From 7c856522069755ab9d163a24ac332cd3cb35fe30 Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Mon, 14 Jan 2013 08:28:28 +1000 Subject: [PATCH] drm/nouveau/clk: implement power state and engine clock control in core User control of this has been hard-coded as disabled for now. Signed-off-by: Ben Skeggs --- drivers/gpu/drm/nouveau/Makefile | 4 + drivers/gpu/drm/nouveau/core/core/option.c | 9 - drivers/gpu/drm/nouveau/core/engine/device/nv50.c | 20 +- drivers/gpu/drm/nouveau/core/engine/device/nve0.c | 10 +- drivers/gpu/drm/nouveau/core/include/core/option.h | 9 + .../gpu/drm/nouveau/core/include/subdev/clock.h | 111 ++++- drivers/gpu/drm/nouveau/core/include/subdev/fb.h | 3 + drivers/gpu/drm/nouveau/core/subdev/clock/base.c | 494 ++++++++++++++++++++ drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c | 2 +- drivers/gpu/drm/nouveau/core/subdev/clock/nv40.c | 183 +++++++- drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c | 520 ++++++++++++++++++++- drivers/gpu/drm/nouveau/core/subdev/clock/nv50.h | 31 ++ drivers/gpu/drm/nouveau/core/subdev/clock/nv84.c | 48 ++ drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c | 271 ++++++++++- drivers/gpu/drm/nouveau/core/subdev/clock/nva3.h | 14 + drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c | 404 +++++++++++++++- drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c | 497 ++++++++++++++++++++ .../gpu/drm/nouveau/core/subdev/clock/pllnv04.c | 8 +- drivers/gpu/drm/nouveau/core/subdev/clock/seq.h | 17 + 19 files changed, 2590 insertions(+), 65 deletions(-) create mode 100644 drivers/gpu/drm/nouveau/core/subdev/clock/base.c create mode 100644 drivers/gpu/drm/nouveau/core/subdev/clock/nv50.h create mode 100644 drivers/gpu/drm/nouveau/core/subdev/clock/nv84.c create mode 100644 drivers/gpu/drm/nouveau/core/subdev/clock/nva3.h create mode 100644 drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c create mode 100644 drivers/gpu/drm/nouveau/core/subdev/clock/seq.h diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile index a6a71663e00e..e65ebb2e3820 100644 --- a/drivers/gpu/drm/nouveau/Makefile +++ b/drivers/gpu/drm/nouveau/Makefile @@ -47,16 +47,20 @@ nouveau-y += core/subdev/bios/therm.o nouveau-y += core/subdev/bios/vmap.o nouveau-y += core/subdev/bios/volt.o nouveau-y += core/subdev/bios/xpio.o +nouveau-y += core/subdev/bus/hwsq.o nouveau-y += core/subdev/bus/nv04.o nouveau-y += core/subdev/bus/nv31.o nouveau-y += core/subdev/bus/nv50.o nouveau-y += core/subdev/bus/nv94.o nouveau-y += core/subdev/bus/nvc0.o +nouveau-y += core/subdev/clock/base.o nouveau-y += core/subdev/clock/nv04.o nouveau-y += core/subdev/clock/nv40.o nouveau-y += core/subdev/clock/nv50.o +nouveau-y += core/subdev/clock/nv84.o nouveau-y += core/subdev/clock/nva3.o nouveau-y += core/subdev/clock/nvc0.o +nouveau-y += core/subdev/clock/nve0.o nouveau-y += core/subdev/clock/pllnv04.o nouveau-y += core/subdev/clock/pllnva3.o nouveau-y += core/subdev/devinit/base.o diff --git a/drivers/gpu/drm/nouveau/core/core/option.c b/drivers/gpu/drm/nouveau/core/core/option.c index d42e72a87651..9f6fcc5f66c2 100644 --- a/drivers/gpu/drm/nouveau/core/core/option.c +++ b/drivers/gpu/drm/nouveau/core/core/option.c @@ -25,15 +25,6 @@ #include #include -/* compares unterminated string 'str' with zero-terminated string 'cmp' */ -static inline int -strncasecmpz(const char *str, const char *cmp, size_t len) -{ - if (strlen(cmp) != len) - return len; - return strncasecmp(str, cmp, len); -} - const char * nouveau_stropt(const char *optstr, const char *opt, int *arglen) { diff --git a/drivers/gpu/drm/nouveau/core/engine/device/nv50.c b/drivers/gpu/drm/nouveau/core/engine/device/nv50.c index cc1b1391d2bc..db139827047c 100644 --- a/drivers/gpu/drm/nouveau/core/engine/device/nv50.c +++ b/drivers/gpu/drm/nouveau/core/engine/device/nv50.c @@ -62,7 +62,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv50_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv50_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv50_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -87,7 +87,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv50_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -115,7 +115,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv50_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -143,7 +143,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv50_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -171,7 +171,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv94_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -199,7 +199,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv94_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -227,7 +227,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv94_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -255,7 +255,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv50_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -283,7 +283,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv94_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; @@ -311,7 +311,7 @@ nv50_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nv50_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nv94_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nv50_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = nv84_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nv84_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass; diff --git a/drivers/gpu/drm/nouveau/core/engine/device/nve0.c b/drivers/gpu/drm/nouveau/core/engine/device/nve0.c index 4107b67c95ed..33f3c92a180c 100644 --- a/drivers/gpu/drm/nouveau/core/engine/device/nve0.c +++ b/drivers/gpu/drm/nouveau/core/engine/device/nve0.c @@ -62,7 +62,7 @@ nve0_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nve0_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nvd0_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nvc0_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = &nve0_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nvd0_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nvc0_devinit_oclass; @@ -95,7 +95,7 @@ nve0_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nve0_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nvd0_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nvc0_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = &nve0_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nvd0_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nvc0_devinit_oclass; @@ -128,7 +128,7 @@ nve0_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nve0_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nvd0_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nvc0_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = &nve0_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nvd0_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nvc0_devinit_oclass; @@ -161,7 +161,7 @@ nve0_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nve0_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nvd0_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nvc0_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = &nve0_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nvd0_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nvc0_devinit_oclass; @@ -196,7 +196,7 @@ nve0_identify(struct nouveau_device *device) device->oclass[NVDEV_SUBDEV_VBIOS ] = &nouveau_bios_oclass; device->oclass[NVDEV_SUBDEV_GPIO ] = &nve0_gpio_oclass; device->oclass[NVDEV_SUBDEV_I2C ] = &nvd0_i2c_oclass; - device->oclass[NVDEV_SUBDEV_CLOCK ] = &nvc0_clock_oclass; + device->oclass[NVDEV_SUBDEV_CLOCK ] = &nve0_clock_oclass; device->oclass[NVDEV_SUBDEV_THERM ] = &nvd0_therm_oclass; device->oclass[NVDEV_SUBDEV_MXM ] = &nv50_mxm_oclass; device->oclass[NVDEV_SUBDEV_DEVINIT] = &nvc0_devinit_oclass; diff --git a/drivers/gpu/drm/nouveau/core/include/core/option.h b/drivers/gpu/drm/nouveau/core/include/core/option.h index 27074957fd21..ed055847887e 100644 --- a/drivers/gpu/drm/nouveau/core/include/core/option.h +++ b/drivers/gpu/drm/nouveau/core/include/core/option.h @@ -8,4 +8,13 @@ bool nouveau_boolopt(const char *optstr, const char *opt, bool value); int nouveau_dbgopt(const char *optstr, const char *sub); +/* compares unterminated string 'str' with zero-terminated string 'cmp' */ +static inline int +strncasecmpz(const char *str, const char *cmp, size_t len) +{ + if (strlen(cmp) != len) + return len; + return strncasecmp(str, cmp, len); +} + #endif diff --git a/drivers/gpu/drm/nouveau/core/include/subdev/clock.h b/drivers/gpu/drm/nouveau/core/include/subdev/clock.h index 89ee289097a6..e2675bc0edba 100644 --- a/drivers/gpu/drm/nouveau/core/include/subdev/clock.h +++ b/drivers/gpu/drm/nouveau/core/include/subdev/clock.h @@ -7,9 +7,78 @@ struct nouveau_pll_vals; struct nvbios_pll; +enum nv_clk_src { + nv_clk_src_crystal, + nv_clk_src_href, + + nv_clk_src_hclk, + nv_clk_src_hclkm3, + nv_clk_src_hclkm3d2, + + nv_clk_src_host, + + nv_clk_src_sppll0, + nv_clk_src_sppll1, + + nv_clk_src_mpllsrcref, + nv_clk_src_mpllsrc, + nv_clk_src_mpll, + nv_clk_src_mdiv, + + nv_clk_src_core, + nv_clk_src_shader, + + nv_clk_src_mem, + + nv_clk_src_gpc, + nv_clk_src_rop, + nv_clk_src_hubk01, + nv_clk_src_hubk06, + nv_clk_src_hubk07, + nv_clk_src_copy, + nv_clk_src_daemon, + nv_clk_src_disp, + nv_clk_src_vdec, + + nv_clk_src_dom6, + + nv_clk_src_max, +}; + +struct nouveau_cstate { + struct list_head head; + u8 voltage; + u32 domain[nv_clk_src_max]; +}; + +struct nouveau_pstate { + struct list_head head; + struct list_head list; /* c-states */ + struct nouveau_cstate base; + u8 pstate; + u8 fanspeed; +}; + struct nouveau_clock { struct nouveau_subdev base; + struct nouveau_clocks *domains; + struct nouveau_pstate bstate; + + struct list_head states; + int state_nr; + + int pstate; /* current */ + int ustate; /* user-requested (-1 disabled, -2 perfmon) */ + int astate; /* perfmon adjustment (base) */ + int tstate; /* thermal adjustment (max-) */ + int dstate; /* display adjustment (min+) */ + + int (*read)(struct nouveau_clock *, enum nv_clk_src); + int (*calc)(struct nouveau_clock *, struct nouveau_cstate *); + int (*prog)(struct nouveau_clock *); + void (*tidy)(struct nouveau_clock *); + /*XXX: die, these are here *only* to support the completely * bat-shit insane what-was-nouveau_hw.c code */ @@ -25,27 +94,42 @@ nouveau_clock(void *obj) return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_CLOCK]; } -#define nouveau_clock_create(p,e,o,d) \ - nouveau_subdev_create((p), (e), (o), 0, "CLOCK", "clock", d) -#define nouveau_clock_destroy(p) \ - nouveau_subdev_destroy(&(p)->base) -#define nouveau_clock_init(p) \ - nouveau_subdev_init(&(p)->base) +struct nouveau_clocks { + enum nv_clk_src name; + u8 bios; /* 0xff for none */ +#define NVKM_CLK_DOM_FLAG_CORE 0x01 + u8 flags; + const char *mname; + int mdiv; +}; + +#define nouveau_clock_create(p,e,o,i,d) \ + nouveau_clock_create_((p), (e), (o), (i), sizeof(**d), (void **)d) +#define nouveau_clock_destroy(p) ({ \ + struct nouveau_clock *clk = (p); \ + _nouveau_clock_dtor(nv_object(clk)); \ +}) +#define nouveau_clock_init(p) ({ \ + struct nouveau_clock *clk = (p); \ + _nouveau_clock_init(nv_object(clk)); \ +}) #define nouveau_clock_fini(p,s) \ nouveau_subdev_fini(&(p)->base, (s)) int nouveau_clock_create_(struct nouveau_object *, struct nouveau_object *, - struct nouveau_oclass *, void *, u32, int, void **); - -#define _nouveau_clock_dtor _nouveau_subdev_dtor -#define _nouveau_clock_init _nouveau_subdev_init + struct nouveau_oclass *, + struct nouveau_clocks *, int, void **); +void _nouveau_clock_dtor(struct nouveau_object *); +int _nouveau_clock_init(struct nouveau_object *); #define _nouveau_clock_fini _nouveau_subdev_fini extern struct nouveau_oclass nv04_clock_oclass; extern struct nouveau_oclass nv40_clock_oclass; -extern struct nouveau_oclass nv50_clock_oclass; +extern struct nouveau_oclass *nv50_clock_oclass; +extern struct nouveau_oclass *nv84_clock_oclass; extern struct nouveau_oclass nva3_clock_oclass; extern struct nouveau_oclass nvc0_clock_oclass; +extern struct nouveau_oclass nve0_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 *, @@ -55,4 +139,9 @@ int nv04_clock_pll_prog(struct nouveau_clock *, u32 reg1, int nva3_clock_pll_calc(struct nouveau_clock *, struct nvbios_pll *, int clk, struct nouveau_pll_vals *); +int nouveau_clock_ustate(struct nouveau_clock *, int req); +int nouveau_clock_astate(struct nouveau_clock *, int req, int rel); +int nouveau_clock_dstate(struct nouveau_clock *, int req, int rel); +int nouveau_clock_tstate(struct nouveau_clock *, int req, int rel); + #endif diff --git a/drivers/gpu/drm/nouveau/core/include/subdev/fb.h b/drivers/gpu/drm/nouveau/core/include/subdev/fb.h index 297c47075dec..33cae4882e73 100644 --- a/drivers/gpu/drm/nouveau/core/include/subdev/fb.h +++ b/drivers/gpu/drm/nouveau/core/include/subdev/fb.h @@ -125,6 +125,9 @@ struct nouveau_ram { int (*get)(struct nouveau_fb *, u64 size, u32 align, u32 size_nc, u32 type, struct nouveau_mem **); void (*put)(struct nouveau_fb *, struct nouveau_mem **); + int (*calc)(struct nouveau_fb *, u32 freq); + int (*prog)(struct nouveau_fb *); + void (*tidy)(struct nouveau_fb *); }; #endif diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/base.c b/drivers/gpu/drm/nouveau/core/subdev/clock/base.c new file mode 100644 index 000000000000..e2938a21b06f --- /dev/null +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/base.c @@ -0,0 +1,494 @@ +/* + * Copyright 2013 Red Hat Inc. + * + * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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: Ben Skeggs + */ + +#include + +#include +#include +#include +#include + +#include +#include +#include +#include + +/****************************************************************************** + * misc + *****************************************************************************/ +static u32 +nouveau_clock_adjust(struct nouveau_clock *clk, bool adjust, + u8 pstate, u8 domain, u32 input) +{ + struct nouveau_bios *bios = nouveau_bios(clk); + struct nvbios_boostE boostE; + u8 ver, hdr, cnt, len; + u16 data; + + data = nvbios_boostEm(bios, pstate, &ver, &hdr, &cnt, &len, &boostE); + if (data) { + struct nvbios_boostS boostS; + u8 idx = 0, sver, shdr; + u16 subd; + + input = max(boostE.min, input); + input = min(boostE.max, input); + do { + sver = ver; + shdr = hdr; + subd = nvbios_boostSp(bios, idx++, data, &sver, &shdr, + cnt, len, &boostS); + if (subd && boostS.domain == domain) { + if (adjust) + input = input * boostS.percent / 100; + input = max(boostS.min, input); + input = min(boostS.max, input); + break; + } + } while (subd); + } + + return input; +} + +/****************************************************************************** + * C-States + *****************************************************************************/ +static int +nouveau_cstate_prog(struct nouveau_clock *clk, + struct nouveau_pstate *pstate, int cstatei) +{ + struct nouveau_therm *ptherm = nouveau_therm(clk); + struct nouveau_volt *volt = nouveau_volt(clk); + struct nouveau_cstate *cstate; + int ret; + + if (!list_empty(&pstate->list)) { + cstate = list_entry(pstate->list.prev, typeof(*cstate), head); + } else { + 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; + } + + 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); + if (ret == 0) { + ret = clk->prog(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); + + 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; +} + +static void +nouveau_cstate_del(struct nouveau_cstate *cstate) +{ + list_del(&cstate->head); + kfree(cstate); +} + +static int +nouveau_cstate_new(struct nouveau_clock *clk, int idx, + struct nouveau_pstate *pstate) +{ + struct nouveau_bios *bios = nouveau_bios(clk); + struct nouveau_clocks *domain = clk->domains; + struct nouveau_cstate *cstate = NULL; + struct nvbios_cstepX cstepX; + u8 ver, hdr; + u16 data; + + data = nvbios_cstepXp(bios, idx, &ver, &hdr, &cstepX); + if (!data) + return -ENOENT; + + cstate = kzalloc(sizeof(*cstate), GFP_KERNEL); + if (!cstate) + return -ENOMEM; + + *cstate = pstate->base; + cstate->voltage = cstepX.voltage; + + while (domain && domain->name != nv_clk_src_max) { + if (domain->flags & NVKM_CLK_DOM_FLAG_CORE) { + u32 freq = nouveau_clock_adjust(clk, true, + pstate->pstate, + domain->bios, + cstepX.freq); + cstate->domain[domain->name] = freq; + } + domain++; + } + + list_add(&cstate->head, &pstate->list); + return 0; +} + +/****************************************************************************** + * P-States + *****************************************************************************/ +static int +nouveau_pstate_prog(struct nouveau_clock *clk, int pstatei) +{ + struct nouveau_fb *pfb = nouveau_fb(clk); + struct nouveau_pstate *pstate; + int ret, idx = 0; + + list_for_each_entry(pstate, &clk->states, head) { + if (idx++ == pstatei) + break; + } + + nv_debug(clk, "setting performance state %d\n", pstatei); + clk->pstate = pstatei; + + if (pfb->ram->calc) { + ret = pfb->ram->calc(pfb, pstate->base.domain[nv_clk_src_mem]); + if (ret == 0) + ret = pfb->ram->prog(pfb); + pfb->ram->tidy(pfb); + } + + return nouveau_cstate_prog(clk, pstate, 0); +} + +static int +nouveau_pstate_calc(struct nouveau_clock *clk) +{ + int pstate, ret = 0; + + 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 (clk->state_nr && clk->ustate != -1) { + pstate = (clk->ustate < 0) ? clk->astate : clk->ustate; + pstate = min(pstate, clk->state_nr - 1 - clk->tstate); + pstate = max(pstate, clk->dstate); + } else { + pstate = clk->pstate = -1; + } + + nv_trace(clk, "-> %d\n", pstate); + if (pstate != clk->pstate) + ret = nouveau_pstate_prog(clk, pstate); + return ret; +} + +static void +nouveau_pstate_info(struct nouveau_clock *clk, struct nouveau_pstate *pstate) +{ + struct nouveau_clocks *clock = clk->domains - 1; + struct nouveau_cstate *cstate; + char info[3][32] = { "", "", "" }; + char name[4] = "--"; + int i = -1; + + if (pstate->pstate != 0xff) + snprintf(name, sizeof(name), "%02x", pstate->pstate); + + while ((++clock)->name != nv_clk_src_max) { + u32 lo = pstate->base.domain[clock->name]; + u32 hi = lo; + if (hi == 0) + continue; + + nv_debug(clk, "%02x: %10d KHz\n", clock->name, lo); + list_for_each_entry(cstate, &pstate->list, head) { + u32 freq = cstate->domain[clock->name]; + lo = min(lo, freq); + hi = max(hi, freq); + nv_debug(clk, "%10d KHz\n", freq); + } + + if (clock->mname && ++i < ARRAY_SIZE(info)) { + lo /= clock->mdiv; + hi /= clock->mdiv; + if (lo == hi) { + snprintf(info[i], sizeof(info[i]), "%s %d MHz", + clock->mname, lo); + } else { + snprintf(info[i], sizeof(info[i]), + "%s %d-%d MHz", clock->mname, lo, hi); + } + } + } + + nv_info(clk, "%s: %s %s %s\n", name, info[0], info[1], info[2]); +} + +static void +nouveau_pstate_del(struct nouveau_pstate *pstate) +{ + struct nouveau_cstate *cstate, *temp; + + list_for_each_entry_safe(cstate, temp, &pstate->list, head) { + nouveau_cstate_del(cstate); + } + + list_del(&pstate->head); + kfree(pstate); +} + +static int +nouveau_pstate_new(struct nouveau_clock *clk, int idx) +{ + struct nouveau_bios *bios = nouveau_bios(clk); + struct nouveau_clocks *domain = clk->domains - 1; + struct nouveau_pstate *pstate; + struct nouveau_cstate *cstate; + struct nvbios_cstepE cstepE; + struct nvbios_perfE perfE; + u8 ver, hdr, cnt, len; + u16 data; + + data = nvbios_perfEp(bios, idx, &ver, &hdr, &cnt, &len, &perfE); + if (!data) + return -EINVAL; + if (perfE.pstate == 0xff) + return 0; + + pstate = kzalloc(sizeof(*pstate), GFP_KERNEL); + cstate = &pstate->base; + if (!pstate) + return -ENOMEM; + + INIT_LIST_HEAD(&pstate->list); + + pstate->pstate = perfE.pstate; + pstate->fanspeed = perfE.fanspeed; + cstate->voltage = perfE.voltage; + cstate->domain[nv_clk_src_core] = perfE.core; + cstate->domain[nv_clk_src_shader] = perfE.shader; + cstate->domain[nv_clk_src_mem] = perfE.memory; + cstate->domain[nv_clk_src_vdec] = perfE.vdec; + cstate->domain[nv_clk_src_dom6] = perfE.disp; + + while (ver >= 0x40 && (++domain)->name != nv_clk_src_max) { + struct nvbios_perfS perfS; + u8 sver = ver, shdr = hdr; + u32 perfSe = nvbios_perfSp(bios, data, domain->bios, + &sver, &shdr, cnt, len, &perfS); + if (perfSe == 0 || sver != 0x40) + continue; + + if (domain->flags & NVKM_CLK_DOM_FLAG_CORE) { + perfS.v40.freq = nouveau_clock_adjust(clk, false, + pstate->pstate, + domain->bios, + perfS.v40.freq); + } + + cstate->domain[domain->name] = perfS.v40.freq; + } + + data = nvbios_cstepEm(bios, pstate->pstate, &ver, &hdr, &cstepE); + if (data) { + int idx = cstepE.index; + do { + nouveau_cstate_new(clk, idx, pstate); + } while(idx--); + } + + nouveau_pstate_info(clk, pstate); + list_add_tail(&pstate->head, &clk->states); + clk->state_nr++; + return 0; +} + +/****************************************************************************** + * Adjustment triggers + *****************************************************************************/ +static int +nouveau_clock_ustate_update(struct nouveau_clock *clk, int req) +{ + struct nouveau_pstate *pstate; + int i = 0; + + /* YKW repellant */ + return -ENOSYS; + + if (req != -1 && req != -2) { + list_for_each_entry(pstate, &clk->states, head) { + if (pstate->pstate == req) + break; + i++; + } + + if (pstate->pstate != req) + return -EINVAL; + req = i; + } + + clk->ustate = req; + return 0; +} + +int +nouveau_clock_ustate(struct nouveau_clock *clk, int req) +{ + int ret = nouveau_clock_ustate_update(clk, req); + if (ret) + return ret; + return nouveau_pstate_calc(clk); +} + +int +nouveau_clock_astate(struct nouveau_clock *clk, int req, int rel) +{ + if (!rel) clk->astate = req; + 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); +} + +int +nouveau_clock_tstate(struct nouveau_clock *clk, int req, int rel) +{ + if (!rel) clk->tstate = req; + 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); +} + +int +nouveau_clock_dstate(struct nouveau_clock *clk, int req, int rel) +{ + if (!rel) clk->dstate = req; + 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); +} + +/****************************************************************************** + * subdev base class implementation + *****************************************************************************/ +int +_nouveau_clock_init(struct nouveau_object *object) +{ + struct nouveau_clock *clk = (void *)object; + struct nouveau_clocks *clock = clk->domains; + int ret; + + memset(&clk->bstate, 0x00, sizeof(clk->bstate)); + INIT_LIST_HEAD(&clk->bstate.list); + clk->bstate.pstate = 0xff; + + while (clock->name != nv_clk_src_max) { + ret = clk->read(clk, clock->name); + if (ret < 0) { + nv_error(clk, "%02x freq unknown\n", clock->name); + return ret; + } + clk->bstate.base.domain[clock->name] = ret; + clock++; + } + + nouveau_pstate_info(clk, &clk->bstate); + + clk->astate = clk->state_nr - 1; + clk->tstate = 0; + clk->dstate = 0; + clk->pstate = -1; + nouveau_pstate_calc(clk); + return 0; +} + +void +_nouveau_clock_dtor(struct nouveau_object *object) +{ + struct nouveau_clock *clk = (void *)object; + struct nouveau_pstate *pstate, *temp; + + list_for_each_entry_safe(pstate, temp, &clk->states, head) { + nouveau_pstate_del(pstate); + } + + nouveau_subdev_destroy(&clk->base); +} + +int +nouveau_clock_create_(struct nouveau_object *parent, + struct nouveau_object *engine, + struct nouveau_oclass *oclass, + struct nouveau_clocks *clocks, + int length, void **object) +{ + struct nouveau_device *device = nv_device(parent); + struct nouveau_clock *clk; + int ret, idx, arglen; + const char *mode; + + ret = nouveau_subdev_create_(parent, engine, oclass, 0, "CLK", + "clock", length, object); + clk = *object; + if (ret) + return ret; + + INIT_LIST_HEAD(&clk->states); + clk->domains = clocks; + clk->ustate = -1; + + idx = 0; + do { + ret = nouveau_pstate_new(clk, idx++); + } while (ret == 0); + + 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; + } + } + + return 0; +} diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c index a14277586595..da50c1b12928 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nv04.c @@ -77,7 +77,7 @@ 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, &priv); + ret = nouveau_clock_create(parent, engine, oclass, NULL, &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 0db5dbfd91b5..db7346f79080 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nv40.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nv40.c @@ -23,11 +23,188 @@ */ #include +#include +#include + +#include "pll.h" struct nv40_clock_priv { struct nouveau_clock base; + u32 ctrl; + u32 npll_ctrl; + u32 npll_coef; + u32 spll; +}; + +static struct nouveau_clocks +nv40_domain[] = { + { nv_clk_src_crystal, 0xff }, + { nv_clk_src_href , 0xff }, + { nv_clk_src_core , 0xff, 0, "core", 1000 }, + { nv_clk_src_shader , 0xff, 0, "shader", 1000 }, + { nv_clk_src_mem , 0xff, 0, "memory", 1000 }, + { nv_clk_src_max } }; +static u32 +read_pll_1(struct nv40_clock_priv *priv, u32 reg) +{ + u32 ctrl = nv_rd32(priv, reg + 0x00); + int P = (ctrl & 0x00070000) >> 16; + int N = (ctrl & 0x0000ff00) >> 8; + int M = (ctrl & 0x000000ff) >> 0; + u32 ref = 27000, clk = 0; + + if (ctrl & 0x80000000) + clk = ref * N / M; + + return clk >> P; +} + +static u32 +read_pll_2(struct nv40_clock_priv *priv, u32 reg) +{ + u32 ctrl = nv_rd32(priv, reg + 0x00); + u32 coef = nv_rd32(priv, reg + 0x04); + int N2 = (coef & 0xff000000) >> 24; + int M2 = (coef & 0x00ff0000) >> 16; + int N1 = (coef & 0x0000ff00) >> 8; + int M1 = (coef & 0x000000ff) >> 0; + int P = (ctrl & 0x00070000) >> 16; + u32 ref = 27000, clk = 0; + + if ((ctrl & 0x80000000) && M1) { + clk = ref * N1 / M1; + if ((ctrl & 0x40000100) == 0x40000000) { + if (M2) + clk = clk * N2 / M2; + else + clk = 0; + } + } + + return clk >> P; +} + +static u32 +read_clk(struct nv40_clock_priv *priv, u32 src) +{ + switch (src) { + case 3: + return read_pll_2(priv, 0x004000); + case 2: + return read_pll_1(priv, 0x004008); + default: + break; + } + + return 0; +} + +static int +nv40_clock_read(struct nouveau_clock *clk, enum nv_clk_src src) +{ + struct nv40_clock_priv *priv = (void *)clk; + u32 mast = nv_rd32(priv, 0x00c040); + + switch (src) { + case nv_clk_src_crystal: + return nv_device(priv)->crystal; + case nv_clk_src_href: + return 100000; /*XXX: PCIE/AGP differ*/ + case nv_clk_src_core: + return read_clk(priv, (mast & 0x00000003) >> 0); + case nv_clk_src_shader: + return read_clk(priv, (mast & 0x00000030) >> 4); + case nv_clk_src_mem: + return read_pll_2(priv, 0x4020); + default: + break; + } + + nv_debug(priv, "unknown clock source %d 0x%08x\n", src, mast); + return -EINVAL; +} + +static int +nv40_clock_calc_pll(struct nv40_clock_priv *priv, u32 reg, u32 clk, + int *N1, int *M1, int *N2, int *M2, int *log2P) +{ + struct nouveau_bios *bios = nouveau_bios(priv); + struct nvbios_pll pll; + int ret; + + ret = nvbios_pll_parse(bios, reg, &pll); + if (ret) + return ret; + + if (clk < pll.vco1.max_freq) + pll.vco2.max_freq = 0; + + ret = nv04_pll_calc(nv_subdev(priv), &pll, clk, N1, M1, N2, M2, log2P); + if (ret == 0) + return -ERANGE; + return ret; +} + +static int +nv40_clock_calc(struct nouveau_clock *clk, struct nouveau_cstate *cstate) +{ + struct nv40_clock_priv *priv = (void *)clk; + int gclk = cstate->domain[nv_clk_src_core]; + int sclk = cstate->domain[nv_clk_src_shader]; + int N1, M1, N2, M2, log2P; + int ret; + + /* core/geometric clock */ + ret = nv40_clock_calc_pll(priv, 0x004000, gclk, + &N1, &M1, &N2, &M2, &log2P); + if (ret < 0) + return ret; + + if (N2 == M2) { + priv->npll_ctrl = 0x80000100 | (log2P << 16); + priv->npll_coef = (N1 << 8) | M1; + } else { + priv->npll_ctrl = 0xc0000000 | (log2P << 16); + priv->npll_coef = (N2 << 24) | (M2 << 16) | (N1 << 8) | M1; + } + + /* use the second pll for shader/rop clock, if it differs from core */ + if (sclk && sclk != gclk) { + ret = nv40_clock_calc_pll(priv, 0x004008, sclk, + &N1, &M1, NULL, NULL, &log2P); + if (ret < 0) + return ret; + + priv->spll = 0xc0000000 | (log2P << 16) | (N1 << 8) | M1; + priv->ctrl = 0x00000223; + } else { + priv->spll = 0x00000000; + priv->ctrl = 0x00000333; + } + + return 0; +} + +static int +nv40_clock_prog(struct nouveau_clock *clk) +{ + struct nv40_clock_priv *priv = (void *)clk; + nv_mask(priv, 0x00c040, 0x00000333, 0x00000000); + nv_wr32(priv, 0x004004, priv->npll_coef); + nv_mask(priv, 0x004000, 0xc0070100, priv->npll_ctrl); + nv_mask(priv, 0x004008, 0xc007ffff, priv->spll); + mdelay(5); + nv_mask(priv, 0x00c040, 0x00000333, priv->ctrl); + return 0; +} + +static void +nv40_clock_tidy(struct nouveau_clock *clk) +{ +} + static int nv40_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, struct nouveau_oclass *oclass, void *data, u32 size, @@ -36,13 +213,17 @@ 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, &priv); + ret = nouveau_clock_create(parent, engine, oclass, nv40_domain, &priv); *pobject = nv_object(priv); if (ret) return ret; priv->base.pll_calc = nv04_clock_pll_calc; priv->base.pll_prog = nv04_clock_pll_prog; + priv->base.read = nv40_clock_read; + priv->base.calc = nv40_clock_calc; + priv->base.prog = nv40_clock_prog; + priv->base.tidy = nv40_clock_tidy; return 0; } diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c index d09d3e78040c..1a016df4e0e8 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.c @@ -22,40 +22,538 @@ * Authors: Ben Skeggs */ -#include #include #include +#include "nv50.h" #include "pll.h" +#include "seq.h" -struct nv50_clock_priv { - struct nouveau_clock base; -}; +static u32 +read_div(struct nv50_clock_priv *priv) +{ + switch (nv_device(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(priv, 0x004700); + case 0x92: + case 0x94: + case 0x96: + return nv_rd32(priv, 0x004800); + default: + return 0x00000000; + } +} + +static u32 +read_pll_src(struct nv50_clock_priv *priv, u32 base) +{ + struct nouveau_clock *clk = &priv->base; + u32 coef, ref = clk->read(clk, nv_clk_src_crystal); + u32 rsel = nv_rd32(priv, 0x00e18c); + int P, N, M, id; + + switch (nv_device(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(priv, "ref: bad pll 0x%06x\n", base); + return 0; + } + + coef = nv_rd32(priv, 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(priv, 0x00e81c); + P = (coef & 0x00070000) >> 16; + N = (coef & 0x0000ff00) >> 8; + M = (coef & 0x000000ff) >> 0; + break; + case 0x94: + case 0x96: + case 0x98: + rsel = nv_rd32(priv, 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(priv, "ref: bad pll 0x%06x\n", base); + return 0; + } + + switch (rsel) { + case 0: id = 1; break; + case 1: return clk->read(clk, nv_clk_src_crystal); + case 2: return clk->read(clk, nv_clk_src_href); + case 3: id = 0; break; + } + + coef = nv_rd32(priv, 0x00e81c + (id * 0x28)); + P = (nv_rd32(priv, 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_ref(struct nv50_clock_priv *priv, u32 base) +{ + struct nouveau_clock *clk = &priv->base; + u32 src, mast = nv_rd32(priv, 0x00c040); + + switch (base) { + case 0x004028: + src = !!(mast & 0x00200000); + break; + case 0x004020: + src = !!(mast & 0x00400000); + break; + case 0x004008: + src = !!(mast & 0x00010000); + break; + case 0x004030: + src = !!(mast & 0x02000000); + break; + case 0x00e810: + return clk->read(clk, nv_clk_src_crystal); + default: + nv_error(priv, "bad pll 0x%06x\n", base); + return 0; + } + + if (src) + return clk->read(clk, nv_clk_src_href); + return read_pll_src(priv, base); +} + +static u32 +read_pll(struct nv50_clock_priv *priv, u32 base) +{ + struct nouveau_clock *clk = &priv->base; + u32 mast = nv_rd32(priv, 0x00c040); + u32 ctrl = nv_rd32(priv, base + 0); + u32 coef = nv_rd32(priv, base + 4); + u32 ref = read_pll_ref(priv, base); + u32 freq = 0; + int N1, N2, M1, M2; + + if (base == 0x004028 && (mast & 0x00100000)) { + /* wtf, appears to only disable post-divider on nva0 */ + if (nv_device(priv)->chipset != 0xa0) + return clk->read(clk, nv_clk_src_dom6); + } + + N2 = (coef & 0xff000000) >> 24; + M2 = (coef & 0x00ff0000) >> 16; + N1 = (coef & 0x0000ff00) >> 8; + M1 = (coef & 0x000000ff); + if ((ctrl & 0x80000000) && M1) { + freq = ref * N1 / M1; + if ((ctrl & 0x40000100) == 0x40000000) { + if (M2) + freq = freq * N2 / M2; + else + freq = 0; + } + } + + return freq; +} static int +nv50_clock_read(struct nouveau_clock *clk, enum nv_clk_src src) +{ + struct nv50_clock_priv *priv = (void *)clk; + u32 mast = nv_rd32(priv, 0x00c040); + u32 P = 0; + + switch (src) { + case nv_clk_src_crystal: + return nv_device(priv)->crystal; + case nv_clk_src_href: + return 100000; /* PCIE reference clock */ + case nv_clk_src_hclk: + return (u64)clk->read(clk, nv_clk_src_href) * 27778 / 10000; + case nv_clk_src_hclkm3: + return clk->read(clk, nv_clk_src_hclk) * 3; + case nv_clk_src_hclkm3d2: + return clk->read(clk, nv_clk_src_hclk) * 3 / 2; + case nv_clk_src_host: + switch (mast & 0x30000000) { + case 0x00000000: return clk->read(clk, nv_clk_src_href); + case 0x10000000: break; + case 0x20000000: /* !0x50 */ + case 0x30000000: return clk->read(clk, nv_clk_src_hclk); + } + break; + case nv_clk_src_core: + if (!(mast & 0x00100000)) + P = (nv_rd32(priv, 0x004028) & 0x00070000) >> 16; + switch (mast & 0x00000003) { + case 0x00000000: return clk->read(clk, nv_clk_src_crystal) >> P; + case 0x00000001: return clk->read(clk, nv_clk_src_dom6); + case 0x00000002: return read_pll(priv, 0x004020) >> P; + case 0x00000003: return read_pll(priv, 0x004028) >> P; + } + break; + case nv_clk_src_shader: + P = (nv_rd32(priv, 0x004020) & 0x00070000) >> 16; + switch (mast & 0x00000030) { + case 0x00000000: + if (mast & 0x00000080) + return clk->read(clk, nv_clk_src_host) >> P; + return clk->read(clk, nv_clk_src_crystal) >> P; + case 0x00000010: break; + case 0x00000020: return read_pll(priv, 0x004028) >> P; + case 0x00000030: return read_pll(priv, 0x004020) >> P; + } + break; + case nv_clk_src_mem: + P = (nv_rd32(priv, 0x004008) & 0x00070000) >> 16; + if (nv_rd32(priv, 0x004008) & 0x00000200) { + switch (mast & 0x0000c000) { + case 0x00000000: + return clk->read(clk, nv_clk_src_crystal) >> P; + case 0x00008000: + case 0x0000c000: + return clk->read(clk, nv_clk_src_href) >> P; + } + } else { + return read_pll(priv, 0x004008) >> P; + } + break; + case nv_clk_src_vdec: + P = (read_div(priv) & 0x00000700) >> 8; + switch (nv_device(priv)->chipset) { + case 0x84: + case 0x86: + case 0x92: + case 0x94: + case 0x96: + case 0xa0: + switch (mast & 0x00000c00) { + case 0x00000000: + if (nv_device(priv)->chipset == 0xa0) /* wtf?? */ + return clk->read(clk, nv_clk_src_core) >> P; + return clk->read(clk, nv_clk_src_crystal) >> P; + case 0x00000400: + return 0; + case 0x00000800: + if (mast & 0x01000000) + return read_pll(priv, 0x004028) >> P; + return read_pll(priv, 0x004030) >> P; + case 0x00000c00: + return clk->read(clk, nv_clk_src_core) >> P; + } + break; + case 0x98: + switch (mast & 0x00000c00) { + case 0x00000000: + return clk->read(clk, nv_clk_src_core) >> P; + case 0x00000400: + return 0; + case 0x00000800: + return clk->read(clk, nv_clk_src_hclkm3d2) >> P; + case 0x00000c00: + return clk->read(clk, nv_clk_src_mem) >> P; + } + break; + } + break; + case nv_clk_src_dom6: + switch (nv_device(priv)->chipset) { + case 0x50: + case 0xa0: + return read_pll(priv, 0x00e810) >> 2; + case 0x84: + case 0x86: + case 0x92: + case 0x94: + case 0x96: + case 0x98: + P = (read_div(priv) & 0x00000007) >> 0; + switch (mast & 0x0c000000) { + case 0x00000000: return clk->read(clk, nv_clk_src_href); + case 0x04000000: break; + case 0x08000000: return clk->read(clk, nv_clk_src_hclk); + case 0x0c000000: + return clk->read(clk, nv_clk_src_hclkm3) >> P; + } + break; + default: + break; + } + default: + break; + } + + nv_debug(priv, "unknown clock source %d 0x%08x\n", src, mast); + return -EINVAL; +} + +static u32 +calc_pll(struct nv50_clock_priv *priv, u32 reg, u32 clk, int *N, int *M, int *P) +{ + struct nouveau_bios *bios = nouveau_bios(priv); + struct nvbios_pll pll; + int ret; + + ret = nvbios_pll_parse(bios, reg, &pll); + if (ret) + return 0; + + pll.vco2.max_freq = 0; + pll.refclk = read_pll_ref(priv, reg); + if (!pll.refclk) + return 0; + + return nv04_pll_calc(nv_subdev(priv), &pll, clk, N, M, NULL, NULL, P); +} + +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; + } + + if (target - clk0 <= clk1 - target) + return clk0; + (*div)--; + return clk1; +} + +static inline u32 +clk_same(u32 a, u32 b) +{ + return ((a / 1000) == (b / 1000)); +} + +static int +nv50_clock_calc(struct nouveau_clock *clk, struct nouveau_cstate *cstate) +{ + struct nv50_clock_priv *priv = (void *)clk; + struct nv50_clock_hwsq *hwsq = &priv->hwsq; + const int shader = cstate->domain[nv_clk_src_shader]; + const int core = cstate->domain[nv_clk_src_core]; + const int vdec = cstate->domain[nv_clk_src_vdec]; + const int dom6 = cstate->domain[nv_clk_src_dom6]; + u32 mastm = 0, mastv = 0; + u32 divsm = 0, divsv = 0; + int N, M, P1, P2; + int freq, out; + + /* prepare a hwsq script from which we'll perform the reclock */ + out = clk_init(hwsq, nv_subdev(clk)); + if (out) + return out; + + clk_wr32(hwsq, fifo, 0x00000001); /* block fifo */ + clk_nsec(hwsq, 8000); + clk_setf(hwsq, 0x10, 0x00); /* disable fb */ + clk_wait(hwsq, 0x00, 0x01); /* wait for fb disabled */ + + /* 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... + */ + if (vdec) { + /* see how close we can get using nvclk as a source */ + freq = calc_div(core, vdec, &P1); + + /* see how close we can get using xpll/hclk as a source */ + if (nv_device(priv)->chipset != 0x98) + out = read_pll(priv, 0x004030); + else + out = clk->read(clk, nv_clk_src_hclkm3d2); + out = calc_div(out, vdec, &P2); + + /* select whichever gets us closest */ + if (abs(vdec - freq) <= abs(vdec - out)) { + if (nv_device(priv)->chipset != 0x98) + mastv |= 0x00000c00; + divsv |= P1 << 8; + } else { + mastv |= 0x00000800; + divsv |= P2 << 8; + } + + mastm |= 0x00000c00; + divsm |= 0x00000700; + } + + /* dom6: nfi what this is, but we're limited to various combinations + * of the host clock frequency + */ + if (dom6) { + if (clk_same(dom6, clk->read(clk, nv_clk_src_href))) { + mastv |= 0x00000000; + } else + if (clk_same(dom6, clk->read(clk, nv_clk_src_hclk))) { + mastv |= 0x08000000; + } else { + freq = clk->read(clk, nv_clk_src_hclk) * 3; + freq = calc_div(freq, dom6, &P1); + + mastv |= 0x0c000000; + divsv |= P1; + } + + mastm |= 0x0c000000; + divsm |= 0x00000007; + } + + /* vdec/dom6: switch to "safe" clocks temporarily, update dividers + * and then switch to target clocks + */ + clk_mask(hwsq, mast, mastm, 0x00000000); + clk_mask(hwsq, divs, divsm, divsv); + clk_mask(hwsq, mast, mastm, mastv); + + /* core/shader: disconnect nvclk/sclk from their PLLs (nvclk to dom6, + * sclk to hclk) before reprogramming + */ + if (nv_device(priv)->chipset < 0x92) + clk_mask(hwsq, mast, 0x001000b0, 0x00100080); + else + clk_mask(hwsq, mast, 0x000000b3, 0x00000081); + + /* core: for the moment at least, always use nvpll */ + freq = calc_pll(priv, 0x4028, core, &N, &M, &P1); + if (freq == 0) + return -ERANGE; + + clk_mask(hwsq, nvpll[0], 0xc03f0100, + 0x80000000 | (P1 << 19) | (P1 << 16)); + clk_mask(hwsq, nvpll[1], 0x0000ffff, (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-- && shader == (core << 1)) { + clk_mask(hwsq, spll[0], 0xc03f0100, (P1 << 19) | (P1 << 16)); + clk_mask(hwsq, mast, 0x00100033, 0x00000023); + } else { + freq = calc_pll(priv, 0x4020, shader, &N, &M, &P1); + if (freq == 0) + return -ERANGE; + + clk_mask(hwsq, spll[0], 0xc03f0100, + 0x80000000 | (P1 << 19) | (P1 << 16)); + clk_mask(hwsq, spll[1], 0x0000ffff, (N << 8) | M); + clk_mask(hwsq, mast, 0x00100033, 0x00000033); + } + + /* restore normal operation */ + clk_setf(hwsq, 0x10, 0x01); /* enable fb */ + clk_wait(hwsq, 0x00, 0x00); /* wait for fb enabled */ + clk_wr32(hwsq, fifo, 0x00000000); /* un-block fifo */ + return 0; +} + +static int +nv50_clock_prog(struct nouveau_clock *clk) +{ + struct nv50_clock_priv *priv = (void *)clk; + return clk_exec(&priv->hwsq, true); +} + +static void +nv50_clock_tidy(struct nouveau_clock *clk) +{ + struct nv50_clock_priv *priv = (void *)clk; + clk_exec(&priv->hwsq, false); +} + +int nv50_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, struct nouveau_oclass *oclass, void *data, u32 size, struct nouveau_object **pobject) { + struct nv50_clock_oclass *pclass = (void *)oclass; struct nv50_clock_priv *priv; int ret; - ret = nouveau_clock_create(parent, engine, oclass, &priv); + ret = nouveau_clock_create(parent, engine, oclass, pclass->domains, + &priv); *pobject = nv_object(priv); if (ret) return ret; - priv->base.pll_calc = nv04_clock_pll_calc; + priv->hwsq.r_fifo = hwsq_reg(0x002504); + priv->hwsq.r_spll[0] = hwsq_reg(0x004020); + priv->hwsq.r_spll[1] = hwsq_reg(0x004024); + priv->hwsq.r_nvpll[0] = hwsq_reg(0x004028); + priv->hwsq.r_nvpll[1] = hwsq_reg(0x00402c); + switch (nv_device(priv)->chipset) { + case 0x92: + case 0x94: + case 0x96: + priv->hwsq.r_divs = hwsq_reg(0x004800); + break; + default: + priv->hwsq.r_divs = hwsq_reg(0x004700); + break; + } + priv->hwsq.r_mast = hwsq_reg(0x00c040); + + priv->base.read = nv50_clock_read; + priv->base.calc = nv50_clock_calc; + priv->base.prog = nv50_clock_prog; + priv->base.tidy = nv50_clock_tidy; return 0; } -struct nouveau_oclass -nv50_clock_oclass = { - .handle = NV_SUBDEV(CLOCK, 0x50), - .ofuncs = &(struct nouveau_ofuncs) { +static struct nouveau_clocks +nv50_domains[] = { + { nv_clk_src_crystal, 0xff }, + { nv_clk_src_href , 0xff }, + { nv_clk_src_core , 0xff, 0, "core", 1000 }, + { nv_clk_src_shader , 0xff, 0, "shader", 1000 }, + { nv_clk_src_mem , 0xff, 0, "memory", 1000 }, + { nv_clk_src_max } +}; + +struct nouveau_oclass * +nv50_clock_oclass = &(struct nv50_clock_oclass) { + .base.handle = NV_SUBDEV(CLOCK, 0x50), + .base.ofuncs = &(struct nouveau_ofuncs) { .ctor = nv50_clock_ctor, .dtor = _nouveau_clock_dtor, .init = _nouveau_clock_init, .fini = _nouveau_clock_fini, }, -}; + .domains = nv50_domains, +}.base; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.h b/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.h new file mode 100644 index 000000000000..f10917d789e8 --- /dev/null +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nv50.h @@ -0,0 +1,31 @@ +#ifndef __NVKM_CLK_NV50_H__ +#define __NVKM_CLK_NV50_H__ + +#include +#include +#include + +struct nv50_clock_hwsq { + struct hwsq base; + struct hwsq_reg r_fifo; + struct hwsq_reg r_spll[2]; + struct hwsq_reg r_nvpll[2]; + struct hwsq_reg r_divs; + struct hwsq_reg r_mast; +}; + +struct nv50_clock_priv { + struct nouveau_clock base; + struct nv50_clock_hwsq hwsq; +}; + +int nv50_clock_ctor(struct nouveau_object *, struct nouveau_object *, + struct nouveau_oclass *, void *, u32, + struct nouveau_object **); + +struct nv50_clock_oclass { + struct nouveau_oclass base; + struct nouveau_clocks *domains; +}; + +#endif diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nv84.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nv84.c new file mode 100644 index 000000000000..b0b7c1437f10 --- /dev/null +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nv84.c @@ -0,0 +1,48 @@ +/* + * Copyright 2013 Red Hat Inc. + * + * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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: Ben Skeggs + */ + +#include "nv50.h" + +static struct nouveau_clocks +nv84_domains[] = { + { nv_clk_src_crystal, 0xff }, + { nv_clk_src_href , 0xff }, + { nv_clk_src_core , 0xff, 0, "core", 1000 }, + { nv_clk_src_shader , 0xff, 0, "shader", 1000 }, + { nv_clk_src_mem , 0xff, 0, "memory", 1000 }, + { nv_clk_src_vdec , 0xff }, + { nv_clk_src_max } +}; + +struct nouveau_oclass * +nv84_clock_oclass = &(struct nv50_clock_oclass) { + .base.handle = NV_SUBDEV(CLOCK, 0x84), + .base.ofuncs = &(struct nouveau_ofuncs) { + .ctor = nv50_clock_ctor, + .dtor = _nouveau_clock_dtor, + .init = _nouveau_clock_init, + .fini = _nouveau_clock_fini, + }, + .domains = nv84_domains, +}.base; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c index f074cd20bc9c..4f5a1373f002 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.c @@ -22,33 +22,277 @@ * Authors: Ben Skeggs */ -#include #include #include +#include #include "pll.h" +#include "nva3.h" + struct nva3_clock_priv { struct nouveau_clock base; + struct nva3_clock_info eng[nv_clk_src_max]; }; +static u32 read_clk(struct nva3_clock_priv *, int, bool); +static u32 read_pll(struct nva3_clock_priv *, int, u32); + +static u32 +read_vco(struct nva3_clock_priv *priv, int clk) +{ + u32 sctl = nv_rd32(priv, 0x4120 + (clk * 4)); + if ((sctl & 0x00000030) != 0x00000030) + return read_pll(priv, 0x41, 0x00e820); + return read_pll(priv, 0x42, 0x00e8a0); +} + +static u32 +read_clk(struct nva3_clock_priv *priv, int clk, bool ignore_en) +{ + u32 sctl, sdiv, sclk; + + /* refclk for the 0xe8xx plls is a fixed frequency */ + if (clk >= 0x40) { + if (nv_device(priv)->chipset == 0xaf) { + /* no joke.. seriously.. sigh.. */ + return nv_rd32(priv, 0x00471c) * 1000; + } + + return nv_device(priv)->crystal; + } + + sctl = nv_rd32(priv, 0x4120 + (clk * 4)); + if (!ignore_en && !(sctl & 0x00000100)) + return 0; + + switch (sctl & 0x00003000) { + case 0x00000000: + return nv_device(priv)->crystal; + case 0x00002000: + if (sctl & 0x00000040) + return 108000; + return 100000; + case 0x00003000: + sclk = read_vco(priv, clk); + sdiv = ((sctl & 0x003f0000) >> 16) + 2; + return (sclk * 2) / sdiv; + default: + return 0; + } +} + +static u32 +read_pll(struct nva3_clock_priv *priv, int clk, u32 pll) +{ + u32 ctrl = nv_rd32(priv, pll + 0); + u32 sclk = 0, P = 1, N = 1, M = 1; + + if (!(ctrl & 0x00000008)) { + if (ctrl & 0x00000001) { + u32 coef = nv_rd32(priv, pll + 4); + M = (coef & 0x000000ff) >> 0; + N = (coef & 0x0000ff00) >> 8; + P = (coef & 0x003f0000) >> 16; + + /* no post-divider on these.. */ + if ((pll & 0x00ff00) == 0x00e800) + P = 1; + + sclk = read_clk(priv, 0x00 + clk, false); + } + } else { + sclk = read_clk(priv, 0x10 + clk, false); + } + + if (M * P) + return sclk * N / (M * P); + return 0; +} + +static int +nva3_clock_read(struct nouveau_clock *clk, enum nv_clk_src src) +{ + struct nva3_clock_priv *priv = (void *)clk; + + switch (src) { + case nv_clk_src_crystal: + return nv_device(priv)->crystal; + case nv_clk_src_href: + return 100000; + case nv_clk_src_core: + return read_pll(priv, 0x00, 0x4200); + case nv_clk_src_shader: + return read_pll(priv, 0x01, 0x4220); + case nv_clk_src_mem: + return read_pll(priv, 0x02, 0x4000); + case nv_clk_src_disp: + return read_clk(priv, 0x20, false); + case nv_clk_src_vdec: + return read_clk(priv, 0x21, false); + case nv_clk_src_daemon: + return read_clk(priv, 0x25, false); + default: + nv_error(clk, "invalid clock source %d\n", src); + return -EINVAL; + } +} + int -nva3_clock_pll_calc(struct nouveau_clock *clock, struct nvbios_pll *info, - int clk, struct nouveau_pll_vals *pv) +nva3_clock_info(struct nouveau_clock *clock, int clk, u32 pll, u32 khz, + struct nva3_clock_info *info) { - int ret, N, M, P; + struct nouveau_bios *bios = nouveau_bios(clock); + struct nva3_clock_priv *priv = (void *)clock; + struct nvbios_pll limits; + u32 oclk, sclk, sdiv; + int P, N, M, diff; + int ret; + + info->pll = 0; + info->clk = 0; + + switch (khz) { + case 27000: + info->clk = 0x00000100; + return khz; + case 100000: + info->clk = 0x00002100; + return khz; + case 108000: + info->clk = 0x00002140; + return khz; + default: + sclk = read_vco(priv, clk); + sdiv = min((sclk * 2) / (khz - 2999), (u32)65); + /* if the clock has a PLL attached, and we can get a within + * [-2, 3) MHz of a divider, we'll disable the PLL and use + * the divider instead. + * + * divider can go as low as 2, limited here because NVIDIA + * and the VBIOS on my NVA8 seem to prefer using the PLL + * for 810MHz - is there a good reason? + */ + if (sdiv > 4) { + oclk = (sclk * 2) / sdiv; + diff = khz - oclk; + if (!pll || (diff >= -2000 && diff < 3000)) { + info->clk = (((sdiv - 2) << 16) | 0x00003100); + return oclk; + } + } + + if (!pll) + return -ERANGE; + break; + } - ret = nva3_pll_calc(nv_subdev(clock), info, clk, &N, NULL, &M, &P); + ret = nvbios_pll_parse(bios, pll, &limits); + if (ret) + return ret; + + limits.refclk = read_clk(priv, clk - 0x10, true); + if (!limits.refclk) + return -EINVAL; - if (ret > 0) { - pv->refclk = info->refclk; - pv->N1 = N; - pv->M1 = M; - pv->log2P = P; + ret = nva3_pll_calc(nv_subdev(priv), &limits, khz, &N, NULL, &M, &P); + if (ret >= 0) { + info->clk = nv_rd32(priv, 0x4120 + (clk * 4)); + info->pll = (P << 16) | (N << 8) | M; } + + return ret ? ret : -ERANGE; +} + +static int +calc_clk(struct nva3_clock_priv *priv, struct nouveau_cstate *cstate, + int clk, u32 pll, int idx) +{ + int ret = nva3_clock_info(&priv->base, clk, pll, cstate->domain[idx], + &priv->eng[idx]); + if (ret >= 0) + return 0; return ret; } +static void +prog_pll(struct nva3_clock_priv *priv, int clk, u32 pll, int idx) +{ + struct nva3_clock_info *info = &priv->eng[idx]; + const u32 src0 = 0x004120 + (clk * 4); + const u32 src1 = 0x004160 + (clk * 4); + const u32 ctrl = pll + 0; + const u32 coef = pll + 4; + + if (info->pll) { + nv_mask(priv, src0, 0x00000101, 0x00000101); + nv_wr32(priv, coef, info->pll); + nv_mask(priv, ctrl, 0x00000015, 0x00000015); + nv_mask(priv, ctrl, 0x00000010, 0x00000000); + nv_wait(priv, ctrl, 0x00020000, 0x00020000); + nv_mask(priv, ctrl, 0x00000010, 0x00000010); + nv_mask(priv, ctrl, 0x00000008, 0x00000000); + nv_mask(priv, src1, 0x00000100, 0x00000000); + nv_mask(priv, src1, 0x00000001, 0x00000000); + } else { + nv_mask(priv, src1, 0x003f3141, 0x00000101 | info->clk); + nv_mask(priv, ctrl, 0x00000018, 0x00000018); + udelay(20); + nv_mask(priv, ctrl, 0x00000001, 0x00000000); + nv_mask(priv, src0, 0x00000100, 0x00000000); + nv_mask(priv, src0, 0x00000001, 0x00000000); + } +} + +static void +prog_clk(struct nva3_clock_priv *priv, int clk, int idx) +{ + struct nva3_clock_info *info = &priv->eng[idx]; + nv_mask(priv, 0x004120 + (clk * 4), 0x003f3141, 0x00000101 | info->clk); +} + +static int +nva3_clock_calc(struct nouveau_clock *clk, struct nouveau_cstate *cstate) +{ + struct nva3_clock_priv *priv = (void *)clk; + int ret; + + if ((ret = calc_clk(priv, cstate, 0x10, 0x4200, nv_clk_src_core)) || + (ret = calc_clk(priv, cstate, 0x11, 0x4220, nv_clk_src_shader)) || + (ret = calc_clk(priv, cstate, 0x20, 0x0000, nv_clk_src_disp)) || + (ret = calc_clk(priv, cstate, 0x21, 0x0000, nv_clk_src_vdec))) + return ret; + + return 0; +} + +static int +nva3_clock_prog(struct nouveau_clock *clk) +{ + struct nva3_clock_priv *priv = (void *)clk; + prog_pll(priv, 0x00, 0x004200, nv_clk_src_core); + prog_pll(priv, 0x01, 0x004220, nv_clk_src_shader); + prog_clk(priv, 0x20, nv_clk_src_disp); + prog_clk(priv, 0x21, nv_clk_src_vdec); + return 0; +} + +static void +nva3_clock_tidy(struct nouveau_clock *clk) +{ +} + +static struct nouveau_clocks +nva3_domain[] = { + { nv_clk_src_crystal, 0xff }, + { nv_clk_src_href , 0xff }, + { nv_clk_src_core , 0x00, 0, "core", 1000 }, + { nv_clk_src_shader , 0x01, 0, "shader", 1000 }, + { nv_clk_src_mem , 0x02, 0, "memory", 1000 }, + { nv_clk_src_vdec , 0x03 }, + { nv_clk_src_disp , 0x04 }, + { nv_clk_src_max } +}; static int nva3_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, @@ -58,12 +302,15 @@ 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, &priv); + ret = nouveau_clock_create(parent, engine, oclass, nva3_domain, &priv); *pobject = nv_object(priv); if (ret) return ret; - priv->base.pll_calc = nva3_clock_pll_calc; + priv->base.read = nva3_clock_read; + priv->base.calc = nva3_clock_calc; + priv->base.prog = nva3_clock_prog; + priv->base.tidy = nva3_clock_tidy; return 0; } diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.h b/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.h new file mode 100644 index 000000000000..6229a509b42e --- /dev/null +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nva3.h @@ -0,0 +1,14 @@ +#ifndef __NVKM_CLK_NVA3_H__ +#define __NVKM_CLK_NVA3_H__ + +#include + +struct nva3_clock_info { + u32 clk; + u32 pll; +}; + +int nva3_clock_info(struct nouveau_clock *, int, u32, u32, + struct nva3_clock_info *); + +#endif diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c index 439d81c26130..c3105720ed24 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nvc0.c @@ -25,11 +25,408 @@ #include #include #include +#include #include "pll.h" +struct nvc0_clock_info { + u32 freq; + u32 ssel; + u32 mdiv; + u32 dsrc; + u32 ddiv; + u32 coef; +}; + struct nvc0_clock_priv { struct nouveau_clock base; + struct nvc0_clock_info eng[16]; +}; + +static u32 read_div(struct nvc0_clock_priv *, int, u32, u32); + +static u32 +read_vco(struct nvc0_clock_priv *priv, u32 dsrc) +{ + struct nouveau_clock *clk = &priv->base; + u32 ssrc = nv_rd32(priv, dsrc); + if (!(ssrc & 0x00000100)) + return clk->read(clk, nv_clk_src_sppll0); + return clk->read(clk, nv_clk_src_sppll1); +} + +static u32 +read_pll(struct nvc0_clock_priv *priv, u32 pll) +{ + struct nouveau_clock *clk = &priv->base; + u32 ctrl = nv_rd32(priv, pll + 0x00); + u32 coef = nv_rd32(priv, pll + 0x04); + u32 P = (coef & 0x003f0000) >> 16; + u32 N = (coef & 0x0000ff00) >> 8; + u32 M = (coef & 0x000000ff) >> 0; + u32 sclk; + + if (!(ctrl & 0x00000001)) + return 0; + + switch (pll) { + case 0x00e800: + case 0x00e820: + sclk = nv_device(priv)->crystal; + P = 1; + break; + case 0x132000: + sclk = clk->read(clk, nv_clk_src_mpllsrc); + break; + case 0x132020: + sclk = clk->read(clk, nv_clk_src_mpllsrcref); + break; + case 0x137000: + case 0x137020: + case 0x137040: + case 0x1370e0: + sclk = read_div(priv, (pll & 0xff) / 0x20, 0x137120, 0x137140); + break; + default: + return 0; + } + + return sclk * N / M / P; +} + +static u32 +read_div(struct nvc0_clock_priv *priv, int doff, u32 dsrc, u32 dctl) +{ + u32 ssrc = nv_rd32(priv, dsrc + (doff * 4)); + u32 sctl = nv_rd32(priv, dctl + (doff * 4)); + + switch (ssrc & 0x00000003) { + case 0: + if ((ssrc & 0x00030000) != 0x00030000) + return nv_device(priv)->crystal; + return 108000; + case 2: + return 100000; + case 3: + if (sctl & 0x80000000) { + u32 sclk = read_vco(priv, dsrc + (doff * 4)); + u32 sdiv = (sctl & 0x0000003f) + 2; + return (sclk * 2) / sdiv; + } + + return read_vco(priv, dsrc + (doff * 4)); + default: + return 0; + } +} + +static u32 +read_clk(struct nvc0_clock_priv *priv, int clk) +{ + u32 sctl = nv_rd32(priv, 0x137250 + (clk * 4)); + u32 ssel = nv_rd32(priv, 0x137100); + u32 sclk, sdiv; + + if (ssel & (1 << clk)) { + if (clk < 7) + sclk = read_pll(priv, 0x137000 + (clk * 0x20)); + else + sclk = read_pll(priv, 0x1370e0); + sdiv = ((sctl & 0x00003f00) >> 8) + 2; + } else { + sclk = read_div(priv, clk, 0x137160, 0x1371d0); + sdiv = ((sctl & 0x0000003f) >> 0) + 2; + } + + if (sctl & 0x80000000) + return (sclk * 2) / sdiv; + + return sclk; +} + +static int +nvc0_clock_read(struct nouveau_clock *clk, enum nv_clk_src src) +{ + struct nouveau_device *device = nv_device(clk); + struct nvc0_clock_priv *priv = (void *)clk; + + switch (src) { + case nv_clk_src_crystal: + return device->crystal; + case nv_clk_src_href: + return 100000; + case nv_clk_src_sppll0: + return read_pll(priv, 0x00e800); + case nv_clk_src_sppll1: + return read_pll(priv, 0x00e820); + + case nv_clk_src_mpllsrcref: + return read_div(priv, 0, 0x137320, 0x137330); + case nv_clk_src_mpllsrc: + return read_pll(priv, 0x132020); + case nv_clk_src_mpll: + return read_pll(priv, 0x132000); + case nv_clk_src_mdiv: + return read_div(priv, 0, 0x137300, 0x137310); + case nv_clk_src_mem: + if (nv_rd32(priv, 0x1373f0) & 0x00000002) + return clk->read(clk, nv_clk_src_mpll); + return clk->read(clk, nv_clk_src_mdiv); + + case nv_clk_src_gpc: + return read_clk(priv, 0x00); + case nv_clk_src_rop: + return read_clk(priv, 0x01); + case nv_clk_src_hubk07: + return read_clk(priv, 0x02); + case nv_clk_src_hubk06: + return read_clk(priv, 0x07); + case nv_clk_src_hubk01: + return read_clk(priv, 0x08); + case nv_clk_src_copy: + return read_clk(priv, 0x09); + case nv_clk_src_daemon: + return read_clk(priv, 0x0c); + case nv_clk_src_vdec: + return read_clk(priv, 0x0e); + default: + nv_error(clk, "invalid clock source %d\n", src); + return -EINVAL; + } +} + +static u32 +calc_div(struct nvc0_clock_priv *priv, int clk, u32 ref, u32 freq, u32 *ddiv) +{ + u32 div = min((ref * 2) / freq, (u32)65); + if (div < 2) + div = 2; + + *ddiv = div - 2; + return (ref * 2) / div; +} + +static u32 +calc_src(struct nvc0_clock_priv *priv, int clk, u32 freq, u32 *dsrc, u32 *ddiv) +{ + u32 sclk; + + /* use one of the fixed frequencies if possible */ + *ddiv = 0x00000000; + switch (freq) { + case 27000: + case 108000: + *dsrc = 0x00000000; + if (freq == 108000) + *dsrc |= 0x00030000; + return freq; + case 100000: + *dsrc = 0x00000002; + return freq; + default: + *dsrc = 0x00000003; + break; + } + + /* otherwise, calculate the closest divider */ + sclk = read_vco(priv, 0x137160 + (clk * 4)); + if (clk < 7) + sclk = calc_div(priv, clk, sclk, freq, ddiv); + return sclk; +} + +static u32 +calc_pll(struct nvc0_clock_priv *priv, int clk, u32 freq, u32 *coef) +{ + struct nouveau_bios *bios = nouveau_bios(priv); + struct nvbios_pll limits; + int N, M, P, ret; + + ret = nvbios_pll_parse(bios, 0x137000 + (clk * 0x20), &limits); + if (ret) + return 0; + + limits.refclk = read_div(priv, clk, 0x137120, 0x137140); + if (!limits.refclk) + return 0; + + ret = nva3_pll_calc(nv_subdev(priv), &limits, freq, &N, NULL, &M, &P); + if (ret <= 0) + return 0; + + *coef = (P << 16) | (N << 8) | M; + return ret; +} + +static int +calc_clk(struct nvc0_clock_priv *priv, + struct nouveau_cstate *cstate, int clk, int dom) +{ + struct nvc0_clock_info *info = &priv->eng[clk]; + u32 freq = cstate->domain[dom]; + u32 src0, div0, div1D, div1P = 0; + u32 clk0, clk1 = 0; + + /* invalid clock domain */ + if (!freq) + return 0; + + /* first possible path, using only dividers */ + clk0 = calc_src(priv, clk, freq, &src0, &div0); + clk0 = calc_div(priv, clk, clk0, freq, &div1D); + + /* see if we can get any closer using PLLs */ + if (clk0 != freq && (0x00004387 & (1 << clk))) { + if (clk <= 7) + clk1 = calc_pll(priv, clk, freq, &info->coef); + else + clk1 = cstate->domain[nv_clk_src_hubk06]; + clk1 = calc_div(priv, clk, clk1, freq, &div1P); + } + + /* select the method which gets closest to target freq */ + if (abs((int)freq - clk0) <= abs((int)freq - clk1)) { + info->dsrc = src0; + if (div0) { + info->ddiv |= 0x80000000; + info->ddiv |= div0 << 8; + info->ddiv |= div0; + } + if (div1D) { + info->mdiv |= 0x80000000; + info->mdiv |= div1D; + } + info->ssel = info->coef = 0; + info->freq = clk0; + } else { + if (div1P) { + info->mdiv |= 0x80000000; + info->mdiv |= div1P << 8; + } + info->ssel = (1 << clk); + info->freq = clk1; + } + + return 0; +} + +static int +nvc0_clock_calc(struct nouveau_clock *clk, struct nouveau_cstate *cstate) +{ + struct nvc0_clock_priv *priv = (void *)clk; + int ret; + + if ((ret = calc_clk(priv, cstate, 0x00, nv_clk_src_gpc)) || + (ret = calc_clk(priv, cstate, 0x01, nv_clk_src_rop)) || + (ret = calc_clk(priv, cstate, 0x02, nv_clk_src_hubk07)) || + (ret = calc_clk(priv, cstate, 0x07, nv_clk_src_hubk06)) || + (ret = calc_clk(priv, cstate, 0x08, nv_clk_src_hubk01)) || + (ret = calc_clk(priv, cstate, 0x09, nv_clk_src_copy)) || + (ret = calc_clk(priv, cstate, 0x0c, nv_clk_src_daemon)) || + (ret = calc_clk(priv, cstate, 0x0e, nv_clk_src_vdec))) + return ret; + + return 0; +} + +static void +nvc0_clock_prog_0(struct nvc0_clock_priv *priv, int clk) +{ + struct nvc0_clock_info *info = &priv->eng[clk]; + if (clk < 7 && !info->ssel) { + nv_mask(priv, 0x1371d0 + (clk * 0x04), 0x80003f3f, info->ddiv); + nv_wr32(priv, 0x137160 + (clk * 0x04), info->dsrc); + } +} + +static void +nvc0_clock_prog_1(struct nvc0_clock_priv *priv, int clk) +{ + nv_mask(priv, 0x137100, (1 << clk), 0x00000000); + nv_wait(priv, 0x137100, (1 << clk), 0x00000000); +} + +static void +nvc0_clock_prog_2(struct nvc0_clock_priv *priv, int clk) +{ + struct nvc0_clock_info *info = &priv->eng[clk]; + const u32 addr = 0x137000 + (clk * 0x20); + if (clk <= 7) { + nv_mask(priv, addr + 0x00, 0x00000004, 0x00000000); + nv_mask(priv, addr + 0x00, 0x00000001, 0x00000000); + if (info->coef) { + nv_wr32(priv, addr + 0x04, info->coef); + nv_mask(priv, addr + 0x00, 0x00000001, 0x00000001); + nv_wait(priv, addr + 0x00, 0x00020000, 0x00020000); + nv_mask(priv, addr + 0x00, 0x00020004, 0x00000004); + } + } +} + +static void +nvc0_clock_prog_3(struct nvc0_clock_priv *priv, int clk) +{ + struct nvc0_clock_info *info = &priv->eng[clk]; + if (info->ssel) { + nv_mask(priv, 0x137100, (1 << clk), info->ssel); + nv_wait(priv, 0x137100, (1 << clk), info->ssel); + } +} + +static void +nvc0_clock_prog_4(struct nvc0_clock_priv *priv, int clk) +{ + struct nvc0_clock_info *info = &priv->eng[clk]; + nv_mask(priv, 0x137250 + (clk * 0x04), 0x00003f3f, info->mdiv); +} + +static int +nvc0_clock_prog(struct nouveau_clock *clk) +{ + struct nvc0_clock_priv *priv = (void *)clk; + struct { + void (*exec)(struct nvc0_clock_priv *, int); + } stage[] = { + { nvc0_clock_prog_0 }, /* div programming */ + { nvc0_clock_prog_1 }, /* select div mode */ + { nvc0_clock_prog_2 }, /* (maybe) program pll */ + { nvc0_clock_prog_3 }, /* (maybe) select pll mode */ + { nvc0_clock_prog_4 }, /* final divider */ + }; + int i, j; + + for (i = 0; i < ARRAY_SIZE(stage); i++) { + for (j = 0; j < ARRAY_SIZE(priv->eng); j++) { + if (!priv->eng[j].freq) + continue; + stage[i].exec(priv, j); + } + } + + return 0; +} + +static void +nvc0_clock_tidy(struct nouveau_clock *clk) +{ + struct nvc0_clock_priv *priv = (void *)clk; + memset(priv->eng, 0x00, sizeof(priv->eng)); +} + +static struct nouveau_clocks +nvc0_domain[] = { + { nv_clk_src_crystal, 0xff }, + { nv_clk_src_href , 0xff }, + { nv_clk_src_hubk06 , 0x00 }, + { nv_clk_src_hubk01 , 0x01 }, + { nv_clk_src_copy , 0x02 }, + { nv_clk_src_gpc , 0x03, 0, "core", 2000 }, + { nv_clk_src_rop , 0x04 }, + { nv_clk_src_mem , 0x05, 0, "memory", 1000 }, + { nv_clk_src_vdec , 0x06 }, + { nv_clk_src_daemon , 0x0a }, + { nv_clk_src_hubk07 , 0x0b }, + { nv_clk_src_max } }; static int @@ -40,12 +437,15 @@ 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, &priv); + ret = nouveau_clock_create(parent, engine, oclass, nvc0_domain, &priv); *pobject = nv_object(priv); if (ret) return ret; - priv->base.pll_calc = nva3_clock_pll_calc; + priv->base.read = nvc0_clock_read; + priv->base.calc = nvc0_clock_calc; + priv->base.prog = nvc0_clock_prog; + priv->base.tidy = nvc0_clock_tidy; return 0; } diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c b/drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c new file mode 100644 index 000000000000..4c62e84b96f5 --- /dev/null +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/nve0.c @@ -0,0 +1,497 @@ +/* + * Copyright 2013 Red Hat Inc. + * + * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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: Ben Skeggs + */ + +#include +#include +#include +#include + +#include "pll.h" + +struct nve0_clock_info { + u32 freq; + u32 ssel; + u32 mdiv; + u32 dsrc; + u32 ddiv; + u32 coef; +}; + +struct nve0_clock_priv { + struct nouveau_clock base; + struct nve0_clock_info eng[16]; +}; + +static u32 read_div(struct nve0_clock_priv *, int, u32, u32); +static u32 read_pll(struct nve0_clock_priv *, u32); + +static u32 +read_vco(struct nve0_clock_priv *priv, u32 dsrc) +{ + u32 ssrc = nv_rd32(priv, dsrc); + if (!(ssrc & 0x00000100)) + return read_pll(priv, 0x00e800); + return read_pll(priv, 0x00e820); +} + +static u32 +read_pll(struct nve0_clock_priv *priv, u32 pll) +{ + u32 ctrl = nv_rd32(priv, pll + 0x00); + u32 coef = nv_rd32(priv, pll + 0x04); + u32 P = (coef & 0x003f0000) >> 16; + u32 N = (coef & 0x0000ff00) >> 8; + u32 M = (coef & 0x000000ff) >> 0; + u32 sclk; + u16 fN = 0xf000; + + if (!(ctrl & 0x00000001)) + return 0; + + switch (pll) { + case 0x00e800: + case 0x00e820: + sclk = nv_device(priv)->crystal; + P = 1; + break; + case 0x132000: + sclk = read_pll(priv, 0x132020); + P = (coef & 0x10000000) ? 2 : 1; + break; + case 0x132020: + sclk = read_div(priv, 0, 0x137320, 0x137330); + fN = nv_rd32(priv, pll + 0x10) >> 16; + break; + case 0x137000: + case 0x137020: + case 0x137040: + case 0x1370e0: + sclk = read_div(priv, (pll & 0xff) / 0x20, 0x137120, 0x137140); + break; + default: + return 0; + } + + if (P == 0) + P = 1; + + sclk = (sclk * N) + (((u16)(fN + 4096) * sclk) >> 13); + return sclk / (M * P); +} + +static u32 +read_div(struct nve0_clock_priv *priv, int doff, u32 dsrc, u32 dctl) +{ + u32 ssrc = nv_rd32(priv, dsrc + (doff * 4)); + u32 sctl = nv_rd32(priv, dctl + (doff * 4)); + + switch (ssrc & 0x00000003) { + case 0: + if ((ssrc & 0x00030000) != 0x00030000) + return nv_device(priv)->crystal; + return 108000; + case 2: + return 100000; + case 3: + if (sctl & 0x80000000) { + u32 sclk = read_vco(priv, dsrc + (doff * 4)); + u32 sdiv = (sctl & 0x0000003f) + 2; + return (sclk * 2) / sdiv; + } + + return read_vco(priv, dsrc + (doff * 4)); + default: + return 0; + } +} + +static u32 +read_mem(struct nve0_clock_priv *priv) +{ + switch (nv_rd32(priv, 0x1373f4) & 0x0000000f) { + case 1: return read_pll(priv, 0x132020); + case 2: return read_pll(priv, 0x132000); + default: + return 0; + } +} + +static u32 +read_clk(struct nve0_clock_priv *priv, int clk) +{ + u32 sctl = nv_rd32(priv, 0x137250 + (clk * 4)); + u32 sclk, sdiv; + + if (clk < 7) { + u32 ssel = nv_rd32(priv, 0x137100); + if (ssel & (1 << clk)) { + sclk = read_pll(priv, 0x137000 + (clk * 0x20)); + sdiv = 1; + } else { + sclk = read_div(priv, clk, 0x137160, 0x1371d0); + sdiv = 0; + } + } else { + u32 ssrc = nv_rd32(priv, 0x137160 + (clk * 0x04)); + if ((ssrc & 0x00000003) == 0x00000003) { + sclk = read_div(priv, clk, 0x137160, 0x1371d0); + if (ssrc & 0x00000100) { + if (ssrc & 0x40000000) + sclk = read_pll(priv, 0x1370e0); + sdiv = 1; + } else { + sdiv = 0; + } + } else { + sclk = read_div(priv, clk, 0x137160, 0x1371d0); + sdiv = 0; + } + } + + if (sctl & 0x80000000) { + if (sdiv) + sdiv = ((sctl & 0x00003f00) >> 8) + 2; + else + sdiv = ((sctl & 0x0000003f) >> 0) + 2; + return (sclk * 2) / sdiv; + } + + return sclk; +} + +static int +nve0_clock_read(struct nouveau_clock *clk, enum nv_clk_src src) +{ + struct nouveau_device *device = nv_device(clk); + struct nve0_clock_priv *priv = (void *)clk; + + switch (src) { + case nv_clk_src_crystal: + return device->crystal; + case nv_clk_src_href: + return 100000; + case nv_clk_src_mem: + return read_mem(priv); + case nv_clk_src_gpc: + return read_clk(priv, 0x00); + case nv_clk_src_rop: + return read_clk(priv, 0x01); + case nv_clk_src_hubk07: + return read_clk(priv, 0x02); + case nv_clk_src_hubk06: + return read_clk(priv, 0x07); + case nv_clk_src_hubk01: + return read_clk(priv, 0x08); + case nv_clk_src_daemon: + return read_clk(priv, 0x0c); + case nv_clk_src_vdec: + return read_clk(priv, 0x0e); + default: + nv_error(clk, "invalid clock source %d\n", src); + return -EINVAL; + } +} + +static u32 +calc_div(struct nve0_clock_priv *priv, int clk, u32 ref, u32 freq, u32 *ddiv) +{ + u32 div = min((ref * 2) / freq, (u32)65); + if (div < 2) + div = 2; + + *ddiv = div - 2; + return (ref * 2) / div; +} + +static u32 +calc_src(struct nve0_clock_priv *priv, int clk, u32 freq, u32 *dsrc, u32 *ddiv) +{ + u32 sclk; + + /* use one of the fixed frequencies if possible */ + *ddiv = 0x00000000; + switch (freq) { + case 27000: + case 108000: + *dsrc = 0x00000000; + if (freq == 108000) + *dsrc |= 0x00030000; + return freq; + case 100000: + *dsrc = 0x00000002; + return freq; + default: + *dsrc = 0x00000003; + break; + } + + /* otherwise, calculate the closest divider */ + sclk = read_vco(priv, 0x137160 + (clk * 4)); + if (clk < 7) + sclk = calc_div(priv, clk, sclk, freq, ddiv); + return sclk; +} + +static u32 +calc_pll(struct nve0_clock_priv *priv, int clk, u32 freq, u32 *coef) +{ + struct nouveau_bios *bios = nouveau_bios(priv); + struct nvbios_pll limits; + int N, M, P, ret; + + ret = nvbios_pll_parse(bios, 0x137000 + (clk * 0x20), &limits); + if (ret) + return 0; + + limits.refclk = read_div(priv, clk, 0x137120, 0x137140); + if (!limits.refclk) + return 0; + + ret = nva3_pll_calc(nv_subdev(priv), &limits, freq, &N, NULL, &M, &P); + if (ret <= 0) + return 0; + + *coef = (P << 16) | (N << 8) | M; + return ret; +} + +static int +calc_clk(struct nve0_clock_priv *priv, + struct nouveau_cstate *cstate, int clk, int dom) +{ + struct nve0_clock_info *info = &priv->eng[clk]; + u32 freq = cstate->domain[dom]; + u32 src0, div0, div1D, div1P = 0; + u32 clk0, clk1 = 0; + + /* invalid clock domain */ + if (!freq) + return 0; + + /* first possible path, using only dividers */ + clk0 = calc_src(priv, clk, freq, &src0, &div0); + clk0 = calc_div(priv, clk, clk0, freq, &div1D); + + /* see if we can get any closer using PLLs */ + if (clk0 != freq && (0x0000ff87 & (1 << clk))) { + if (clk <= 7) + clk1 = calc_pll(priv, clk, freq, &info->coef); + else + clk1 = cstate->domain[nv_clk_src_hubk06]; + clk1 = calc_div(priv, clk, clk1, freq, &div1P); + } + + /* select the method which gets closest to target freq */ + if (abs((int)freq - clk0) <= abs((int)freq - clk1)) { + info->dsrc = src0; + if (div0) { + info->ddiv |= 0x80000000; + info->ddiv |= div0 << 8; + info->ddiv |= div0; + } + if (div1D) { + info->mdiv |= 0x80000000; + info->mdiv |= div1D; + } + info->ssel = 0; + info->freq = clk0; + } else { + if (div1P) { + info->mdiv |= 0x80000000; + info->mdiv |= div1P << 8; + } + info->ssel = (1 << clk); + info->dsrc = 0x40000100; + info->freq = clk1; + } + + return 0; +} + +static int +nve0_clock_calc(struct nouveau_clock *clk, struct nouveau_cstate *cstate) +{ + struct nve0_clock_priv *priv = (void *)clk; + int ret; + + if ((ret = calc_clk(priv, cstate, 0x00, nv_clk_src_gpc)) || + (ret = calc_clk(priv, cstate, 0x01, nv_clk_src_rop)) || + (ret = calc_clk(priv, cstate, 0x02, nv_clk_src_hubk07)) || + (ret = calc_clk(priv, cstate, 0x07, nv_clk_src_hubk06)) || + (ret = calc_clk(priv, cstate, 0x08, nv_clk_src_hubk01)) || + (ret = calc_clk(priv, cstate, 0x0c, nv_clk_src_daemon)) || + (ret = calc_clk(priv, cstate, 0x0e, nv_clk_src_vdec))) + return ret; + + return 0; +} + +static void +nve0_clock_prog_0(struct nve0_clock_priv *priv, int clk) +{ + struct nve0_clock_info *info = &priv->eng[clk]; + if (!info->ssel) { + nv_mask(priv, 0x1371d0 + (clk * 0x04), 0x80003f3f, info->ddiv); + nv_wr32(priv, 0x137160 + (clk * 0x04), info->dsrc); + } +} + +static void +nve0_clock_prog_1_0(struct nve0_clock_priv *priv, int clk) +{ + nv_mask(priv, 0x137100, (1 << clk), 0x00000000); + nv_wait(priv, 0x137100, (1 << clk), 0x00000000); +} + +static void +nve0_clock_prog_1_1(struct nve0_clock_priv *priv, int clk) +{ + nv_mask(priv, 0x137160 + (clk * 0x04), 0x00000100, 0x00000000); +} + +static void +nve0_clock_prog_2(struct nve0_clock_priv *priv, int clk) +{ + struct nve0_clock_info *info = &priv->eng[clk]; + const u32 addr = 0x137000 + (clk * 0x20); + nv_mask(priv, addr + 0x00, 0x00000004, 0x00000000); + nv_mask(priv, addr + 0x00, 0x00000001, 0x00000000); + if (info->coef) { + nv_wr32(priv, addr + 0x04, info->coef); + nv_mask(priv, addr + 0x00, 0x00000001, 0x00000001); + nv_wait(priv, addr + 0x00, 0x00020000, 0x00020000); + nv_mask(priv, addr + 0x00, 0x00020004, 0x00000004); + } +} + +static void +nve0_clock_prog_3(struct nve0_clock_priv *priv, int clk) +{ + struct nve0_clock_info *info = &priv->eng[clk]; + nv_mask(priv, 0x137250 + (clk * 0x04), 0x00003f3f, info->mdiv); +} + +static void +nve0_clock_prog_4_0(struct nve0_clock_priv *priv, int clk) +{ + struct nve0_clock_info *info = &priv->eng[clk]; + if (info->ssel) { + nv_mask(priv, 0x137100, (1 << clk), info->ssel); + nv_wait(priv, 0x137100, (1 << clk), info->ssel); + } +} + +static void +nve0_clock_prog_4_1(struct nve0_clock_priv *priv, int clk) +{ + struct nve0_clock_info *info = &priv->eng[clk]; + if (info->ssel) { + nv_mask(priv, 0x137160 + (clk * 0x04), 0x40000000, 0x40000000); + nv_mask(priv, 0x137160 + (clk * 0x04), 0x00000100, 0x00000100); + } +} + +static int +nve0_clock_prog(struct nouveau_clock *clk) +{ + struct nve0_clock_priv *priv = (void *)clk; + struct { + u32 mask; + void (*exec)(struct nve0_clock_priv *, int); + } stage[] = { + { 0x007f, nve0_clock_prog_0 }, /* div programming */ + { 0x007f, nve0_clock_prog_1_0 }, /* select div mode */ + { 0xff80, nve0_clock_prog_1_1 }, + { 0x00ff, nve0_clock_prog_2 }, /* (maybe) program pll */ + { 0xff80, nve0_clock_prog_3 }, /* final divider */ + { 0x007f, nve0_clock_prog_4_0 }, /* (maybe) select pll mode */ + { 0xff80, nve0_clock_prog_4_1 }, + }; + int i, j; + + for (i = 0; i < ARRAY_SIZE(stage); i++) { + for (j = 0; j < ARRAY_SIZE(priv->eng); j++) { + if (!(stage[i].mask & (1 << j))) + continue; + if (!priv->eng[j].freq) + continue; + stage[i].exec(priv, j); + } + } + + return 0; +} + +static void +nve0_clock_tidy(struct nouveau_clock *clk) +{ + struct nve0_clock_priv *priv = (void *)clk; + memset(priv->eng, 0x00, sizeof(priv->eng)); +} + +static struct nouveau_clocks +nve0_domain[] = { + { nv_clk_src_crystal, 0xff }, + { nv_clk_src_href , 0xff }, + { nv_clk_src_gpc , 0x00, NVKM_CLK_DOM_FLAG_CORE, "core", 2000 }, + { nv_clk_src_hubk07 , 0x01, NVKM_CLK_DOM_FLAG_CORE }, + { nv_clk_src_rop , 0x02, NVKM_CLK_DOM_FLAG_CORE }, + { nv_clk_src_mem , 0x03, 0, "memory", 1000 }, + { nv_clk_src_hubk06 , 0x04, NVKM_CLK_DOM_FLAG_CORE }, + { nv_clk_src_hubk01 , 0x05 }, + { nv_clk_src_vdec , 0x06 }, + { nv_clk_src_daemon , 0x07 }, + { nv_clk_src_max } +}; + +static int +nve0_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine, + struct nouveau_oclass *oclass, void *data, u32 size, + struct nouveau_object **pobject) +{ + struct nve0_clock_priv *priv; + int ret; + + ret = nouveau_clock_create(parent, engine, oclass, nve0_domain, &priv); + *pobject = nv_object(priv); + if (ret) + return ret; + + priv->base.read = nve0_clock_read; + priv->base.calc = nve0_clock_calc; + priv->base.prog = nve0_clock_prog; + priv->base.tidy = nve0_clock_tidy; + return 0; +} + +struct nouveau_oclass +nve0_clock_oclass = { + .handle = NV_SUBDEV(CLOCK, 0xe0), + .ofuncs = &(struct nouveau_ofuncs) { + .ctor = nve0_clock_ctor, + .dtor = _nouveau_clock_dtor, + .init = _nouveau_clock_init, + .fini = _nouveau_clock_fini, + }, +}; diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/pllnv04.c b/drivers/gpu/drm/nouveau/core/subdev/clock/pllnv04.c index 7302219492dc..b47d543ab2e3 100644 --- a/drivers/gpu/drm/nouveau/core/subdev/clock/pllnv04.c +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/pllnv04.c @@ -230,10 +230,12 @@ nv04_pll_calc(struct nouveau_subdev *subdev, struct nvbios_pll *info, u32 freq, { int ret; - if (!info->vco2.max_freq) { + if (!info->vco2.max_freq || !N2) { ret = getMNP_single(subdev, info, freq, N1, M1, P); - *N2 = 1; - *M2 = 1; + if (N2) { + *N2 = 1; + *M2 = 1; + } } else { ret = getMNP_double(subdev, info, freq, N1, M1, N2, M2, P); } diff --git a/drivers/gpu/drm/nouveau/core/subdev/clock/seq.h b/drivers/gpu/drm/nouveau/core/subdev/clock/seq.h new file mode 100644 index 000000000000..fb33f06ebd59 --- /dev/null +++ b/drivers/gpu/drm/nouveau/core/subdev/clock/seq.h @@ -0,0 +1,17 @@ +#ifndef __NVKM_CLK_SEQ_H__ +#define __NVKM_CLK_SEQ_H__ + +#include +#include + +#define clk_init(s,p) hwsq_init(&(s)->base, (p)) +#define clk_exec(s,e) hwsq_exec(&(s)->base, (e)) +#define clk_have(s,r) ((s)->r_##r.addr != 0x000000) +#define clk_rd32(s,r) hwsq_rd32(&(s)->base, &(s)->r_##r) +#define clk_wr32(s,r,d) hwsq_wr32(&(s)->base, &(s)->r_##r, (d)) +#define clk_mask(s,r,m,d) hwsq_mask(&(s)->base, &(s)->r_##r, (m), (d)) +#define clk_setf(s,f,d) hwsq_setf(&(s)->base, (f), (d)) +#define clk_wait(s,f,d) hwsq_wait(&(s)->base, (f), (d)) +#define clk_nsec(s,n) hwsq_nsec(&(s)->base, (n)) + +#endif -- 2.11.0