && fifo_left != 0);

	/* give it all to the first plane if none are active */
	if (active_planes == 0) {
		WARN_ON(fifo_left != fifo_size);
		fifo_state->plane[PLANE_PRIMARY] = fifo_left;
	}

	return 0;
}

/* mark all levels starting from 'level' as invalid */
static void vlv_invalidate_wms(struct intel_crtc *crtc,
			       struct vlv_wm_state *wm_state, int level)
{
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);

	for (; level < intel_wm_num_levels(dev_priv); level++) {
		enum plane_id plane_id;

		for_each_plane_id_on_crtc(crtc, plane_id)
			wm_state->wm[level].plane[plane_id] = USHRT_MAX;

		wm_state->sr[level].cursor = USHRT_MAX;
		wm_state->sr[level].plane = USHRT_MAX;
	}
}

static u16 vlv_invert_wm_value(u16 wm, u16 fifo_size)
{
	if (wm > fifo_size)
		return USHRT_MAX;
	else
		return fifo_size - wm;
}

/*
 * Starting from 'level' set all higher
 * levels to 'value' in the "raw" watermarks.
 */
static bool vlv_raw_plane_wm_set(struct intel_crtc_state *crtc_state,
				 int level, enum plane_id plane_id, u16 value)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	int num_levels = intel_wm_num_levels(dev_priv);
	bool dirty = false;

	for (; level < num_levels; level++) {
		struct g4x_pipe_wm *raw = &crtc_state->wm.vlv.raw[level];

		dirty |= raw->plane[plane_id] != value;
		raw->plane[plane_id] = value;
	}

	return dirty;
}

static bool vlv_raw_plane_wm_compute(struct intel_crtc_state *crtc_state,
				     const struct intel_plane_state *plane_state)
{
	struct intel_plane *plane = to_intel_plane(plane_state->base.plane);
	enum plane_id plane_id = plane->id;
	int num_levels = intel_wm_num_levels(to_i915(plane->base.dev));
	int level;
	bool dirty = false;

	if (!intel_wm_plane_visible(crtc_state, plane_state)) {
		dirty |= vlv_raw_plane_wm_set(crtc_state, 0, plane_id, 0);
		goto out;
	}

	for (level = 0; level < num_levels; level++) {
		struct g4x_pipe_wm *raw = &crtc_state->wm.vlv.raw[level];
		int wm = vlv_compute_wm_level(crtc_state, plane_state, level);
		int max_wm = plane_id == PLANE_CURSOR ? 63 : 511;

		if (wm > max_wm)
			break;

		dirty |= raw->plane[plane_id] != wm;
		raw->plane[plane_id] = wm;
	}

	/* mark all higher levels as invalid */
	dirty |= vlv_raw_plane_wm_set(crtc_state, level, plane_id, USHRT_MAX);

out:
	if (dirty)
		DRM_DEBUG_KMS("%s watermarks: PM2=%d, PM5=%d, DDR DVFS=%d\n",
			      plane->base.name,
			      crtc_state->wm.vlv.raw[VLV_WM_LEVEL_PM2].plane[plane_id],
			      crtc_state->wm.vlv.raw[VLV_WM_LEVEL_PM5].plane[plane_id],
			      crtc_state->wm.vlv.raw[VLV_WM_LEVEL_DDR_DVFS].plane[plane_id]);

	return dirty;
}

static bool vlv_raw_plane_wm_is_valid(const struct intel_crtc_state *crtc_state,
				      enum plane_id plane_id, int level)
{
	const struct g4x_pipe_wm *raw =
		&crtc_state->wm.vlv.raw[level];
	const struct vlv_fifo_state *fifo_state =
		&crtc_state->wm.vlv.fifo_state;

	return raw->plane[plane_id] <= fifo_state->plane[plane_id];
}

static bool vlv_raw_crtc_wm_is_valid(const struct intel_crtc_state *crtc_state, int level)
{
	return vlv_raw_plane_wm_is_valid(crtc_state, PLANE_PRIMARY, level) &&
		vlv_raw_plane_wm_is_valid(crtc_state, PLANE_SPRITE0, level) &&
		vlv_raw_plane_wm_is_valid(crtc_state, PLANE_SPRITE1, level) &&
		vlv_raw_plane_wm_is_valid(crtc_state, PLANE_CURSOR, level);
}

static int vlv_compute_pipe_wm(struct intel_crtc_state *crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	struct intel_atomic_state *state =
		to_intel_atomic_state(crtc_state->base.state);
	struct vlv_wm_state *wm_state = &crtc_state->wm.vlv.optimal;
	const struct vlv_fifo_state *fifo_state =
		&crtc_state->wm.vlv.fifo_state;
	int num_active_planes = hweight8(crtc_state->active_planes &
					 ~BIT(PLANE_CURSOR));
	bool needs_modeset = drm_atomic_crtc_needs_modeset(&crtc_state->base);
	const struct intel_plane_state *old_plane_state;
	const struct intel_plane_state *new_plane_state;
	struct intel_plane *plane;
	enum plane_id plane_id;
	int level, ret, i;
	unsigned int dirty = 0;

	for_each_oldnew_intel_plane_in_state(state, plane,
					     old_plane_state,
					     new_plane_state, i) {
		if (new_plane_state->base.crtc != &crtc->base &&
		    old_plane_state->base.crtc != &crtc->base)
			continue;

		if (vlv_raw_plane_wm_compute(crtc_state, new_plane_state))
			dirty |= BIT(plane->id);
	}

	/*
	 * DSPARB registers may have been reset due to the
	 * power well being turned off. Make sure we restore
	 * them to a consistent state even if no primary/sprite
	 * planes are initially active.
	 */
	if (needs_modeset)
		crtc_state->fifo_changed = true;

	if (!dirty)
		return 0;

	/* cursor changes don't warrant a FIFO recompute */
	if (dirty & ~BIT(PLANE_CURSOR)) {
		const struct intel_crtc_state *old_crtc_state =
			intel_atomic_get_old_crtc_state(state, crtc);
		const struct vlv_fifo_state *old_fifo_state =
			&old_crtc_state->wm.vlv.fifo_state;

		ret = vlv_compute_fifo(crtc_state);
		if (ret)
			return ret;

		if (needs_modeset ||
		    memcmp(old_fifo_state, fifo_state,
			   sizeof(*fifo_state)) != 0)
			crtc_state->fifo_changed = true;
	}

	/* initially allow all levels */
	wm_state->num_levels = intel_wm_num_levels(dev_priv);
	/*
	 * Note that enabling cxsr with no primary/sprite planes
	 * enabled can wedge the pipe. Hence we only allow cxsr
	 * with exactly one enabled primary/sprite plane.
	 */
	wm_state->cxsr = crtc->pipe != PIPE_C && num_active_planes == 1;

	for (level = 0; level < wm_state->num_levels; level++) {
		const struct g4x_pipe_wm *raw = &crtc_state->wm.vlv.raw[level];
		const int sr_fifo_size = INTEL_NUM_PIPES(dev_priv) * 512 - 1;

		if (!vlv_raw_crtc_wm_is_valid(crtc_state, level))
			break;

		for_each_plane_id_on_crtc(crtc, plane_id) {
			wm_state->wm[level].plane[plane_id] =
				vlv_invert_wm_value(raw->plane[plane_id],
						    fifo_state->plane[plane_id]);
		}

		wm_state->sr[level].plane =
			vlv_invert_wm_value(max3(raw->plane[PLANE_PRIMARY],
						 raw->plane[PLANE_SPRITE0],
						 raw->plane[PLANE_SPRITE1]),
					    sr_fifo_size);

		wm_state->sr[level].cursor =
			vlv_invert_wm_value(raw->plane[PLANE_CURSOR],
					    63);
	}

	if (level == 0)
		return -EINVAL;

	/* limit to only levels we can actually handle */
	wm_state->num_levels = level;

	/* invalidate the higher levels */
	vlv_invalidate_wms(crtc, wm_state, level);

	return 0;
}

#define VLV_FIFO(plane, value) \
	(((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV)

static void vlv_atomic_update_fifo(struct intel_atomic_state *state,
				   struct intel_crtc_state *crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	struct intel_uncore *uncore = &dev_priv->uncore;
	const struct vlv_fifo_state *fifo_state =
		&crtc_state->wm.vlv.fifo_state;
	int sprite0_start, sprite1_start, fifo_size;

	if (!crtc_state->fifo_changed)
		return;

	sprite0_start = fifo_state->plane[PLANE_PRIMARY];
	sprite1_start = fifo_state->plane[PLANE_SPRITE0] + sprite0_start;
	fifo_size = fifo_state->plane[PLANE_SPRITE1] + sprite1_start;

	WARN_ON(fifo_state->plane[PLANE_CURSOR] != 63);
	WARN_ON(fifo_size != 511);

	trace_vlv_fifo_size(crtc, sprite0_start, sprite1_start, fifo_size);

	/*
	 * uncore.lock serves a double purpose here. It allows us to
	 * use the less expensive I915_{READ,WRITE}_FW() functions, and
	 * it protects the DSPARB registers from getting clobbered by
	 * parallel updates from multiple pipes.
	 *
	 * intel_pipe_update_start() has already disabled interrupts
	 * for us, so a plain spin_lock() is sufficient here.
	 */
	spin_lock(&uncore->lock);

	switch (crtc->pipe) {
		u32 dsparb, dsparb2, dsparb3;
	case PIPE_A:
		dsparb = intel_uncore_read_fw(uncore, DSPARB);
		dsparb2 = intel_uncore_read_fw(uncore, DSPARB2);

		dsparb &= ~(VLV_FIFO(SPRITEA, 0xff) |
			    VLV_FIFO(SPRITEB, 0xff));
		dsparb |= (VLV_FIFO(SPRITEA, sprite0_start) |
			   VLV_FIFO(SPRITEB, sprite1_start));

		dsparb2 &= ~(VLV_FIFO(SPRITEA_HI, 0x1) |
			     VLV_FIFO(SPRITEB_HI, 0x1));
		dsparb2 |= (VLV_FIFO(SPRITEA_HI, sprite0_start >> 8) |
			   VLV_FIFO(SPRITEB_HI, sprite1_start >> 8));

		intel_uncore_write_fw(uncore, DSPARB, dsparb);
		intel_uncore_write_fw(uncore, DSPARB2, dsparb2);
		break;
	case PIPE_B:
		dsparb = intel_uncore_read_fw(uncore, DSPARB);
		dsparb2 = intel_uncore_read_fw(uncore, DSPARB2);

		dsparb &= ~(VLV_FIFO(SPRITEC, 0xff) |
			    VLV_FIFO(SPRITED, 0xff));
		dsparb |= (VLV_FIFO(SPRITEC, sprite0_start) |
			   VLV_FIFO(SPRITED, sprite1_start));

		dsparb2 &= ~(VLV_FIFO(SPRITEC_HI, 0xff) |
			     VLV_FIFO(SPRITED_HI, 0xff));
		dsparb2 |= (VLV_FIFO(SPRITEC_HI, sprite0_start >> 8) |
			   VLV_FIFO(SPRITED_HI, sprite1_start >> 8));

		intel_uncore_write_fw(uncore, DSPARB, dsparb);
		intel_uncore_write_fw(uncore, DSPARB2, dsparb2);
		break;
	case PIPE_C:
		dsparb3 = intel_uncore_read_fw(uncore, DSPARB3);
		dsparb2 = intel_uncore_read_fw(uncore, DSPARB2);

		dsparb3 &= ~(VLV_FIFO(SPRITEE, 0xff) |
			     VLV_FIFO(SPRITEF, 0xff));
		dsparb3 |= (VLV_FIFO(SPRITEE, sprite0_start) |
			    VLV_FIFO(SPRITEF, sprite1_start));

		dsparb2 &= ~(VLV_FIFO(SPRITEE_HI, 0xff) |
			     VLV_FIFO(SPRITEF_HI, 0xff));
		dsparb2 |= (VLV_FIFO(SPRITEE_HI, sprite0_start >> 8) |
			   VLV_FIFO(SPRITEF_HI, sprite1_start >> 8));

		intel_uncore_write_fw(uncore, DSPARB3, dsparb3);
		intel_uncore_write_fw(uncore, DSPARB2, dsparb2);
		break;
	default:
		break;
	}

	intel_uncore_posting_read_fw(uncore, DSPARB);

	spin_unlock(&uncore->lock);
}

#undef VLV_FIFO

static int vlv_compute_intermediate_wm(struct intel_crtc_state *new_crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(new_crtc_state->base.crtc);
	struct vlv_wm_state *intermediate = &new_crtc_state->wm.vlv.intermediate;
	const struct vlv_wm_state *optimal = &new_crtc_state->wm.vlv.optimal;
	struct intel_atomic_state *intel_state =
		to_intel_atomic_state(new_crtc_state->base.state);
	const struct intel_crtc_state *old_crtc_state =
		intel_atomic_get_old_crtc_state(intel_state, crtc);
	const struct vlv_wm_state *active = &old_crtc_state->wm.vlv.optimal;
	int level;

	if (!new_crtc_state->base.active || drm_atomic_crtc_needs_modeset(&new_crtc_state->base)) {
		*intermediate = *optimal;

		intermediate->cxsr = false;
		goto out;
	}

	intermediate->num_levels = min(optimal->num_levels, active->num_levels);
	intermediate->cxsr = optimal->cxsr && active->cxsr &&
		!new_crtc_state->disable_cxsr;

	for (level = 0; level < intermediate->num_levels; level++) {
		enum plane_id plane_id;

		for_each_plane_id_on_crtc(crtc, plane_id) {
			intermediate->wm[level].plane[plane_id] =
				min(optimal->wm[level].plane[plane_id],
				    active->wm[level].plane[plane_id]);
		}

		intermediate->sr[level].plane = min(optimal->sr[level].plane,
						    active->sr[level].plane);
		intermediate->sr[level].cursor = min(optimal->sr[level].cursor,
						     active->sr[level].cursor);
	}

	vlv_invalidate_wms(crtc, intermediate, level);

out:
	/*
	 * If our intermediate WM are identical to the final WM, then we can
	 * omit the post-vblank programming; only update if it's different.
	 */
	if (memcmp(intermediate, optimal, sizeof(*intermediate)) != 0)
		new_crtc_state->wm.need_postvbl_update = true;

	return 0;
}

static void vlv_merge_wm(struct drm_i915_private *dev_priv,
			 struct vlv_wm_values *wm)
{
	struct intel_crtc *crtc;
	int num_active_pipes = 0;

	wm->level = dev_priv->wm.max_level;
	wm->cxsr = true;

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		const struct vlv_wm_state *wm_state = &crtc->wm.active.vlv;

		if (!crtc->active)
			continue;

		if (!wm_state->cxsr)
			wm->cxsr = false;

		num_active_pipes++;
		wm->level = min_t(int, wm->level, wm_state->num_levels - 1);
	}

	if (num_active_pipes != 1)
		wm->cxsr = false;

	if (num_active_pipes > 1)
		wm->level = VLV_WM_LEVEL_PM2;

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		const struct vlv_wm_state *wm_state = &crtc->wm.active.vlv;
		enum pipe pipe = crtc->pipe;

		wm->pipe[pipe] = wm_state->wm[wm->level];
		if (crtc->active && wm->cxsr)
			wm->sr = wm_state->sr[wm->level];

		wm->ddl[pipe].plane[PLANE_PRIMARY] = DDL_PRECISION_HIGH | 2;
		wm->ddl[pipe].plane[PLANE_SPRITE0] = DDL_PRECISION_HIGH | 2;
		wm->ddl[pipe].plane[PLANE_SPRITE1] = DDL_PRECISION_HIGH | 2;
		wm->ddl[pipe].plane[PLANE_CURSOR] = DDL_PRECISION_HIGH | 2;
	}
}

static void vlv_program_watermarks(struct drm_i915_private *dev_priv)
{
	struct vlv_wm_values *old_wm = &dev_priv->wm.vlv;
	struct vlv_wm_values new_wm = {};

	vlv_merge_wm(dev_priv, &new_wm);

	if (memcmp(old_wm, &new_wm, sizeof(new_wm)) == 0)
		return;

	if (is_disabling(old_wm->level, new_wm.level, VLV_WM_LEVEL_DDR_DVFS))
		chv_set_memory_dvfs(dev_priv, false);

	if (is_disabling(old_wm->level, new_wm.level, VLV_WM_LEVEL_PM5))
		chv_set_memory_pm5(dev_priv, false);

	if (is_disabling(old_wm->cxsr, new_wm.cxsr, true))
		_intel_set_memory_cxsr(dev_priv, false);

	vlv_write_wm_values(dev_priv, &new_wm);

	if (is_enabling(old_wm->cxsr, new_wm.cxsr, true))
		_intel_set_memory_cxsr(dev_priv, true);

	if (is_enabling(old_wm->level, new_wm.level, VLV_WM_LEVEL_PM5))
		chv_set_memory_pm5(dev_priv, true);

	if (is_enabling(old_wm->level, new_wm.level, VLV_WM_LEVEL_DDR_DVFS))
		chv_set_memory_dvfs(dev_priv, true);

	*old_wm = new_wm;
}

static void vlv_initial_watermarks(struct intel_atomic_state *state,
				   struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);

	mutex_lock(&dev_priv->wm.wm_mutex);
	crtc->wm.active.vlv = crtc_state->wm.vlv.intermediate;
	vlv_program_watermarks(dev_priv);
	mutex_unlock(&dev_priv->wm.wm_mutex);
}

static void vlv_optimize_watermarks(struct intel_atomic_state *state,
				    struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);

	if (!crtc_state->wm.need_postvbl_update)
		return;

	mutex_lock(&dev_priv->wm.wm_mutex);
	crtc->wm.active.vlv = crtc_state->wm.vlv.optimal;
	vlv_program_watermarks(dev_priv);
	mutex_unlock(&dev_priv->wm.wm_mutex);
}

static void i965_update_wm(struct intel_crtc *unused_crtc)
{
	struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev);
	struct intel_crtc *crtc;
	int srwm = 1;
	int cursor_sr = 16;
	bool cxsr_enabled;

	/* Calc sr entries for one plane configs */
	crtc = single_enabled_crtc(dev_priv);
	if (crtc) {
		/* self-refresh has much higher latency */
		static const int sr_latency_ns = 12000;
		const struct drm_display_mode *adjusted_mode =
			&crtc->config->base.adjusted_mode;
		const struct drm_framebuffer *fb =
			crtc->base.primary->state->fb;
		int clock = adjusted_mode->crtc_clock;
		int htotal = adjusted_mode->crtc_htotal;
		int hdisplay = crtc->config->pipe_src_w;
		int cpp = fb->format->cpp[0];
		int entries;

		entries = intel_wm_method2(clock, htotal,
					   hdisplay, cpp, sr_latency_ns / 100);
		entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
		srwm = I965_FIFO_SIZE - entries;
		if (srwm < 0)
			srwm = 1;
		srwm &= 0x1ff;
		DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
			      entries, srwm);

		entries = intel_wm_method2(clock, htotal,
					   crtc->base.cursor->state->crtc_w, 4,
					   sr_latency_ns / 100);
		entries = DIV_ROUND_UP(entries,
				       i965_cursor_wm_info.cacheline_size) +
			i965_cursor_wm_info.guard_size;

		cursor_sr = i965_cursor_wm_info.fifo_size - entries;
		if (cursor_sr > i965_cursor_wm_info.max_wm)
			cursor_sr = i965_cursor_wm_info.max_wm;

		DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
			      "cursor %d\n", srwm, cursor_sr);

		cxsr_enabled = true;
	} else {
		cxsr_enabled = false;
		/* Turn off self refresh if both pipes are enabled */
		intel_set_memory_cxsr(dev_priv, false);
	}

	DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
		      srwm);

	/* 965 has limitations... */
	I915_WRITE(DSPFW1, FW_WM(srwm, SR) |
		   FW_WM(8, CURSORB) |
		   FW_WM(8, PLANEB) |
		   FW_WM(8, PLANEA));
	I915_WRITE(DSPFW2, FW_WM(8, CURSORA) |
		   FW_WM(8, PLANEC_OLD));
	/* update cursor SR watermark */
	I915_WRITE(DSPFW3, FW_WM(cursor_sr, CURSOR_SR));

	if (cxsr_enabled)
		intel_set_memory_cxsr(dev_priv, true);
}

#undef FW_WM

static void i9xx_update_wm(struct intel_crtc *unused_crtc)
{
	struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev);
	const struct intel_watermark_params *wm_info;
	u32 fwater_lo;
	u32 fwater_hi;
	int cwm, srwm = 1;
	int fifo_size;
	int planea_wm, planeb_wm;
	struct intel_crtc *crtc, *enabled = NULL;

	if (IS_I945GM(dev_priv))
		wm_info = &i945_wm_info;
	else if (!IS_GEN(dev_priv, 2))
		wm_info = &i915_wm_info;
	else
		wm_info = &i830_a_wm_info;

	fifo_size = dev_priv->display.get_fifo_size(dev_priv, PLANE_A);
	crtc = intel_get_crtc_for_plane(dev_priv, PLANE_A);
	if (intel_crtc_active(crtc)) {
		const struct drm_display_mode *adjusted_mode =
			&crtc->config->base.adjusted_mode;
		const struct drm_framebuffer *fb =
			crtc->base.primary->state->fb;
		int cpp;

		if (IS_GEN(dev_priv, 2))
			cpp = 4;
		else
			cpp = fb->format->cpp[0];

		planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
					       wm_info, fifo_size, cpp,
					       pessimal_latency_ns);
		enabled = crtc;
	} else {
		planea_wm = fifo_size - wm_info->guard_size;
		if (planea_wm > (long)wm_info->max_wm)
			planea_wm = wm_info->max_wm;
	}

	if (IS_GEN(dev_priv, 2))
		wm_info = &i830_bc_wm_info;

	fifo_size = dev_priv->display.get_fifo_size(dev_priv, PLANE_B);
	crtc = intel_get_crtc_for_plane(dev_priv, PLANE_B);
	if (intel_crtc_active(crtc)) {
		const struct drm_display_mode *adjusted_mode =
			&crtc->config->base.adjusted_mode;
		const struct drm_framebuffer *fb =
			crtc->base.primary->state->fb;
		int cpp;

		if (IS_GEN(dev_priv, 2))
			cpp = 4;
		else
			cpp = fb->format->cpp[0];

		planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
					       wm_info, fifo_size, cpp,
					       pessimal_latency_ns);
		if (enabled == NULL)
			enabled = crtc;
		else
			enabled = NULL;
	} else {
		planeb_wm = fifo_size - wm_info->guard_size;
		if (planeb_wm > (long)wm_info->max_wm)
			planeb_wm = wm_info->max_wm;
	}

	DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);

	if (IS_I915GM(dev_priv) && enabled) {
		struct drm_i915_gem_object *obj;

		obj = intel_fb_obj(enabled->base.primary->state->fb);

		/* self-refresh seems busted with untiled */
		if (!i915_gem_object_is_tiled(obj))
			enabled = NULL;
	}

	/*
	 * Overlay gets an aggressive default since video jitter is bad.
	 */
	cwm = 2;

	/* Play safe and disable self-refresh before adjusting watermarks. */
	intel_set_memory_cxsr(dev_priv, false);

	/* Calc sr entries for one plane configs */
	if (HAS_FW_BLC(dev_priv) && enabled) {
		/* self-refresh has much higher latency */
		static const int sr_latency_ns = 6000;
		const struct drm_display_mode *adjusted_mode =
			&enabled->config->base.adjusted_mode;
		const struct drm_framebuffer *fb =
			enabled->base.primary->state->fb;
		int clock = adjusted_mode->crtc_clock;
		int htotal = adjusted_mode->crtc_htotal;
		int hdisplay = enabled->config->pipe_src_w;
		int cpp;
		int entries;

		if (IS_I915GM(dev_priv) || IS_I945GM(dev_priv))
			cpp = 4;
		else
			cpp = fb->format->cpp[0];

		entries = intel_wm_method2(clock, htotal, hdisplay, cpp,
					   sr_latency_ns / 100);
		entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
		DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
		srwm = wm_info->fifo_size - entries;
		if (srwm < 0)
			srwm = 1;

		if (IS_I945G(dev_priv) || IS_I945GM(dev_priv))
			I915_WRITE(FW_BLC_SELF,
				   FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
		else
			I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
	}

	DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
		      planea_wm, planeb_wm, cwm, srwm);

	fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
	fwater_hi = (cwm & 0x1f);

	/* Set request length to 8 cachelines per fetch */
	fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
	fwater_hi = fwater_hi | (1 << 8);

	I915_WRITE(FW_BLC, fwater_lo);
	I915_WRITE(FW_BLC2, fwater_hi);

	if (enabled)
		intel_set_memory_cxsr(dev_priv, true);
}

static void i845_update_wm(struct intel_crtc *unused_crtc)
{
	struct drm_i915_private *dev_priv = to_i915(unused_crtc->base.dev);
	struct intel_crtc *crtc;
	const struct drm_display_mode *adjusted_mode;
	u32 fwater_lo;
	int planea_wm;

	crtc = single_enabled_crtc(dev_priv);
	if (crtc == NULL)
		return;

	adjusted_mode = &crtc->config->base.adjusted_mode;
	planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
				       &i845_wm_info,
				       dev_priv->display.get_fifo_size(dev_priv, PLANE_A),
				       4, pessimal_latency_ns);
	fwater_lo = I915_READ(FW_BLC) & ~0xfff;
	fwater_lo |= (3<<8) | planea_wm;

	DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);

	I915_WRITE(FW_BLC, fwater_lo);
}

/* latency must be in 0.1us units. */
static unsigned int ilk_wm_method1(unsigned int pixel_rate,
				   unsigned int cpp,
				   unsigned int latency)
{
	unsigned int ret;

	ret = intel_wm_method1(pixel_rate, cpp, latency);
	ret = DIV_ROUND_UP(ret, 64) + 2;

	return ret;
}

/* latency must be in 0.1us units. */
static unsigned int ilk_wm_method2(unsigned int pixel_rate,
				   unsigned int htotal,
				   unsigned int width,
				   unsigned int cpp,
				   unsigned int latency)
{
	unsigned int ret;

	ret = intel_wm_method2(pixel_rate, htotal,
			       width, cpp, latency);
	ret = DIV_ROUND_UP(ret, 64) + 2;

	return ret;
}

static u32 ilk_wm_fbc(u32 pri_val, u32 horiz_pixels, u8 cpp)
{
	/*
	 * Neither of these should be possible since this function shouldn't be
	 * called if the CRTC is off or the plane is invisible.  But let's be
	 * extra paranoid to avoid a potential divide-by-zero if we screw up
	 * elsewhere in the driver.
	 */
	if (WARN_ON(!cpp))
		return 0;
	if (WARN_ON(!horiz_pixels))
		return 0;

	return DIV_ROUND_UP(pri_val * 64, horiz_pixels * cpp) + 2;
}

struct ilk_wm_maximums {
	u16 pri;
	u16 spr;
	u16 cur;
	u16 fbc;
};

/*
 * For both WM_PIPE and WM_LP.
 * mem_value must be in 0.1us units.
 */
static u32 ilk_compute_pri_wm(const struct intel_crtc_state *crtc_state,
			      const struct intel_plane_state *plane_state,
			      u32 mem_value, bool is_lp)
{
	u32 method1, method2;
	int cpp;

	if (mem_value == 0)
		return U32_MAX;

	if (!intel_wm_plane_visible(crtc_state, plane_state))
		return 0;

	cpp = plane_state->base.fb->format->cpp[0];

	method1 = ilk_wm_method1(crtc_state->pixel_rate, cpp, mem_value);

	if (!is_lp)
		return method1;

	method2 = ilk_wm_method2(crtc_state->pixel_rate,
				 crtc_state->base.adjusted_mode.crtc_htotal,
				 drm_rect_width(&plane_state->base.dst),
				 cpp, mem_value);

	return min(method1, method2);
}

/*
 * For both WM_PIPE and WM_LP.
 * mem_value must be in 0.1us units.
 */
static u32 ilk_compute_spr_wm(const struct intel_crtc_state *crtc_state,
			      const struct intel_plane_state *plane_state,
			      u32 mem_value)
{
	u32 method1, method2;
	int cpp;

	if (mem_value == 0)
		return U32_MAX;

	if (!intel_wm_plane_visible(crtc_state, plane_state))
		return 0;

	cpp = plane_state->base.fb->format->cpp[0];

	method1 = ilk_wm_method1(crtc_state->pixel_rate, cpp, mem_value);
	method2 = ilk_wm_method2(crtc_state->pixel_rate,
				 crtc_state->base.adjusted_mode.crtc_htotal,
				 drm_rect_width(&plane_state->base.dst),
				 cpp, mem_value);
	return min(method1, method2);
}

/*
 * For both WM_PIPE and WM_LP.
 * mem_value must be in 0.1us units.
 */
static u32 ilk_compute_cur_wm(const struct intel_crtc_state *crtc_state,
			      const struct intel_plane_state *plane_state,
			      u32 mem_value)
{
	int cpp;

	if (mem_value == 0)
		return U32_MAX;

	if (!intel_wm_plane_visible(crtc_state, plane_state))
		return 0;

	cpp = plane_state->base.fb->format->cpp[0];

	return ilk_wm_method2(crtc_state->pixel_rate,
			      crtc_state->base.adjusted_mode.crtc_htotal,
			      drm_rect_width(&plane_state->base.dst),
			      cpp, mem_value);
}

/* Only for WM_LP. */
static u32 ilk_compute_fbc_wm(const struct intel_crtc_state *crtc_state,
			      const struct intel_plane_state *plane_state,
			      u32 pri_val)
{
	int cpp;

	if (!intel_wm_plane_visible(crtc_state, plane_state))
		return 0;

	cpp = plane_state->base.fb->format->cpp[0];

	return ilk_wm_fbc(pri_val, drm_rect_width(&plane_state->base.dst), cpp);
}

static unsigned int
ilk_display_fifo_size(const struct drm_i915_private *dev_priv)
{
	if (INTEL_GEN(dev_priv) >= 8)
		return 3072;
	else if (INTEL_GEN(dev_priv) >= 7)
		return 768;
	else
		return 512;
}

static unsigned int
ilk_plane_wm_reg_max(const struct drm_i915_private *dev_priv,
		     int level, bool is_sprite)
{
	if (INTEL_GEN(dev_priv) >= 8)
		/* BDW primary/sprite plane watermarks */
		return level == 0 ? 255 : 2047;
	else if (INTEL_GEN(dev_priv) >= 7)
		/* IVB/HSW primary/sprite plane watermarks */
		return level == 0 ? 127 : 1023;
	else if (!is_sprite)
		/* ILK/SNB primary plane watermarks */
		return level == 0 ? 127 : 511;
	else
		/* ILK/SNB sprite plane watermarks */
		return level == 0 ? 63 : 255;
}

static unsigned int
ilk_cursor_wm_reg_max(const struct drm_i915_private *dev_priv, int level)
{
	if (INTEL_GEN(dev_priv) >= 7)
		return level == 0 ? 63 : 255;
	else
		return level == 0 ? 31 : 63;
}

static unsigned int ilk_fbc_wm_reg_max(const struct drm_i915_private *dev_priv)
{
	if (INTEL_GEN(dev_priv) >= 8)
		return 31;
	else
		return 15;
}

/* Calculate the maximum primary/sprite plane watermark */
static unsigned int ilk_plane_wm_max(const struct drm_i915_private *dev_priv,
				     int level,
				     const struct intel_wm_config *config,
				     enum intel_ddb_partitioning ddb_partitioning,
				     bool is_sprite)
{
	unsigned int fifo_size = ilk_display_fifo_size(dev_priv);

	/* if sprites aren't enabled, sprites get nothing */
	if (is_sprite && !config->sprites_enabled)
		return 0;

	/* HSW allows LP1+ watermarks even with multiple pipes */
	if (level == 0 || config->num_pipes_active > 1) {
		fifo_size /= INTEL_NUM_PIPES(dev_priv);

		/*
		 * For some reason the non self refresh
		 * FIFO size is only half of the self
		 * refresh FIFO size on ILK/SNB.
		 */
		if (INTEL_GEN(dev_priv) <= 6)
			fifo_size /= 2;
	}

	if (config->sprites_enabled) {
		/* level 0 is always calculated with 1:1 split */
		if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
			if (is_sprite)
				fifo_size *= 5;
			fifo_size /= 6;
		} else {
			fifo_size /= 2;
		}
	}

	/* clamp to max that the registers can hold */
	return min(fifo_size, ilk_plane_wm_reg_max(dev_priv, level, is_sprite));
}

/* Calculate the maximum cursor plane watermark */
static unsigned int ilk_cursor_wm_max(const struct drm_i915_private *dev_priv,
				      int level,
				      const struct intel_wm_config *config)
{
	/* HSW LP1+ watermarks w/ multiple pipes */
	if (level > 0 && config->num_pipes_active > 1)
		return 64;

	/* otherwise just report max that registers can hold */
	return ilk_cursor_wm_reg_max(dev_priv, level);
}

static void ilk_compute_wm_maximums(const struct drm_i915_private *dev_priv,
				    int level,
				    const struct intel_wm_config *config,
				    enum intel_ddb_partitioning ddb_partitioning,
				    struct ilk_wm_maximums *max)
{
	max->pri = ilk_plane_wm_max(dev_priv, level, config, ddb_partitioning, false);
	max->spr = ilk_plane_wm_max(dev_priv, level, config, ddb_partitioning, true);
	max->cur = ilk_cursor_wm_max(dev_priv, level, config);
	max->fbc = ilk_fbc_wm_reg_max(dev_priv);
}

static void ilk_compute_wm_reg_maximums(const struct drm_i915_private *dev_priv,
					int level,
					struct ilk_wm_maximums *max)
{
	max->pri = ilk_plane_wm_reg_max(dev_priv, level, false);
	max->spr = ilk_plane_wm_reg_max(dev_priv, level, true);
	max->cur = ilk_cursor_wm_reg_max(dev_priv, level);
	max->fbc = ilk_fbc_wm_reg_max(dev_priv);
}

static bool ilk_validate_wm_level(int level,
				  const struct ilk_wm_maximums *max,
				  struct intel_wm_level *result)
{
	bool ret;

	/* already determined to be invalid? */
	if (!result->enable)
		return false;

	result->enable = result->pri_val <= max->pri &&
			 result->spr_val <= max->spr &&
			 result->cur_val <= max->cur;

	ret = result->enable;

	/*
	 * HACK until we can pre-compute everything,
	 * and thus fail gracefully if LP0 watermarks
	 * are exceeded...
	 */
	if (level == 0 && !result->enable) {
		if (result->pri_val > max->pri)
			DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
				      level, result->pri_val, max->pri);
		if (result->spr_val > max->spr)
			DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
				      level, result->spr_val, max->spr);
		if (result->cur_val > max->cur)
			DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
				      level, result->cur_val, max->cur);

		result->pri_val = min_t(u32, result->pri_val, max->pri);
		result->spr_val = min_t(u32, result->spr_val, max->spr);
		result->cur_val = min_t(u32, result->cur_val, max->cur);
		result->enable = true;
	}

	return ret;
}

static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv,
				 const struct intel_crtc *intel_crtc,
				 int level,
				 struct intel_crtc_state *crtc_state,
				 const struct intel_plane_state *pristate,
				 const struct intel_plane_state *sprstate,
				 const struct intel_plane_state *curstate,
				 struct intel_wm_level *result)
{
	u16 pri_latency = dev_priv->wm.pri_latency[level];
	u16 spr_latency = dev_priv->wm.spr_latency[level];
	u16 cur_latency = dev_priv->wm.cur_latency[level];

	/* WM1+ latency values stored in 0.5us units */
	if (level > 0) {
		pri_latency *= 5;
		spr_latency *= 5;
		cur_latency *= 5;
	}

	if (pristate) {
		result->pri_val = ilk_compute_pri_wm(crtc_state, pristate,
						     pri_latency, level);
		result->fbc_val = ilk_compute_fbc_wm(crtc_state, pristate, result->pri_val);
	}

	if (sprstate)
		result->spr_val = ilk_compute_spr_wm(crtc_state, sprstate, spr_latency);

	if (curstate)
		result->cur_val = ilk_compute_cur_wm(crtc_state, curstate, cur_latency);

	result->enable = true;
}

static u32
hsw_compute_linetime_wm(const struct intel_crtc_state *crtc_state)
{
	const struct intel_atomic_state *intel_state =
		to_intel_atomic_state(crtc_state->base.state);
	const struct drm_display_mode *adjusted_mode =
		&crtc_state->base.adjusted_mode;
	u32 linetime, ips_linetime;

	if (!crtc_state->base.active)
		return 0;
	if (WARN_ON(adjusted_mode->crtc_clock == 0))
		return 0;
	if (WARN_ON(intel_state->cdclk.logical.cdclk == 0))
		return 0;

	/* The WM are computed with base on how long it takes to fill a single
	 * row at the given clock rate, multiplied by 8.
	 * */
	linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8,
				     adjusted_mode->crtc_clock);
	ips_linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8,
					 intel_state->cdclk.logical.cdclk);

	return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
	       PIPE_WM_LINETIME_TIME(linetime);
}

static void intel_read_wm_latency(struct drm_i915_private *dev_priv,
				  u16 wm[8])
{
	struct intel_uncore *uncore = &dev_priv->uncore;

	if (INTEL_GEN(dev_priv) >= 9) {
		u32 val;
		int ret, i;
		int level, max_level = ilk_wm_max_level(dev_priv);

		/* read the first set of memory latencies[0:3] */
		val = 0; /* data0 to be programmed to 0 for first set */
		ret = sandybridge_pcode_read(dev_priv,
					     GEN9_PCODE_READ_MEM_LATENCY,
					     &val, NULL);

		if (ret) {
			DRM_ERROR("SKL Mailbox read error = %d\n", ret);
			return;
		}

		wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK;
		wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) &
				GEN9_MEM_LATENCY_LEVEL_MASK;
		wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) &
				GEN9_MEM_LATENCY_LEVEL_MASK;
		wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) &
				GEN9_MEM_LATENCY_LEVEL_MASK;

		/* read the second set of memory latencies[4:7] */
		val = 1; /* data0 to be programmed to 1 for second set */
		ret = sandybridge_pcode_read(dev_priv,
					     GEN9_PCODE_READ_MEM_LATENCY,
					     &val, NULL);
		if (ret) {
			DRM_ERROR("SKL Mailbox read error = %d\n", ret);
			return;
		}

		wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK;
		wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) &
				GEN9_MEM_LATENCY_LEVEL_MASK;
		wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) &
				GEN9_MEM_LATENCY_LEVEL_MASK;
		wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) &
				GEN9_MEM_LATENCY_LEVEL_MASK;

		/*
		 * If a level n (n > 1) has a 0us latency, all levels m (m >= n)
		 * need to be disabled. We make sure to sanitize the values out
		 * of the punit to satisfy this requirement.
		 */
		for (level = 1; level <= max_level; level++) {
			if (wm[level] == 0) {
				for (i = level + 1; i <= max_level; i++)
					wm[i] = 0;
				break;
			}
		}

		/*
		 * WaWmMemoryReadLatency:skl+,glk
		 *
		 * punit doesn't take into account the read latency so we need
		 * to add 2us to the various latency levels we retrieve from the
		 * punit when level 0 response data us 0us.
		 */
		if (wm[0] == 0) {
			wm[0] += 2;
			for (level = 1; level <= max_level; level++) {
				if (wm[level] == 0)
					break;
				wm[level] += 2;
			}
		}

		/*
		 * WA Level-0 adjustment for 16GB DIMMs: SKL+
		 * If we could not get dimm info enable this WA to prevent from
		 * any underrun. If not able to get Dimm info assume 16GB dimm
		 * to avoid any underrun.
		 */
		if (dev_priv->dram_info.is_16gb_dimm)
			wm[0] += 1;

	} else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) {
		u64 sskpd = intel_uncore_read64(uncore, MCH_SSKPD);

		wm[0] = (sskpd >> 56) & 0xFF;
		if (wm[0] == 0)
			wm[0] = sskpd & 0xF;
		wm[1] = (sskpd >> 4) & 0xFF;
		wm[2] = (sskpd >> 12) & 0xFF;
		wm[3] = (sskpd >> 20) & 0x1FF;
		wm[4] = (sskpd >> 32) & 0x1FF;
	} else if (INTEL_GEN(dev_priv) >= 6) {
		u32 sskpd = intel_uncore_read(uncore, MCH_SSKPD);

		wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
		wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
		wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
		wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
	} else if (INTEL_GEN(dev_priv) >= 5) {
		u32 mltr = intel_uncore_read(uncore, MLTR_ILK);

		/* ILK primary LP0 latency is 700 ns */
		wm[0] = 7;
		wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
		wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
	} else {
		MISSING_CASE(INTEL_DEVID(dev_priv));
	}
}

static void intel_fixup_spr_wm_latency(struct drm_i915_private *dev_priv,
				       u16 wm[5])
{
	/* ILK sprite LP0 latency is 1300 ns */
	if (IS_GEN(dev_priv, 5))
		wm[0] = 13;
}

static void intel_fixup_cur_wm_latency(struct drm_i915_private *dev_priv,
				       u16 wm[5])
{
	/* ILK cursor LP0 latency is 1300 ns */
	if (IS_GEN(dev_priv, 5))
		wm[0] = 13;
}

int ilk_wm_max_level(const struct drm_i915_private *dev_priv)
{
	/* how many WM levels are we expecting */
	if (INTEL_GEN(dev_priv) >= 9)
		return 7;
	else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
		return 4;
	else if (INTEL_GEN(dev_priv) >= 6)
		return 3;
	else
		return 2;
}

static void intel_print_wm_latency(struct drm_i915_private *dev_priv,
				   const char *name,
				   const u16 wm[8])
{
	int level, max_level = ilk_wm_max_level(dev_priv);

	for (level = 0; level <= max_level; level++) {
		unsigned int latency = wm[level];

		if (latency == 0) {
			DRM_DEBUG_KMS("%s WM%d latency not provided\n",
				      name, level);
			continue;
		}

		/*
		 * - latencies are in us on gen9.
		 * - before then, WM1+ latency values are in 0.5us units
		 */
		if (INTEL_GEN(dev_priv) >= 9)
			latency *= 10;
		else if (level > 0)
			latency *= 5;

		DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
			      name, level, wm[level],
			      latency / 10, latency % 10);
	}
}

static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
				    u16 wm[5], u16 min)
{
	int level, max_level = ilk_wm_max_level(dev_priv);

	if (wm[0] >= min)
		return false;

	wm[0] = max(wm[0], min);
	for (level = 1; level <= max_level; level++)
		wm[level] = max_t(u16, wm[level], DIV_ROUND_UP(min, 5));

	return true;
}

static void snb_wm_latency_quirk(struct drm_i915_private *dev_priv)
{
	bool changed;

	/*
	 * The BIOS provided WM memory latency values are often
	 * inadequate for high resolution displays. Adjust them.
	 */
	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);

	if (!changed)
		return;

	DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
	intel_print_wm_latency(dev_priv, "Primary", dev_priv->wm.pri_latency);
	intel_print_wm_latency(dev_priv, "Sprite", dev_priv->wm.spr_latency);
	intel_print_wm_latency(dev_priv, "Cursor", dev_priv->wm.cur_latency);
}

static void snb_wm_lp3_irq_quirk(struct drm_i915_private *dev_priv)
{
	/*
	 * On some SNB machines (Thinkpad X220 Tablet at least)
	 * LP3 usage can cause vblank interrupts to be lost.
	 * The DEIIR bit will go high but it looks like the CPU
	 * never gets interrupted.
	 *
	 * It's not clear whether other interrupt source could
	 * be affected or if this is somehow limited to vblank
	 * interrupts only. To play it safe we disable LP3
	 * watermarks entirely.
	 */
	if (dev_priv->wm.pri_latency[3] == 0 &&
	    dev_priv->wm.spr_latency[3] == 0 &&
	    dev_priv->wm.cur_latency[3] == 0)
		return;

	dev_priv->wm.pri_latency[3] = 0;
	dev_priv->wm.spr_latency[3] = 0;
	dev_priv->wm.cur_latency[3] = 0;

	DRM_DEBUG_KMS("LP3 watermarks disabled due to potential for lost interrupts\n");
	intel_print_wm_latency(dev_priv, "Primary", dev_priv->wm.pri_latency);
	intel_print_wm_latency(dev_priv, "Sprite", dev_priv->wm.spr_latency);
	intel_print_wm_latency(dev_priv, "Cursor", dev_priv->wm.cur_latency);
}

static void ilk_setup_wm_latency(struct drm_i915_private *dev_priv)
{
	intel_read_wm_latency(dev_priv, dev_priv->wm.pri_latency);

	memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
	       sizeof(dev_priv->wm.pri_latency));
	memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
	       sizeof(dev_priv->wm.pri_latency));

	intel_fixup_spr_wm_latency(dev_priv, dev_priv->wm.spr_latency);
	intel_fixup_cur_wm_latency(dev_priv, dev_priv->wm.cur_latency);

	intel_print_wm_latency(dev_priv, "Primary", dev_priv->wm.pri_latency);
	intel_print_wm_latency(dev_priv, "Sprite", dev_priv->wm.spr_latency);
	intel_print_wm_latency(dev_priv, "Cursor", dev_priv->wm.cur_latency);

	if (IS_GEN(dev_priv, 6)) {
		snb_wm_latency_quirk(dev_priv);
		snb_wm_lp3_irq_quirk(dev_priv);
	}
}

static void skl_setup_wm_latency(struct drm_i915_private *dev_priv)
{
	intel_read_wm_latency(dev_priv, dev_priv->wm.skl_latency);
	intel_print_wm_latency(dev_priv, "Gen9 Plane", dev_priv->wm.skl_latency);
}

static bool ilk_validate_pipe_wm(const struct drm_i915_private *dev_priv,
				 struct intel_pipe_wm *pipe_wm)
{
	/* LP0 watermark maximums depend on this pipe alone */
	const struct intel_wm_config config = {
		.num_pipes_active = 1,
		.sprites_enabled = pipe_wm->sprites_enabled,
		.sprites_scaled = pipe_wm->sprites_scaled,
	};
	struct ilk_wm_maximums max;

	/* LP0 watermarks always use 1/2 DDB partitioning */
	ilk_compute_wm_maximums(dev_priv, 0, &config, INTEL_DDB_PART_1_2, &max);

	/* At least LP0 must be valid */
	if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0])) {
		DRM_DEBUG_KMS("LP0 watermark invalid\n");
		return false;
	}

	return true;
}

/* Compute new watermarks for the pipe */
static int ilk_compute_pipe_wm(struct intel_crtc_state *crtc_state)
{
	struct drm_atomic_state *state = crtc_state->base.state;
	struct intel_crtc *intel_crtc = to_intel_crtc(crtc_state->base.crtc);
	struct intel_pipe_wm *pipe_wm;
	struct drm_device *dev = state->dev;
	const struct drm_i915_private *dev_priv = to_i915(dev);
	struct intel_plane *plane;
	const struct intel_plane_state *plane_state;
	const struct intel_plane_state *pristate = NULL;
	const struct intel_plane_state *sprstate = NULL;
	const struct intel_plane_state *curstate = NULL;
	int level, max_level = ilk_wm_max_level(dev_priv), usable_level;
	struct ilk_wm_maximums max;

	pipe_wm = &crtc_state->wm.ilk.optimal;

	intel_atomic_crtc_state_for_each_plane_state(plane, plane_state, crtc_state) {
		if (plane->base.type == DRM_PLANE_TYPE_PRIMARY)
			pristate = plane_state;
		else if (plane->base.type == DRM_PLANE_TYPE_OVERLAY)
			sprstate = plane_state;
		else if (plane->base.type == DRM_PLANE_TYPE_CURSOR)
			curstate = plane_state;
	}

	pipe_wm->pipe_enabled = crtc_state->base.active;
	if (sprstate) {
		pipe_wm->sprites_enabled = sprstate->base.visible;
		pipe_wm->sprites_scaled = sprstate->base.visible &&
			(drm_rect_width(&sprstate->base.dst) != drm_rect_width(&sprstate->base.src) >> 16 ||
			 drm_rect_height(&sprstate->base.dst) != drm_rect_height(&sprstate->base.src) >> 16);
	}

	usable_level = max_level;

	/* ILK/SNB: LP2+ watermarks only w/o sprites */
	if (INTEL_GEN(dev_priv) <= 6 && pipe_wm->sprites_enabled)
		usable_level = 1;

	/* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
	if (pipe_wm->sprites_scaled)
		usable_level = 0;

	memset(&pipe_wm->wm, 0, sizeof(pipe_wm->wm));
	ilk_compute_wm_level(dev_priv, intel_crtc, 0, crtc_state,
			     pristate, sprstate, curstate, &pipe_wm->wm[0]);

	if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
		pipe_wm->linetime = hsw_compute_linetime_wm(crtc_state);

	if (!ilk_validate_pipe_wm(dev_priv, pipe_wm))
		return -EINVAL;

	ilk_compute_wm_reg_maximums(dev_priv, 1, &max);

	for (level = 1; level <= usable_level; level++) {
		struct intel_wm_level *wm = &pipe_wm->wm[level];

		ilk_compute_wm_level(dev_priv, intel_crtc, level, crtc_state,
				     pristate, sprstate, curstate, wm);

		/*
		 * Disable any watermark level that exceeds the
		 * register maximums since such watermarks are
		 * always invalid.
		 */
		if (!ilk_validate_wm_level(level, &max, wm)) {
			memset(wm, 0, sizeof(*wm));
			break;
		}
	}

	return 0;
}

/*
 * Build a set of 'intermediate' watermark values that satisfy both the old
 * state and the new state.  These can be programmed to the hardware
 * immediately.
 */
static int ilk_compute_intermediate_wm(struct intel_crtc_state *newstate)
{
	struct intel_crtc *intel_crtc = to_intel_crtc(newstate->base.crtc);
	struct drm_i915_private *dev_priv = to_i915(intel_crtc->base.dev);
	struct intel_pipe_wm *a = &newstate->wm.ilk.intermediate;
	struct intel_atomic_state *intel_state =
		to_intel_atomic_state(newstate->base.state);
	const struct intel_crtc_state *oldstate =
		intel_atomic_get_old_crtc_state(intel_state, intel_crtc);
	const struct intel_pipe_wm *b = &oldstate->wm.ilk.optimal;
	int level, max_level = ilk_wm_max_level(dev_priv);

	/*
	 * Start with the final, target watermarks, then combine with the
	 * currently active watermarks to get values that are safe both before
	 * and after the vblank.
	 */
	*a = newstate->wm.ilk.optimal;
	if (!newstate->base.active || drm_atomic_crtc_needs_modeset(&newstate->base) ||
	    intel_state->skip_intermediate_wm)
		return 0;

	a->pipe_enabled |= b->pipe_enabled;
	a->sprites_enabled |= b->sprites_enabled;
	a->sprites_scaled |= b->sprites_scaled;

	for (level = 0; level <= max_level; level++) {
		struct intel_wm_level *a_wm = &a->wm[level];
		const struct intel_wm_level *b_wm = &b->wm[level];

		a_wm->enable &= b_wm->enable;
		a_wm->pri_val = max(a_wm->pri_val, b_wm->pri_val);
		a_wm->spr_val = max(a_wm->spr_val, b_wm->spr_val);
		a_wm->cur_val = max(a_wm->cur_val, b_wm->cur_val);
		a_wm->fbc_val = max(a_wm->fbc_val, b_wm->fbc_val);
	}

	/*
	 * We need to make sure that these merged watermark values are
	 * actually a valid configuration themselves.  If they're not,
	 * there's no safe way to transition from the old state to
	 * the new state, so we need to fail the atomic transaction.
	 */
	if (!ilk_validate_pipe_wm(dev_priv, a))
		return -EINVAL;

	/*
	 * If our intermediate WM are identical to the final WM, then we can
	 * omit the post-vblank programming; only update if it's different.
	 */
	if (memcmp(a, &newstate->wm.ilk.optimal, sizeof(*a)) != 0)
		newstate->wm.need_postvbl_update = true;

	return 0;
}

/*
 * Merge the watermarks from all active pipes for a specific level.
 */
static void ilk_merge_wm_level(struct drm_i915_private *dev_priv,
			       int level,
			       struct intel_wm_level *ret_wm)
{
	const struct intel_crtc *intel_crtc;

	ret_wm->enable = true;

	for_each_intel_crtc(&dev_priv->drm, intel_crtc) {
		const struct intel_pipe_wm *active = &intel_crtc->wm.active.ilk;
		const struct intel_wm_level *wm = &active->wm[level];

		if (!active->pipe_enabled)
			continue;

		/*
		 * The watermark values may have been used in the past,
		 * so we must maintain them in the registers for some
		 * time even if the level is now disabled.
		 */
		if (!wm->enable)
			ret_wm->enable = false;

		ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
		ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
		ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
		ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
	}
}

/*
 * Merge all low power watermarks for all active pipes.
 */
static void ilk_wm_merge(struct drm_i915_private *dev_priv,
			 const struct intel_wm_config *config,
			 const struct ilk_wm_maximums *max,
			 struct intel_pipe_wm *merged)
{
	int level, max_level = ilk_wm_max_level(dev_priv);
	int last_enabled_level = max_level;

	/* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
	if ((INTEL_GEN(dev_priv) <= 6 || IS_IVYBRIDGE(dev_priv)) &&
	    config->num_pipes_active > 1)
		last_enabled_level = 0;

	/* ILK: FBC WM must be disabled always */
	merged->fbc_wm_enabled = INTEL_GEN(dev_priv) >= 6;

	/* merge each WM1+ level */
	for (level = 1; level <= max_level; level++) {
		struct intel_wm_level *wm = &merged->wm[level];

		ilk_merge_wm_level(dev_priv, level, wm);

		if (level > last_enabled_level)
			wm->enable = false;
		else if (!ilk_validate_wm_level(level, max, wm))
			/* make sure all following levels get disabled */
			last_enabled_level = level - 1;

		/*
		 * The spec says it is preferred to disable
		 * FBC WMs instead of disabling a WM level.
		 */
		if (wm->fbc_val > max->fbc) {
			if (wm->enable)
				merged->fbc_wm_enabled = false;
			wm->fbc_val = 0;
		}
	}

	/* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
	/*
	 * FIXME this is racy. FBC might get enabled later.
	 * What we should check here is whether FBC can be
	 * enabled sometime later.
	 */
	if (IS_GEN(dev_priv, 5) && !merged->fbc_wm_enabled &&
	    intel_fbc_is_active(dev_priv)) {
		for (level = 2; level <= max_level; level++) {
			struct intel_wm_level *wm = &merged->wm[level];

			wm->enable = false;
		}
	}
}

static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
{
	/* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
	return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
}

/* The value we need to program into the WM_LPx latency field */
static unsigned int ilk_wm_lp_latency(struct drm_i915_private *dev_priv,
				      int level)
{
	if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
		return 2 * level;
	else
		return dev_priv->wm.pri_latency[level];
}

static void ilk_compute_wm_results(struct drm_i915_private *dev_priv,
				   const struct intel_pipe_wm *merged,
				   enum intel_ddb_partitioning partitioning,
				   struct ilk_wm_values *results)
{
	struct intel_crtc *intel_crtc;
	int level, wm_lp;

	results->enable_fbc_wm = merged->fbc_wm_enabled;
	results->partitioning = partitioning;

	/* LP1+ register values */
	for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
		const struct intel_wm_level *r;

		level = ilk_wm_lp_to_level(wm_lp, merged);

		r = &merged->wm[level];

		/*
		 * Maintain the watermark values even if the level is
		 * disabled. Doing otherwise could cause underruns.
		 */
		results->wm_lp[wm_lp - 1] =
			(ilk_wm_lp_latency(dev_priv, level) << WM1_LP_LATENCY_SHIFT) |
			(r->pri_val << WM1_LP_SR_SHIFT) |
			r->cur_val;

		if (r->enable)
			results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN;

		if (INTEL_GEN(dev_priv) >= 8)
			results->wm_lp[wm_lp - 1] |=
				r->fbc_val << WM1_LP_FBC_SHIFT_BDW;
		else
			results->wm_lp[wm_lp - 1] |=
				r->fbc_val << WM1_LP_FBC_SHIFT;

		/*
		 * Always set WM1S_LP_EN when spr_val != 0, even if the
		 * level is disabled. Doing otherwise could cause underruns.
		 */
		if (INTEL_GEN(dev_priv) <= 6 && r->spr_val) {
			WARN_ON(wm_lp != 1);
			results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val;
		} else
			results->wm_lp_spr[wm_lp - 1] = r->spr_val;
	}

	/* LP0 register values */
	for_each_intel_crtc(&dev_priv->drm, intel_crtc) {
		enum pipe pipe = intel_crtc->pipe;
		const struct intel_wm_level *r =
			&intel_crtc->wm.active.ilk.wm[0];

		if (WARN_ON(!r->enable))
			continue;

		results->wm_linetime[pipe] = intel_crtc->wm.active.ilk.linetime;

		results->wm_pipe[pipe] =
			(r->pri_val << WM0_PIPE_PLANE_SHIFT) |
			(r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
			r->cur_val;
	}
}

/* Find the result with the highest level enabled. Check for enable_fbc_wm in
 * case both are at the same level. Prefer r1 in case they're the same. */
static struct intel_pipe_wm *
ilk_find_best_result(struct drm_i915_private *dev_priv,
		     struct intel_pipe_wm *r1,
		     struct intel_pipe_wm *r2)
{
	int level, max_level = ilk_wm_max_level(dev_priv);
	int level1 = 0, level2 = 0;

	for (level = 1; level <= max_level; level++) {
		if (r1->wm[level].enable)
			level1 = level;
		if (r2->wm[level].enable)
			level2 = level;
	}

	if (level1 == level2) {
		if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
			return r2;
		else
			return r1;
	} else if (level1 > level2) {
		return r1;
	} else {
		return r2;
	}
}

/* dirty bits used to track which watermarks need changes */
#define WM_DIRTY_PIPE(pipe) (1 << (pipe))
#define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
#define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
#define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
#define WM_DIRTY_FBC (1 << 24)
#define WM_DIRTY_DDB (1 << 25)

static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv,
					 const struct ilk_wm_values *old,
					 const struct ilk_wm_values *new)
{
	unsigned int dirty = 0;
	enum pipe pipe;
	int wm_lp;

	for_each_pipe(dev_priv, pipe) {
		if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) {
			dirty |= WM_DIRTY_LINETIME(pipe);
			/* Must disable LP1+ watermarks too */
			dirty |= WM_DIRTY_LP_ALL;
		}

		if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) {
			dirty |= WM_DIRTY_PIPE(pipe);
			/* Must disable LP1+ watermarks too */
			dirty |= WM_DIRTY_LP_ALL;
		}
	}

	if (old->enable_fbc_wm != new->enable_fbc_wm) {
		dirty |= WM_DIRTY_FBC;
		/* Must disable LP1+ watermarks too */
		dirty |= WM_DIRTY_LP_ALL;
	}

	if (old->partitioning != new->partitioning) {
		dirty |= WM_DIRTY_DDB;
		/* Must disable LP1+ watermarks too */
		dirty |= WM_DIRTY_LP_ALL;
	}

	/* LP1+ watermarks already deemed dirty, no need to continue */
	if (dirty & WM_DIRTY_LP_ALL)
		return dirty;

	/* Find the lowest numbered LP1+ watermark in need of an update... */
	for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
		if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] ||
		    old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1])
			break;
	}

	/* ...and mark it and all higher numbered LP1+ watermarks as dirty */
	for (; wm_lp <= 3; wm_lp++)
		dirty |= WM_DIRTY_LP(wm_lp);

	return dirty;
}

static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv,
			       unsigned int dirty)
{
	struct ilk_wm_values *previous = &dev_priv->wm.hw;
	bool changed = false;

	if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) {
		previous->wm_lp[2] &= ~WM1_LP_SR_EN;
		I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]);
		changed = true;
	}
	if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) {
		previous->wm_lp[1] &= ~WM1_LP_SR_EN;
		I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]);
		changed = true;
	}
	if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) {
		previous->wm_lp[0] &= ~WM1_LP_SR_EN;
		I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]);
		changed = true;
	}

	/*
	 * Don't touch WM1S_LP_EN here.
	 * Doing so could cause underruns.
	 */

	return changed;
}

/*
 * The spec says we shouldn't write when we don't need, because every write
 * causes WMs to be re-evaluated, expending some power.
 */
static void ilk_write_wm_values(struct drm_i915_private *dev_priv,
				struct ilk_wm_values *results)
{
	struct ilk_wm_values *previous = &dev_priv->wm.hw;
	unsigned int dirty;
	u32 val;

	dirty = ilk_compute_wm_dirty(dev_priv, previous, results);
	if (!dirty)
		return;

	_ilk_disable_lp_wm(dev_priv, dirty);

	if (dirty & WM_DIRTY_PIPE(PIPE_A))
		I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
	if (dirty & WM_DIRTY_PIPE(PIPE_B))
		I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
	if (dirty & WM_DIRTY_PIPE(PIPE_C))
		I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);

	if (dirty & WM_DIRTY_LINETIME(PIPE_A))
		I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
	if (dirty & WM_DIRTY_LINETIME(PIPE_B))
		I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
	if (dirty & WM_DIRTY_LINETIME(PIPE_C))
		I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);

	if (dirty & WM_DIRTY_DDB) {
		if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) {
			val = I915_READ(WM_MISC);
			if (results->partitioning == INTEL_DDB_PART_1_2)
				val &= ~WM_MISC_DATA_PARTITION_5_6;
			else
				val |= WM_MISC_DATA_PARTITION_5_6;
			I915_WRITE(WM_MISC, val);
		} else {
			val = I915_READ(DISP_ARB_CTL2);
			if (results->partitioning == INTEL_DDB_PART_1_2)
				val &= ~DISP_DATA_PARTITION_5_6;
			else
				val |= DISP_DATA_PARTITION_5_6;
			I915_WRITE(DISP_ARB_CTL2, val);
		}
	}

	if (dirty & WM_DIRTY_FBC) {
		val = I915_READ(DISP_ARB_CTL);
		if (results->enable_fbc_wm)
			val &= ~DISP_FBC_WM_DIS;
		else
			val |= DISP_FBC_WM_DIS;
		I915_WRITE(DISP_ARB_CTL, val);
	}

	if (dirty & WM_DIRTY_LP(1) &&
	    previous->wm_lp_spr[0] != results->wm_lp_spr[0])
		I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);

	if (INTEL_GEN(dev_priv) >= 7) {
		if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1])
			I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
		if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2])
			I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
	}

	if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0])
		I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
	if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1])
		I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
	if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2])
		I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);

	dev_priv->wm.hw = *results;
}

bool ilk_disable_lp_wm(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = to_i915(dev);

	return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL);
}

static u8 intel_enabled_dbuf_slices_num(struct drm_i915_private *dev_priv)
{
	u8 enabled_slices;

	/* Slice 1 will always be enabled */
	enabled_slices = 1;

	/* Gen prior to GEN11 have only one DBuf slice */
	if (INTEL_GEN(dev_priv) < 11)
		return enabled_slices;

	/*
	 * FIXME: for now we'll only ever use 1 slice; pretend that we have
	 * only that 1 slice enabled until we have a proper way for on-demand
	 * toggling of the second slice.
	 */
	if (0 && I915_READ(DBUF_CTL_S2) & DBUF_POWER_STATE)
		enabled_slices++;

	return enabled_slices;
}

/*
 * FIXME: We still don't have the proper code detect if we need to apply the WA,
 * so assume we'll always need it in order to avoid underruns.
 */
static bool skl_needs_memory_bw_wa(struct drm_i915_private *dev_priv)
{
	return IS_GEN9_BC(dev_priv) || IS_BROXTON(dev_priv);
}

static bool
intel_has_sagv(struct drm_i915_private *dev_priv)
{
	/* HACK! */
	if (IS_GEN(dev_priv, 12))
		return false;

	return (IS_GEN9_BC(dev_priv) || INTEL_GEN(dev_priv) >= 10) &&
		dev_priv->sagv_status != I915_SAGV_NOT_CONTROLLED;
}

static void
skl_setup_sagv_block_time(struct drm_i915_private *dev_priv)
{
	if (INTEL_GEN(dev_priv) >= 12) {
		u32 val = 0;
		int ret;

		ret = sandybridge_pcode_read(dev_priv,
					     GEN12_PCODE_READ_SAGV_BLOCK_TIME_US,
					     &val, NULL);
		if (!ret) {
			dev_priv->sagv_block_time_us = val;
			return;
		}

		DRM_DEBUG_DRIVER("Couldn't read SAGV block time!\n");
	} else if (IS_GEN(dev_priv, 11)) {
		dev_priv->sagv_block_time_us = 10;
		return;
	} else if (IS_GEN(dev_priv, 10)) {
		dev_priv->sagv_block_time_us = 20;
		return;
	} else if (IS_GEN(dev_priv, 9)) {
		dev_priv->sagv_block_time_us = 30;
		return;
	} else {
		MISSING_CASE(INTEL_GEN(dev_priv));
	}

	/* Default to an unusable block time */
	dev_priv->sagv_block_time_us = -1;
}

/*
 * SAGV dynamically adjusts the system agent voltage and clock frequencies
 * depending on power and performance requirements. The display engine access
 * to system memory is blocked during the adjustment time. Because of the
 * blocking time, having this enabled can cause full system hangs and/or pipe
 * underruns if we don't meet all of the following requirements:
 *
 *  - <= 1 pipe enabled
 *  - All planes can enable watermarks for latencies >= SAGV engine block time
 *  - We're not using an interlaced display configuration
 */
int
intel_enable_sagv(struct drm_i915_private *dev_priv)
{
	int ret;

	if (!intel_has_sagv(dev_priv))
		return 0;

	if (dev_priv->sagv_status == I915_SAGV_ENABLED)
		return 0;

	DRM_DEBUG_KMS("Enabling SAGV\n");
	ret = sandybridge_pcode_write(dev_priv, GEN9_PCODE_SAGV_CONTROL,
				      GEN9_SAGV_ENABLE);

	/* We don't need to wait for SAGV when enabling */

	/*
	 * Some skl systems, pre-release machines in particular,
	 * don't actually have SAGV.
	 */
	if (IS_SKYLAKE(dev_priv) && ret == -ENXIO) {
		DRM_DEBUG_DRIVER("No SAGV found on system, ignoring\n");
		dev_priv->sagv_status = I915_SAGV_NOT_CONTROLLED;
		return 0;
	} else if (ret < 0) {
		DRM_ERROR("Failed to enable SAGV\n");
		return ret;
	}

	dev_priv->sagv_status = I915_SAGV_ENABLED;
	return 0;
}

int
intel_disable_sagv(struct drm_i915_private *dev_priv)
{
	int ret;

	if (!intel_has_sagv(dev_priv))
		return 0;

	if (dev_priv->sagv_status == I915_SAGV_DISABLED)
		return 0;

	DRM_DEBUG_KMS("Disabling SAGV\n");
	/* bspec says to keep retrying for at least 1 ms */
	ret = skl_pcode_request(dev_priv, GEN9_PCODE_SAGV_CONTROL,
				GEN9_SAGV_DISABLE,
				GEN9_SAGV_IS_DISABLED, GEN9_SAGV_IS_DISABLED,
				1);
	/*
	 * Some skl systems, pre-release machines in particular,
	 * don't actually have SAGV.
	 */
	if (IS_SKYLAKE(dev_priv) && ret == -ENXIO) {
		DRM_DEBUG_DRIVER("No SAGV found on system, ignoring\n");
		dev_priv->sagv_status = I915_SAGV_NOT_CONTROLLED;
		return 0;
	} else if (ret < 0) {
		DRM_ERROR("Failed to disable SAGV (%d)\n", ret);
		return ret;
	}

	dev_priv->sagv_status = I915_SAGV_DISABLED;
	return 0;
}

bool intel_can_enable_sagv(struct intel_atomic_state *state)
{
	struct drm_device *dev = state->base.dev;
	struct drm_i915_private *dev_priv = to_i915(dev);
	struct intel_crtc *crtc;
	struct intel_plane *plane;
	struct intel_crtc_state *crtc_state;
	enum pipe pipe;
	int level, latency;

	if (!intel_has_sagv(dev_priv))
		return false;

	/*
	 * If there are no active CRTCs, no additional checks need be performed
	 */
	if (hweight8(state->active_pipes) == 0)
		return true;

	/*
	 * SKL+ workaround: bspec recommends we disable SAGV when we have
	 * more then one pipe enabled
	 */
	if (hweight8(state->active_pipes) > 1)
		return false;

	/* Since we're now guaranteed to only have one active CRTC... */
	pipe = ffs(state->active_pipes) - 1;
	crtc = intel_get_crtc_for_pipe(dev_priv, pipe);
	crtc_state = to_intel_crtc_state(crtc->base.state);

	if (crtc->base.state->adjusted_mode.flags & DRM_MODE_FLAG_INTERLACE)
		return false;

	for_each_intel_plane_on_crtc(dev, crtc, plane) {
		struct skl_plane_wm *wm =
			&crtc_state->wm.skl.optimal.planes[plane->id];

		/* Skip this plane if it's not enabled */
		if (!wm->wm[0].plane_en)
			continue;

		/* Find the highest enabled wm level for this plane */
		for (level = ilk_wm_max_level(dev_priv);
		     !wm->wm[level].plane_en; --level)
		     { }

		latency = dev_priv->wm.skl_latency[level];

		if (skl_needs_memory_bw_wa(dev_priv) &&
		    plane->base.state->fb->modifier ==
		    I915_FORMAT_MOD_X_TILED)
			latency += 15;

		/*
		 * If any of the planes on this pipe don't enable wm levels that
		 * incur memory latencies higher than sagv_block_time_us we
		 * can't enable SAGV.
		 */
		if (latency < dev_priv->sagv_block_time_us)
			return false;
	}

	return true;
}

static u16 intel_get_ddb_size(struct drm_i915_private *dev_priv,
			      const struct intel_crtc_state *crtc_state,
			      const u64 total_data_rate,
			      const int num_active,
			      struct skl_ddb_allocation *ddb)
{
	const struct drm_display_mode *adjusted_mode;
	u64 total_data_bw;
	u16 ddb_size = INTEL_INFO(dev_priv)->ddb_size;

	WARN_ON(ddb_size == 0);

	if (INTEL_GEN(dev_priv) < 11)
		return ddb_size - 4; /* 4 blocks for bypass path allocation */

	adjusted_mode = &crtc_state->base.adjusted_mode;
	total_data_bw = total_data_rate * drm_mode_vrefresh(adjusted_mode);

	/*
	 * 12GB/s is maximum BW supported by single DBuf slice.
	 *
	 * FIXME dbuf slice code is broken:
	 * - must wait for planes to stop using the slice before powering it off
	 * - plane straddling both slices is illegal in multi-pipe scenarios
	 * - should validate we stay within the hw bandwidth limits
	 */
	if (0 && (num_active > 1 || total_data_bw >= GBps(12))) {
		ddb->enabled_slices = 2;
	} else {
		ddb->enabled_slices = 1;
		ddb_size /= 2;
	}

	return ddb_size;
}

static void
skl_ddb_get_pipe_allocation_limits(struct drm_i915_private *dev_priv,
				   const struct intel_crtc_state *crtc_state,
				   const u64 total_data_rate,
				   struct skl_ddb_allocation *ddb,
				   struct skl_ddb_entry *alloc, /* out */
				   int *num_active /* out */)
{
	struct drm_atomic_state *state = crtc_state->base.state;
	struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
	struct drm_crtc *for_crtc = crtc_state->base.crtc;
	const struct intel_crtc *crtc;
	u32 pipe_width = 0, total_width = 0, width_before_pipe = 0;
	enum pipe for_pipe = to_intel_crtc(for_crtc)->pipe;
	u16 ddb_size;
	u32 i;

	if (WARN_ON(!state) || !crtc_state->base.active) {
		alloc->start = 0;
		alloc->end = 0;
		*num_active = hweight8(dev_priv->active_pipes);
		return;
	}

	if (intel_state->active_pipe_changes)
		*num_active = hweight8(intel_state->active_pipes);
	else
		*num_active = hweight8(dev_priv->active_pipes);

	ddb_size = intel_get_ddb_size(dev_priv, crtc_state, total_data_rate,
				      *num_active, ddb);

	/*
	 * If the state doesn't change the active CRTC's or there is no
	 * modeset request, then there's no need to recalculate;
	 * the existing pipe allocation limits should remain unchanged.
	 * Note that we're safe from racing commits since any racing commit
	 * that changes the active CRTC list or do modeset would need to
	 * grab _all_ crtc locks, including the one we currently hold.
	 */
	if (!intel_state->active_pipe_changes && !intel_state->modeset) {
		/*
		 * alloc may be cleared by clear_intel_crtc_state,
		 * copy from old state to be sure
		 */
		*alloc = to_intel_crtc_state(for_crtc->state)->wm.skl.ddb;
		return;
	}

	/*
	 * Watermark/ddb requirement highly depends upon width of the
	 * framebuffer, So instead of allocating DDB equally among pipes
	 * distribute DDB based on resolution/width of the display.
	 */
	for_each_new_intel_crtc_in_state(intel_state, crtc, crtc_state, i) {
		const struct drm_display_mode *adjusted_mode =
			&crtc_state->base.adjusted_mode;
		enum pipe pipe = crtc->pipe;
		int hdisplay, vdisplay;

		if (!crtc_state->base.enable)
			continue;

		drm_mode_get_hv_timing(adjusted_mode, &hdisplay, &vdisplay);
		total_width += hdisplay;

		if (pipe < for_pipe)
			width_before_pipe += hdisplay;
		else if (pipe == for_pipe)
			pipe_width = hdisplay;
	}

	alloc->start = ddb_size * width_before_pipe / total_width;
	alloc->end = ddb_size * (width_before_pipe + pipe_width) / total_width;
}

static int skl_compute_wm_params(const struct intel_crtc_state *crtc_state,
				 int width, const struct drm_format_info *format,
				 u64 modifier, unsigned int rotation,
				 u32 plane_pixel_rate, struct skl_wm_params *wp,
				 int color_plane);
static void skl_compute_plane_wm(const struct intel_crtc_state *crtc_state,
				 int level,
				 const struct skl_wm_params *wp,
				 const struct skl_wm_level *result_prev,
				 struct skl_wm_level *result /* out */);

static unsigned int
skl_cursor_allocation(const struct intel_crtc_state *crtc_state,
		      int num_active)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	int level, max_level = ilk_wm_max_level(dev_priv);
	struct skl_wm_level wm = {};
	int ret, min_ddb_alloc = 0;
	struct skl_wm_params wp;

	ret = skl_compute_wm_params(crtc_state, 256,
				    drm_format_info(DRM_FORMAT_ARGB8888),
				    DRM_FORMAT_MOD_LINEAR,
				    DRM_MODE_ROTATE_0,
				    crtc_state->pixel_rate, &wp, 0);
	WARN_ON(ret);

	for (level = 0; level <= max_level; level++) {
		skl_compute_plane_wm(crtc_state, level, &wp, &wm, &wm);
		if (wm.min_ddb_alloc == U16_MAX)
			break;

		min_ddb_alloc = wm.min_ddb_alloc;
	}

	return max(num_active == 1 ? 32 : 8, min_ddb_alloc);
}

static void skl_ddb_entry_init_from_hw(struct drm_i915_private *dev_priv,
				       struct skl_ddb_entry *entry, u32 reg)
{

	entry->start = reg & DDB_ENTRY_MASK;
	entry->end = (reg >> DDB_ENTRY_END_SHIFT) & DDB_ENTRY_MASK;

	if (entry->end)
		entry->end += 1;
}

static void
skl_ddb_get_hw_plane_state(struct drm_i915_private *dev_priv,
			   const enum pipe pipe,
			   const enum plane_id plane_id,
			   struct skl_ddb_entry *ddb_y,
			   struct skl_ddb_entry *ddb_uv)
{
	u32 val, val2;
	u32 fourcc = 0;

	/* Cursor doesn't support NV12/planar, so no extra calculation needed */
	if (plane_id == PLANE_CURSOR) {
		val = I915_READ(CUR_BUF_CFG(pipe));
		skl_ddb_entry_init_from_hw(dev_priv, ddb_y, val);
		return;
	}

	val = I915_READ(PLANE_CTL(pipe, plane_id));

	/* No DDB allocated for disabled planes */
	if (val & PLANE_CTL_ENABLE)
		fourcc = skl_format_to_fourcc(val & PLANE_CTL_FORMAT_MASK,
					      val & PLANE_CTL_ORDER_RGBX,
					      val & PLANE_CTL_ALPHA_MASK);

	if (INTEL_GEN(dev_priv) >= 11) {
		val = I915_READ(PLANE_BUF_CFG(pipe, plane_id));
		skl_ddb_entry_init_from_hw(dev_priv, ddb_y, val);
	} else {
		val = I915_READ(PLANE_BUF_CFG(pipe, plane_id));
		val2 = I915_READ(PLANE_NV12_BUF_CFG(pipe, plane_id));

		if (fourcc &&
		    drm_format_info_is_yuv_semiplanar(drm_format_info(fourcc)))
			swap(val, val2);

		skl_ddb_entry_init_from_hw(dev_priv, ddb_y, val);
		skl_ddb_entry_init_from_hw(dev_priv, ddb_uv, val2);
	}
}

void skl_pipe_ddb_get_hw_state(struct intel_crtc *crtc,
			       struct skl_ddb_entry *ddb_y,
			       struct skl_ddb_entry *ddb_uv)
{
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	enum intel_display_power_domain power_domain;
	enum pipe pipe = crtc->pipe;
	intel_wakeref_t wakeref;
	enum plane_id plane_id;

	power_domain = POWER_DOMAIN_PIPE(pipe);
	wakeref = intel_display_power_get_if_enabled(dev_priv, power_domain);
	if (!wakeref)
		return;

	for_each_plane_id_on_crtc(crtc, plane_id)
		skl_ddb_get_hw_plane_state(dev_priv, pipe,
					   plane_id,
					   &ddb_y[plane_id],
					   &ddb_uv[plane_id]);

	intel_display_power_put(dev_priv, power_domain, wakeref);
}

void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv,
			  struct skl_ddb_allocation *ddb /* out */)
{
	ddb->enabled_slices = intel_enabled_dbuf_slices_num(dev_priv);
}

/*
 * Determines the downscale amount of a plane for the purposes of watermark calculations.
 * The bspec defines downscale amount as:
 *
 * """
 * Horizontal down scale amount = maximum[1, Horizontal source size /
 *                                           Horizontal destination size]
 * Vertical down scale amount = maximum[1, Vertical source size /
 *                                         Vertical destination size]
 * Total down scale amount = Horizontal down scale amount *
 *                           Vertical down scale amount
 * """
 *
 * Return value is provided in 16.16 fixed point form to retain fractional part.
 * Caller should take care of dividing & rounding off the value.
 */
static uint_fixed_16_16_t
skl_plane_downscale_amount(const struct intel_crtc_state *crtc_state,
			   const struct intel_plane_state *plane_state)
{
	u32 src_w, src_h, dst_w, dst_h;
	uint_fixed_16_16_t fp_w_ratio, fp_h_ratio;
	uint_fixed_16_16_t downscale_h, downscale_w;

	if (WARN_ON(!intel_wm_plane_visible(crtc_state, plane_state)))
		return u32_to_fixed16(0);

	/*
	 * Src coordinates are already rotated by 270 degrees for
	 * the 90/270 degree plane rotation cases (to match the
	 * GTT mapping), hence no need to account for rotation here.
	 *
	 * n.b., src is 16.16 fixed point, dst is whole integer.
	 */
	src_w = drm_rect_width(&plane_state->base.src) >> 16;
	src_h = drm_rect_height(&plane_state->base.src) >> 16;
	dst_w = drm_rect_width(&plane_state->base.dst);
	dst_h = drm_rect_height(&plane_state->base.dst);

	fp_w_ratio = div_fixed16(src_w, dst_w);
	fp_h_ratio = div_fixed16(src_h, dst_h);
	downscale_w = max_fixed16(fp_w_ratio, u32_to_fixed16(1));
	downscale_h = max_fixed16(fp_h_ratio, u32_to_fixed16(1));

	return mul_fixed16(downscale_w, downscale_h);
}

static u64
skl_plane_relative_data_rate(const struct intel_crtc_state *crtc_state,
			     const struct intel_plane_state *plane_state,
			     int color_plane)
{
	struct intel_plane *plane = to_intel_plane(plane_state->base.plane);
	const struct drm_framebuffer *fb = plane_state->base.fb;
	u32 data_rate;
	u32 width = 0, height = 0;
	uint_fixed_16_16_t down_scale_amount;
	u64 rate;

	if (!plane_state->base.visible)
		return 0;

	if (plane->id == PLANE_CURSOR)
		return 0;

	if (color_plane == 1 &&
	    !drm_format_info_is_yuv_semiplanar(fb->format))
		return 0;

	/*
	 * Src coordinates are already rotated by 270 degrees for
	 * the 90/270 degree plane rotation cases (to match the
	 * GTT mapping), hence no need to account for rotation here.
	 */
	width = drm_rect_width(&plane_state->base.src) >> 16;
	height = drm_rect_height(&plane_state->base.src) >> 16;

	/* UV plane does 1/2 pixel sub-sampling */
	if (color_plane == 1) {
		width /= 2;
		height /= 2;
	}

	data_rate = width * height;

	down_scale_amount = skl_plane_downscale_amount(crtc_state, plane_state);

	rate = mul_round_up_u32_fixed16(data_rate, down_scale_amount);

	rate *= fb->format->cpp[color_plane];
	return rate;
}

static u64
skl_get_total_relative_data_rate(struct intel_crtc_state *crtc_state,
				 u64 *plane_data_rate,
				 u64 *uv_plane_data_rate)
{
	struct drm_atomic_state *state = crtc_state->base.state;
	struct intel_plane *plane;
	const struct intel_plane_state *plane_state;
	u64 total_data_rate = 0;

	if (WARN_ON(!state))
		return 0;

	/* Calculate and cache data rate for each plane */
	intel_atomic_crtc_state_for_each_plane_state(plane, plane_state, crtc_state) {
		enum plane_id plane_id = plane->id;
		u64 rate;

		/* packed/y */
		rate = skl_plane_relative_data_rate(crtc_state, plane_state, 0);
		plane_data_rate[plane_id] = rate;
		total_data_rate += rate;

		/* uv-plane */
		rate = skl_plane_relative_data_rate(crtc_state, plane_state, 1);
		uv_plane_data_rate[plane_id] = rate;
		total_data_rate += rate;
	}

	return total_data_rate;
}

static u64
icl_get_total_relative_data_rate(struct intel_crtc_state *crtc_state,
				 u64 *plane_data_rate)
{
	struct intel_plane *plane;
	const struct intel_plane_state *plane_state;
	u64 total_data_rate = 0;

	if (WARN_ON(!crtc_state->base.state))
		return 0;

	/* Calculate and cache data rate for each plane */
	intel_atomic_crtc_state_for_each_plane_state(plane, plane_state, crtc_state) {
		enum plane_id plane_id = plane->id;
		u64 rate;

		if (!plane_state->planar_linked_plane) {
			rate = skl_plane_relative_data_rate(crtc_state, plane_state, 0);
			plane_data_rate[plane_id] = rate;
			total_data_rate += rate;
		} else {
			enum plane_id y_plane_id;

			/*
			 * The slave plane might not iterate in
			 * intel_atomic_crtc_state_for_each_plane_state(),
			 * and needs the master plane state which may be
			 * NULL if we try get_new_plane_state(), so we
			 * always calculate from the master.
			 */
			if (plane_state->planar_slave)
				continue;

			/* Y plane rate is calculated on the slave */
			rate = skl_plane_relative_data_rate(crtc_state, plane_state, 0);
			y_plane_id = plane_state->planar_linked_plane->id;
			plane_data_rate[y_plane_id] = rate;
			total_data_rate += rate;

			rate = skl_plane_relative_data_rate(crtc_state, plane_state, 1);
			plane_data_rate[plane_id] = rate;
			total_data_rate += rate;
		}
	}

	return total_data_rate;
}

static int
skl_allocate_pipe_ddb(struct intel_crtc_state *crtc_state,
		      struct skl_ddb_allocation *ddb /* out */)
{
	struct drm_atomic_state *state = crtc_state->base.state;
	struct drm_crtc *crtc = crtc_state->base.crtc;
	struct drm_i915_private *dev_priv = to_i915(crtc->dev);
	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
	struct skl_ddb_entry *alloc = &crtc_state->wm.skl.ddb;
	u16 alloc_size, start = 0;
	u16 total[I915_MAX_PLANES] = {};
	u16 uv_total[I915_MAX_PLANES] = {};
	u64 total_data_rate;
	enum plane_id plane_id;
	int num_active;
	u64 plane_data_rate[I915_MAX_PLANES] = {};
	u64 uv_plane_data_rate[I915_MAX_PLANES] = {};
	u32 blocks;
	int level;

	/* Clear the partitioning for disabled planes. */
	memset(crtc_state->wm.skl.plane_ddb_y, 0, sizeof(crtc_state->wm.skl.plane_ddb_y));
	memset(crtc_state->wm.skl.plane_ddb_uv, 0, sizeof(crtc_state->wm.skl.plane_ddb_uv));

	if (WARN_ON(!state))
		return 0;

	if (!crtc_state->base.active) {
		alloc->start = alloc->end = 0;
		return 0;
	}

	if (INTEL_GEN(dev_priv) >= 11)
		total_data_rate =
			icl_get_total_relative_data_rate(crtc_state,
							 plane_data_rate);
	else
		total_data_rate =
			skl_get_total_relative_data_rate(crtc_state,
							 plane_data_rate,
							 uv_plane_data_rate);


	skl_ddb_get_pipe_allocation_limits(dev_priv, crtc_state, total_data_rate,
					   ddb, alloc, &num_active);
	alloc_size = skl_ddb_entry_size(alloc);
	if (alloc_size == 0)
		return 0;

	/* Allocate fixed number of blocks for cursor. */
	total[PLANE_CURSOR] = skl_cursor_allocation(crtc_state, num_active);
	alloc_size -= total[PLANE_CURSOR];
	crtc_state->wm.skl.plane_ddb_y[PLANE_CURSOR].start =
		alloc->end - total[PLANE_CURSOR];
	crtc_state->wm.skl.plane_ddb_y[PLANE_CURSOR].end = alloc->end;

	if (total_data_rate == 0)
		return 0;

	/*
	 * Find the highest watermark level for which we can satisfy the block
	 * requirement of active planes.
	 */
	for (level = ilk_wm_max_level(dev_priv); level >= 0; level--) {
		blocks = 0;
		for_each_plane_id_on_crtc(intel_crtc, plane_id) {
			const struct skl_plane_wm *wm =
				&crtc_state->wm.skl.optimal.planes[plane_id];

			if (plane_id == PLANE_CURSOR) {
				if (WARN_ON(wm->wm[level].min_ddb_alloc >
					    total[PLANE_CURSOR])) {
					blocks = U32_MAX;
					break;
				}
				continue;
			}

			blocks += wm->wm[level].min_ddb_alloc;
			blocks += wm->uv_wm[level].min_ddb_alloc;
		}

		if (blocks <= alloc_size) {
			alloc_size -= blocks;
			break;
		}
	}

	if (level < 0) {
		DRM_DEBUG_KMS("Requested display configuration exceeds system DDB limitations");
		DRM_DEBUG_KMS("minimum required %d/%d\n", blocks,
			      alloc_size);
		return -EINVAL;
	}

	/*
	 * Grant each plane the blocks it requires at the highest achievable
	 * watermark level, plus an extra share of the leftover blocks
	 * proportional to its relative data rate.
	 */
	for_each_plane_id_on_crtc(intel_crtc, plane_id) {
		const struct skl_plane_wm *wm =
			&crtc_state->wm.skl.optimal.planes[plane_id];
		u64 rate;
		u16 extra;

		if (plane_id == PLANE_CURSOR)
			continue;

		/*
		 * We've accounted for all active planes; remaining planes are
		 * all disabled.
		 */
		if (total_data_rate == 0)
			break;

		rate = plane_data_rate[plane_id];
		extra = min_t(u16, alloc_size,
			      DIV64_U64_ROUND_UP(alloc_size * rate,
						 total_data_rate));
		total[plane_id] = wm->wm[level].min_ddb_alloc + extra;
		alloc_size -= extra;
		total_data_rate -= rate;

		if (total_data_rate == 0)
			break;

		rate = uv_plane_data_rate[plane_id];
		extra = min_t(u16, alloc_size,
			      DIV64_U64_ROUND_UP(alloc_size * rate,
						 total_data_rate));
		uv_total[plane_id] = wm->uv_wm[level].min_ddb_alloc + extra;
		alloc_size -= extra;
		total_data_rate -= rate;
	}
	WARN_ON(alloc_size != 0 || total_data_rate != 0);

	/* Set the actual DDB start/end points for each plane */
	start = alloc->start;
	for_each_plane_id_on_crtc(intel_crtc, plane_id) {
		struct skl_ddb_entry *plane_alloc =
			&crtc_state->wm.skl.plane_ddb_y[plane_id];
		struct skl_ddb_entry *uv_plane_alloc =
			&crtc_state->wm.skl.plane_ddb_uv[plane_id];

		if (plane_id == PLANE_CURSOR)
			continue;

		/* Gen11+ uses a separate plane for UV watermarks */
		WARN_ON(INTEL_GEN(dev_priv) >= 11 && uv_total[plane_id]);

		/* Leave disabled planes at (0,0) */
		if (total[plane_id]) {
			plane_alloc->start = start;
			start += total[plane_id];
			plane_alloc->end = start;
		}

		if (uv_total[plane_id]) {
			uv_plane_alloc->start = start;
			start += uv_total[plane_id];
			uv_plane_alloc->end = start;
		}
	}

	/*
	 * When we calculated watermark values we didn't know how high
	 * of a level we'd actually be able to hit, so we just marked
	 * all levels as "enabled."  Go back now and disable the ones
	 * that aren't actually possible.
	 */
	for (level++; level <= ilk_wm_max_level(dev_priv); level++) {
		for_each_plane_id_on_crtc(intel_crtc, plane_id) {
			struct skl_plane_wm *wm =
				&crtc_state->wm.skl.optimal.planes[plane_id];

			/*
			 * We only disable the watermarks for each plane if
			 * they exceed the ddb allocation of said plane. This
			 * is done so that we don't end up touching cursor
			 * watermarks needlessly when some other plane reduces
			 * our max possible watermark level.
			 *
			 * Bspec has this to say about the PLANE_WM enable bit:
			 * "All the watermarks at this level for all enabled
			 *  planes must be enabled before the level will be used."
			 * So this is actually safe to do.
			 */
			if (wm->wm[level].min_ddb_alloc > total[plane_id] ||
			    wm->uv_wm[level].min_ddb_alloc > uv_total[plane_id])
				memset(&wm->wm[level], 0, sizeof(wm->wm[level]));

			/*
			 * Wa_1408961008:icl, ehl
			 * Underruns with WM1+ disabled
			 */
			if (IS_GEN(dev_priv, 11) &&
			    level == 1 && wm->wm[0].plane_en) {
				wm->wm[level].plane_res_b = wm->wm[0].plane_res_b;
				wm->wm[level].plane_res_l = wm->wm[0].plane_res_l;
				wm->wm[level].ignore_lines = wm->wm[0].ignore_lines;
			}
		}
	}

	/*
	 * Go back and disable the transition watermark if it turns out we
	 * don't have enough DDB blocks for it.
	 */
	for_each_plane_id_on_crtc(intel_crtc, plane_id) {
		struct skl_plane_wm *wm =
			&crtc_state->wm.skl.optimal.planes[plane_id];

		if (wm->trans_wm.plane_res_b >= total[plane_id])
			memset(&wm->trans_wm, 0, sizeof(wm->trans_wm));
	}

	return 0;
}

/*
 * The max latency should be 257 (max the punit can code is 255 and we add 2us
 * for the read latency) and cpp should always be <= 8, so that
 * should allow pixel_rate up to ~2 GHz which seems sufficient since max
 * 2xcdclk is 1350 MHz and the pixel rate should never exceed that.
*/
static uint_fixed_16_16_t
skl_wm_method1(const struct drm_i915_private *dev_priv, u32 pixel_rate,
	       u8 cpp, u32 latency, u32 dbuf_block_size)
{
	u32 wm_intermediate_val;
	uint_fixed_16_16_t ret;

	if (latency == 0)
		return FP_16_16_MAX;

	wm_intermediate_val = latency * pixel_rate * cpp;
	ret = div_fixed16(wm_intermediate_val, 1000 * dbuf_block_size);

	if (INTEL_GEN(dev_priv) >= 10)
		ret = add_fixed16_u32(ret, 1);

	return ret;
}

static uint_fixed_16_16_t
skl_wm_method2(u32 pixel_rate, u32 pipe_htotal, u32 latency,
	       uint_fixed_16_16_t plane_blocks_per_line)
{
	u32 wm_intermediate_val;
	uint_fixed_16_16_t ret;

	if (latency == 0)
		return FP_16_16_MAX;

	wm_intermediate_val = latency * pixel_rate;
	wm_intermediate_val = DIV_ROUND_UP(wm_intermediate_val,
					   pipe_htotal * 1000);
	ret = mul_u32_fixed16(wm_intermediate_val, plane_blocks_per_line);
	return ret;
}

static uint_fixed_16_16_t
intel_get_linetime_us(const struct intel_crtc_state *crtc_state)
{
	u32 pixel_rate;
	u32 crtc_htotal;
	uint_fixed_16_16_t linetime_us;

	if (!crtc_state->base.active)
		return u32_to_fixed16(0);

	pixel_rate = crtc_state->pixel_rate;

	if (WARN_ON(pixel_rate == 0))
		return u32_to_fixed16(0);

	crtc_htotal = crtc_state->base.adjusted_mode.crtc_htotal;
	linetime_us = div_fixed16(crtc_htotal * 1000, pixel_rate);

	return linetime_us;
}

static u32
skl_adjusted_plane_pixel_rate(const struct intel_crtc_state *crtc_state,
			      const struct intel_plane_state *plane_state)
{
	u64 adjusted_pixel_rate;
	uint_fixed_16_16_t downscale_amount;

	/* Shouldn't reach here on disabled planes... */
	if (WARN_ON(!intel_wm_plane_visible(crtc_state, plane_state)))
		return 0;

	/*
	 * Adjusted plane pixel rate is just the pipe's adjusted pixel rate
	 * with additional adjustments for plane-specific scaling.
	 */
	adjusted_pixel_rate = crtc_state->pixel_rate;
	downscale_amount = skl_plane_downscale_amount(crtc_state, plane_state);

	return mul_round_up_u32_fixed16(adjusted_pixel_rate,
					    downscale_amount);
}

static int
skl_compute_wm_params(const struct intel_crtc_state *crtc_state,
		      int width, const struct drm_format_info *format,
		      u64 modifier, unsigned int rotation,
		      u32 plane_pixel_rate, struct skl_wm_params *wp,
		      int color_plane)
{
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	u32 interm_pbpl;

	/* only planar format has two planes */
	if (color_plane == 1 && !drm_format_info_is_yuv_semiplanar(format)) {
		DRM_DEBUG_KMS("Non planar format have single plane\n");
		return -EINVAL;
	}

	wp->y_tiled = modifier == I915_FORMAT_MOD_Y_TILED ||
		      modifier == I915_FORMAT_MOD_Yf_TILED ||
		      modifier == I915_FORMAT_MOD_Y_TILED_CCS ||
		      modifier == I915_FORMAT_MOD_Yf_TILED_CCS;
	wp->x_tiled = modifier == I915_FORMAT_MOD_X_TILED;
	wp->rc_surface = modifier == I915_FORMAT_MOD_Y_TILED_CCS ||
			 modifier == I915_FORMAT_MOD_Yf_TILED_CCS;
	wp->is_planar = drm_format_info_is_yuv_semiplanar(format);

	wp->width = width;
	if (color_plane == 1 && wp->is_planar)
		wp->width /= 2;

	wp->cpp = format->cpp[color_plane];
	wp->plane_pixel_rate = plane_pixel_rate;

	if (INTEL_GEN(dev_priv) >= 11 &&
	    modifier == I915_FORMAT_MOD_Yf_TILED  && wp->cpp == 1)
		wp->dbuf_block_size = 256;
	else
		wp->dbuf_block_size = 512;

	if (drm_rotation_90_or_270(rotation)) {
		switch (wp->cpp) {
		case 1:
			wp->y_min_scanlines = 16;
			break;
		case 2:
			wp->y_min_scanlines = 8;
			break;
		case 4:
			wp->y_min_scanlines = 4;
			break;
		default:
			MISSING_CASE(wp->cpp);
			return -EINVAL;
		}
	} else {
		wp->y_min_scanlines = 4;
	}

	if (skl_needs_memory_bw_wa(dev_priv))
		wp->y_min_scanlines *= 2;

	wp->plane_bytes_per_line = wp->width * wp->cpp;
	if (wp->y_tiled) {
		interm_pbpl = DIV_ROUND_UP(wp->plane_bytes_per_line *
					   wp->y_min_scanlines,
					   wp->dbuf_block_size);

		if (INTEL_GEN(dev_priv) >= 10)
			interm_pbpl++;

		wp->plane_blocks_per_line = div_fixed16(interm_pbpl,
							wp->y_min_scanlines);
	} else if (wp->x_tiled && IS_GEN(dev_priv, 9)) {
		interm_pbpl = DIV_ROUND_UP(wp->plane_bytes_per_line,
					   wp->dbuf_block_size);
		wp->plane_blocks_per_line = u32_to_fixed16(interm_pbpl);
	} else {
		interm_pbpl = DIV_ROUND_UP(wp->plane_bytes_per_line,
					   wp->dbuf_block_size) + 1;
		wp->plane_blocks_per_line = u32_to_fixed16(interm_pbpl);
	}

	wp->y_tile_minimum = mul_u32_fixed16(wp->y_min_scanlines,
					     wp->plane_blocks_per_line);

	wp->linetime_us = fixed16_to_u32_round_up(
					intel_get_linetime_us(crtc_state));

	return 0;
}

static int
skl_compute_plane_wm_params(const struct intel_crtc_state *crtc_state,
			    const struct intel_plane_state *plane_state,
			    struct skl_wm_params *wp, int color_plane)
{
	const struct drm_framebuffer *fb = plane_state->base.fb;
	int width;

	/*
	 * Src coordinates are already rotated by 270 degrees for
	 * the 90/270 degree plane rotation cases (to match the
	 * GTT mapping), hence no need to account for rotation here.
	 */
	width = drm_rect_width(&plane_state->base.src) >> 16;

	return skl_compute_wm_params(crtc_state, width,
				     fb->format, fb->modifier,
				     plane_state->base.rotation,
				     skl_adjusted_plane_pixel_rate(crtc_state, plane_state),
				     wp, color_plane);
}

static bool skl_wm_has_lines(struct drm_i915_private *dev_priv, int level)
{
	if (INTEL_GEN(dev_priv) >= 10 || IS_GEMINILAKE(dev_priv))
		return true;

	/* The number of lines are ignored for the level 0 watermark. */
	return level > 0;
}

static void skl_compute_plane_wm(const struct intel_crtc_state *crtc_state,
				 int level,
				 const struct skl_wm_params *wp,
				 const struct skl_wm_level *result_prev,
				 struct skl_wm_level *result /* out */)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	u32 latency = dev_priv->wm.skl_latency[level];
	uint_fixed_16_16_t method1, method2;
	uint_fixed_16_16_t selected_result;
	u32 res_blocks, res_lines, min_ddb_alloc = 0;

	if (latency == 0) {
		/* reject it */
		result->min_ddb_alloc = U16_MAX;
		return;
	}

	/*
	 * WaIncreaseLatencyIPCEnabled: kbl,cfl
	 * Display WA #1141: kbl,cfl
	 */
	if ((IS_KABYLAKE(dev_priv) || IS_COFFEELAKE(dev_priv)) ||
	    dev_priv->ipc_enabled)
		latency += 4;

	if (skl_needs_memory_bw_wa(dev_priv) && wp->x_tiled)
		latency += 15;

	method1 = skl_wm_method1(dev_priv, wp->plane_pixel_rate,
				 wp->cpp, latency, wp->dbuf_block_size);
	method2 = skl_wm_method2(wp->plane_pixel_rate,
				 crtc_state->base.adjusted_mode.crtc_htotal,
				 latency,
				 wp->plane_blocks_per_line);

	if (wp->y_tiled) {
		selected_result = max_fixed16(method2, wp->y_tile_minimum);
	} else {
		if ((wp->cpp * crtc_state->base.adjusted_mode.crtc_htotal /
		     wp->dbuf_block_size < 1) &&
		     (wp->plane_bytes_per_line / wp->dbuf_block_size < 1)) {
			selected_result = method2;
		} else if (latency >= wp->linetime_us) {
			if (IS_GEN(dev_priv, 9) &&
			    !IS_GEMINILAKE(dev_priv))
				selected_result = min_fixed16(method1, method2);
			else
				selected_result = method2;
		} else {
			selected_result = method1;
		}
	}

	res_blocks = fixed16_to_u32_round_up(selected_result) + 1;
	res_lines = div_round_up_fixed16(selected_result,
					 wp->plane_blocks_per_line);

	if (IS_GEN9_BC(dev_priv) || IS_BROXTON(dev_priv)) {
		/* Display WA #1125: skl,bxt,kbl */
		if (level == 0 && wp->rc_surface)
			res_blocks +=
				fixed16_to_u32_round_up(wp->y_tile_minimum);

		/* Display WA #1126: skl,bxt,kbl */
		if (level >= 1 && level <= 7) {
			if (wp->y_tiled) {
				res_blocks +=
				    fixed16_to_u32_round_up(wp->y_tile_minimum);
				res_lines += wp->y_min_scanlines;
			} else {
				res_blocks++;
			}

			/*
			 * Make sure result blocks for higher latency levels are
			 * atleast as high as level below the current level.
			 * Assumption in DDB algorithm optimization for special
			 * cases. Also covers Display WA #1125 for RC.
			 */
			if (result_prev->plane_res_b > res_blocks)
				res_blocks = result_prev->plane_res_b;
		}
	}

	if (INTEL_GEN(dev_priv) >= 11) {
		if (wp->y_tiled) {
			int extra_lines;

			if (res_lines % wp->y_min_scanlines == 0)
				extra_lines = wp->y_min_scanlines;
			else
				extra_lines = wp->y_min_scanlines * 2 -
					res_lines % wp->y_min_scanlines;

			min_ddb_alloc = mul_round_up_u32_fixed16(res_lines + extra_lines,
								 wp->plane_blocks_per_line);
		} else {
			min_ddb_alloc = res_blocks +
				DIV_ROUND_UP(res_blocks, 10);
		}
	}

	if (!skl_wm_has_lines(dev_priv, level))
		res_lines = 0;

	if (res_lines > 31) {
		/* reject it */
		result->min_ddb_alloc = U16_MAX;
		return;
	}

	/*
	 * If res_lines is valid, assume we can use this watermark level
	 * for now.  We'll come back and disable it after we calculate the
	 * DDB allocation if it turns out we don't actually have enough
	 * blocks to satisfy it.
	 */
	result->plane_res_b = res_blocks;
	result->plane_res_l = res_lines;
	/* Bspec says: value >= plane ddb allocation -> invalid, hence the +1 here */
	result->min_ddb_alloc = max(min_ddb_alloc, res_blocks) + 1;
	result->plane_en = true;
}

static void
skl_compute_wm_levels(const struct intel_crtc_state *crtc_state,
		      const struct skl_wm_params *wm_params,
		      struct skl_wm_level *levels)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	int level, max_level = ilk_wm_max_level(dev_priv);
	struct skl_wm_level *result_prev = &levels[0];

	for (level = 0; level <= max_level; level++) {
		struct skl_wm_level *result = &levels[level];

		skl_compute_plane_wm(crtc_state, level, wm_params,
				     result_prev, result);

		result_prev = result;
	}
}

static u32
skl_compute_linetime_wm(const struct intel_crtc_state *crtc_state)
{
	struct drm_atomic_state *state = crtc_state->base.state;
	struct drm_i915_private *dev_priv = to_i915(state->dev);
	uint_fixed_16_16_t linetime_us;
	u32 linetime_wm;

	linetime_us = intel_get_linetime_us(crtc_state);
	linetime_wm = fixed16_to_u32_round_up(mul_u32_fixed16(8, linetime_us));

	/* Display WA #1135: BXT:ALL GLK:ALL */
	if (IS_GEN9_LP(dev_priv) && dev_priv->ipc_enabled)
		linetime_wm /= 2;

	return linetime_wm;
}

static void skl_compute_transition_wm(const struct intel_crtc_state *crtc_state,
				      const struct skl_wm_params *wp,
				      struct skl_plane_wm *wm)
{
	struct drm_device *dev = crtc_state->base.crtc->dev;
	const struct drm_i915_private *dev_priv = to_i915(dev);
	u16 trans_min, trans_y_tile_min;
	const u16 trans_amount = 10; /* This is configurable amount */
	u16 wm0_sel_res_b, trans_offset_b, res_blocks;

	/* Transition WM are not recommended by HW team for GEN9 */
	if (INTEL_GEN(dev_priv) <= 9)
		return;

	/* Transition WM don't make any sense if ipc is disabled */
	if (!dev_priv->ipc_enabled)
		return;

	trans_min = 14;
	if (INTEL_GEN(dev_priv) >= 11)
		trans_min = 4;

	trans_offset_b = trans_min + trans_amount;

	/*
	 * The spec asks for Selected Result Blocks for wm0 (the real value),
	 * not Result Blocks (the integer value). Pay attention to the capital
	 * letters. The value wm_l0->plane_res_b is actually Result Blocks, but
	 * since Result Blocks is the ceiling of Selected Result Blocks plus 1,
	 * and since we later will have to get the ceiling of the sum in the
	 * transition watermarks calculation, we can just pretend Selected
	 * Result Blocks is Result Blocks minus 1 and it should work for the
	 * current platforms.
	 */
	wm0_sel_res_b = wm->wm[0].plane_res_b - 1;

	if (wp->y_tiled) {
		trans_y_tile_min =
			(u16)mul_round_up_u32_fixed16(2, wp->y_tile_minimum);
		res_blocks = max(wm0_sel_res_b, trans_y_tile_min) +
				trans_offset_b;
	} else {
		res_blocks = wm0_sel_res_b + trans_offset_b;

		/* WA BUG:1938466 add one block for non y-tile planes */
		if (IS_CNL_REVID(dev_priv, CNL_REVID_A0, CNL_REVID_A0))
			res_blocks += 1;

	}

	/*
	 * Just assume we can enable the transition watermark.  After
	 * computing the DDB we'll come back and disable it if that
	 * assumption turns out to be false.
	 */
	wm->trans_wm.plane_res_b = res_blocks + 1;
	wm->trans_wm.plane_en = true;
}

static int skl_build_plane_wm_single(struct intel_crtc_state *crtc_state,
				     const struct intel_plane_state *plane_state,
				     enum plane_id plane_id, int color_plane)
{
	struct skl_plane_wm *wm = &crtc_state->wm.skl.optimal.planes[plane_id];
	struct skl_wm_params wm_params;
	int ret;

	ret = skl_compute_plane_wm_params(crtc_state, plane_state,
					  &wm_params, color_plane);
	if (ret)
		return ret;

	skl_compute_wm_levels(crtc_state, &wm_params, wm->wm);
	skl_compute_transition_wm(crtc_state, &wm_params, wm);

	return 0;
}

static int skl_build_plane_wm_uv(struct intel_crtc_state *crtc_state,
				 const struct intel_plane_state *plane_state,
				 enum plane_id plane_id)
{
	struct skl_plane_wm *wm = &crtc_state->wm.skl.optimal.planes[plane_id];
	struct skl_wm_params wm_params;
	int ret;

	wm->is_planar = true;

	/* uv plane watermarks must also be validated for NV12/Planar */
	ret = skl_compute_plane_wm_params(crtc_state, plane_state,
					  &wm_params, 1);
	if (ret)
		return ret;

	skl_compute_wm_levels(crtc_state, &wm_params, wm->uv_wm);

	return 0;
}

static int skl_build_plane_wm(struct intel_crtc_state *crtc_state,
			      const struct intel_plane_state *plane_state)
{
	struct intel_plane *plane = to_intel_plane(plane_state->base.plane);
	const struct drm_framebuffer *fb = plane_state->base.fb;
	enum plane_id plane_id = plane->id;
	int ret;

	if (!intel_wm_plane_visible(crtc_state, plane_state))
		return 0;

	ret = skl_build_plane_wm_single(crtc_state, plane_state,
					plane_id, 0);
	if (ret)
		return ret;

	if (fb->format->is_yuv && fb->format->num_planes > 1) {
		ret = skl_build_plane_wm_uv(crtc_state, plane_state,
					    plane_id);
		if (ret)
			return ret;
	}

	return 0;
}

static int icl_build_plane_wm(struct intel_crtc_state *crtc_state,
			      const struct intel_plane_state *plane_state)
{
	enum plane_id plane_id = to_intel_plane(plane_state->base.plane)->id;
	int ret;

	/* Watermarks calculated in master */
	if (plane_state->planar_slave)
		return 0;

	if (plane_state->planar_linked_plane) {
		const struct drm_framebuffer *fb = plane_state->base.fb;
		enum plane_id y_plane_id = plane_state->planar_linked_plane->id;

		WARN_ON(!intel_wm_plane_visible(crtc_state, plane_state));
		WARN_ON(!fb->format->is_yuv ||
			fb->format->num_planes == 1);

		ret = skl_build_plane_wm_single(crtc_state, plane_state,
						y_plane_id, 0);
		if (ret)
			return ret;

		ret = skl_build_plane_wm_single(crtc_state, plane_state,
						plane_id, 1);
		if (ret)
			return ret;
	} else if (intel_wm_plane_visible(crtc_state, plane_state)) {
		ret = skl_build_plane_wm_single(crtc_state, plane_state,
						plane_id, 0);
		if (ret)
			return ret;
	}

	return 0;
}

static int skl_build_pipe_wm(struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	struct skl_pipe_wm *pipe_wm = &crtc_state->wm.skl.optimal;
	struct intel_plane *plane;
	const struct intel_plane_state *plane_state;
	int ret;

	/*
	 * We'll only calculate watermarks for planes that are actually
	 * enabled, so make sure all other planes are set as disabled.
	 */
	memset(pipe_wm->planes, 0, sizeof(pipe_wm->planes));

	intel_atomic_crtc_state_for_each_plane_state(plane, plane_state,
						     crtc_state) {

		if (INTEL_GEN(dev_priv) >= 11)
			ret = icl_build_plane_wm(crtc_state, plane_state);
		else
			ret = skl_build_plane_wm(crtc_state, plane_state);
		if (ret)
			return ret;
	}

	pipe_wm->linetime = skl_compute_linetime_wm(crtc_state);

	return 0;
}

static void skl_ddb_entry_write(struct drm_i915_private *dev_priv,
				i915_reg_t reg,
				const struct skl_ddb_entry *entry)
{
	if (entry->end)
		I915_WRITE_FW(reg, (entry->end - 1) << 16 | entry->start);
	else
		I915_WRITE_FW(reg, 0);
}

static void skl_write_wm_level(struct drm_i915_private *dev_priv,
			       i915_reg_t reg,
			       const struct skl_wm_level *level)
{
	u32 val = 0;

	if (level->plane_en)
		val |= PLANE_WM_EN;
	if (level->ignore_lines)
		val |= PLANE_WM_IGNORE_LINES;
	val |= level->plane_res_b;
	val |= level->plane_res_l << PLANE_WM_LINES_SHIFT;

	I915_WRITE_FW(reg, val);
}

void skl_write_plane_wm(struct intel_plane *plane,
			const struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
	int level, max_level = ilk_wm_max_level(dev_priv);
	enum plane_id plane_id = plane->id;
	enum pipe pipe = plane->pipe;
	const struct skl_plane_wm *wm =
		&crtc_state->wm.skl.optimal.planes[plane_id];
	const struct skl_ddb_entry *ddb_y =
		&crtc_state->wm.skl.plane_ddb_y[plane_id];
	const struct skl_ddb_entry *ddb_uv =
		&crtc_state->wm.skl.plane_ddb_uv[plane_id];

	for (level = 0; level <= max_level; level++) {
		skl_write_wm_level(dev_priv, PLANE_WM(pipe, plane_id, level),
				   &wm->wm[level]);
	}
	skl_write_wm_level(dev_priv, PLANE_WM_TRANS(pipe, plane_id),
			   &wm->trans_wm);

	if (INTEL_GEN(dev_priv) >= 11) {
		skl_ddb_entry_write(dev_priv,
				    PLANE_BUF_CFG(pipe, plane_id), ddb_y);
		return;
	}

	if (wm->is_planar)
		swap(ddb_y, ddb_uv);

	skl_ddb_entry_write(dev_priv,
			    PLANE_BUF_CFG(pipe, plane_id), ddb_y);
	skl_ddb_entry_write(dev_priv,
			    PLANE_NV12_BUF_CFG(pipe, plane_id), ddb_uv);
}

void skl_write_cursor_wm(struct intel_plane *plane,
			 const struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
	int level, max_level = ilk_wm_max_level(dev_priv);
	enum plane_id plane_id = plane->id;
	enum pipe pipe = plane->pipe;
	const struct skl_plane_wm *wm =
		&crtc_state->wm.skl.optimal.planes[plane_id];
	const struct skl_ddb_entry *ddb =
		&crtc_state->wm.skl.plane_ddb_y[plane_id];

	for (level = 0; level <= max_level; level++) {
		skl_write_wm_level(dev_priv, CUR_WM(pipe, level),
				   &wm->wm[level]);
	}
	skl_write_wm_level(dev_priv, CUR_WM_TRANS(pipe), &wm->trans_wm);

	skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe), ddb);
}

bool skl_wm_level_equals(const struct skl_wm_level *l1,
			 const struct skl_wm_level *l2)
{
	return l1->plane_en == l2->plane_en &&
		l1->ignore_lines == l2->ignore_lines &&
		l1->plane_res_l == l2->plane_res_l &&
		l1->plane_res_b == l2->plane_res_b;
}

static bool skl_plane_wm_equals(struct drm_i915_private *dev_priv,
				const struct skl_plane_wm *wm1,
				const struct skl_plane_wm *wm2)
{
	int level, max_level = ilk_wm_max_level(dev_priv);

	for (level = 0; level <= max_level; level++) {
		if (!skl_wm_level_equals(&wm1->wm[level], &wm2->wm[level]) ||
		    !skl_wm_level_equals(&wm1->uv_wm[level], &wm2->uv_wm[level]))
			return false;
	}

	return skl_wm_level_equals(&wm1->trans_wm, &wm2->trans_wm);
}

static bool skl_pipe_wm_equals(struct intel_crtc *crtc,
			       const struct skl_pipe_wm *wm1,
			       const struct skl_pipe_wm *wm2)
{
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	enum plane_id plane_id;

	for_each_plane_id_on_crtc(crtc, plane_id) {
		if (!skl_plane_wm_equals(dev_priv,
					 &wm1->planes[plane_id],
					 &wm2->planes[plane_id]))
			return false;
	}

	return wm1->linetime == wm2->linetime;
}

static inline bool skl_ddb_entries_overlap(const struct skl_ddb_entry *a,
					   const struct skl_ddb_entry *b)
{
	return a->start < b->end && b->start < a->end;
}

bool skl_ddb_allocation_overlaps(const struct skl_ddb_entry *ddb,
				 const struct skl_ddb_entry *entries,
				 int num_entries, int ignore_idx)
{
	int i;

	for (i = 0; i < num_entries; i++) {
		if (i != ignore_idx &&
		    skl_ddb_entries_overlap(ddb, &entries[i]))
			return true;
	}

	return false;
}

static int
skl_ddb_add_affected_planes(const struct intel_crtc_state *old_crtc_state,
			    struct intel_crtc_state *new_crtc_state)
{
	struct intel_atomic_state *state = to_intel_atomic_state(new_crtc_state->base.state);
	struct intel_crtc *crtc = to_intel_crtc(new_crtc_state->base.crtc);
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	struct intel_plane *plane;

	for_each_intel_plane_on_crtc(&dev_priv->drm, crtc, plane) {
		struct intel_plane_state *plane_state;
		enum plane_id plane_id = plane->id;

		if (skl_ddb_entry_equal(&old_crtc_state->wm.skl.plane_ddb_y[plane_id],
					&new_crtc_state->wm.skl.plane_ddb_y[plane_id]) &&
		    skl_ddb_entry_equal(&old_crtc_state->wm.skl.plane_ddb_uv[plane_id],
					&new_crtc_state->wm.skl.plane_ddb_uv[plane_id]))
			continue;

		plane_state = intel_atomic_get_plane_state(state, plane);
		if (IS_ERR(plane_state))
			return PTR_ERR(plane_state);

		new_crtc_state->update_planes |= BIT(plane_id);
	}

	return 0;
}

static int
skl_compute_ddb(struct intel_atomic_state *state)
{
	const struct drm_i915_private *dev_priv = to_i915(state->base.dev);
	struct skl_ddb_allocation *ddb = &state->wm_results.ddb;
	struct intel_crtc_state *old_crtc_state;
	struct intel_crtc_state *new_crtc_state;
	struct intel_crtc *crtc;
	int ret, i;

	memcpy(ddb, &dev_priv->wm.skl_hw.ddb, sizeof(*ddb));

	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
					    new_crtc_state, i) {
		ret = skl_allocate_pipe_ddb(new_crtc_state, ddb);
		if (ret)
			return ret;

		ret = skl_ddb_add_affected_planes(old_crtc_state,
						  new_crtc_state);
		if (ret)
			return ret;
	}

	return 0;
}

static char enast(bool enable)
{
	return enable ? '*' : ' ';
}

static void
skl_print_wm_changes(struct intel_atomic_state *state)
{
	struct drm_i915_private *dev_priv = to_i915(state->base.dev);
	const struct intel_crtc_state *old_crtc_state;
	const struct intel_crtc_state *new_crtc_state;
	struct intel_plane *plane;
	struct intel_crtc *crtc;
	int i;

	if ((drm_debug & DRM_UT_KMS) == 0)
		return;

	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
					    new_crtc_state, i) {
		const struct skl_pipe_wm *old_pipe_wm, *new_pipe_wm;

		old_pipe_wm = &old_crtc_state->wm.skl.optimal;
		new_pipe_wm = &new_crtc_state->wm.skl.optimal;

		for_each_intel_plane_on_crtc(&dev_priv->drm, crtc, plane) {
			enum plane_id plane_id = plane->id;
			const struct skl_ddb_entry *old, *new;

			old = &old_crtc_state->wm.skl.plane_ddb_y[plane_id];
			new = &new_crtc_state->wm.skl.plane_ddb_y[plane_id];

			if (skl_ddb_entry_equal(old, new))
				continue;

			DRM_DEBUG_KMS("[PLANE:%d:%s] ddb (%4d - %4d) -> (%4d - %4d), size %4d -> %4d\n",
				      plane->base.base.id, plane->base.name,
				      old->start, old->end, new->start, new->end,
				      skl_ddb_entry_size(old), skl_ddb_entry_size(new));
		}

		for_each_intel_plane_on_crtc(&dev_priv->drm, crtc, plane) {
			enum plane_id plane_id = plane->id;
			const struct skl_plane_wm *old_wm, *new_wm;

			old_wm = &old_pipe_wm->planes[plane_id];
			new_wm = &new_pipe_wm->planes[plane_id];

			if (skl_plane_wm_equals(dev_priv, old_wm, new_wm))
				continue;

			DRM_DEBUG_KMS("[PLANE:%d:%s]   level %cwm0,%cwm1,%cwm2,%cwm3,%cwm4,%cwm5,%cwm6,%cwm7,%ctwm"
				      " -> %cwm0,%cwm1,%cwm2,%cwm3,%cwm4,%cwm5,%cwm6,%cwm7,%ctwm\n",
				      plane->base.base.id, plane->base.name,
				      enast(old_wm->wm[0].plane_en), enast(old_wm->wm[1].plane_en),
				      enast(old_wm->wm[2].plane_en), enast(old_wm->wm[3].plane_en),
				      enast(old_wm->wm[4].plane_en), enast(old_wm->wm[5].plane_en),
				      enast(old_wm->wm[6].plane_en), enast(old_wm->wm[7].plane_en),
				      enast(old_wm->trans_wm.plane_en),
				      enast(new_wm->wm[0].plane_en), enast(new_wm->wm[1].plane_en),
				      enast(new_wm->wm[2].plane_en), enast(new_wm->wm[3].plane_en),
				      enast(new_wm->wm[4].plane_en), enast(new_wm->wm[5].plane_en),
				      enast(new_wm->wm[6].plane_en), enast(new_wm->wm[7].plane_en),
				      enast(new_wm->trans_wm.plane_en));

			DRM_DEBUG_KMS("[PLANE:%d:%s]   lines %c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d"
				      " -> %c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d,%c%3d\n",
				      plane->base.base.id, plane->base.name,
				      enast(old_wm->wm[0].ignore_lines), old_wm->wm[0].plane_res_l,
				      enast(old_wm->wm[1].ignore_lines), old_wm->wm[1].plane_res_l,
				      enast(old_wm->wm[2].ignore_lines), old_wm->wm[2].plane_res_l,
				      enast(old_wm->wm[3].ignore_lines), old_wm->wm[3].plane_res_l,
				      enast(old_wm->wm[4].ignore_lines), old_wm->wm[4].plane_res_l,
				      enast(old_wm->wm[5].ignore_lines), old_wm->wm[5].plane_res_l,
				      enast(old_wm->wm[6].ignore_lines), old_wm->wm[6].plane_res_l,
				      enast(old_wm->wm[7].ignore_lines), old_wm->wm[7].plane_res_l,
				      enast(old_wm->trans_wm.ignore_lines), old_wm->trans_wm.plane_res_l,

				      enast(new_wm->wm[0].ignore_lines), new_wm->wm[0].plane_res_l,
				      enast(new_wm->wm[1].ignore_lines), new_wm->wm[1].plane_res_l,
				      enast(new_wm->wm[2].ignore_lines), new_wm->wm[2].plane_res_l,
				      enast(new_wm->wm[3].ignore_lines), new_wm->wm[3].plane_res_l,
				      enast(new_wm->wm[4].ignore_lines), new_wm->wm[4].plane_res_l,
				      enast(new_wm->wm[5].ignore_lines), new_wm->wm[5].plane_res_l,
				      enast(new_wm->wm[6].ignore_lines), new_wm->wm[6].plane_res_l,
				      enast(new_wm->wm[7].ignore_lines), new_wm->wm[7].plane_res_l,
				      enast(new_wm->trans_wm.ignore_lines), new_wm->trans_wm.plane_res_l);

			DRM_DEBUG_KMS("[PLANE:%d:%s]  blocks %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d"
				      " -> %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d\n",
				      plane->base.base.id, plane->base.name,
				      old_wm->wm[0].plane_res_b, old_wm->wm[1].plane_res_b,
				      old_wm->wm[2].plane_res_b, old_wm->wm[3].plane_res_b,
				      old_wm->wm[4].plane_res_b, old_wm->wm[5].plane_res_b,
				      old_wm->wm[6].plane_res_b, old_wm->wm[7].plane_res_b,
				      old_wm->trans_wm.plane_res_b,
				      new_wm->wm[0].plane_res_b, new_wm->wm[1].plane_res_b,
				      new_wm->wm[2].plane_res_b, new_wm->wm[3].plane_res_b,
				      new_wm->wm[4].plane_res_b, new_wm->wm[5].plane_res_b,
				      new_wm->wm[6].plane_res_b, new_wm->wm[7].plane_res_b,
				      new_wm->trans_wm.plane_res_b);

			DRM_DEBUG_KMS("[PLANE:%d:%s] min_ddb %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d"
				      " -> %4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d,%4d\n",
				      plane->base.base.id, plane->base.name,
				      old_wm->wm[0].min_ddb_alloc, old_wm->wm[1].min_ddb_alloc,
				      old_wm->wm[2].min_ddb_alloc, old_wm->wm[3].min_ddb_alloc,
				      old_wm->wm[4].min_ddb_alloc, old_wm->wm[5].min_ddb_alloc,
				      old_wm->wm[6].min_ddb_alloc, old_wm->wm[7].min_ddb_alloc,
				      old_wm->trans_wm.min_ddb_alloc,
				      new_wm->wm[0].min_ddb_alloc, new_wm->wm[1].min_ddb_alloc,
				      new_wm->wm[2].min_ddb_alloc, new_wm->wm[3].min_ddb_alloc,
				      new_wm->wm[4].min_ddb_alloc, new_wm->wm[5].min_ddb_alloc,
				      new_wm->wm[6].min_ddb_alloc, new_wm->wm[7].min_ddb_alloc,
				      new_wm->trans_wm.min_ddb_alloc);
		}
	}
}

static int intel_add_all_pipes(struct intel_atomic_state *state)
{
	struct drm_i915_private *dev_priv = to_i915(state->base.dev);
	struct intel_crtc *crtc;

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		struct intel_crtc_state *crtc_state;

		crtc_state = intel_atomic_get_crtc_state(&state->base, crtc);
		if (IS_ERR(crtc_state))
			return PTR_ERR(crtc_state);
	}

	return 0;
}

static int
skl_ddb_add_affected_pipes(struct intel_atomic_state *state)
{
	struct drm_i915_private *dev_priv = to_i915(state->base.dev);
	int ret;

	/*
	 * If this is our first atomic update following hardware readout,
	 * we can't trust the DDB that the BIOS programmed for us.  Let's
	 * pretend that all pipes switched active status so that we'll
	 * ensure a full DDB recompute.
	 */
	if (dev_priv->wm.distrust_bios_wm) {
		ret = drm_modeset_lock(&dev_priv->drm.mode_config.connection_mutex,
				       state->base.acquire_ctx);
		if (ret)
			return ret;

		state->active_pipe_changes = ~0;

		/*
		 * We usually only initialize state->active_pipes if we
		 * we're doing a modeset; make sure this field is always
		 * initialized during the sanitization process that happens
		 * on the first commit too.
		 */
		if (!state->modeset)
			state->active_pipes = dev_priv->active_pipes;
	}

	/*
	 * If the modeset changes which CRTC's are active, we need to
	 * recompute the DDB allocation for *all* active pipes, even
	 * those that weren't otherwise being modified in any way by this
	 * atomic commit.  Due to the shrinking of the per-pipe allocations
	 * when new active CRTC's are added, it's possible for a pipe that
	 * we were already using and aren't changing at all here to suddenly
	 * become invalid if its DDB needs exceeds its new allocation.
	 *
	 * Note that if we wind up doing a full DDB recompute, we can't let
	 * any other display updates race with this transaction, so we need
	 * to grab the lock on *all* CRTC's.
	 */
	if (state->active_pipe_changes || state->modeset) {
		state->wm_results.dirty_pipes = ~0;

		ret = intel_add_all_pipes(state);
		if (ret)
			return ret;
	}

	return 0;
}

/*
 * To make sure the cursor watermark registers are always consistent
 * with our computed state the following scenario needs special
 * treatment:
 *
 * 1. enable cursor
 * 2. move cursor entirely offscreen
 * 3. disable cursor
 *
 * Step 2. does call .disable_plane() but does not zero the watermarks
 * (since we consider an offscreen cursor still active for the purposes
 * of watermarks). Step 3. would not normally call .disable_plane()
 * because the actual plane visibility isn't changing, and we don't
 * deallocate the cursor ddb until the pipe gets disabled. So we must
 * force step 3. to call .disable_plane() to update the watermark
 * registers properly.
 *
 * Other planes do not suffer from this issues as their watermarks are
 * calculated based on the actual plane visibility. The only time this
 * can trigger for the other planes is during the initial readout as the
 * default value of the watermarks registers is not zero.
 */
static int skl_wm_add_affected_planes(struct intel_atomic_state *state,
				      struct intel_crtc *crtc)
{
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	const struct intel_crtc_state *old_crtc_state =
		intel_atomic_get_old_crtc_state(state, crtc);
	struct intel_crtc_state *new_crtc_state =
		intel_atomic_get_new_crtc_state(state, crtc);
	struct intel_plane *plane;

	for_each_intel_plane_on_crtc(&dev_priv->drm, crtc, plane) {
		struct intel_plane_state *plane_state;
		enum plane_id plane_id = plane->id;

		/*
		 * Force a full wm update for every plane on modeset.
		 * Required because the reset value of the wm registers
		 * is non-zero, whereas we want all disabled planes to
		 * have zero watermarks. So if we turn off the relevant
		 * power well the hardware state will go out of sync
		 * with the software state.
		 */
		if (!drm_atomic_crtc_needs_modeset(&new_crtc_state->base) &&
		    skl_plane_wm_equals(dev_priv,
					&old_crtc_state->wm.skl.optimal.planes[plane_id],
					&new_crtc_state->wm.skl.optimal.planes[plane_id]))
			continue;

		plane_state = intel_atomic_get_plane_state(state, plane);
		if (IS_ERR(plane_state))
			return PTR_ERR(plane_state);

		new_crtc_state->update_planes |= BIT(plane_id);
	}

	return 0;
}

static int
skl_compute_wm(struct intel_atomic_state *state)
{
	struct intel_crtc *crtc;
	struct intel_crtc_state *new_crtc_state;
	struct intel_crtc_state *old_crtc_state;
	struct skl_ddb_values *results = &state->wm_results;
	int ret, i;

	/* Clear all dirty flags */
	results->dirty_pipes = 0;

	ret = skl_ddb_add_affected_pipes(state);
	if (ret)
		return ret;

	/*
	 * Calculate WM's for all pipes that are part of this transaction.
	 * Note that skl_ddb_add_affected_pipes may have added more CRTC's that
	 * weren't otherwise being modified (and set bits in dirty_pipes) if
	 * pipe allocations had to change.
	 */
	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
					    new_crtc_state, i) {
		ret = skl_build_pipe_wm(new_crtc_state);
		if (ret)
			return ret;

		ret = skl_wm_add_affected_planes(state, crtc);
		if (ret)
			return ret;

		if (!skl_pipe_wm_equals(crtc,
					&old_crtc_state->wm.skl.optimal,
					&new_crtc_state->wm.skl.optimal))
			results->dirty_pipes |= BIT(crtc->pipe);
	}

	ret = skl_compute_ddb(state);
	if (ret)
		return ret;

	skl_print_wm_changes(state);

	return 0;
}

static void skl_atomic_update_crtc_wm(struct intel_atomic_state *state,
				      struct intel_crtc_state *crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
	struct drm_i915_private *dev_priv = to_i915(state->base.dev);
	struct skl_pipe_wm *pipe_wm = &crtc_state->wm.skl.optimal;
	enum pipe pipe = crtc->pipe;

	if ((state->wm_results.dirty_pipes & BIT(crtc->pipe)) == 0)
		return;

	I915_WRITE(PIPE_WM_LINETIME(pipe), pipe_wm->linetime);
}

static void skl_initial_wm(struct intel_atomic_state *state,
			   struct intel_crtc_state *crtc_state)
{
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	struct skl_ddb_values *results = &state->wm_results;

	if ((results->dirty_pipes & BIT(crtc->pipe)) == 0)
		return;

	mutex_lock(&dev_priv->wm.wm_mutex);

	if (crtc_state->base.active_changed)
		skl_atomic_update_crtc_wm(state, crtc_state);

	mutex_unlock(&dev_priv->wm.wm_mutex);
}

static void ilk_compute_wm_config(struct drm_i915_private *dev_priv,
				  struct intel_wm_config *config)
{
	struct intel_crtc *crtc;

	/* Compute the currently _active_ config */
	for_each_intel_crtc(&dev_priv->drm, crtc) {
		const struct intel_pipe_wm *wm = &crtc->wm.active.ilk;

		if (!wm->pipe_enabled)
			continue;

		config->sprites_enabled |= wm->sprites_enabled;
		config->sprites_scaled |= wm->sprites_scaled;
		config->num_pipes_active++;
	}
}

static void ilk_program_watermarks(struct drm_i915_private *dev_priv)
{
	struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
	struct ilk_wm_maximums max;
	struct intel_wm_config config = {};
	struct ilk_wm_values results = {};
	enum intel_ddb_partitioning partitioning;

	ilk_compute_wm_config(dev_priv, &config);

	ilk_compute_wm_maximums(dev_priv, 1, &config, INTEL_DDB_PART_1_2, &max);
	ilk_wm_merge(dev_priv, &config, &max, &lp_wm_1_2);

	/* 5/6 split only in single pipe config on IVB+ */
	if (INTEL_GEN(dev_priv) >= 7 &&
	    config.num_pipes_active == 1 && config.sprites_enabled) {
		ilk_compute_wm_maximums(dev_priv, 1, &config, INTEL_DDB_PART_5_6, &max);
		ilk_wm_merge(dev_priv, &config, &max, &lp_wm_5_6);

		best_lp_wm = ilk_find_best_result(dev_priv, &lp_wm_1_2, &lp_wm_5_6);
	} else {
		best_lp_wm = &lp_wm_1_2;
	}

	partitioning = (best_lp_wm == &lp_wm_1_2) ?
		       INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;

	ilk_compute_wm_results(dev_priv, best_lp_wm, partitioning, &results);

	ilk_write_wm_values(dev_priv, &results);
}

static void ilk_initial_watermarks(struct intel_atomic_state *state,
				   struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);

	mutex_lock(&dev_priv->wm.wm_mutex);
	crtc->wm.active.ilk = crtc_state->wm.ilk.intermediate;
	ilk_program_watermarks(dev_priv);
	mutex_unlock(&dev_priv->wm.wm_mutex);
}

static void ilk_optimize_watermarks(struct intel_atomic_state *state,
				    struct intel_crtc_state *crtc_state)
{
	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
	struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);

	if (!crtc_state->wm.need_postvbl_update)
		return;

	mutex_lock(&dev_priv->wm.wm_mutex);
	crtc->wm.active.ilk = crtc_state->wm.ilk.optimal;
	ilk_program_watermarks(dev_priv);
	mutex_unlock(&dev_priv->wm.wm_mutex);
}

static inline void skl_wm_level_from_reg_val(u32 val,
					     struct skl_wm_level *level)
{
	level->plane_en = val & PLANE_WM_EN;
	level->ignore_lines = val & PLANE_WM_IGNORE_LINES;
	level->plane_res_b = val & PLANE_WM_BLOCKS_MASK;
	level->plane_res_l = (val >> PLANE_WM_LINES_SHIFT) &
		PLANE_WM_LINES_MASK;
}

void skl_pipe_wm_get_hw_state(struct intel_crtc *crtc,
			      struct skl_pipe_wm *out)
{
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
	enum pipe pipe = crtc->pipe;
	int level, max_level;
	enum plane_id plane_id;
	u32 val;

	max_level = ilk_wm_max_level(dev_priv);

	for_each_plane_id_on_crtc(crtc, plane_id) {
		struct skl_plane_wm *wm = &out->planes[plane_id];

		for (level = 0; level <= max_level; level++) {
			if (plane_id != PLANE_CURSOR)
				val = I915_READ(PLANE_WM(pipe, plane_id, level));
			else
				val = I915_READ(CUR_WM(pipe, level));

			skl_wm_level_from_reg_val(val, &wm->wm[level]);
		}

		if (plane_id != PLANE_CURSOR)
			val = I915_READ(PLANE_WM_TRANS(pipe, plane_id));
		else
			val = I915_READ(CUR_WM_TRANS(pipe));

		skl_wm_level_from_reg_val(val, &wm->trans_wm);
	}

	if (!crtc->active)
		return;

	out->linetime = I915_READ(PIPE_WM_LINETIME(pipe));
}

void skl_wm_get_hw_state(struct drm_i915_private *dev_priv)
{
	struct skl_ddb_values *hw = &dev_priv->wm.skl_hw;
	struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb;
	struct intel_crtc *crtc;
	struct intel_crtc_state *crtc_state;

	skl_ddb_get_hw_state(dev_priv, ddb);
	for_each_intel_crtc(&dev_priv->drm, crtc) {
		crtc_state = to_intel_crtc_state(crtc->base.state);

		skl_pipe_wm_get_hw_state(crtc, &crtc_state->wm.skl.optimal);

		if (crtc->active)
			hw->dirty_pipes |= BIT(crtc->pipe);
	}

	if (dev_priv->active_pipes) {
		/* Fully recompute DDB on first atomic commit */
		dev_priv->wm.distrust_bios_wm = true;
	}
}

static void ilk_pipe_wm_get_hw_state(struct intel_crtc *crtc)
{
	struct drm_device *dev = crtc->base.dev;
	struct drm_i915_private *dev_priv = to_i915(dev);
	struct ilk_wm_values *hw = &dev_priv->wm.hw;
	struct intel_crtc_state *crtc_state = to_intel_crtc_state(crtc->base.state);
	struct intel_pipe_wm *active = &crtc_state->wm.ilk.optimal;
	enum pipe pipe = crtc->pipe;
	static const i915_reg_t wm0_pipe_reg[] = {
		[PIPE_A] = WM0_PIPEA_ILK,
		[PIPE_B] = WM0_PIPEB_ILK,
		[PIPE_C] = WM0_PIPEC_IVB,
	};

	hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]);
	if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
		hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));

	memset(active, 0, sizeof(*active));

	active->pipe_enabled = crtc->active;

	if (active->pipe_enabled) {
		u32 tmp = hw->wm_pipe[pipe];

		/*
		 * For active pipes LP0 watermark is marked as
		 * enabled, and LP1+ watermaks as disabled since
		 * we can't really reverse compute them in case
		 * multiple pipes are active.
		 */
		active->wm[0].enable = true;
		active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT;
		active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT;
		active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK;
		active->linetime = hw->wm_linetime[pipe];
	} else {
		int level, max_level = ilk_wm_max_level(dev_priv);

		/*
		 * For inactive pipes, all watermark levels
		 * should be marked as enabled but zeroed,
		 * which is what we'd compute them to.
		 */
		for (level = 0; level <= max_level; level++)
			active->wm[level].enable = true;
	}

	crtc->wm.active.ilk = *active;
}

#define _FW_WM(value, plane) \
	(((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT)
#define _FW_WM_VLV(value, plane) \
	(((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT)

static void g4x_read_wm_values(struct drm_i915_private *dev_priv,
			       struct g4x_wm_values *wm)
{
	u32 tmp;

	tmp = I915_READ(DSPFW1);
	wm->sr.plane = _FW_WM(tmp, SR);
	wm->pipe[PIPE_B].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORB);
	wm->pipe[PIPE_B].plane[PLANE_PRIMARY] = _FW_WM(tmp, PLANEB);
	wm->pipe[PIPE_A].plane[PLANE_PRIMARY] = _FW_WM(tmp, PLANEA);

	tmp = I915_READ(DSPFW2);
	wm->fbc_en = tmp & DSPFW_FBC_SR_EN;
	wm->sr.fbc = _FW_WM(tmp, FBC_SR);
	wm->hpll.fbc = _FW_WM(tmp, FBC_HPLL_SR);
	wm->pipe[PIPE_B].plane[PLANE_SPRITE0] = _FW_WM(tmp, SPRITEB);
	wm->pipe[PIPE_A].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORA);
	wm->pipe[PIPE_A].plane[PLANE_SPRITE0] = _FW_WM(tmp, SPRITEA);

	tmp = I915_READ(DSPFW3);
	wm->hpll_en = tmp & DSPFW_HPLL_SR_EN;
	wm->sr.cursor = _FW_WM(tmp, CURSOR_SR);
	wm->hpll.cursor = _FW_WM(tmp, HPLL_CURSOR);
	wm->hpll.plane = _FW_WM(tmp, HPLL_SR);
}

static void vlv_read_wm_values(struct drm_i915_private *dev_priv,
			       struct vlv_wm_values *wm)
{
	enum pipe pipe;
	u32 tmp;

	for_each_pipe(dev_priv, pipe) {
		tmp = I915_READ(VLV_DDL(pipe));

		wm->ddl[pipe].plane[PLANE_PRIMARY] =
			(tmp >> DDL_PLANE_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
		wm->ddl[pipe].plane[PLANE_CURSOR] =
			(tmp >> DDL_CURSOR_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
		wm->ddl[pipe].plane[PLANE_SPRITE0] =
			(tmp >> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
		wm->ddl[pipe].plane[PLANE_SPRITE1] =
			(tmp >> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
	}

	tmp = I915_READ(DSPFW1);
	wm->sr.plane = _FW_WM(tmp, SR);
	wm->pipe[PIPE_B].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORB);
	wm->pipe[PIPE_B].plane[PLANE_PRIMARY] = _FW_WM_VLV(tmp, PLANEB);
	wm->pipe[PIPE_A].plane[PLANE_PRIMARY] = _FW_WM_VLV(tmp, PLANEA);

	tmp = I915_READ(DSPFW2);
	wm->pipe[PIPE_A].plane[PLANE_SPRITE1] = _FW_WM_VLV(tmp, SPRITEB);
	wm->pipe[PIPE_A].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORA);
	wm->pipe[PIPE_A].plane[PLANE_SPRITE0] = _FW_WM_VLV(tmp, SPRITEA);

	tmp = I915_READ(DSPFW3);
	wm->sr.cursor = _FW_WM(tmp, CURSOR_SR);

	if (IS_CHERRYVIEW(dev_priv)) {
		tmp = I915_READ(DSPFW7_CHV);
		wm->pipe[PIPE_B].plane[PLANE_SPRITE1] = _FW_WM_VLV(tmp, SPRITED);
		wm->pipe[PIPE_B].plane[PLANE_SPRITE0] = _FW_WM_VLV(tmp, SPRITEC);

		tmp = I915_READ(DSPFW8_CHV);
		wm->pipe[PIPE_C].plane[PLANE_SPRITE1] = _FW_WM_VLV(tmp, SPRITEF);
		wm->pipe[PIPE_C].plane[PLANE_SPRITE0] = _FW_WM_VLV(tmp, SPRITEE);

		tmp = I915_READ(DSPFW9_CHV);
		wm->pipe[PIPE_C].plane[PLANE_PRIMARY] = _FW_WM_VLV(tmp, PLANEC);
		wm->pipe[PIPE_C].plane[PLANE_CURSOR] = _FW_WM(tmp, CURSORC);

		tmp = I915_READ(DSPHOWM);
		wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9;
		wm->pipe[PIPE_C].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITEF_HI) << 8;
		wm->pipe[PIPE_C].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEE_HI) << 8;
		wm->pipe[PIPE_C].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEC_HI) << 8;
		wm->pipe[PIPE_B].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITED_HI) << 8;
		wm->pipe[PIPE_B].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEC_HI) << 8;
		wm->pipe[PIPE_B].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEB_HI) << 8;
		wm->pipe[PIPE_A].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITEB_HI) << 8;
		wm->pipe[PIPE_A].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEA_HI) << 8;
		wm->pipe[PIPE_A].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEA_HI) << 8;
	} else {
		tmp = I915_READ(DSPFW7);
		wm->pipe[PIPE_B].plane[PLANE_SPRITE1] = _FW_WM_VLV(tmp, SPRITED);
		wm->pipe[PIPE_B].plane[PLANE_SPRITE0] = _FW_WM_VLV(tmp, SPRITEC);

		tmp = I915_READ(DSPHOWM);
		wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9;
		wm->pipe[PIPE_B].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITED_HI) << 8;
		wm->pipe[PIPE_B].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEC_HI) << 8;
		wm->pipe[PIPE_B].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEB_HI) << 8;
		wm->pipe[PIPE_A].plane[PLANE_SPRITE1] |= _FW_WM(tmp, SPRITEB_HI) << 8;
		wm->pipe[PIPE_A].plane[PLANE_SPRITE0] |= _FW_WM(tmp, SPRITEA_HI) << 8;
		wm->pipe[PIPE_A].plane[PLANE_PRIMARY] |= _FW_WM(tmp, PLANEA_HI) << 8;
	}
}

#undef _FW_WM
#undef _FW_WM_VLV

void g4x_wm_get_hw_state(struct drm_i915_private *dev_priv)
{
	struct g4x_wm_values *wm = &dev_priv->wm.g4x;
	struct intel_crtc *crtc;

	g4x_read_wm_values(dev_priv, wm);

	wm->cxsr = I915_READ(FW_BLC_SELF) & FW_BLC_SELF_EN;

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		struct intel_crtc_state *crtc_state =
			to_intel_crtc_state(crtc->base.state);
		struct g4x_wm_state *active = &crtc->wm.active.g4x;
		struct g4x_pipe_wm *raw;
		enum pipe pipe = crtc->pipe;
		enum plane_id plane_id;
		int level, max_level;

		active->cxsr = wm->cxsr;
		active->hpll_en = wm->hpll_en;
		active->fbc_en = wm->fbc_en;

		active->sr = wm->sr;
		active->hpll = wm->hpll;

		for_each_plane_id_on_crtc(crtc, plane_id) {
			active->wm.plane[plane_id] =
				wm->pipe[pipe].plane[plane_id];
		}

		if (wm->cxsr && wm->hpll_en)
			max_level = G4X_WM_LEVEL_HPLL;
		else if (wm->cxsr)
			max_level = G4X_WM_LEVEL_SR;
		else
			max_level = G4X_WM_LEVEL_NORMAL;

		level = G4X_WM_LEVEL_NORMAL;
		raw = &crtc_state->wm.g4x.raw[level];
		for_each_plane_id_on_crtc(crtc, plane_id)
			raw->plane[plane_id] = active->wm.plane[plane_id];

		if (++level > max_level)
			goto out;

		raw = &crtc_state->wm.g4x.raw[level];
		raw->plane[PLANE_PRIMARY] = active->sr.plane;
		raw->plane[PLANE_CURSOR] = active->sr.cursor;
		raw->plane[PLANE_SPRITE0] = 0;
		raw->fbc = active->sr.fbc;

		if (++level > max_level)
			goto out;

		raw = &crtc_state->wm.g4x.raw[level];
		raw->plane[PLANE_PRIMARY] = active->hpll.plane;
		raw->plane[PLANE_CURSOR] = active->hpll.cursor;
		raw->plane[PLANE_SPRITE0] = 0;
		raw->fbc = active->hpll.fbc;

	out:
		for_each_plane_id_on_crtc(crtc, plane_id)
			g4x_raw_plane_wm_set(crtc_state, level,
					     plane_id, USHRT_MAX);
		g4x_raw_fbc_wm_set(crtc_state, level, USHRT_MAX);

		crtc_state->wm.g4x.optimal = *active;
		crtc_state->wm.g4x.intermediate = *active;

		DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite=%d\n",
			      pipe_name(pipe),
			      wm->pipe[pipe].plane[PLANE_PRIMARY],
			      wm->pipe[pipe].plane[PLANE_CURSOR],
			      wm->pipe[pipe].plane[PLANE_SPRITE0]);
	}

	DRM_DEBUG_KMS("Initial SR watermarks: plane=%d, cursor=%d fbc=%d\n",
		      wm->sr.plane, wm->sr.cursor, wm->sr.fbc);
	DRM_DEBUG_KMS("Initial HPLL watermarks: plane=%d, SR cursor=%d fbc=%d\n",
		      wm->hpll.plane, wm->hpll.cursor, wm->hpll.fbc);
	DRM_DEBUG_KMS("Initial SR=%s HPLL=%s FBC=%s\n",
		      yesno(wm->cxsr), yesno(wm->hpll_en), yesno(wm->fbc_en));
}

void g4x_wm_sanitize(struct drm_i915_private *dev_priv)
{
	struct intel_plane *plane;
	struct intel_crtc *crtc;

	mutex_lock(&dev_priv->wm.wm_mutex);

	for_each_intel_plane(&dev_priv->drm, plane) {
		struct intel_crtc *crtc =
			intel_get_crtc_for_pipe(dev_priv, plane->pipe);
		struct intel_crtc_state *crtc_state =
			to_intel_crtc_state(crtc->base.state);
		struct intel_plane_state *plane_state =
			to_intel_plane_state(plane->base.state);
		struct g4x_wm_state *wm_state = &crtc_state->wm.g4x.optimal;
		enum plane_id plane_id = plane->id;
		int level;

		if (plane_state->base.visible)
			continue;

		for (level = 0; level < 3; level++) {
			struct g4x_pipe_wm *raw =
				&crtc_state->wm.g4x.raw[level];

			raw->plane[plane_id] = 0;
			wm_state->wm.plane[plane_id] = 0;
		}

		if (plane_id == PLANE_PRIMARY) {
			for (level = 0; level < 3; level++) {
				struct g4x_pipe_wm *raw =
					&crtc_state->wm.g4x.raw[level];
				raw->fbc = 0;
			}

			wm_state->sr.fbc = 0;
			wm_state->hpll.fbc = 0;
			wm_state->fbc_en = false;
		}
	}

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		struct intel_crtc_state *crtc_state =
			to_intel_crtc_state(crtc->base.state);

		crtc_state->wm.g4x.intermediate =
			crtc_state->wm.g4x.optimal;
		crtc->wm.active.g4x = crtc_state->wm.g4x.optimal;
	}

	g4x_program_watermarks(dev_priv);

	mutex_unlock(&dev_priv->wm.wm_mutex);
}

void vlv_wm_get_hw_state(struct drm_i915_private *dev_priv)
{
	struct vlv_wm_values *wm = &dev_priv->wm.vlv;
	struct intel_crtc *crtc;
	u32 val;

	vlv_read_wm_values(dev_priv, wm);

	wm->cxsr = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN;
	wm->level = VLV_WM_LEVEL_PM2;

	if (IS_CHERRYVIEW(dev_priv)) {
		vlv_punit_get(dev_priv);

		val = vlv_punit_read(dev_priv, PUNIT_REG_DSPSSPM);
		if (val & DSP_MAXFIFO_PM5_ENABLE)
			wm->level = VLV_WM_LEVEL_PM5;

		/*
		 * If DDR DVFS is disabled in the BIOS, Punit
		 * will never ack the request. So if that happens
		 * assume we don't have to enable/disable DDR DVFS
		 * dynamically. To test that just set the REQ_ACK
		 * bit to poke the Punit, but don't change the
		 * HIGH/LOW bits so that we don't actually change
		 * the current state.
		 */
		val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
		val |= FORCE_DDR_FREQ_REQ_ACK;
		vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val);

		if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) &
			      FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) {
			DRM_DEBUG_KMS("Punit not acking DDR DVFS request, "
				      "assuming DDR DVFS is disabled\n");
			dev_priv->wm.max_level = VLV_WM_LEVEL_PM5;
		} else {
			val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
			if ((val & FORCE_DDR_HIGH_FREQ) == 0)
				wm->level = VLV_WM_LEVEL_DDR_DVFS;
		}

		vlv_punit_put(dev_priv);
	}

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		struct intel_crtc_state *crtc_state =
			to_intel_crtc_state(crtc->base.state);
		struct vlv_wm_state *active = &crtc->wm.active.vlv;
		const struct vlv_fifo_state *fifo_state =
			&crtc_state->wm.vlv.fifo_state;
		enum pipe pipe = crtc->pipe;
		enum plane_id plane_id;
		int level;

		vlv_get_fifo_size(crtc_state);

		active->num_levels = wm->level + 1;
		active->cxsr = wm->cxsr;

		for (level = 0; level < active->num_levels; level++) {
			struct g4x_pipe_wm *raw =
				&crtc_state->wm.vlv.raw[level];

			active->sr[level].plane = wm->sr.plane;
			active->sr[level].cursor = wm->sr.cursor;

			for_each_plane_id_on_crtc(crtc, plane_id) {
				active->wm[level].plane[plane_id] =
					wm->pipe[pipe].plane[plane_id];

				raw->plane[plane_id] =
					vlv_invert_wm_value(active->wm[level].plane[plane_id],
							    fifo_state->plane[plane_id]);
			}
		}

		for_each_plane_id_on_crtc(crtc, plane_id)
			vlv_raw_plane_wm_set(crtc_state, level,
					     plane_id, USHRT_MAX);
		vlv_invalidate_wms(crtc, active, level);

		crtc_state->wm.vlv.optimal = *active;
		crtc_state->wm.vlv.intermediate = *active;

		DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n",
			      pipe_name(pipe),
			      wm->pipe[pipe].plane[PLANE_PRIMARY],
			      wm->pipe[pipe].plane[PLANE_CURSOR],
			      wm->pipe[pipe].plane[PLANE_SPRITE0],
			      wm->pipe[pipe].plane[PLANE_SPRITE1]);
	}

	DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n",
		      wm->sr.plane, wm->sr.cursor, wm->level, wm->cxsr);
}

void vlv_wm_sanitize(struct drm_i915_private *dev_priv)
{
	struct intel_plane *plane;
	struct intel_crtc *crtc;

	mutex_lock(&dev_priv->wm.wm_mutex);

	for_each_intel_plane(&dev_priv->drm, plane) {
		struct intel_crtc *crtc =
			intel_get_crtc_for_pipe(dev_priv, plane->pipe);
		struct intel_crtc_state *crtc_state =
			to_intel_crtc_state(crtc->base.state);
		struct intel_plane_state *plane_state =
			to_intel_plane_state(plane->base.state);
		struct vlv_wm_state *wm_state = &crtc_state->wm.vlv.optimal;
		const struct vlv_fifo_state *fifo_state =
			&crtc_state->wm.vlv.fifo_state;
		enum plane_id plane_id = plane->id;
		int level;

		if (plane_state->base.visible)
			continue;

		for (level = 0; level < wm_state->num_levels; level++) {
			struct g4x_pipe_wm *raw =
				&crtc_state->wm.vlv.raw[level];

			raw->plane[plane_id] = 0;

			wm_state->wm[level].plane[plane_id] =
				vlv_invert_wm_value(raw->plane[plane_id],
						    fifo_state->plane[plane_id]);
		}
	}

	for_each_intel_crtc(&dev_priv->drm, crtc) {
		struct intel_crtc_state *crtc_state =
			to_intel_crtc_state(crtc->base.state);

		crtc_state->wm.vlv.intermediate =
			crtc_state->wm.vlv.optimal;
		crtc->wm.active.vlv = crtc_state->wm.vlv.optimal;
	}

	vlv_program_watermarks(dev_priv);

	mutex_unlock(&dev_priv->wm.wm_mutex);
}

/*
 * FIXME should probably kill this and improve
 * the real watermark readout/sanitation instead
 */
static void ilk_init_lp_watermarks(struct drm_i915_private *dev_priv)
{
	I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN);
	I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN);
	I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN);

	/*
	 * Don't touch WM1S_LP_EN here.
	 * Doing so could cause underruns.
	 */
}

void ilk_wm_get_hw_state(struct drm_i915_private *dev_priv)
{
	struct ilk_wm_values *hw = &dev_priv->wm.hw;
	struct intel_crtc *crtc;

	ilk_init_lp_watermarks(dev_priv);

	for_each_intel_crtc(&dev_priv->drm, crtc)
		ilk_pipe_wm_get_hw_state(crtc);

	hw->wm_lp[0] = I915_READ(WM1_LP_ILK);
	hw->wm_lp[1] = I915_READ(WM2_LP_ILK);
	hw->wm_lp[2] = I915_READ(WM3_LP_ILK);

	hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
	if (INTEL_GEN(dev_priv) >= 7) {
		hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
		hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
	}

	if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
		hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
			INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
	else if (IS_IVYBRIDGE(dev_priv))
		hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ?
			INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;

	hw->enable_fbc_wm =
		!(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
}

/**
 * intel_update_watermarks - update FIFO watermark values based on current modes
 * @crtc: the #intel_crtc on which to compute the WM
 *
 * Calculate watermark values for the various WM regs based on current mode
 * and plane configuration.
 *
 * There are several cases to deal with here:
 *   - normal (i.e. non-self-refresh)
 *   - self-refresh (SR) mode
 *   - lines are large relative to FIFO size (buffer can hold up to 2)
 *   - lines are small relative to FIFO size (buffer can hold more than 2
 *     lines), so need to account for TLB latency
 *
 *   The normal calculation is:
 *     watermark = dotclock * bytes per pixel * latency
 *   where latency is platform & configuration dependent (we assume pessimal
 *   values here).
 *
 *   The SR calculation is:
 *     watermark = (trunc(latency/line time)+1) * surface width *
 *       bytes per pixel
 *   where
 *     line time = htotal / dotclock
 *     surface width = hdisplay for normal plane and 64 for cursor
 *   and latency is assumed to be high, as above.
 *
 * The final value programmed to the register should always be rounded up,
 * and include an extra 2 entries to account for clock crossings.
 *
 * We don't use the sprite, so we can ignore that.  And on Crestline we have
 * to set the non-SR watermarks to 8.
 */
void intel_update_watermarks(struct intel_crtc *crtc)
{
	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);

	if (dev_priv->display.update_wm)
		dev_priv->display.update_wm(crtc);
}

void intel_enable_ipc(struct drm_i915_private *dev_priv)
{
	u32 val;

	if (!HAS_IPC(dev_priv))
		return;

	val = I915_READ(DISP_ARB_CTL2);

	if (dev_priv->ipc_enabled)
		val |= DISP_IPC_ENABLE;
	else
		val &= ~DISP_IPC_ENABLE;

	I915_WRITE(DISP_ARB_CTL2, val);
}

static bool intel_can_enable_ipc(struct drm_i915_private *dev_priv)
{
	/* Display WA #0477 WaDisableIPC: skl */
	if (IS_SKYLAKE(dev_priv))
		return false;

	/* Display WA #1141: SKL:all KBL:all CFL */
	if (IS_KABYLAKE(dev_priv) || IS_COFFEELAKE(dev_priv))
		return dev_priv->dram_info.symmetric_memory;

	return true;
}

void intel_init_ipc(struct drm_i915_private *dev_priv)
{
	if (!HAS_IPC(dev_priv))
		return;

	dev_priv->ipc_enabled = intel_can_enable_ipc(dev_priv);

	intel_enable_ipc(dev_priv);
}

static void ibx_init_clock_gating(struct drm_i915_private *dev_priv)
{
	/*
	 * On Ibex Peak and Cougar Point, we need to disable clock
	 * gating for the panel power sequencer or it will fail to
	 * start up when no ports are active.
	 */
	I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
}

static void g4x_disable_trickle_feed(struct drm_i915_private *dev_priv)
{
	enum pipe pipe;

	for_each_pipe(dev_priv, pipe) {
		I915_WRITE(DSPCNTR(pipe),
			   I915_READ(DSPCNTR(pipe)) |
			   DISPPLANE_TRICKLE_FEED_DISABLE);

		I915_WRITE(DSPSURF(pipe), I915_READ(DSPSURF(pipe)));
		POSTING_READ(DSPSURF(pipe));
	}
}

static void ilk_init_clock_gating(struct drm_i915_private *dev_priv)
{
	u32 dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;

	/*
	 * Required for FBC
	 * WaFbcDisableDpfcClockGating:ilk
	 */
	dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
		   ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
		   ILK_DPFDUNIT_CLOCK_GATE_ENABLE;

	I915_WRITE(PCH_3DCGDIS0,
		   MARIUNIT_CLOCK_GATE_DISABLE |
		   SVSMUNIT_CLOCK_GATE_DISABLE);
	I915_WRITE(PCH_3DCGDIS1,
		   VFMUNIT_CLOCK_GATE_DISABLE);

	/*
	 * According to the spec the following bits should be set in
	 * order to enable memory self-refresh
	 * The bit 22/21 of 0x42004
	 * The bit 5 of 0x42020
	 * The bit 15 of 0x45000
	 */
	I915_WRITE(ILK_DISPLAY_CHICKEN2,
		   (I915_READ(ILK_DISPLAY_CHICKEN2) |
		    ILK_DPARB_GATE | ILK_VSDPFD_FULL));
	dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
	I915_WRITE(DISP_ARB_CTL,
		   (I915_READ(DISP_ARB_CTL) |
		    DISP_FBC_WM_DIS));

	/*
	 * Based on the document from hardware guys the following bits
	 * should be set unconditionally in order to enable FBC.
	 * The bit 22 of 0x42000
	 * The bit 22 of 0x42004
	 * The bit 7,8,9 of 0x42020.
	 */
	if (IS_IRONLAKE_M(dev_priv)) {
		/* WaFbcAsynchFlipDisableFbcQueue:ilk */
		I915_WRITE(ILK_DISPLAY_CHICKEN1,
			   I915_READ(ILK_DISPLAY_CHICKEN1) |
			   ILK_FBCQ_DIS);
		I915_WRITE(ILK_DISPLAY_CHICKEN2,
			   I915_READ(ILK_DISPLAY_CHICKEN2) |
			   ILK_DPARB_GATE);
	}

	I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);

	I915_WRITE(ILK_DISPLAY_CHICKEN2,
		   I915_READ(ILK_DISPLAY_CHICKEN2) |
		   ILK_ELPIN_409_SELECT);
	I915_WRITE(_3D_CHICKEN2,
		   _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
		   _3D_CHICKEN2_WM_READ_PIPELINED);

	/* WaDisableRenderCachePipelinedFlush:ilk */
	I915_WRITE(CACHE_MODE_0,
		   _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));

	/* WaDisable_RenderCache_OperationalFlush:ilk */
	I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));

	g4x_disable_trickle_feed(dev_priv);

	ibx_init_clock_gating(dev_priv);
}

static void cpt_init_clock_gating(struct drm_i915_private *dev_priv)
{
	enum pipe pipe;
	u32 val;

	/*
	 * On Ibex Peak and Cougar Point, we need to disable clock
	 * gating for the panel power sequencer or it will fail to
	 * start up when no ports are active.
	 */
	I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE |
		   PCH_DPLUNIT_CLOCK_GATE_DISABLE |
		   PCH_CPUNIT_CLOCK_GATE_DISABLE);
	I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
		   DPLS_EDP_PPS_FIX_DIS);
	/* The below fixes the weird display corruption, a few pixels shifted
	 * downward, on (only) LVDS of some HP laptops with IVY.
	 */
	for_each_pipe(dev_priv, pipe) {
		val = I915_READ(TRANS_CHICKEN2(pipe));
		val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
		val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
		if (dev_priv->vbt.fdi_rx_polarity_inverted)
			val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
		val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
		val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
		val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
		I915_WRITE(TRANS_CHICKEN2(pipe), val);
	}
	/* WADP0ClockGatingDisable */
	for_each_pipe(dev_priv, pipe) {
		I915_WRITE(TRANS_CHICKEN1(pipe),
			   TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
	}
}

static void gen6_check_mch_setup(struct drm_i915_private *dev_priv)
{
	u32 tmp;

	tmp = I915_READ(MCH_SSKPD);
	if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL)
		DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n",
			      tmp);
}

static void gen6_init_clock_gating(struct drm_i915_private *dev_priv)
{
	u32 dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;

	I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);

	I915_WRITE(ILK_DISPLAY_CHICKEN2,
		   I915_READ(ILK_DISPLAY_CHICKEN2) |
		   ILK_ELPIN_409_SELECT);

	/* WaDisableHiZPlanesWhenMSAAEnabled:snb */
	I915_WRITE(_3D_CHICKEN,
		   _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));

	/* WaDisable_RenderCache_OperationalFlush:snb */
	I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));

	/*
	 * BSpec recoomends 8x4 when MSAA is used,
	 * however in practice 16x4 seems fastest.
	 *
	 * Note that PS/WM thread counts depend on the WIZ hashing
	 * disable bit, which we don't touch here, but it's good
	 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
	 */
	I915_WRITE(GEN6_GT_MODE,
		   _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));

	I915_WRITE(CACHE_MODE_0,
		   _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));

	I915_WRITE(GEN6_UCGCTL1,
		   I915_READ(GEN6_UCGCTL1) |
		   GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
		   GEN6_CSUNIT_CLOCK_GATE_DISABLE);

	/* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
	 * gating disable must be set.  Failure to set it results in
	 * flickering pixels due to Z write ordering failures after
	 * some amount of runtime in the Mesa "fire" demo, and Unigine
	 * Sanctuary and Tropics, and apparently anything else with
	 * alpha test or pixel discard.
	 *
	 * According to the spec, bit 11 (RCCUNIT) must also be set,
	 * but we didn't debug actual testcases to find it out.
	 *
	 * WaDisableRCCUnitClockGating:snb
	 * WaDisableRCPBUnitClockGating:snb
	 */
	I915_WRITE(GEN6_UCGCTL2,
		   GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
		   GEN6_RCCUNIT_CLOCK_GATE_DISABLE);

	/* WaStripsFansDisableFastClipPerformanceFix:snb */
	I915_WRITE(_3D_CHICKEN3,
		   _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL));

	/*
	 * Bspec says:
	 * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
	 * 3DSTATE_SF number of SF output attributes is more than 16."
	 */
	I915_WRITE(_3D_CHICKEN3,
		   _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH));

	/*
	 * According to the spec the following bits should be
	 * set in order to enable memory self-refresh and fbc:
	 * The bit21 and bit22 of 0x42000
	 * The bit21 and bit22 of 0x42004
	 * The bit5 and bit7 of 0x42020
	 * The bit14 of 0x70180
	 * The bit14 of 0x71180
	 *
	 * WaFbcAsynchFlipDisableFbcQueue:snb
	 */
	I915_WRITE(ILK_DISPLAY_CHICKEN1,
		   I915_READ(ILK_DISPLAY_CHICKEN1) |
		   ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
	I915_WRITE(ILK_DISPLAY_CHICKEN2,
		   I915_READ(ILK_DISPLAY_CHICKEN2) |
		   ILK_DPARB_GATE | ILK_VSDPFD_FULL);
	I915_WRITE(ILK_DSPCLK_GATE_D,
		   I915_READ(ILK_DSPCLK_GATE_D) |
		   ILK_DPARBUNIT_CLOCK_GATE_ENABLE  |
		   ILK_DPFDUNIT_CLOCK_GATE_ENABLE);

	g4x_disable_trickle_feed(dev_priv);

	cpt_init_clock_gating(dev_priv);

	gen6_check_mch_setup(dev_priv);
}

static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
{
	u32 reg = I915_READ(GEN7_FF_THREAD_MODE);

	/*
	 * WaVSThreadDispatchOverride:ivb,vlv
	 *
	 * This actually overrides the dispatch
	 * mode for all thread types.
	 */
	reg &= ~GEN7_FF_SCHED_MASK;
	reg |= GEN7_FF_TS_SCHED_HW;
	reg |= GEN7_FF_VS_SCHED_HW;
	reg |= GEN7_FF_DS_SCHED_HW;

	I915_WRITE(GEN7_FF_THREAD_MODE, reg);
}

static void lpt_init_clock_gating(struct drm_i915_private *dev_priv)
{
	/*
	 * TODO: this bit should only be enabled when really needed, then
	 * disabled when not needed anymore in order to save power.
	 */
	if (HAS_PCH_LPT_LP(dev_priv))
		I915_WRITE(SOUTH_DSPCLK_GATE_D,
			   I915_READ(SOUTH_DSPCLK_GATE_D) |
			   PCH_LP_PARTITION_LEVEL_DISABLE);

	/* WADPOClockGatingDisable:hsw */
	I915_WRITE(TRANS_CHICKEN1(PIPE_A),
		   I915_READ(TRANS_CHICKEN1(PIPE_A)) |
		   TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
}

static void lpt_suspend_hw(struct drm_i915_private *dev_priv)
{
	if (HAS_PCH_LPT_LP(dev_priv)) {
		u32 val = I915_READ(SOUTH_DSPCLK_GATE_D);

		val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
		I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
	}
}

static void gen8_set_l3sqc_credits(struct drm_i915_private *dev_priv,
				   int general_prio_credits,
				   int high_prio_credits)
{
	u32 misccpctl;
	u32 val;

	/* WaTempDisableDOPClkGating:bdw */
	misccpctl = I915_READ(GEN7_MISCCPCTL);
	I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE);

	val = I915_READ(GEN8_L3SQCREG1);
	val &= ~L3_PRIO_CREDITS_MASK;
	val |= L3_GENERAL_PRIO_CREDITS(general_prio_credits);
	val |= L3_HIGH_PRIO_CREDITS(high_prio_credits);
	I915_WRITE(GEN8_L3SQCREG1, val);

	/*
	 * Wait at least 100 clocks before re-enabling clock gating.
	 * See the definition of L3SQCREG1 in BSpec.
	 */
	POSTING_READ(GEN8_L3SQCREG1);
	udelay(1);
	I915_WRITE(GEN7_MISCCPCTL, misccpctl);
}

static void icl_init_clock_gating(struct drm_i915_private *dev_priv)
{
	/* This is not an Wa. Enable to reduce Sampler power */
	I915_WRITE(GEN10_DFR_RATIO_EN_AND_CHICKEN,
		   I915_READ(GEN10_DFR_RATIO_EN_AND_CHICKEN) & ~DFR_DISABLE);

	/* WaEnable32PlaneMode:icl */
	I915_WRITE(GEN9_CSFE_CHICKEN1_RCS,
		   _MASKED_BIT_ENABLE(GEN11_ENABLE_32_PLANE_MODE));
}

static void tgl_init_clock_gating(struct drm_i915_private *dev_priv)
{
	u32 vd_pg_enable = 0;
	unsigned int i;

	/* This is not a WA. Enable VD HCP & MFX_ENC powergate */
	for (i = 0; i < I915_MAX_VCS; i++) {
		if (HAS_ENGINE(dev_priv, _VCS(i)))
			vd_pg_enable |= VDN_HCP_POWERGATE_ENABLE(i) |
					VDN_MFX_POWERGATE_ENABLE(i);
	}

	I915_WRITE(POWERGATE_ENABLE,
		   I915_READ(POWERGATE_ENABLE) | vd_pg_enable);
}

static void cnp_init_clock_gating(struct drm_i915_private *dev_priv)
{
	if (!HAS_PCH_CNP(dev_priv))
		return;

	/* Display WA #1181 WaSouthDisplayDisablePWMCGEGating: cnp */
	I915_WRITE(SOUTH_DSPCLK_GATE_D, I915_READ(SOUTH_DSPCLK_GATE_D) |
		   CNP_PWM_CGE_GATING_DISABLE);
}

static void cnl_init_clock_gating(struct drm_i915_private *dev_priv)
{
	u32 val;
	cnp_init_clock_gating(dev_priv);

	/* This is not an Wa. Enable for better image quality */
	I915_WRITE(_3D_CHICKEN3,
		   _MASKED_BIT_ENABLE(_3D_CHICKEN3_AA_LINE_QUALITY_FIX_ENABLE));

	/* WaEnableChickenDCPR:cnl */
	I915_WRITE(GEN8_CHICKEN_DCPR_1,
		   I915_READ(GEN8_CHICKEN_DCPR_1) | MASK_WAKEMEM);

	/* WaFbcWakeMemOn:cnl */
	I915_WRITE(DISP_ARB_CTL, I915_READ(DISP_ARB_CTL) |
		   DISP_FBC_MEMORY_WAKE);

	val = I915_READ(SLICE_UNIT_LEVEL_CLKGATE);
	/* ReadHitWriteOnlyDisable:cnl */
	val |= RCCUNIT_CLKGATE_DIS;
	/* WaSarbUnitClockGatingDisable:cnl (pre-prod) */
	if (IS_CNL_REVID(dev_priv, CNL_REVID_A0, CNL_REVID_B0))
		val |= SARBUNIT_CLKGATE_DIS;
	I915_WRITE(SLICE_UNIT_LEVEL_CLKGATE, val);

	/* Wa_2201832410:cnl */
	val = I915_READ(SUBSLICE_UNIT_LEVEL_CLKGATE);
	val |= GWUNIT_CLKGATE_DIS;
	I915_WRITE(SUBSLICE_UNIT_LEVEL_CLKGATE, val);

	/* WaDisableVFclkgate:cnl */
	/* WaVFUnitClockGatingDisable:cnl */
	val = I915_READ(UNSLICE_UNIT_LEVEL_CLKGATE);
	val |= VFUNIT_CLKGATE_DIS;
	I915_WRITE(UNSLICE_UNIT_LEVEL_CLKGATE, val);
}

static void cfl_init_clock_gating(struct drm_i915_private *dev_priv)
{
	cnp_init_clock_gating(dev_priv);
	gen9_init_clock_gating(dev_priv);

	/* WaFbcNukeOnHostModify:cfl */
	I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) |
		   ILK_DPFC_NUKE_ON_ANY_MODIFICATION);
}

static void kbl_init_clock_gating(struct drm_i915_private *dev_priv)
{
	gen9_init_clock_gating(dev_priv);

	/* WaDisableSDEUnitClockGating:kbl */
	if (IS_KBL_REVID(dev_priv, 0, KBL_REVID_B0))
		I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
			   GEN8_SDEUNIT_CLOCK_GATE_DISABLE);

	/* WaDisableGamClockGating:kbl */
	if (IS_KBL_REVID(dev_priv, 0, KBL_REVID_B0))
		I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) |
			   GEN6_GAMUNIT_CLOCK_GATE_DISABLE);

	/* WaFbcNukeOnHostModify:kbl */
	I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) |
		   ILK_DPFC_NUKE_ON_ANY_MODIFICATION);
}

static void skl_init_clock_gating(struct drm_i915_private *dev_priv)
{
	gen9_init_clock_gating(dev_priv);

	/* WAC6entrylatency:skl */
	I915_WRITE(FBC_LLC_READ_CTRL, I915_READ(FBC_LLC_READ_CTRL) |
		   FBC_LLC_FULLY_OPEN);

	/* WaFbcNukeOnHostModify:skl */
	I915_WRITE(ILK_DPFC_CHICKEN, I915_READ(ILK_DPFC_CHICKEN) |
		   ILK_DPFC_NUKE_ON_ANY_MODIFICATION);
}

static void bdw_init_clock_gating(struct drm_i915_private *dev_priv)
{
	enum pipe pipe;

	/* WaSwitchSolVfFArbitrationPriority:bdw */
	I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);

	/* WaPsrDPAMaskVBlankInSRD:bdw */
	I915_WRITE(CHICKEN_PAR1_1,
		   I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD);

	/* WaPsrDPRSUnmaskVBlankInSRD:bdw */
	for_each_pipe(dev_priv, pipe) {
		I915_WRITE(CHICKEN_PIPESL_1(pipe),
			   I915_READ(CHICKEN_PIPESL_1(pipe)) |
			   BDW_DPRS_MASK_VBLANK_SRD);
	}

	/* WaVSRefCountFullforceMissDisable:bdw */
	/* WaDSRefCountFullforceMissDisable:bdw */
	I915_WRITE(GEN7_FF_THREAD_MODE,
		   I915_READ(GEN7_FF_THREAD_MODE) &
		   ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));

	I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
		   _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));

	/* WaDisableSDEUnitClockGating:bdw */
	I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
		   GEN8_SDEUNIT_CLOCK_GATE_DISABLE);

	/* WaProgramL3SqcReg1Default:bdw */
	gen8_set_l3sqc_credits(dev_priv, 30, 2);

	/* WaKVMNotificationOnConfigChange:bdw */
	I915_WRITE(CHICKEN_PAR2_1, I915_READ(CHICKEN_PAR2_1)
		   | KVM_CONFIG_CHANGE_NOTIFICATION_SELECT);

	lpt_init_clock_gating(dev_priv);

	/* WaDisableDopClockGating:bdw
	 *
	 * Also see the CHICKEN2 write in bdw_init_workarounds() to disable DOP
	 * clock gating.
	 */
	I915_WRITE(GEN6_UCGCTL1,
		   I915_READ(GEN6_UCGCTL1) | GEN6_EU_TCUNIT_CLOCK_GATE_DISABLE);
}

static void hsw_init_clock_gating(struct drm_i915_private *dev_priv)
{
	/* L3 caching of data atomics doesn't work -- disable it. */
	I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE);
	I915_WRITE(HSW_ROW_CHICKEN3,
		   _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE));

	/* This is required by WaCatErrorRejectionIssue:hsw */
	I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
			I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
			GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);

	/* WaVSRefCountFullforceMissDisable:hsw */
	I915_WRITE(GEN7_FF_THREAD_MODE,
		   I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME);

	/* WaDisable_RenderCache_OperationalFlush:hsw */
	I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));

	/* enable HiZ Raw Stall Optimization */
	I915_WRITE(CACHE_MODE_0_GEN7,
		   _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));

	/* WaDisable4x2SubspanOptimization:hsw */
	I915_WRITE(CACHE_MODE_1,
		   _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));

	/*
	 * BSpec recommends 8x4 when MSAA is used,
	 * however in practice 16x4 seems fastest.
	 *
	 * Note that PS/WM thread counts depend on the WIZ hashing
	 * disable bit, which we don't touch here, but it's good
	 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
	 */
	I915_WRITE(GEN7_GT_MODE,
		   _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));

	/* WaSampleCChickenBitEnable:hsw */
	I915_WRITE(HALF_SLICE_CHICKEN3,
		   _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE));

	/* WaSwitchSolVfFArbitrationPriority:hsw */
	I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);

	lpt_init_clock_gating(dev_priv);
}

static void ivb_init_clock_gating(struct drm_i915_private *dev_priv)
{
	u32 snpcr;

	I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);

	/* WaDisableEarlyCull:ivb */
	I915_WRITE(_3D_CHICKEN3,
		   _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));

	/* WaDisableBackToBackFlipFix:ivb */
	I915_WRITE(IVB_CHICKEN3,
		   CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
		   CHICKEN3_DGMG_DONE_FIX_DISABLE);

	/* WaDisablePSDDualDispatchEnable:ivb */
	if (IS_IVB_GT1(dev_priv))
		I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
			   _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));

	/* WaDisable_RenderCache_OperationalFlush:ivb */
	I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));

	/* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
	I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
		   GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);

	/* WaApplyL3ControlAndL3ChickenMode:ivb */
	I915_WRITE(GEN7_L3CNTLREG1,
			GEN7_WA_FOR_GEN7_L3_CONTROL);
	I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
		   GEN7_WA_L3_CHICKEN_MODE);
	if (IS_IVB_GT1(dev_priv))
		I915_WRITE(GEN7_ROW_CHICKEN2,
			   _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
	else {
		/* must write both registers */
		I915_WRITE(GEN7_ROW_CHICKEN2,
			   _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
		I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
			   _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
	}

	/* WaForceL3Serialization:ivb */
	I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
		   ~L3SQ_URB_READ_CAM_MATCH_DISABLE);

	/*
	 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
	 * This implements the WaDisableRCZUnitClockGating:ivb workaround.
	 */
	I915_WRITE(GEN6_UCGCTL2,
		   GEN6_RCZUNIT_CLOCK_GATE_DISABLE);

	/* This is required by WaCatErrorRejectionIssue:ivb */
	I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
			I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
			GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);

	g4x_disable_trickle_feed(dev_priv);

	gen7_setup_fixed_func_scheduler(dev_priv);

	if (0) { /* causes HiZ corruption on ivb:gt1 */
		/* enable HiZ Raw Stall Optimization */
		I915_WRITE(CACHE_MODE_0_GEN7,
			   _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
	}

	/* WaDisable4x2SubspanOptimization:ivb */
	I915_WRITE(CACHE_MODE_1,
		   _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));

	/*
	 * BSpec recommends 8x4 when MSAA is used,
	 * however in practice 16x4 seems fastest.
	 *
	 * Note that PS/WM thread counts depend on the WIZ hashing
	 * disable bit, which we don't touch here, but it's good
	 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
	 */
	I915_WRITE(GEN7_GT_MODE,
		   _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));

	snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
	snpcr &= ~GEN6_MBC_SNPCR_MASK;
	snpcr |= GEN6_MBC_SNPCR_MED;
	I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);

	if (!HAS_PCH_NOP(dev_priv))
		cpt_init_clock_gating(dev_priv);

	gen6_check_mch_setup(dev_priv);
}

static void vlv_init_clock_gating(struct drm_i915_private *dev_priv)
{
	/* WaDisableEarlyCull:vlv */
	I915_WRITE(_3D_CHICKEN3,
		   _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));

	/* WaDisableBackToBackFlipFix:vlv */
	I915_WRITE(IVB_CHICKEN3,
		   CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
		   CHICKEN3_DGMG_DONE_FIX_DISABLE);

	/* WaPsdDispatchEnable:vlv */
	/* WaDisablePSDDualDispatchEnable:vlv */
	I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
		   _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
				      GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));

	/* WaDisable_RenderCache_OperationalFlush:vlv */
	I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));

	/* WaForceL3Serialization:vlv */
	I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
		   ~L3SQ_URB_READ_CAM_MATCH_DISABLE);

	/* WaDisableDopClockGating:vlv */
	I915_WRITE(GEN7_ROW_CHICKEN2,
		   _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));

	/* This is required by WaCatErrorRejectionIssue:vlv */
	I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
		   I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
		   GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);

	gen7_setup_fixed_func_scheduler(dev_priv);

	/*
	 * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
	 * This implements the WaDisableRCZUnitClockGating:vlv workaround.
	 */
	I915_WRITE(GEN6_UCGCTL2,
		   GEN6_RCZUNIT_CLOCK_GATE_DISABLE);

	/* WaDisableL3Bank2xClockGate:vlv
	 * Disabling L3 clock gating- MMIO 940c[25] = 1
	 * Set bit 25, to disable L3_BANK_2x_CLK_GATING */
	I915_WRITE(GEN7_UCGCTL4,
		   I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE);

	/*
	 * BSpec says this must be set, even though
	 * WaDisable4x2SubspanOptimization isn't listed for VLV.
	 */
	I915_WRITE(CACHE_MODE_1,
		   _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));

	/*
	 * BSpec recommends 8x4 when MSAA is used,
	 * however in practice 16x4 seems fastest.
	 *
	 * Note that PS/WM thread counts depend on the WIZ hashing
	 * disable bit, which we don't touch here, but it's good
	 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
	 */
	I915_WRITE(GEN7_GT_MODE,
		   _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));

	/*
	 * WaIncreaseL3CreditsForVLVB0:vlv
	 * This is the hardware default actually.
	 */
	I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE);

	/*
	 * WaDisableVLVClockGating_VBIIssue:vlv
	 * Disable clock gating on th GCFG unit to prevent a delay
	 * in the reporting of vblank events.
	 */
	I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS);
}

static void chv_init_clock_gating(struct drm_i915_private *dev_priv)
{
	/* WaVSRefCountFullforceMissDisable:chv */
	/* WaDSRefCountFullforceMissDisable:chv */
	I915_WRITE(GEN7_FF_THREAD_MODE,
		   I915_READ(GEN7_FF_THREAD_MODE) &
		   ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));

	/* WaDisableSemaphoreAndSyncFlipWait:chv */
	I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
		   _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));

	/* WaDisableCSUnitClockGating:chv */
	I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) |
		   GEN6_CSUNIT_CLOCK_GATE_DISABLE);

	/* WaDisableSDEUnitClockGating:chv */
	I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
		   GEN8_SDEUNIT_CLOCK_GATE_DISABLE);

	/*
	 * WaProgramL3SqcReg1Default:chv
	 * See gfxspecs/Related Documents/Performance Guide/
	 * LSQC Setting Recommendations.
	 */
	gen8_set_l3sqc_credits(dev_priv, 38, 2);
}

static void g4x_init_clock_gating(struct drm_i915_private *dev_priv)
{
	u32 dspclk_gate;

	I915_WRITE(RENCLK_GATE_D1, 0);
	I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
		   GS_UNIT_CLOCK_GATE_DISABLE |
		   CL_UNIT_CLOCK_GATE_DISABLE);
	I915_WRITE(RAMCLK_GATE_D, 0);
	dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
		OVRUNIT_CLOCK_GATE_DISABLE |
		OVCUNIT_CLOCK_GATE_DISABLE;
	if (IS_GM45(dev_priv))
		dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
	I915_WRITE(DSPCLK_GATE_D, dspclk_gate);

	/* WaDisableRenderCachePipelinedFlush */
	I915_WRITE(CACHE_MODE_0,
		   _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));

	/* WaDisable_RenderCache_OperationalFlush:g4x */
	I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));

	g4x_disable_trickle_feed(dev_priv);
}

static void i965gm_init_clock_gating(struct drm_i915_private *dev_priv)
{
	struct intel_uncore *uncore = &dev_priv->uncore;

	intel_uncore_write(uncore, RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
	intel_uncore_write(uncore, RENCLK_GATE_D2, 0);
	intel_uncore_write(uncore, DSPCLK_GATE_D, 0);
	intel_uncore_write(uncore, RAMCLK_GATE_D, 0);
	intel_uncore_write16(uncore, DEUC, 0);
	intel_uncore_write(uncore,
			   MI_ARB_STATE,
			   _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));

	/* WaDisable_RenderCache_OperationalFlush:gen4 */
	intel_uncore_write(uncore,
			   CACHE_MODE_0,
			   _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
}

static void i965g_init_clock_gating(struct drm_i915_private *dev_priv)
{
	I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
		   I965_RCC_CLOCK_GATE_DISABLE |
		   I965_RCPB_CLOCK_GATE_DISABLE |
		   I965_ISC_CLOCK_GATE_DISABLE |
		   I965_FBC_CLOCK_GATE_DISABLE);
	I915_WRITE(RENCLK_GATE_D2, 0);
	I915_WRITE(MI_ARB_STATE,
		   _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));

	/* WaDisable_RenderCache_OperationalFlush:gen4 */
	I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
}

static void gen3_init_clock_gating(struct drm_i915_private *dev_priv)
{
	u32 dstate = I915_READ(D_STATE);

	dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
		DSTATE_DOT_CLOCK_GATING;
	I915_WRITE(D_STATE, dstate);

	if (IS_PINEVIEW(dev_priv))
		I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));

	/* IIR "flip pending" means done if this bit is set */
	I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));

	/* interrupts should cause a wake up from C3 */
	I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN));

	/* On GEN3 we really need to make sure the ARB C3 LP bit is set */
	I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE));

	I915_WRITE(MI_ARB_STATE,
		   _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
}

static void i85x_init_clock_gating(struct drm_i915_private *dev_priv)
{
	I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);

	/* interrupts should cause a wake up from C3 */
	I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) |
		   _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE));

	I915_WRITE(MEM_MODE,
		   _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE));
}

static void i830_init_clock_gating(struct drm_i915_private *dev_priv)
{
	I915_WRITE(MEM_MODE,
		   _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) |
		   _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE));
}

void intel_init_clock_gating(struct drm_i915_private *dev_priv)
{
	dev_priv->display.init_clock_gating(dev_priv);
}

void intel_suspend_hw(struct drm_i915_private *dev_priv)
{
	if (HAS_PCH_LPT(dev_priv))
		lpt_suspend_hw(dev_priv);
}

static void nop_init_clock_gating(struct drm_i915_private *dev_priv)
{
	DRM_DEBUG_KMS("No clock gating settings or workarounds applied.\n");
}

/**
 * intel_init_clock_gating_hooks - setup the clock gating hooks
 * @dev_priv: device private
 *
 * Setup the hooks that configure which clocks of a given platform can be
 * gated and also apply various GT and display specific workarounds for these
 * platforms. Note that some GT specific workarounds are applied separately
 * when GPU contexts or batchbuffers start their execution.
 */
void intel_init_clock_gating_hooks(struct drm_i915_private *dev_priv)
{
	if (IS_GEN(dev_priv, 12))
		dev_priv->display.init_clock_gating = tgl_init_clock_gating;
	else if (IS_GEN(dev_priv, 11))
		dev_priv->display.init_clock_gating = icl_init_clock_gating;
	else if (IS_CANNONLAKE(dev_priv))
		dev_priv->display.init_clock_gating = cnl_init_clock_gating;
	else if (IS_COFFEELAKE(dev_priv))
		dev_priv->display.init_clock_gating = cfl_init_clock_gating;
	else if (IS_SKYLAKE(dev_priv))
		dev_priv->display.init_clock_gating = skl_init_clock_gating;
	else if (IS_KABYLAKE(dev_priv))
		dev_priv->display.init_clock_gating = kbl_init_clock_gating;
	else if (IS_BROXTON(dev_priv))
		dev_priv->display.init_clock_gating = bxt_init_clock_gating;
	else if (IS_GEMINILAKE(dev_priv))
		dev_priv->display.init_clock_gating = glk_init_clock_gating;
	else if (IS_BROADWELL(dev_priv))
		dev_priv->display.init_clock_gating = bdw_init_clock_gating;
	else if (IS_CHERRYVIEW(dev_priv))
		dev_priv->display.init_clock_gating = chv_init_clock_gating;
	else if (IS_HASWELL(dev_priv))
		dev_priv->display.init_clock_gating = hsw_init_clock_gating;
	else if (IS_IVYBRIDGE(dev_priv))
		dev_priv->display.init_clock_gating = ivb_init_clock_gating;
	else if (IS_VALLEYVIEW(dev_priv))
		dev_priv->display.init_clock_gating = vlv_init_clock_gating;
	else if (IS_GEN(dev_priv, 6))
		dev_priv->display.init_clock_gating = gen6_init_clock_gating;
	else if (IS_GEN(dev_priv, 5))
		dev_priv->display.init_clock_gating = ilk_init_clock_gating;
	else if (IS_G4X(dev_priv))
		dev_priv->display.init_clock_gating = g4x_init_clock_gating;
	else if (IS_I965GM(dev_priv))
		dev_priv->display.init_clock_gating = i965gm_init_clock_gating;
	else if (IS_I965G(dev_priv))
		dev_priv->display.init_clock_gating = i965g_init_clock_gating;
	else if (IS_GEN(dev_priv, 3))
		dev_priv->display.init_clock_gating = gen3_init_clock_gating;
	else if (IS_I85X(dev_priv) || IS_I865G(dev_priv))
		dev_priv->display.init_clock_gating = i85x_init_clock_gating;
	else if (IS_GEN(dev_priv, 2))
		dev_priv->display.init_clock_gating = i830_init_clock_gating;
	else {
		MISSING_CASE(INTEL_DEVID(dev_priv));
		dev_priv->display.init_clock_gating = nop_init_clock_gating;
	}
}

/* Set up chip specific power management-related functions */
void intel_init_pm(struct drm_i915_private *dev_priv)
{
	/* For cxsr */
	if (IS_PINEVIEW(dev_priv))
		i915_pineview_get_mem_freq(dev_priv);
	else if (IS_GEN(dev_priv, 5))
		i915_ironlake_get_mem_freq(dev_priv);

	if (intel_has_sagv(dev_priv))
		skl_setup_sagv_block_time(dev_priv);

	/* For FIFO watermark updates */
	if (INTEL_GEN(dev_priv) >= 9) {
		skl_setup_wm_latency(dev_priv);
		dev_priv->display.initial_watermarks = skl_initial_wm;
		dev_priv->display.atomic_update_watermarks = skl_atomic_update_crtc_wm;
		dev_priv->display.compute_global_watermarks = skl_compute_wm;
	} else if (HAS_PCH_SPLIT(dev_priv)) {
		ilk_setup_wm_latency(dev_priv);

		if ((IS_GEN(dev_priv, 5) && dev_priv->wm.pri_latency[1] &&
		     dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) ||
		    (!IS_GEN(dev_priv, 5) && dev_priv->wm.pri_latency[0] &&
		     dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) {
			dev_priv->display.compute_pipe_wm = ilk_compute_pipe_wm;
			dev_priv->display.compute_intermediate_wm =
				ilk_compute_intermediate_wm;
			dev_priv->display.initial_watermarks =
				ilk_initial_watermarks;
			dev_priv->display.optimize_watermarks =
				ilk_optimize_watermarks;
		} else {
			DRM_DEBUG_KMS("Failed to read display plane latency. "
				      "Disable CxSR\n");
		}
	} else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
		vlv_setup_wm_latency(dev_priv);
		dev_priv->display.compute_pipe_wm = vlv_compute_pipe_wm;
		dev_priv->display.compute_intermediate_wm = vlv_compute_intermediate_wm;
		dev_priv->display.initial_watermarks = vlv_initial_watermarks;
		dev_priv->display.optimize_watermarks = vlv_optimize_watermarks;
		dev_priv->display.atomic_update_watermarks = vlv_atomic_update_fifo;
	} else if (IS_G4X(dev_priv)) {
		g4x_setup_wm_latency(dev_priv);
		dev_priv->display.compute_pipe_wm = g4x_compute_pipe_wm;
		dev_priv->display.compute_intermediate_wm = g4x_compute_intermediate_wm;
		dev_priv->display.initial_watermarks = g4x_initial_watermarks;
		dev_priv->display.optimize_watermarks = g4x_optimize_watermarks;
	} else if (IS_PINEVIEW(dev_priv)) {
		if (!intel_get_cxsr_latency(!IS_MOBILE(dev_priv),
					    dev_priv->is_ddr3,
					    dev_priv->fsb_freq,
					    dev_priv->mem_freq)) {
			DRM_INFO("failed to find known CxSR latency "
				 "(found ddr%s fsb freq %d, mem freq %d), "
				 "disabling CxSR\n",
				 (dev_priv->is_ddr3 == 1) ? "3" : "2",
				 dev_priv->fsb_freq, dev_priv->mem_freq);
			/* Disable CxSR and never update its watermark again */
			intel_set_memory_cxsr(dev_priv, false);
			dev_priv->display.update_wm = NULL;
		} else
			dev_priv->display.update_wm = pineview_update_wm;
	} else if (IS_GEN(dev_priv, 4)) {
		dev_priv->display.update_wm = i965_update_wm;
	} else if (IS_GEN(dev_priv, 3)) {
		dev_priv->display.update_wm = i9xx_update_wm;
		dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
	} else if (IS_GEN(dev_priv, 2)) {
		if (INTEL_NUM_PIPES(dev_priv) == 1) {
			dev_priv->display.update_wm = i845_update_wm;
			dev_priv->display.get_fifo_size = i845_get_fifo_size;
		} else {
			dev_priv->display.update_wm = i9xx_update_wm;
			dev_priv->display.get_fifo_size = i830_get_fifo_size;
		}
	} else {
		DRM_ERROR("unexpected fall-through in intel_init_pm\n");
	}
}

void intel_pm_setup(struct drm_i915_private *dev_priv)
{
	dev_priv->runtime_pm.suspended = false;
	atomic_set(&dev_priv->runtime_pm.wakeref_count, 0);
}