drm/nouveau/gk20a: reclocking support
authorAlexandre Courbot <acourbot@nvidia.com>
Sat, 26 Jul 2014 09:41:41 +0000 (18:41 +0900)
committerBen Skeggs <bskeggs@redhat.com>
Sat, 9 Aug 2014 19:12:48 +0000 (05:12 +1000)
Add support for reclocking on GK20A, using a statically-defined pstates
table. The algorithms for calculating the coefficients and setting the
clocks are directly taken from the ChromeOS kernel.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drivers/gpu/drm/nouveau/Makefile
drivers/gpu/drm/nouveau/core/engine/device/nve0.c
drivers/gpu/drm/nouveau/core/include/subdev/clock.h
drivers/gpu/drm/nouveau/core/os.h
drivers/gpu/drm/nouveau/core/subdev/clock/gk20a.c [new file with mode: 0644]

index bff9ed6c6d19585ca841f49a6f26b381c88f422c..fcd915ca8a4a52e41efe70e629716731548c1392 100644 (file)
@@ -65,6 +65,7 @@ nouveau-y += core/subdev/clock/nva3.o
 nouveau-y += core/subdev/clock/nvaa.o
 nouveau-y += core/subdev/clock/nvc0.o
 nouveau-y += core/subdev/clock/nve0.o
+nouveau-y += core/subdev/clock/gk20a.o
 nouveau-y += core/subdev/clock/pllnv04.o
 nouveau-y += core/subdev/clock/pllnva3.o
 nouveau-y += core/subdev/devinit/base.o
index bc0dca9d22123918de0bebea590ef52d49a7d1f8..6c243ae3dbec32fe64015a782ef980ebc8f056bd 100644 (file)
@@ -158,6 +158,7 @@ nve0_identify(struct nouveau_device *device)
                break;
        case 0xea:
                device->cname = "GK20A";
+               device->oclass[NVDEV_SUBDEV_CLOCK  ] = &gk20a_clock_oclass;
                device->oclass[NVDEV_SUBDEV_MC     ] =  gk20a_mc_oclass;
                device->oclass[NVDEV_SUBDEV_BUS    ] =  nvc0_bus_oclass;
                device->oclass[NVDEV_SUBDEV_TIMER  ] = &gk20a_timer_oclass;
index a7581478e4ea6b2428f7e3831dde25655119451f..0d30fa002db29fbc24de1808eaf3e2ca38138b06 100644 (file)
@@ -147,6 +147,7 @@ extern struct nouveau_oclass *nvaa_clock_oclass;
 extern struct nouveau_oclass nva3_clock_oclass;
 extern struct nouveau_oclass nvc0_clock_oclass;
 extern struct nouveau_oclass nve0_clock_oclass;
+extern struct nouveau_oclass gk20a_clock_oclass;
 
 int nv04_clock_pll_set(struct nouveau_clock *, u32 type, u32 freq);
 int nv04_clock_pll_calc(struct nouveau_clock *, struct nvbios_pll *,
index e50decfebf30c2646679b43d2e6f507afa7139cb..ccfa21d72ddc950e3a2f583f7e9194382e0a08ed 100644 (file)
@@ -22,6 +22,7 @@
 #include <linux/log2.h>
 #include <linux/pm_runtime.h>
 #include <linux/power_supply.h>
+#include <linux/clk.h>
 
 #include <asm/unaligned.h>
 
diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/gk20a.c b/drivers/gpu/drm/nouveau/core/subdev/clock/gk20a.c
new file mode 100644 (file)
index 0000000..425a8d5
--- /dev/null
@@ -0,0 +1,665 @@
+/*
+ * Copyright (c) 2014, NVIDIA CORPORATION. All rights reserved.
+ *
+ * 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 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.
+ *
+ * Shamelessly ripped off from ChromeOS's gk20a/clk_pllg.c
+ *
+ */
+
+#define MHZ (1000 * 1000)
+
+#define MASK(w)        ((1 << w) - 1)
+
+#define SYS_GPCPLL_CFG_BASE                    0x00137000
+#define GPC_BCASE_GPCPLL_CFG_BASE              0x00132800
+
+#define GPCPLL_CFG             (SYS_GPCPLL_CFG_BASE + 0)
+#define GPCPLL_CFG_ENABLE      BIT(0)
+#define GPCPLL_CFG_IDDQ                BIT(1)
+#define GPCPLL_CFG_LOCK_DET_OFF        BIT(4)
+#define GPCPLL_CFG_LOCK                BIT(17)
+
+#define GPCPLL_COEFF           (SYS_GPCPLL_CFG_BASE + 4)
+#define GPCPLL_COEFF_M_SHIFT   0
+#define GPCPLL_COEFF_M_WIDTH   8
+#define GPCPLL_COEFF_N_SHIFT   8
+#define GPCPLL_COEFF_N_WIDTH   8
+#define GPCPLL_COEFF_P_SHIFT   16
+#define GPCPLL_COEFF_P_WIDTH   6
+
+#define GPCPLL_CFG2                    (SYS_GPCPLL_CFG_BASE + 0xc)
+#define GPCPLL_CFG2_SETUP2_SHIFT       16
+#define GPCPLL_CFG2_PLL_STEPA_SHIFT    24
+
+#define GPCPLL_CFG3                    (SYS_GPCPLL_CFG_BASE + 0x18)
+#define GPCPLL_CFG3_PLL_STEPB_SHIFT    16
+
+#define GPCPLL_NDIV_SLOWDOWN                   (SYS_GPCPLL_CFG_BASE + 0x1c)
+#define GPCPLL_NDIV_SLOWDOWN_NDIV_LO_SHIFT     0
+#define GPCPLL_NDIV_SLOWDOWN_NDIV_MID_SHIFT    8
+#define GPCPLL_NDIV_SLOWDOWN_STEP_SIZE_LO2MID_SHIFT    16
+#define GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT  22
+#define GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT  31
+
+#define SEL_VCO                                (SYS_GPCPLL_CFG_BASE + 0x100)
+#define SEL_VCO_GPC2CLK_OUT_SHIFT      0
+
+#define GPC2CLK_OUT                    (SYS_GPCPLL_CFG_BASE + 0x250)
+#define GPC2CLK_OUT_SDIV14_INDIV4_WIDTH        1
+#define GPC2CLK_OUT_SDIV14_INDIV4_SHIFT        31
+#define GPC2CLK_OUT_SDIV14_INDIV4_MODE 1
+#define GPC2CLK_OUT_VCODIV_WIDTH       6
+#define GPC2CLK_OUT_VCODIV_SHIFT       8
+#define GPC2CLK_OUT_VCODIV1            0
+#define GPC2CLK_OUT_VCODIV_MASK                (MASK(GPC2CLK_OUT_VCODIV_WIDTH) << \
+                                       GPC2CLK_OUT_VCODIV_SHIFT)
+#define        GPC2CLK_OUT_BYPDIV_WIDTH        6
+#define GPC2CLK_OUT_BYPDIV_SHIFT       0
+#define GPC2CLK_OUT_BYPDIV31           0x3c
+#define GPC2CLK_OUT_INIT_MASK  ((MASK(GPC2CLK_OUT_SDIV14_INDIV4_WIDTH) << \
+               GPC2CLK_OUT_SDIV14_INDIV4_SHIFT)\
+               | (MASK(GPC2CLK_OUT_VCODIV_WIDTH) << GPC2CLK_OUT_VCODIV_SHIFT)\
+               | (MASK(GPC2CLK_OUT_BYPDIV_WIDTH) << GPC2CLK_OUT_BYPDIV_SHIFT))
+#define GPC2CLK_OUT_INIT_VAL   ((GPC2CLK_OUT_SDIV14_INDIV4_MODE << \
+               GPC2CLK_OUT_SDIV14_INDIV4_SHIFT) \
+               | (GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT) \
+               | (GPC2CLK_OUT_BYPDIV31 << GPC2CLK_OUT_BYPDIV_SHIFT))
+
+#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG  (GPC_BCASE_GPCPLL_CFG_BASE + 0xa0)
+#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT    24
+#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK \
+           (0x1 << GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT)
+
+#include <subdev/clock.h>
+#include <subdev/timer.h>
+
+#ifdef __KERNEL__
+#include <nouveau_platform.h>
+#endif
+
+static const u8 pl_to_div[] = {
+/* PL:   0, 1, 2, 3, 4, 5, 6,  7,  8,  9, 10, 11, 12, 13, 14 */
+/* p: */ 1, 2, 3, 4, 5, 6, 8, 10, 12, 16, 12, 16, 20, 24, 32,
+};
+
+/* All frequencies in Mhz */
+struct gk20a_clk_pllg_params {
+       u32 min_vco, max_vco;
+       u32 min_u, max_u;
+       u32 min_m, max_m;
+       u32 min_n, max_n;
+       u32 min_pl, max_pl;
+};
+
+static const struct gk20a_clk_pllg_params gk20a_pllg_params = {
+       .min_vco = 1000, .max_vco = 1700,
+       .min_u = 12, .max_u = 38,
+       .min_m = 1, .max_m = 255,
+       .min_n = 8, .max_n = 255,
+       .min_pl = 1, .max_pl = 32,
+};
+
+struct gk20a_clock_priv {
+       struct nouveau_clock base;
+       const struct gk20a_clk_pllg_params *params;
+       u32 m, n, pl;
+       u32 parent_rate;
+};
+#define to_gk20a_clock(base) container_of(base, struct gk20a_clock_priv, base)
+
+static void
+gk20a_pllg_read_mnp(struct gk20a_clock_priv *priv)
+{
+       u32 val;
+
+       val = nv_rd32(priv, GPCPLL_COEFF);
+       priv->m = (val >> GPCPLL_COEFF_M_SHIFT) & MASK(GPCPLL_COEFF_M_WIDTH);
+       priv->n = (val >> GPCPLL_COEFF_N_SHIFT) & MASK(GPCPLL_COEFF_N_WIDTH);
+       priv->pl = (val >> GPCPLL_COEFF_P_SHIFT) & MASK(GPCPLL_COEFF_P_WIDTH);
+}
+
+static u32
+gk20a_pllg_calc_rate(struct gk20a_clock_priv *priv)
+{
+       u32 rate;
+       u32 divider;
+
+       rate = priv->parent_rate * priv->n;
+       divider = priv->m * pl_to_div[priv->pl];
+       do_div(rate, divider);
+
+       return rate / 2;
+}
+
+static int
+gk20a_pllg_calc_mnp(struct gk20a_clock_priv *priv, unsigned long rate)
+{
+       u32 target_clk_f, ref_clk_f, target_freq;
+       u32 min_vco_f, max_vco_f;
+       u32 low_pl, high_pl, best_pl;
+       u32 target_vco_f, vco_f;
+       u32 best_m, best_n;
+       u32 u_f;
+       u32 m, n, n2;
+       u32 delta, lwv, best_delta = ~0;
+       u32 pl;
+
+       target_clk_f = rate * 2 / MHZ;
+       ref_clk_f = priv->parent_rate / MHZ;
+
+       max_vco_f = priv->params->max_vco;
+       min_vco_f = priv->params->min_vco;
+       best_m = priv->params->max_m;
+       best_n = priv->params->min_n;
+       best_pl = priv->params->min_pl;
+
+       target_vco_f = target_clk_f + target_clk_f / 50;
+       if (max_vco_f < target_vco_f)
+               max_vco_f = target_vco_f;
+
+       /* min_pl <= high_pl <= max_pl */
+       high_pl = (max_vco_f + target_vco_f - 1) / target_vco_f;
+       high_pl = min(high_pl, priv->params->max_pl);
+       high_pl = max(high_pl, priv->params->min_pl);
+
+       /* min_pl <= low_pl <= max_pl */
+       low_pl = min_vco_f / target_vco_f;
+       low_pl = min(low_pl, priv->params->max_pl);
+       low_pl = max(low_pl, priv->params->min_pl);
+
+       /* Find Indices of high_pl and low_pl */
+       for (pl = 0; pl < ARRAY_SIZE(pl_to_div) - 1; pl++) {
+               if (pl_to_div[pl] >= low_pl) {
+                       low_pl = pl;
+                       break;
+               }
+       }
+       for (pl = 0; pl < ARRAY_SIZE(pl_to_div) - 1; pl++) {
+               if (pl_to_div[pl] >= high_pl) {
+                       high_pl = pl;
+                       break;
+               }
+       }
+
+       nv_debug(priv, "low_PL %d(div%d), high_PL %d(div%d)", low_pl,
+                pl_to_div[low_pl], high_pl, pl_to_div[high_pl]);
+
+       /* Select lowest possible VCO */
+       for (pl = low_pl; pl <= high_pl; pl++) {
+               target_vco_f = target_clk_f * pl_to_div[pl];
+               for (m = priv->params->min_m; m <= priv->params->max_m; m++) {
+                       u_f = ref_clk_f / m;
+
+                       if (u_f < priv->params->min_u)
+                               break;
+                       if (u_f > priv->params->max_u)
+                               continue;
+
+                       n = (target_vco_f * m) / ref_clk_f;
+                       n2 = ((target_vco_f * m) + (ref_clk_f - 1)) / ref_clk_f;
+
+                       if (n > priv->params->max_n)
+                               break;
+
+                       for (; n <= n2; n++) {
+                               if (n < priv->params->min_n)
+                                       continue;
+                               if (n > priv->params->max_n)
+                                       break;
+
+                               vco_f = ref_clk_f * n / m;
+
+                               if (vco_f >= min_vco_f && vco_f <= max_vco_f) {
+                                       lwv = (vco_f + (pl_to_div[pl] / 2))
+                                               / pl_to_div[pl];
+                                       delta = abs(lwv - target_clk_f);
+
+                                       if (delta < best_delta) {
+                                               best_delta = delta;
+                                               best_m = m;
+                                               best_n = n;
+                                               best_pl = pl;
+
+                                               if (best_delta == 0)
+                                                       goto found_match;
+                                       }
+                               }
+                       }
+               }
+       }
+
+found_match:
+       WARN_ON(best_delta == ~0);
+
+       if (best_delta != 0)
+               nv_debug(priv, "no best match for target @ %dMHz on gpc_pll",
+                        target_clk_f);
+
+       priv->m = best_m;
+       priv->n = best_n;
+       priv->pl = best_pl;
+
+       target_freq = gk20a_pllg_calc_rate(priv) / MHZ;
+
+       nv_debug(priv, "actual target freq %d MHz, M %d, N %d, PL %d(div%d)\n",
+                target_freq, priv->m, priv->n, priv->pl, pl_to_div[priv->pl]);
+
+       return 0;
+}
+
+static int
+gk20a_pllg_slide(struct gk20a_clock_priv *priv, u32 n)
+{
+       u32 val;
+       int ramp_timeout;
+
+       /* get old coefficients */
+       val = nv_rd32(priv, GPCPLL_COEFF);
+       /* do nothing if NDIV is the same */
+       if (n == ((val >> GPCPLL_COEFF_N_SHIFT) & MASK(GPCPLL_COEFF_N_WIDTH)))
+               return 0;
+
+       /* setup */
+       nv_mask(priv, GPCPLL_CFG2, 0xff << GPCPLL_CFG2_PLL_STEPA_SHIFT,
+               0x2b << GPCPLL_CFG2_PLL_STEPA_SHIFT);
+       nv_mask(priv, GPCPLL_CFG3, 0xff << GPCPLL_CFG3_PLL_STEPB_SHIFT,
+               0xb << GPCPLL_CFG3_PLL_STEPB_SHIFT);
+
+       /* pll slowdown mode */
+       nv_mask(priv, GPCPLL_NDIV_SLOWDOWN,
+               BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT),
+               BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT));
+
+       /* new ndiv ready for ramp */
+       val = nv_rd32(priv, GPCPLL_COEFF);
+       val &= ~(MASK(GPCPLL_COEFF_N_WIDTH) << GPCPLL_COEFF_N_SHIFT);
+       val |= (n & MASK(GPCPLL_COEFF_N_WIDTH)) << GPCPLL_COEFF_N_SHIFT;
+       udelay(1);
+       nv_wr32(priv, GPCPLL_COEFF, val);
+
+       /* dynamic ramp to new ndiv */
+       val = nv_rd32(priv, GPCPLL_NDIV_SLOWDOWN);
+       val |= 0x1 << GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT;
+       udelay(1);
+       nv_wr32(priv, GPCPLL_NDIV_SLOWDOWN, val);
+
+       for (ramp_timeout = 500; ramp_timeout > 0; ramp_timeout--) {
+               udelay(1);
+               val = nv_rd32(priv, GPC_BCAST_NDIV_SLOWDOWN_DEBUG);
+               if (val & GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK)
+                       break;
+       }
+
+       /* exit slowdown mode */
+       nv_mask(priv, GPCPLL_NDIV_SLOWDOWN,
+               BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT) |
+               BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT), 0);
+       nv_rd32(priv, GPCPLL_NDIV_SLOWDOWN);
+
+       if (ramp_timeout <= 0) {
+               nv_error(priv, "gpcpll dynamic ramp timeout\n");
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+
+static void
+_gk20a_pllg_enable(struct gk20a_clock_priv *priv)
+{
+       nv_mask(priv, GPCPLL_CFG, GPCPLL_CFG_ENABLE, GPCPLL_CFG_ENABLE);
+       nv_rd32(priv, GPCPLL_CFG);
+}
+
+static void
+_gk20a_pllg_disable(struct gk20a_clock_priv *priv)
+{
+       nv_mask(priv, GPCPLL_CFG, GPCPLL_CFG_ENABLE, 0);
+       nv_rd32(priv, GPCPLL_CFG);
+}
+
+static int
+_gk20a_pllg_program_mnp(struct gk20a_clock_priv *priv, bool allow_slide)
+{
+       u32 val, cfg;
+       u32 m_old, pl_old, n_lo;
+
+       /* get old coefficients */
+       val = nv_rd32(priv, GPCPLL_COEFF);
+       m_old = (val >> GPCPLL_COEFF_M_SHIFT) & MASK(GPCPLL_COEFF_M_WIDTH);
+       pl_old = (val >> GPCPLL_COEFF_P_SHIFT) & MASK(GPCPLL_COEFF_P_WIDTH);
+
+       /* do NDIV slide if there is no change in M and PL */
+       cfg = nv_rd32(priv, GPCPLL_CFG);
+       if (allow_slide && priv->m == m_old && priv->pl == pl_old &&
+           (cfg & GPCPLL_CFG_ENABLE)) {
+               return gk20a_pllg_slide(priv, priv->n);
+       }
+
+       /* slide down to NDIV_LO */
+       n_lo = DIV_ROUND_UP(m_old * priv->params->min_vco,
+                           priv->parent_rate / MHZ);
+       if (allow_slide && (cfg & GPCPLL_CFG_ENABLE)) {
+               int ret = gk20a_pllg_slide(priv, n_lo);
+
+               if (ret)
+                       return ret;
+       }
+
+       /* split FO-to-bypass jump in halfs by setting out divider 1:2 */
+       nv_mask(priv, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+               0x2 << GPC2CLK_OUT_VCODIV_SHIFT);
+
+       /* put PLL in bypass before programming it */
+       val = nv_rd32(priv, SEL_VCO);
+       val &= ~(BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+       udelay(2);
+       nv_wr32(priv, SEL_VCO, val);
+
+       /* get out from IDDQ */
+       val = nv_rd32(priv, GPCPLL_CFG);
+       if (val & GPCPLL_CFG_IDDQ) {
+               val &= ~GPCPLL_CFG_IDDQ;
+               nv_wr32(priv, GPCPLL_CFG, val);
+               nv_rd32(priv, GPCPLL_CFG);
+               udelay(2);
+       }
+
+       _gk20a_pllg_disable(priv);
+
+       nv_debug(priv, "%s: m=%d n=%d pl=%d\n", __func__, priv->m, priv->n,
+                priv->pl);
+
+       n_lo = DIV_ROUND_UP(priv->m * priv->params->min_vco,
+                           priv->parent_rate / MHZ);
+       val = priv->m << GPCPLL_COEFF_M_SHIFT;
+       val |= (allow_slide ? n_lo : priv->n) << GPCPLL_COEFF_N_SHIFT;
+       val |= priv->pl << GPCPLL_COEFF_P_SHIFT;
+       nv_wr32(priv, GPCPLL_COEFF, val);
+
+       _gk20a_pllg_enable(priv);
+
+       val = nv_rd32(priv, GPCPLL_CFG);
+       if (val & GPCPLL_CFG_LOCK_DET_OFF) {
+               val &= ~GPCPLL_CFG_LOCK_DET_OFF;
+               nv_wr32(priv, GPCPLL_CFG, val);
+       }
+
+       if (!nouveau_timer_wait_eq(priv, 300000, GPCPLL_CFG, GPCPLL_CFG_LOCK,
+                                  GPCPLL_CFG_LOCK)) {
+               nv_error(priv, "%s: timeout waiting for pllg lock\n", __func__);
+               return -ETIMEDOUT;
+       }
+
+       /* switch to VCO mode */
+       nv_mask(priv, SEL_VCO, 0, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+
+       /* restore out divider 1:1 */
+       val = nv_rd32(priv, GPC2CLK_OUT);
+       val &= ~GPC2CLK_OUT_VCODIV_MASK;
+       udelay(2);
+       nv_wr32(priv, GPC2CLK_OUT, val);
+
+       /* slide up to new NDIV */
+       return allow_slide ? gk20a_pllg_slide(priv, priv->n) : 0;
+}
+
+static int
+gk20a_pllg_program_mnp(struct gk20a_clock_priv *priv)
+{
+       int err;
+
+       err = _gk20a_pllg_program_mnp(priv, true);
+       if (err)
+               err = _gk20a_pllg_program_mnp(priv, false);
+
+       return err;
+}
+
+static void
+gk20a_pllg_disable(struct gk20a_clock_priv *priv)
+{
+       u32 val;
+
+       /* slide to VCO min */
+       val = nv_rd32(priv, GPCPLL_CFG);
+       if (val & GPCPLL_CFG_ENABLE) {
+               u32 coeff, m, n_lo;
+
+               coeff = nv_rd32(priv, GPCPLL_COEFF);
+               m = (coeff >> GPCPLL_COEFF_M_SHIFT) & MASK(GPCPLL_COEFF_M_WIDTH);
+               n_lo = DIV_ROUND_UP(m * priv->params->min_vco,
+                                   priv->parent_rate / MHZ);
+               gk20a_pllg_slide(priv, n_lo);
+       }
+
+       /* put PLL in bypass before disabling it */
+       nv_mask(priv, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
+
+       _gk20a_pllg_disable(priv);
+}
+
+#define GK20A_CLK_GPC_MDIV 1000
+
+static struct nouveau_clocks
+gk20a_domains[] = {
+       { nv_clk_src_crystal, 0xff },
+       { nv_clk_src_gpc, 0xff, 0, "core", GK20A_CLK_GPC_MDIV },
+       { nv_clk_src_max }
+};
+
+static struct nouveau_pstate
+gk20a_pstates[] = {
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 72000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 108000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 180000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 252000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 324000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 396000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 468000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 540000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 612000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 648000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 684000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 708000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 756000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 804000,
+               },
+       },
+       {
+               .base = {
+                       .domain[nv_clk_src_gpc] = 852000,
+               },
+       },
+};
+
+static int
+gk20a_clock_read(struct nouveau_clock *clk, enum nv_clk_src src)
+{
+       struct gk20a_clock_priv *priv = (void *)clk;
+
+       switch (src) {
+       case nv_clk_src_crystal:
+               return nv_device(clk)->crystal;
+       case nv_clk_src_gpc:
+               gk20a_pllg_read_mnp(priv);
+               return gk20a_pllg_calc_rate(priv) / GK20A_CLK_GPC_MDIV;
+       default:
+               nv_error(clk, "invalid clock source %d\n", src);
+               return -EINVAL;
+       }
+}
+
+static int
+gk20a_clock_calc(struct nouveau_clock *clk, struct nouveau_cstate *cstate)
+{
+       struct gk20a_clock_priv *priv = (void *)clk;
+
+       return gk20a_pllg_calc_mnp(priv, cstate->domain[nv_clk_src_gpc] *
+                                        GK20A_CLK_GPC_MDIV);
+}
+
+static int
+gk20a_clock_prog(struct nouveau_clock *clk)
+{
+       struct gk20a_clock_priv *priv = (void *)clk;
+
+       return gk20a_pllg_program_mnp(priv);
+}
+
+static void
+gk20a_clock_tidy(struct nouveau_clock *clk)
+{
+}
+
+static int
+gk20a_clock_fini(struct nouveau_object *object, bool suspend)
+{
+       struct gk20a_clock_priv *priv = (void *)object;
+       int ret;
+
+       ret = nouveau_clock_fini(&priv->base, false);
+
+       gk20a_pllg_disable(priv);
+
+       return ret;
+}
+
+static int
+gk20a_clock_init(struct nouveau_object *object)
+{
+       struct gk20a_clock_priv *priv = (void *)object;
+       int ret;
+
+       nv_mask(priv, GPC2CLK_OUT, GPC2CLK_OUT_INIT_MASK, GPC2CLK_OUT_INIT_VAL);
+
+       ret = nouveau_clock_init(&priv->base);
+       if (ret)
+               return ret;
+
+       ret = gk20a_clock_prog(&priv->base);
+       if (ret) {
+               nv_error(priv, "cannot initialize clock\n");
+               return ret;
+       }
+
+       return 0;
+}
+
+static int
+gk20a_clock_ctor(struct nouveau_object *parent,  struct nouveau_object *engine,
+                struct nouveau_oclass *oclass, void *data, u32 size,
+                struct nouveau_object **pobject)
+{
+       struct gk20a_clock_priv *priv;
+       struct nouveau_platform_device *plat;
+       int ret;
+       int i;
+
+       /* Finish initializing the pstates */
+       for (i = 0; i < ARRAY_SIZE(gk20a_pstates); i++) {
+               INIT_LIST_HEAD(&gk20a_pstates[i].list);
+               gk20a_pstates[i].pstate = i + 1;
+       }
+
+       ret = nouveau_clock_create(parent, engine, oclass, gk20a_domains,
+                       gk20a_pstates, ARRAY_SIZE(gk20a_pstates), true, &priv);
+       *pobject = nv_object(priv);
+       if (ret)
+               return ret;
+
+       priv->params = &gk20a_pllg_params;
+
+       plat = nv_device_to_platform(nv_device(parent));
+       priv->parent_rate = clk_get_rate(plat->gpu->clk);
+       nv_info(priv, "parent clock rate: %d Mhz\n", priv->parent_rate / MHZ);
+
+       priv->base.read = gk20a_clock_read;
+       priv->base.calc = gk20a_clock_calc;
+       priv->base.prog = gk20a_clock_prog;
+       priv->base.tidy = gk20a_clock_tidy;
+
+       return 0;
+}
+
+struct nouveau_oclass
+gk20a_clock_oclass = {
+       .handle = NV_SUBDEV(CLOCK, 0xea),
+       .ofuncs = &(struct nouveau_ofuncs) {
+               .ctor = gk20a_clock_ctor,
+               .dtor = _nouveau_subdev_dtor,
+               .init = gk20a_clock_init,
+               .fini = gk20a_clock_fini,
+       },
+};