drm/nv50/pm: rewrite clock management, and switch to the new pm hooks

This area is horrifically complicated on these chipsets, and it's likely we
will need at least a few more tweaks yet.

Oh yes, and it's completely disabled on IGPs for the moment.  From traces,
things look potentially different there yet again.  Sigh...
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/nouveau_drv.h

@@ -493,6 +493,7 @@ struct nouveau_pm_level {
 	u32 copy;
 	u32 daemon;
 	u32 vdec;
+	u32 dom6;
 	u32 unka0;	/* nva3:nvc0 */
 	u32 hub01;	/* nvc0- */
 	u32 hub06;	/* nvc0- */

drivers/gpu/drm/nouveau/nouveau_perf.c

@@ -302,6 +302,7 @@ nouveau_perf_init(struct drm_device *dev)
 			perflvl->shader = ROM16(entry[10]) * 1000;
 			perflvl->memory = ROM16(entry[12]) * 1000;
 			perflvl->vdec = ROM16(entry[16]) * 1000;
+			perflvl->dom6 = ROM16(entry[20]) * 1000;
 			break;
 		case 0x40:
 #define subent(n) (ROM16(entry[perf[2] + ((n) * perf[3])]) & 0xfff) * 1000

drivers/gpu/drm/nouveau/nouveau_pm.h

@@ -60,10 +60,9 @@ int nv40_pm_pwm_get(struct drm_device *, struct dcb_gpio_entry *, u32*, u32*);
 int nv40_pm_pwm_set(struct drm_device *, struct dcb_gpio_entry *, u32, u32);
 
 /* nv50_pm.c */
-int nv50_pm_clock_get(struct drm_device *, u32 id);
-void *nv50_pm_clock_pre(struct drm_device *, struct nouveau_pm_level *,
-			u32 id, int khz);
-void nv50_pm_clock_set(struct drm_device *, void *);
+int nv50_pm_clocks_get(struct drm_device *, struct nouveau_pm_level *);
+void *nv50_pm_clocks_pre(struct drm_device *, struct nouveau_pm_level *);
+int nv50_pm_clocks_set(struct drm_device *, void *);
 int nv50_pm_pwm_get(struct drm_device *, struct dcb_gpio_entry *, u32*, u32*);
 int nv50_pm_pwm_set(struct drm_device *, struct dcb_gpio_entry *, u32, u32);
 

drivers/gpu/drm/nouveau/nouveau_state.c

@@ -356,9 +356,9 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		case 0xaa:
 		case 0xac:
 		case 0x50:
-			engine->pm.clock_get	= nv50_pm_clock_get;
-			engine->pm.clock_pre	= nv50_pm_clock_pre;
-			engine->pm.clock_set	= nv50_pm_clock_set;
+			engine->pm.clocks_get	= nv50_pm_clocks_get;
+			engine->pm.clocks_pre	= nv50_pm_clocks_pre;
+			engine->pm.clocks_set	= nv50_pm_clocks_set;
 			break;
 		default:
 			engine->pm.clocks_get	= nva3_pm_clocks_get;

drivers/gpu/drm/nouveau/nv50_pm.c

@@ -25,123 +25,601 @@
 #include "drmP.h"
 #include "nouveau_drv.h"
 #include "nouveau_bios.h"
+#include "nouveau_hw.h"
 #include "nouveau_pm.h"
 
-struct nv50_pm_state {
-	struct nouveau_pm_level *perflvl;
-	struct pll_lims pll;
-	enum pll_types type;
-	int N, M, P;
+enum clk_src {
+	clk_src_crystal,
+	clk_src_href,
+	clk_src_hclk,
+	clk_src_hclkm3,
+	clk_src_hclkm3d2,
+	clk_src_host,
+	clk_src_nvclk,
+	clk_src_sclk,
+	clk_src_mclk,
+	clk_src_vdec,
+	clk_src_dom6
 };
 
+static u32 read_clk(struct drm_device *, enum clk_src);
+
+static u32
+read_div(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	switch (dev_priv->chipset) {
+	case 0x50: /* it exists, but only has bit 31, not the dividers.. */
+	case 0x84:
+	case 0x86:
+	case 0x98:
+	case 0xa0:
+		return nv_rd32(dev, 0x004700);
+	case 0x92:
+	case 0x94:
+	case 0x96:
+		return nv_rd32(dev, 0x004800);
+	default:
+		return 0x00000000;
+	}
+}
+
+static u32
+read_pll_ref(struct drm_device *dev, u32 base)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 coef, ref = read_clk(dev, clk_src_crystal);
+	u32 rsel = nv_rd32(dev, 0x00e18c);
+	int P, N, M, id;
+
+	switch (dev_priv->chipset) {
+	case 0x50:
+	case 0xa0:
+		switch (base) {
+		case 0x4020:
+		case 0x4028: id = !!(rsel & 0x00000004); break;
+		case 0x4008: id = !!(rsel & 0x00000008); break;
+		case 0x4030: id = 0; break;
+		default:
+			NV_ERROR(dev, "ref: bad pll 0x%06x\n", base);
+			return 0;
+		}
+
+		coef = nv_rd32(dev, 0x00e81c + (id * 0x0c));
+		ref *=  (coef & 0x01000000) ? 2 : 4;
+		P    =  (coef & 0x00070000) >> 16;
+		N    = ((coef & 0x0000ff00) >> 8) + 1;
+		M    = ((coef & 0x000000ff) >> 0) + 1;
+		break;
+	case 0x84:
+	case 0x86:
+	case 0x92:
+		coef = nv_rd32(dev, 0x00e81c);
+		P    = (coef & 0x00070000) >> 16;
+		N    = (coef & 0x0000ff00) >> 8;
+		M    = (coef & 0x000000ff) >> 0;
+		break;
+	case 0x94:
+	case 0x96:
+	case 0x98:
+		rsel = nv_rd32(dev, 0x00c050);
+		switch (base) {
+		case 0x4020: rsel = (rsel & 0x00000003) >> 0; break;
+		case 0x4008: rsel = (rsel & 0x0000000c) >> 2; break;
+		case 0x4028: rsel = (rsel & 0x00001800) >> 11; break;
+		case 0x4030: rsel = 3; break;
+		default:
+			NV_ERROR(dev, "ref: bad pll 0x%06x\n", base);
+			return 0;
+		}
+
+		switch (rsel) {
+		case 0: id = 1; break;
+		case 1: return read_clk(dev, clk_src_crystal);
+		case 2: return read_clk(dev, clk_src_href);
+		case 3: id = 0; break;
+		}
+
+		coef =  nv_rd32(dev, 0x00e81c + (id * 0x28));
+		P    = (nv_rd32(dev, 0x00e824 + (id * 0x28)) >> 16) & 7;
+		P   += (coef & 0x00070000) >> 16;
+		N    = (coef & 0x0000ff00) >> 8;
+		M    = (coef & 0x000000ff) >> 0;
+		break;
+	default:
+		BUG_ON(1);
+	}
+
+	if (M)
+		return (ref * N / M) >> P;
+	return 0;
+}
+
+static u32
+read_pll(struct drm_device *dev, u32 base)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 mast = nv_rd32(dev, 0x00c040);
+	u32 src = 0, ref = 0, clk = 0;
+	u32 ctrl, coef;
+	int N1, N2, M1, M2;
+
+	switch (base) {
+	case 0x004028:
+		if (mast & 0x00100000) {
+			/* wtf, appears to only disable post-divider on nva0 */
+			if (dev_priv->chipset != 0xa0)
+				return read_clk(dev, clk_src_dom6);
+		}
+		src = !!(mast & 0x00200000);
+		break;
+	case 0x004020:
+		src = !!(mast & 0x00400000);
+		break;
+	case 0x004008:
+		src = !!(mast & 0x00010000);
+		break;
+	case 0x004030:
+		src = !!(mast & 0x02000000);
+		break;
+	case 0x00e810:
+		ref = read_clk(dev, clk_src_crystal);
+		break;
+	default:
+		NV_ERROR(dev, "bad pll 0x%06x\n", base);
+		return 0;
+	}
+
+	if (ref == 0) {
+		if (src)
+			ref = read_clk(dev, clk_src_href);
+		else
+			ref = read_pll_ref(dev, base);
+	}
+
+	ctrl = nv_rd32(dev, base + 0);
+	coef = nv_rd32(dev, base + 4);
+
+	N2 = (coef & 0xff000000) >> 24;
+	M2 = (coef & 0x00ff0000) >> 16;
+	N1 = (coef & 0x0000ff00) >> 8;
+	M1 = (coef & 0x000000ff);
+	if ((ctrl & 0x80000000) && M1) {
+		clk = ref * N1 / M1;
+		if ((ctrl & 0x40000100) == 0x40000000) {
+			if (M2)
+				clk = clk * N2 / M2;
+			else
+				clk = 0;
+		}
+	}
+
+	return clk;
+}
+
+static u32
+read_clk(struct drm_device *dev, enum clk_src src)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	u32 mast = nv_rd32(dev, 0x00c040);
+	u32 P = 0;
+
+	switch (src) {
+	case clk_src_crystal:
+		return dev_priv->crystal;
+	case clk_src_href:
+		return 100000; /* PCIE reference clock */
+	case clk_src_hclk:
+		return read_clk(dev, clk_src_href) * 27778 / 10000;
+	case clk_src_hclkm3:
+		return read_clk(dev, clk_src_hclk) * 3;
+	case clk_src_hclkm3d2:
+		return read_clk(dev, clk_src_hclk) * 3 / 2;
+	case clk_src_host:
+		switch (mast & 0x30000000) {
+		case 0x00000000: return read_clk(dev, clk_src_href);
+		case 0x10000000: break;
+		case 0x20000000: /* !0x50 */
+		case 0x30000000: return read_clk(dev, clk_src_hclk);
+		}
+		break;
+	case clk_src_nvclk:
+		if (!(mast & 0x00100000))
+			P = (nv_rd32(dev, 0x004028) & 0x00070000) >> 16;
+		switch (mast & 0x00000003) {
+		case 0x00000000: return read_clk(dev, clk_src_crystal) >> P;
+		case 0x00000001: return read_clk(dev, clk_src_dom6);
+		case 0x00000002: return read_pll(dev, 0x004020) >> P;
+		case 0x00000003: return read_pll(dev, 0x004028) >> P;
+		}
+		break;
+	case clk_src_sclk:
+		P = (nv_rd32(dev, 0x004020) & 0x00070000) >> 16;
+		switch (mast & 0x00000030) {
+		case 0x00000000:
+			if (mast & 0x00000080)
+				return read_clk(dev, clk_src_host) >> P;
+			return read_clk(dev, clk_src_crystal) >> P;
+		case 0x00000010: break;
+		case 0x00000020: return read_pll(dev, 0x004028) >> P;
+		case 0x00000030: return read_pll(dev, 0x004020) >> P;
+		}
+		break;
+	case clk_src_mclk:
+		P = (nv_rd32(dev, 0x004008) & 0x00070000) >> 16;
+		if (nv_rd32(dev, 0x004008) & 0x00000200) {
+			switch (mast & 0x0000c000) {
+			case 0x00000000:
+				return read_clk(dev, clk_src_crystal) >> P;
+			case 0x00008000:
+			case 0x0000c000:
+				return read_clk(dev, clk_src_href) >> P;
+			}
+		} else {
+			return read_pll(dev, 0x004008) >> P;
+		}
+		break;
+	case clk_src_vdec:
+		P = (read_div(dev) & 0x00000700) >> 8;
+		switch (dev_priv->chipset) {
+		case 0x84:
+		case 0x86:
+		case 0x92:
+		case 0x94:
+		case 0x96:
+		case 0xa0:
+			switch (mast & 0x00000c00) {
+			case 0x00000000:
+				if (dev_priv->chipset == 0xa0) /* wtf?? */
+					return read_clk(dev, clk_src_nvclk) >> P;
+				return read_clk(dev, clk_src_crystal) >> P;
+			case 0x00000400:
+				return 0;
+			case 0x00000800:
+				if (mast & 0x01000000)
+					return read_pll(dev, 0x004028) >> P;
+				return read_pll(dev, 0x004030) >> P;
+			case 0x00000c00:
+				return read_clk(dev, clk_src_nvclk) >> P;
+			}
+			break;
+		case 0x98:
+			switch (mast & 0x00000c00) {
+			case 0x00000000:
+				return read_clk(dev, clk_src_nvclk) >> P;
+			case 0x00000400:
+				return 0;
+			case 0x00000800:
+				return read_clk(dev, clk_src_hclkm3d2) >> P;
+			case 0x00000c00:
+				return read_pll(dev, clk_src_mclk) >> P;
+			}
+			break;
+		}
+		break;
+	case clk_src_dom6:
+		switch (dev_priv->chipset) {
+		case 0x50:
+		case 0xa0:
+			return read_pll(dev, 0x00e810) >> 2;
+		case 0x84:
+		case 0x86:
+		case 0x92:
+		case 0x94:
+		case 0x96:
+		case 0x98:
+			P = (read_div(dev) & 0x00000007) >> 0;
+			switch (mast & 0x0c000000) {
+			case 0x00000000: return read_clk(dev, clk_src_href);
+			case 0x04000000: break;
+			case 0x08000000: return read_clk(dev, clk_src_hclk);
+			case 0x0c000000:
+				return read_clk(dev, clk_src_hclkm3) >> P;
+			}
+			break;
+		default:
+			break;
+		}
+	default:
+		break;
+	}
+
+	NV_DEBUG(dev, "unknown clock source %d 0x%08x\n", src, mast);
+	return 0;
+}
+
 int
-nv50_pm_clock_get(struct drm_device *dev, u32 id)
+nv50_pm_clocks_get(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 {
-	struct pll_lims pll;
-	int P, N, M, ret;
-	u32 reg0, reg1;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	if (dev_priv->chipset == 0xaa ||
+	    dev_priv->chipset == 0xac)
+		return 0;
 
-	ret = get_pll_limits(dev, id, &pll);
+	perflvl->core   = read_clk(dev, clk_src_nvclk);
+	perflvl->shader = read_clk(dev, clk_src_sclk);
+	perflvl->memory = read_clk(dev, clk_src_mclk);
+	if (dev_priv->chipset != 0x50) {
+		perflvl->vdec = read_clk(dev, clk_src_vdec);
+		perflvl->dom6 = read_clk(dev, clk_src_dom6);
+	}
+
+	return 0;
+}
+
+struct nv50_pm_state {
+	u32 emast;
+	u32 nctrl;
+	u32 ncoef;
+	u32 sctrl;
+	u32 scoef;
+
+	u32 amast;
+	u32 pdivs;
+
+	u32 mscript;
+	u32 mctrl;
+	u32 mcoef;
+};
+
+static u32
+calc_pll(struct drm_device *dev, u32 reg, struct pll_lims *pll,
+	 u32 clk, int *N1, int *M1, int *log2P)
+{
+	struct nouveau_pll_vals coef;
+	int ret;
+
+	ret = get_pll_limits(dev, reg, pll);
 	if (ret)
-		return ret;
+		return 0;
+
+	pll->vco2.maxfreq = 0;
+	pll->refclk = read_pll_ref(dev, reg);
+	if (!pll->refclk)
+		return 0;
 
-	reg0 = nv_rd32(dev, pll.reg + 0);
-	reg1 = nv_rd32(dev, pll.reg + 4);
+	ret = nouveau_calc_pll_mnp(dev, pll, clk, &coef);
+	if (ret == 0)
+		return 0;
 
-	if ((reg0 & 0x80000000) == 0) {
-		if (id == PLL_SHADER) {
-			NV_DEBUG(dev, "Shader PLL is disabled. "
-				"Shader clock is twice the core\n");
-			ret = nv50_pm_clock_get(dev, PLL_CORE);
-			if (ret > 0)
-				return ret << 1;
-		} else if (id == PLL_MEMORY) {
-			NV_DEBUG(dev, "Memory PLL is disabled. "
-				"Memory clock is equal to the ref_clk\n");
-			return pll.refclk;
+	*N1 = coef.N1;
+	*M1 = coef.M1;
+	*log2P = coef.log2P;
+	return ret;
+}
+
+static inline u32
+calc_div(u32 src, u32 target, int *div)
+{
+	u32 clk0 = src, clk1 = src;
+	for (*div = 0; *div <= 7; (*div)++) {
+		if (clk0 <= target) {
+			clk1 = clk0 << (*div ? 1 : 0);
+			break;
 		}
+		clk0 >>= 1;
 	}
 
-	P = (reg0 & 0x00070000) >> 16;
-	N = (reg1 & 0x0000ff00) >> 8;
-	M = (reg1 & 0x000000ff);
+	if (target - clk0 <= clk1 - target)
+		return clk0;
+	(*div)--;
+	return clk1;
+}
 
-	return ((pll.refclk * N / M) >> P);
+static inline u32
+clk_same(u32 a, u32 b)
+{
+	return ((a / 1000) == (b / 1000));
 }
 
 void *
-nv50_pm_clock_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl,
-		  u32 id, int khz)
+nv50_pm_clocks_pre(struct drm_device *dev, struct nouveau_pm_level *perflvl)
 {
-	struct nv50_pm_state *state;
-	int dummy, ret;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv50_pm_state *info;
+	struct pll_lims pll;
+	int ret = -EINVAL;
+	int N, M, P1, P2;
+	u32 clk, out;
+
+	if (dev_priv->chipset == 0xaa ||
+	    dev_priv->chipset == 0xac)
+		return ERR_PTR(-ENODEV);
 
-	state = kzalloc(sizeof(*state), GFP_KERNEL);
-	if (!state)
+	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	if (!info)
 		return ERR_PTR(-ENOMEM);
-	state->type = id;
-	state->perflvl = perflvl;
 
-	ret = get_pll_limits(dev, id, &state->pll);
-	if (ret < 0) {
-		kfree(state);
-		return (ret == -ENOENT) ? NULL : ERR_PTR(ret);
+	/* core: for the moment at least, always use nvpll */
+	clk = calc_pll(dev, 0x4028, &pll, perflvl->core, &N, &M, &P1);
+	if (clk == 0)
+		goto error;
+
+	info->emast = 0x00000003;
+	info->nctrl = 0x80000000 | (P1 << 19) | (P1 << 16);
+	info->ncoef = (N << 8) | M;
+
+	/* shader: tie to nvclk if possible, otherwise use spll.  have to be
+	 * very careful that the shader clock is at least twice the core, or
+	 * some chipsets will be very unhappy.  i expect most or all of these
+	 * cases will be handled by tying to nvclk, but it's possible there's
+	 * corners
+	 */
+	if (P1-- && perflvl->shader == (perflvl->core << 1)) {
+		info->emast |= 0x00000020;
+		info->sctrl  = 0x00000000 | (P1 << 19) | (P1 << 16);
+		info->scoef  = nv_rd32(dev, 0x004024);
+	} else {
+		clk = calc_pll(dev, 0x4020, &pll, perflvl->shader, &N, &M, &P1);
+		if (clk == 0)
+			goto error;
+
+		info->emast |= 0x00000030;
+		info->sctrl  = 0x80000000 | (P1 << 19) | (P1 << 16);
+		info->scoef  = (N << 8) | M;
 	}
 
-	ret = nv50_calc_pll(dev, &state->pll, khz, &state->N, &state->M,
-			    &dummy, &dummy, &state->P);
-	if (ret < 0) {
-		kfree(state);
-		return ERR_PTR(ret);
+	/* memory: use pcie refclock if possible, otherwise use mpll */
+	info->mscript = perflvl->memscript;
+	if (clk_same(perflvl->memory, read_clk(dev, clk_src_href))) {
+		info->mctrl = nv_rd32(dev, 0x4008) | 0x00000200;
+		info->mcoef = nv_rd32(dev, 0x400c);
+	} else
+	if (perflvl->memory) {
+		clk = calc_pll(dev, 0x4008, &pll, perflvl->memory,
+			       &N, &M, &P1);
+		if (clk == 0)
+			goto error;
+
+		info->mctrl  = 0x80000000 | (P1 << 22) | (P1 << 16);
+		info->mctrl |= pll.log2p_bias << 19;
+		info->mcoef  = (N << 8) | M;
+	} else {
+		info->mctrl = 0x00000000;
 	}
 
-	return state;
+	/* vdec: avoid modifying xpll until we know exactly how the other
+	 * clock domains work, i suspect at least some of them can also be
+	 * tied to xpll...
+	 */
+	info->amast = info->pdivs = 0;
+	if (perflvl->vdec) {
+		/* see how close we can get using nvclk as a source */
+		clk = calc_div(perflvl->core, perflvl->vdec, &P1);
+
+		/* see how close we can get using xpll/hclk as a source */
+		if (dev_priv->chipset != 0x98)
+			out = read_pll(dev, 0x004030);
+		else
+			out = read_clk(dev, clk_src_hclkm3d2);
+		out = calc_div(out, perflvl->vdec, &P2);
+
+		/* select whichever gets us closest */
+		if (abs((int)perflvl->vdec - clk) <=
+		    abs((int)perflvl->vdec - out)) {
+			if (dev_priv->chipset != 0x98)
+				info->amast |= 0x00000c00;
+			else
+				info->amast |= 0x00000000;
+			info->pdivs |= P1 << 8;
+		} else {
+			info->amast |= 0x00000800;
+			info->pdivs |= P2 << 8;
+		}
+	}
+
+	/* dom6: nfi what this is, but we're limited to various combinations
+	 * of the host clock frequency
+	 */
+	if (clk_same(perflvl->dom6, read_clk(dev, clk_src_href))) {
+		info->amast |= 0x00000000;
+		info->pdivs |= read_div(dev) & 0x00000007;
+	} else
+	if (clk_same(perflvl->dom6, read_clk(dev, clk_src_hclk))) {
+		info->amast |= 0x08000000;
+		info->pdivs |= read_div(dev) & 0x00000007;
+	} else
+	if (perflvl->dom6) {
+		clk = read_clk(dev, clk_src_hclk) * 3;
+		clk = calc_div(clk, perflvl->dom6, &P1);
+
+		info->amast |= 0x0c000000;
+		info->pdivs |= P1;
+	}
+
+
+	return info;
+error:
+	kfree(info);
+	return ERR_PTR(ret);
 }
 
-void
-nv50_pm_clock_set(struct drm_device *dev, void *pre_state)
+int
+nv50_pm_clocks_set(struct drm_device *dev, void *data)
 {
-	struct nv50_pm_state *state = pre_state;
-	struct nouveau_pm_level *perflvl = state->perflvl;
-	u32 reg = state->pll.reg, tmp;
-	struct bit_entry BIT_M;
-	u16 script;
-	int N = state->N;
-	int M = state->M;
-	int P = state->P;
-
-	if (state->type == PLL_MEMORY && perflvl->memscript &&
-	    bit_table(dev, 'M', &BIT_M) == 0 &&
-	    BIT_M.version == 1 && BIT_M.length >= 0x0b) {
-		script = ROM16(BIT_M.data[0x05]);
-		if (script)
-			nouveau_bios_run_init_table(dev, script, NULL, -1);
-		script = ROM16(BIT_M.data[0x07]);
-		if (script)
-			nouveau_bios_run_init_table(dev, script, NULL, -1);
-		script = ROM16(BIT_M.data[0x09]);
-		if (script)
-			nouveau_bios_run_init_table(dev, script, NULL, -1);
-
-		nouveau_bios_run_init_table(dev, perflvl->memscript, NULL, -1);
-	}
-
-	if (state->type == PLL_MEMORY) {
-		nv_wr32(dev, 0x100210, 0);
-		nv_wr32(dev, 0x1002dc, 1);
-	}
-
-	tmp  = nv_rd32(dev, reg + 0) & 0xfff8ffff;
-	tmp |= 0x80000000 | (P << 16);
-	nv_wr32(dev, reg + 0, tmp);
-	nv_wr32(dev, reg + 4, (N << 8) | M);
-
-	if (state->type == PLL_MEMORY) {
-		nv_wr32(dev, 0x1002dc, 0);
-		nv_wr32(dev, 0x100210, 0x80000000);
-	}
-
-	kfree(state);
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv50_pm_state *info = data;
+	struct bit_entry M;
+	int ret = 0;
+
+	/* halt and idle execution engines */
+	nv_mask(dev, 0x002504, 0x00000001, 0x00000001);
+	if (!nv_wait(dev, 0x002504, 0x00000010, 0x00000010))
+		goto error;
+
+	/* reclock vdec/dom6 */
+	nv_mask(dev, 0x00c040, 0x00000c00, 0x00000000);
+	switch (dev_priv->chipset) {
+	case 0x92:
+	case 0x94:
+	case 0x96:
+		nv_mask(dev, 0x004800, 0x00000707, info->pdivs);
+		break;
+	default:
+		nv_mask(dev, 0x004700, 0x00000707, info->pdivs);
+		break;
+	}
+	nv_mask(dev, 0x00c040, 0x0c000c00, info->amast);
+
+	/* core/shader: switch core to dom6, shader to hclk */
+	if (dev_priv->chipset == 0x50)
+		nv_mask(dev, 0x00c040, 0x001000b0, 0x00100080); /* grrr! */
+	else
+		nv_mask(dev, 0x00c040, 0x000000b3, 0x00000081);
+	nv_mask(dev, 0x004020, 0xc03f0100, info->sctrl);
+	nv_wr32(dev, 0x004024, info->scoef);
+	nv_mask(dev, 0x004028, 0xc03f0100, info->nctrl);
+	nv_wr32(dev, 0x00402c, info->ncoef);
+	nv_mask(dev, 0x00c040, 0x00100033, info->emast);
+
+	/* memory */
+	if (!info->mctrl)
+		goto resume;
+
+	/* execute some scripts that do ??? from the vbios.. */
+	if (!bit_table(dev, 'M', &M) && M.version == 1) {
+		if (M.length >= 6)
+			nouveau_bios_init_exec(dev, ROM16(M.data[5]));
+		if (M.length >= 8)
+			nouveau_bios_init_exec(dev, ROM16(M.data[7]));
+		if (M.length >= 10)
+			nouveau_bios_init_exec(dev, ROM16(M.data[9]));
+		nouveau_bios_init_exec(dev, info->mscript);
+	}
+
+	/* disable display */
+	nv_wr32(dev, 0x611200, 0x00003300);
+	udelay(100);
+
+	/* prepare ram for reclocking */
+	nv_wr32(dev, 0x1002d4, 0x00000001); /* precharge */
+	nv_wr32(dev, 0x1002d0, 0x00000001); /* refresh */
+	nv_wr32(dev, 0x1002d0, 0x00000001); /* refresh */
+	nv_mask(dev, 0x100210, 0x80000000, 0x00000000); /* no auto-refresh */
+	nv_wr32(dev, 0x1002dc, 0x00000001); /* enable self-refresh */
+
+	/* modify mpll */
+	nv_mask(dev, 0x00c040, 0x0000c000, 0x0000c000);
+	nv_mask(dev, 0x004008, 0x81ff0200, 0x00000200 | info->mctrl);
+	nv_wr32(dev, 0x00400c, info->mcoef);
+	udelay(100);
+	nv_mask(dev, 0x004008, 0x81ff0200, info->mctrl);
+
+	/* re-enable normal operation of memory controller */
+	nv_wr32(dev, 0x1002dc, 0x00000000);
+	nv_mask(dev, 0x100210, 0x80000000, 0x80000000);
+	udelay(100);
+
+	/* re-enable display */
+	nv_wr32(dev, 0x611200, 0x00003330);
+
+	goto resume;
+error:
+	ret = -EBUSY;
+resume:
+	nv_mask(dev, 0x002504, 0x00000001, 0x00000000);
+	return ret;
 }
 
 static int