diff options
Diffstat (limited to 'drivers/gpu/drm/nouveau/core/subdev/clock')
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/base.c | 197 | ||||
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/gk20a.c | 665 | ||||
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/nv40.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/nvaa.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c | 4 |
9 files changed, 825 insertions, 63 deletions
diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/base.c b/drivers/gpu/drm/nouveau/core/subdev/clock/base.c index 22351f594d2a..a276a711294a 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/base.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/base.c @@ -90,16 +90,20 @@ nouveau_cstate_prog(struct nouveau_clock *clk, cstate = &pstate->base; } - ret = nouveau_therm_cstate(ptherm, pstate->fanspeed, +1); - if (ret && ret != -ENODEV) { - nv_error(clk, "failed to raise fan speed: %d\n", ret); - return ret; + if (ptherm) { + ret = nouveau_therm_cstate(ptherm, pstate->fanspeed, +1); + if (ret && ret != -ENODEV) { + nv_error(clk, "failed to raise fan speed: %d\n", ret); + return ret; + } } - ret = volt->set_id(volt, cstate->voltage, +1); - if (ret && ret != -ENODEV) { - nv_error(clk, "failed to raise voltage: %d\n", ret); - return ret; + if (volt) { + ret = volt->set_id(volt, cstate->voltage, +1); + if (ret && ret != -ENODEV) { + nv_error(clk, "failed to raise voltage: %d\n", ret); + return ret; + } } ret = clk->calc(clk, cstate); @@ -108,13 +112,17 @@ nouveau_cstate_prog(struct nouveau_clock *clk, clk->tidy(clk); } - ret = volt->set_id(volt, cstate->voltage, -1); - if (ret && ret != -ENODEV) - nv_error(clk, "failed to lower voltage: %d\n", ret); + if (volt) { + ret = volt->set_id(volt, cstate->voltage, -1); + if (ret && ret != -ENODEV) + nv_error(clk, "failed to lower voltage: %d\n", ret); + } - ret = nouveau_therm_cstate(ptherm, pstate->fanspeed, -1); - if (ret && ret != -ENODEV) - nv_error(clk, "failed to lower fan speed: %d\n", ret); + if (ptherm) { + ret = nouveau_therm_cstate(ptherm, pstate->fanspeed, -1); + if (ret && ret != -ENODEV) + nv_error(clk, "failed to lower fan speed: %d\n", ret); + } return 0; } @@ -194,16 +202,23 @@ nouveau_pstate_prog(struct nouveau_clock *clk, int pstatei) return nouveau_cstate_prog(clk, pstate, 0); } -static int -nouveau_pstate_calc(struct nouveau_clock *clk) +static void +nouveau_pstate_work(struct work_struct *work) { - int pstate, ret = 0; + struct nouveau_clock *clk = container_of(work, typeof(*clk), work); + int pstate; - nv_trace(clk, "P %d U %d A %d T %d D %d\n", clk->pstate, - clk->ustate, clk->astate, clk->tstate, clk->dstate); + if (!atomic_xchg(&clk->waiting, 0)) + return; + clk->pwrsrc = power_supply_is_system_supplied(); - if (clk->state_nr && clk->ustate != -1) { - pstate = (clk->ustate < 0) ? clk->astate : clk->ustate; + nv_trace(clk, "P %d PWR %d U(AC) %d U(DC) %d A %d T %d D %d\n", + clk->pstate, clk->pwrsrc, clk->ustate_ac, clk->ustate_dc, + clk->astate, clk->tstate, clk->dstate); + + pstate = clk->pwrsrc ? clk->ustate_ac : clk->ustate_dc; + if (clk->state_nr && pstate != -1) { + pstate = (pstate < 0) ? clk->astate : pstate; pstate = min(pstate, clk->state_nr - 1 - clk->tstate); pstate = max(pstate, clk->dstate); } else { @@ -211,9 +226,26 @@ nouveau_pstate_calc(struct nouveau_clock *clk) } nv_trace(clk, "-> %d\n", pstate); - if (pstate != clk->pstate) - ret = nouveau_pstate_prog(clk, pstate); - return ret; + if (pstate != clk->pstate) { + int ret = nouveau_pstate_prog(clk, pstate); + if (ret) { + nv_error(clk, "error setting pstate %d: %d\n", + pstate, ret); + } + } + + wake_up_all(&clk->wait); + nvkm_notify_get(&clk->pwrsrc_ntfy); +} + +static int +nouveau_pstate_calc(struct nouveau_clock *clk, bool wait) +{ + atomic_set(&clk->waiting, 1); + schedule_work(&clk->work); + if (wait) + wait_event(clk->wait, !atomic_read(&clk->waiting)); + return 0; } static void @@ -361,17 +393,40 @@ nouveau_clock_ustate_update(struct nouveau_clock *clk, int req) req = i; } - clk->ustate = req; - return 0; + return req + 2; +} + +static int +nouveau_clock_nstate(struct nouveau_clock *clk, const char *mode, int arglen) +{ + int ret = 1; + + if (strncasecmpz(mode, "disabled", arglen)) { + char save = mode[arglen]; + long v; + + ((char *)mode)[arglen] = '\0'; + if (!kstrtol(mode, 0, &v)) { + ret = nouveau_clock_ustate_update(clk, v); + if (ret < 0) + ret = 1; + } + ((char *)mode)[arglen] = save; + } + + return ret - 2; } int -nouveau_clock_ustate(struct nouveau_clock *clk, int req) +nouveau_clock_ustate(struct nouveau_clock *clk, int req, int pwr) { int ret = nouveau_clock_ustate_update(clk, req); - if (ret) - return ret; - return nouveau_pstate_calc(clk); + if (ret >= 0) { + if (ret -= 2, pwr) clk->ustate_ac = ret; + else clk->ustate_dc = ret; + return nouveau_pstate_calc(clk, true); + } + return ret; } int @@ -381,7 +436,7 @@ nouveau_clock_astate(struct nouveau_clock *clk, int req, int rel) if ( rel) clk->astate += rel; clk->astate = min(clk->astate, clk->state_nr - 1); clk->astate = max(clk->astate, 0); - return nouveau_pstate_calc(clk); + return nouveau_pstate_calc(clk, true); } int @@ -391,7 +446,7 @@ nouveau_clock_tstate(struct nouveau_clock *clk, int req, int rel) if ( rel) clk->tstate += rel; clk->tstate = min(clk->tstate, 0); clk->tstate = max(clk->tstate, -(clk->state_nr - 1)); - return nouveau_pstate_calc(clk); + return nouveau_pstate_calc(clk, true); } int @@ -401,12 +456,30 @@ nouveau_clock_dstate(struct nouveau_clock *clk, int req, int rel) if ( rel) clk->dstate += rel; clk->dstate = min(clk->dstate, clk->state_nr - 1); clk->dstate = max(clk->dstate, 0); - return nouveau_pstate_calc(clk); + return nouveau_pstate_calc(clk, true); +} + +static int +nouveau_clock_pwrsrc(struct nvkm_notify *notify) +{ + struct nouveau_clock *clk = + container_of(notify, typeof(*clk), pwrsrc_ntfy); + nouveau_pstate_calc(clk, false); + return NVKM_NOTIFY_DROP; } /****************************************************************************** * subdev base class implementation *****************************************************************************/ + +int +_nouveau_clock_fini(struct nouveau_object *object, bool suspend) +{ + struct nouveau_clock *clk = (void *)object; + nvkm_notify_put(&clk->pwrsrc_ntfy); + return nouveau_subdev_fini(&clk->base, suspend); +} + int _nouveau_clock_init(struct nouveau_object *object) { @@ -414,6 +487,10 @@ _nouveau_clock_init(struct nouveau_object *object) struct nouveau_clocks *clock = clk->domains; int ret; + ret = nouveau_subdev_init(&clk->base); + if (ret) + return ret; + memset(&clk->bstate, 0x00, sizeof(clk->bstate)); INIT_LIST_HEAD(&clk->bstate.list); clk->bstate.pstate = 0xff; @@ -434,7 +511,7 @@ _nouveau_clock_init(struct nouveau_object *object) clk->tstate = 0; clk->dstate = 0; clk->pstate = -1; - nouveau_pstate_calc(clk); + nouveau_pstate_calc(clk, true); return 0; } @@ -444,6 +521,8 @@ _nouveau_clock_dtor(struct nouveau_object *object) struct nouveau_clock *clk = (void *)object; struct nouveau_pstate *pstate, *temp; + nvkm_notify_fini(&clk->pwrsrc_ntfy); + list_for_each_entry_safe(pstate, temp, &clk->states, head) { nouveau_pstate_del(pstate); } @@ -456,6 +535,7 @@ nouveau_clock_create_(struct nouveau_object *parent, struct nouveau_object *engine, struct nouveau_oclass *oclass, struct nouveau_clocks *clocks, + struct nouveau_pstate *pstates, int nb_pstates, bool allow_reclock, int length, void **object) { @@ -472,29 +552,46 @@ nouveau_clock_create_(struct nouveau_object *parent, INIT_LIST_HEAD(&clk->states); clk->domains = clocks; - clk->ustate = -1; + clk->ustate_ac = -1; + clk->ustate_dc = -1; + + INIT_WORK(&clk->work, nouveau_pstate_work); + init_waitqueue_head(&clk->wait); + atomic_set(&clk->waiting, 0); - idx = 0; - do { - ret = nouveau_pstate_new(clk, idx++); - } while (ret == 0); + /* If no pstates are provided, try and fetch them from the BIOS */ + if (!pstates) { + idx = 0; + do { + ret = nouveau_pstate_new(clk, idx++); + } while (ret == 0); + } else { + for (idx = 0; idx < nb_pstates; idx++) + list_add_tail(&pstates[idx].head, &clk->states); + clk->state_nr = nb_pstates; + } clk->allow_reclock = allow_reclock; + ret = nvkm_notify_init(&device->event, nouveau_clock_pwrsrc, true, + NULL, 0, 0, &clk->pwrsrc_ntfy); + if (ret) + return ret; + mode = nouveau_stropt(device->cfgopt, "NvClkMode", &arglen); if (mode) { - if (!strncasecmpz(mode, "disabled", arglen)) { - clk->ustate = -1; - } else { - char save = mode[arglen]; - long v; - - ((char *)mode)[arglen] = '\0'; - if (!kstrtol(mode, 0, &v)) - nouveau_clock_ustate_update(clk, v); - ((char *)mode)[arglen] = save; - } + clk->ustate_ac = nouveau_clock_nstate(clk, mode, arglen); + clk->ustate_dc = nouveau_clock_nstate(clk, mode, arglen); } + mode = nouveau_stropt(device->cfgopt, "NvClkModeAC", &arglen); + if (mode) + clk->ustate_ac = nouveau_clock_nstate(clk, mode, arglen); + + mode = nouveau_stropt(device->cfgopt, "NvClkModeDC", &arglen); + if (mode) + clk->ustate_dc = nouveau_clock_nstate(clk, mode, arglen); + + return 0; } 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 index 000000000000..425a8d5e9129 --- /dev/null +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/gk20a.c @@ -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, + }, +}; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c index eb2d4425a49e..4c48232686be 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c @@ -82,8 +82,8 @@ nv04_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, struct nv04_clock_priv *priv; int ret; - ret = nouveau_clock_create(parent, engine, oclass, nv04_domain, false, - &priv); + ret = nouveau_clock_create(parent, engine, oclass, nv04_domain, NULL, 0, + false, &priv); *pobject = nv_object(priv); if (ret) return ret; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nv40.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nv40.c index 8a9e16839791..08368fe97029 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nv40.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nv40.c @@ -213,8 +213,8 @@ nv40_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, struct nv40_clock_priv *priv; int ret; - ret = nouveau_clock_create(parent, engine, oclass, nv40_domain, true, - &priv); + ret = nouveau_clock_create(parent, engine, oclass, nv40_domain, NULL, 0, + true, &priv); *pobject = nv_object(priv); if (ret) return ret; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c index 8c132772ba9e..5070ebc260f8 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c @@ -507,7 +507,7 @@ nv50_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, int ret; ret = nouveau_clock_create(parent, engine, oclass, pclass->domains, - false, &priv); + NULL, 0, false, &priv); *pobject = nv_object(priv); if (ret) return ret; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c index 9fb58354a80b..087012b18956 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c @@ -302,8 +302,8 @@ nva3_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, struct nva3_clock_priv *priv; int ret; - ret = nouveau_clock_create(parent, engine, oclass, nva3_domain, false, - &priv); + ret = nouveau_clock_create(parent, engine, oclass, nva3_domain, NULL, 0, + false, &priv); *pobject = nv_object(priv); if (ret) return ret; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nvaa.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nvaa.c index 6a65fc9e9663..74e19731b1b7 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nvaa.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nvaa.c @@ -421,8 +421,8 @@ nvaa_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, struct nvaa_clock_priv *priv; int ret; - ret = nouveau_clock_create(parent, engine, oclass, nvaa_domains, true, - &priv); + ret = nouveau_clock_create(parent, engine, oclass, nvaa_domains, NULL, + 0, true, &priv); *pobject = nv_object(priv); if (ret) return ret; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c index dbf8517f54da..1234abaab2db 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c @@ -437,8 +437,8 @@ nvc0_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, struct nvc0_clock_priv *priv; int ret; - ret = nouveau_clock_create(parent, engine, oclass, nvc0_domain, false, - &priv); + ret = nouveau_clock_create(parent, engine, oclass, nvc0_domain, NULL, 0, + false, &priv); *pobject = nv_object(priv); if (ret) return ret; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c index 0e62a3240144..7eccad57512e 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c @@ -475,8 +475,8 @@ nve0_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, struct nve0_clock_priv *priv; int ret; - ret = nouveau_clock_create(parent, engine, oclass, nve0_domain, true, - &priv); + ret = nouveau_clock_create(parent, engine, oclass, nve0_domain, NULL, 0, + true, &priv); *pobject = nv_object(priv); if (ret) return ret; |