Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 00/15] clk/tegra: improve code and add DFS support
This series adds support for GM20B PLL's Maxwell features, namely glitchless switch and (more importantly) DFS support. DFS lets the PLL lower its output speed according to input current variations, making the clock more stable and allowing it to run safely at lower voltage. All GM20B additions are done in the last patch, which consequently ends up being considerably big ; fortunately, it only consists of code additions thanks to the big refactoring work done by the first 14 patches. This refactoring makes GK20A's clock code much simpler to read and understand, and flexible enough to be reused by GM20B's when relevant. Note that the sliding and MNP programming had to be re-implemented due to the differences introduced by DFS - trying to factor them out with more abstractions resulted in rather ugly code IMHO. The first few patches also improve GK20A's volt driver a bit, mainly in order to support a minimum voltage that GM20B requires. Alexandre Courbot (15): tegra: fetch gpu_speedo_id volt/gk20a: make unused public functions static volt/gk20a: constify and name v_scale volt/gk20a: rename constructor volt/gm20b: add support for vmin parameter clk/gk20a: properly protect macro argument clk/gk20a: setup slide once during init clk/gk20a: reorganize MNP calculation a bit clk/gk20a: use nvkm_ functions in slide() clk/gk20a: add and use MNP programming functions clk/gk20a: parameterize PLL settings clk/gk20a: factorize n_lo computation code clk/gk20a: improve MNP programming clk/gk20a: rename constructor clk/gm20b: add glitchless and DFS support drm/nouveau/include/nvkm/core/tegra.h | 1 + drm/nouveau/nvkm/engine/device/tegra.c | 1 + drm/nouveau/nvkm/subdev/clk/gk20a.c | 394 +++++++-------- drm/nouveau/nvkm/subdev/clk/gk20a.h | 96 +++- drm/nouveau/nvkm/subdev/clk/gm20b.c | 894 ++++++++++++++++++++++++++++++++- drm/nouveau/nvkm/subdev/volt/gk20a.c | 27 +- drm/nouveau/nvkm/subdev/volt/gk20a.h | 11 +- drm/nouveau/nvkm/subdev/volt/gm20b.c | 40 +- lib/include/nvif/os.h | 1 + 9 files changed, 1216 insertions(+), 249 deletions(-) -- 2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 01/15] tegra: fetch gpu_speedo_id
The GPU speedo ID is required to select the right clk/volt parameters on
GM20B.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/include/nvkm/core/tegra.h | 1 +
drm/nouveau/nvkm/engine/device/tegra.c | 1 +
lib/include/nvif/os.h | 1 +
3 files changed, 3 insertions(+)
diff --git a/drm/nouveau/include/nvkm/core/tegra.h
b/drm/nouveau/include/nvkm/core/tegra.h
index b5370cb56e3c..e5c9b6268dcc 100644
--- a/drm/nouveau/include/nvkm/core/tegra.h
+++ b/drm/nouveau/include/nvkm/core/tegra.h
@@ -28,6 +28,7 @@ struct nvkm_device_tegra {
} iommu;
int gpu_speedo;
+ int gpu_speedo_id;
};
struct nvkm_device_tegra_func {
diff --git a/drm/nouveau/nvkm/engine/device/tegra.c
b/drm/nouveau/nvkm/engine/device/tegra.c
index 8edc0a583ae2..9b638bd905ff 100644
--- a/drm/nouveau/nvkm/engine/device/tegra.c
+++ b/drm/nouveau/nvkm/engine/device/tegra.c
@@ -311,6 +311,7 @@ nvkm_device_tegra_new(const struct nvkm_device_tegra_func
*func,
goto remove;
tdev->gpu_speedo = tegra_sku_info.gpu_speedo_value;
+ tdev->gpu_speedo_id = tegra_sku_info.gpu_speedo_id;
ret = nvkm_device_ctor(&nvkm_device_tegra_func, NULL, &pdev->dev,
NVKM_DEVICE_TEGRA, pdev->id, NULL,
cfg, dbg, detect, mmio, subdev_mask,
diff --git a/lib/include/nvif/os.h b/lib/include/nvif/os.h
index f0710b39c05b..ba253b0f1d37 100644
--- a/lib/include/nvif/os.h
+++ b/lib/include/nvif/os.h
@@ -1546,6 +1546,7 @@ reset_control_deassert(struct reset_control *reset)
*****************************************************************************/
struct tegra_sku_info {
int gpu_speedo_value;
+ int gpu_speedo_id;
};
extern struct tegra_sku_info tegra_sku_info;
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 02/15] volt/gk20a: make unused public functions static
Nobody else is using these, so make them private.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/volt/gk20a.c | 8 ++++----
drm/nouveau/nvkm/subdev/volt/gk20a.h | 5 -----
2 files changed, 4 insertions(+), 9 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/volt/gk20a.c
b/drm/nouveau/nvkm/subdev/volt/gk20a.c
index d554455326da..8255ef4d1bc2 100644
--- a/drm/nouveau/nvkm/subdev/volt/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/volt/gk20a.c
@@ -77,7 +77,7 @@ gk20a_volt_get_cvb_t_voltage(int speedo, int temp, int
s_scale, int t_scale,
return mv;
}
-int
+static int
gk20a_volt_calc_voltage(const struct cvb_coef *coef, int speedo)
{
int mv;
@@ -88,7 +88,7 @@ gk20a_volt_calc_voltage(const struct cvb_coef *coef, int
speedo)
return mv * 1000;
}
-int
+static int
gk20a_volt_vid_get(struct nvkm_volt *base)
{
struct gk20a_volt *volt = gk20a_volt(base);
@@ -103,7 +103,7 @@ gk20a_volt_vid_get(struct nvkm_volt *base)
return -EINVAL;
}
-int
+static int
gk20a_volt_vid_set(struct nvkm_volt *base, u8 vid)
{
struct gk20a_volt *volt = gk20a_volt(base);
@@ -113,7 +113,7 @@ gk20a_volt_vid_set(struct nvkm_volt *base, u8 vid)
return regulator_set_voltage(volt->vdd, volt->base.vid[vid].uv,
1200000);
}
-int
+static int
gk20a_volt_set_id(struct nvkm_volt *base, u8 id, int condition)
{
struct gk20a_volt *volt = gk20a_volt(base);
diff --git a/drm/nouveau/nvkm/subdev/volt/gk20a.h
b/drm/nouveau/nvkm/subdev/volt/gk20a.h
index 0fa3b502bcf8..92479948eabf 100644
--- a/drm/nouveau/nvkm/subdev/volt/gk20a.h
+++ b/drm/nouveau/nvkm/subdev/volt/gk20a.h
@@ -41,9 +41,4 @@ int _gk20a_volt_ctor(struct nvkm_device *device, int index,
const struct cvb_coef *coefs, int nb_coefs,
struct gk20a_volt *volt);
-int gk20a_volt_calc_voltage(const struct cvb_coef *coef, int speedo);
-int gk20a_volt_vid_get(struct nvkm_volt *volt);
-int gk20a_volt_vid_set(struct nvkm_volt *volt, u8 vid);
-int gk20a_volt_set_id(struct nvkm_volt *volt, u8 id, int condition);
-
#endif
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 03/15] volt/gk20a: constify and name v_scale
Give a name to this constant so we at least get an idea of what it is
for.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/volt/gk20a.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drm/nouveau/nvkm/subdev/volt/gk20a.c
b/drm/nouveau/nvkm/subdev/volt/gk20a.c
index 8255ef4d1bc2..8869d53a7082 100644
--- a/drm/nouveau/nvkm/subdev/volt/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/volt/gk20a.c
@@ -80,10 +80,11 @@ gk20a_volt_get_cvb_t_voltage(int speedo, int temp, int
s_scale, int t_scale,
static int
gk20a_volt_calc_voltage(const struct cvb_coef *coef, int speedo)
{
+ static const int v_scale = 1000;
int mv;
mv = gk20a_volt_get_cvb_t_voltage(speedo, -10, 100, 10, coef);
- mv = DIV_ROUND_UP(mv, 1000);
+ mv = DIV_ROUND_UP(mv, v_scale);
return mv * 1000;
}
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 04/15] volt/gk20a: rename constructor
Strip the _ prefix off the gk20a volt constructor.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/volt/gk20a.c | 10 +++++-----
drm/nouveau/nvkm/subdev/volt/gk20a.h | 6 +++---
drm/nouveau/nvkm/subdev/volt/gm20b.c | 4 ++--
3 files changed, 10 insertions(+), 10 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/volt/gk20a.c
b/drm/nouveau/nvkm/subdev/volt/gk20a.c
index 8869d53a7082..a4e9a5027a4c 100644
--- a/drm/nouveau/nvkm/subdev/volt/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/volt/gk20a.c
@@ -144,9 +144,9 @@ gk20a_volt = {
};
int
-_gk20a_volt_ctor(struct nvkm_device *device, int index,
- const struct cvb_coef *coefs, int nb_coefs,
- struct gk20a_volt *volt)
+gk20a_volt_ctor(struct nvkm_device *device, int index,
+ const struct cvb_coef *coefs, int nb_coefs,
+ struct gk20a_volt *volt)
{
struct nvkm_device_tegra *tdev = device->func->tegra(device);
int i, uv;
@@ -181,6 +181,6 @@ gk20a_volt_new(struct nvkm_device *device, int index, struct
nvkm_volt **pvolt)
return -ENOMEM;
*pvolt = &volt->base;
- return _gk20a_volt_ctor(device, index, gk20a_cvb_coef,
- ARRAY_SIZE(gk20a_cvb_coef), volt);
+ return gk20a_volt_ctor(device, index, gk20a_cvb_coef,
+ ARRAY_SIZE(gk20a_cvb_coef), volt);
}
diff --git a/drm/nouveau/nvkm/subdev/volt/gk20a.h
b/drm/nouveau/nvkm/subdev/volt/gk20a.h
index 92479948eabf..1ccca950c41a 100644
--- a/drm/nouveau/nvkm/subdev/volt/gk20a.h
+++ b/drm/nouveau/nvkm/subdev/volt/gk20a.h
@@ -37,8 +37,8 @@ struct gk20a_volt {
struct regulator *vdd;
};
-int _gk20a_volt_ctor(struct nvkm_device *device, int index,
- const struct cvb_coef *coefs, int nb_coefs,
- struct gk20a_volt *volt);
+int gk20a_volt_ctor(struct nvkm_device *device, int index,
+ const struct cvb_coef *coefs, int nb_coefs,
+ struct gk20a_volt *volt);
#endif
diff --git a/drm/nouveau/nvkm/subdev/volt/gm20b.c
b/drm/nouveau/nvkm/subdev/volt/gm20b.c
index 49b5ecb701e4..753074035cb1 100644
--- a/drm/nouveau/nvkm/subdev/volt/gm20b.c
+++ b/drm/nouveau/nvkm/subdev/volt/gm20b.c
@@ -51,6 +51,6 @@ gm20b_volt_new(struct nvkm_device *device, int index, struct
nvkm_volt **pvolt)
return -ENOMEM;
*pvolt = &volt->base;
- return _gk20a_volt_ctor(device, index, gm20b_cvb_coef,
- ARRAY_SIZE(gm20b_cvb_coef), volt);
+ return gk20a_volt_ctor(device, index, gm20b_cvb_coef,
+ ARRAY_SIZE(gm20b_cvb_coef), volt);
}
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 05/15] volt/gm20b: add support for vmin parameter
Chips may be characterized for a minimum voltage. Support this extra
parameter and select the appropriate minimum voltage for the detected
GPU speedo.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/volt/gk20a.c | 10 +++++-----
drm/nouveau/nvkm/subdev/volt/gk20a.h | 2 +-
drm/nouveau/nvkm/subdev/volt/gm20b.c | 17 ++++++++++++++++-
3 files changed, 22 insertions(+), 7 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/volt/gk20a.c
b/drm/nouveau/nvkm/subdev/volt/gk20a.c
index a4e9a5027a4c..ce5d83cdc7cf 100644
--- a/drm/nouveau/nvkm/subdev/volt/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/volt/gk20a.c
@@ -146,7 +146,7 @@ gk20a_volt = {
int
gk20a_volt_ctor(struct nvkm_device *device, int index,
const struct cvb_coef *coefs, int nb_coefs,
- struct gk20a_volt *volt)
+ int vmin, struct gk20a_volt *volt)
{
struct nvkm_device_tegra *tdev = device->func->tegra(device);
int i, uv;
@@ -161,9 +161,9 @@ gk20a_volt_ctor(struct nvkm_device *device, int index,
volt->base.vid_nr = nb_coefs;
for (i = 0; i < volt->base.vid_nr; i++) {
volt->base.vid[i].vid = i;
- volt->base.vid[i].uv - gk20a_volt_calc_voltage(&coefs[i],
- tdev->gpu_speedo);
+ volt->base.vid[i].uv = max(
+ gk20a_volt_calc_voltage(&coefs[i], tdev->gpu_speedo),
+ vmin);
nvkm_debug(&volt->base.subdev, "%2d: vid=%d, uv=%d\n", i,
volt->base.vid[i].vid, volt->base.vid[i].uv);
}
@@ -182,5 +182,5 @@ gk20a_volt_new(struct nvkm_device *device, int index, struct
nvkm_volt **pvolt)
*pvolt = &volt->base;
return gk20a_volt_ctor(device, index, gk20a_cvb_coef,
- ARRAY_SIZE(gk20a_cvb_coef), volt);
+ ARRAY_SIZE(gk20a_cvb_coef), 0, volt);
}
diff --git a/drm/nouveau/nvkm/subdev/volt/gk20a.h
b/drm/nouveau/nvkm/subdev/volt/gk20a.h
index 1ccca950c41a..6a6c97f9684e 100644
--- a/drm/nouveau/nvkm/subdev/volt/gk20a.h
+++ b/drm/nouveau/nvkm/subdev/volt/gk20a.h
@@ -39,6 +39,6 @@ struct gk20a_volt {
int gk20a_volt_ctor(struct nvkm_device *device, int index,
const struct cvb_coef *coefs, int nb_coefs,
- struct gk20a_volt *volt);
+ int vmin, struct gk20a_volt *volt);
#endif
diff --git a/drm/nouveau/nvkm/subdev/volt/gm20b.c
b/drm/nouveau/nvkm/subdev/volt/gm20b.c
index 753074035cb1..d2d39ab74b8e 100644
--- a/drm/nouveau/nvkm/subdev/volt/gm20b.c
+++ b/drm/nouveau/nvkm/subdev/volt/gm20b.c
@@ -41,16 +41,31 @@ const struct cvb_coef gm20b_cvb_coef[] = {
/* 921600 */ { 2647676, -106455, 1632 },
};
+const u32 speedo_to_vmin[] = {
+ /* 0, 1, 2, 3, 4, */
+ 950000, 840000, 818750, 840000, 810000,
+};
+
int
gm20b_volt_new(struct nvkm_device *device, int index, struct nvkm_volt **pvolt)
{
+ struct nvkm_device_tegra *tdev = device->func->tegra(device);
struct gk20a_volt *volt;
+ u32 vmin;
+
+ if (tdev->gpu_speedo_id >= ARRAY_SIZE(speedo_to_vmin)) {
+ nvdev_error(device, "unsupported speedo %d\n",
+ tdev->gpu_speedo_id);
+ return -EINVAL;
+ }
volt = kzalloc(sizeof(*volt), GFP_KERNEL);
if (!volt)
return -ENOMEM;
*pvolt = &volt->base;
+ vmin = speedo_to_vmin[tdev->gpu_speedo_id];
+
return gk20a_volt_ctor(device, index, gm20b_cvb_coef,
- ARRAY_SIZE(gm20b_cvb_coef), volt);
+ ARRAY_SIZE(gm20b_cvb_coef), vmin, volt);
}
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 06/15] clk/gk20a: properly protect macro argument
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com> --- drm/nouveau/nvkm/subdev/clk/gk20a.c | 2 +- drm/nouveau/nvkm/subdev/clk/gm20b.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c b/drm/nouveau/nvkm/subdev/clk/gk20a.c index 5f0ee24e31b8..d633669b52dc 100644 --- a/drm/nouveau/nvkm/subdev/clk/gk20a.c +++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c @@ -31,7 +31,7 @@ #define KHZ (1000) #define MHZ (KHZ * 1000) -#define MASK(w) ((1 << w) - 1) +#define MASK(w) ((1 << (w)) - 1) #define GPCPLL_CFG (SYS_GPCPLL_CFG_BASE + 0) #define GPCPLL_CFG_ENABLE BIT(0) diff --git a/drm/nouveau/nvkm/subdev/clk/gm20b.c b/drm/nouveau/nvkm/subdev/clk/gm20b.c index 71b2bbb61973..274a74c14fe7 100644 --- a/drm/nouveau/nvkm/subdev/clk/gm20b.c +++ b/drm/nouveau/nvkm/subdev/clk/gm20b.c @@ -29,7 +29,7 @@ #define KHZ (1000) #define MHZ (KHZ * 1000) -#define MASK(w) ((1 << w) - 1) +#define MASK(w) ((1 << (w)) - 1) #define BYPASSCTRL_SYS (SYS_GPCPLL_CFG_BASE + 0x340) #define BYPASSCTRL_SYS_GPCPLL_SHIFT 0 -- 2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 07/15] clk/gk20a: setup slide once during init
Slide setup needs to be performed only once, during init. Also
use the proper parameters for different clock speeds.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 46 ++++++++++++++++++++++++++++++++-----
drm/nouveau/nvkm/subdev/clk/gk20a.h | 2 ++
drm/nouveau/nvkm/subdev/clk/gm20b.c | 4 ++++
3 files changed, 46 insertions(+), 6 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index d633669b52dc..166f51cc5621 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -274,12 +274,6 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
if (n == ((val >> GPCPLL_COEFF_N_SHIFT) &
MASK(GPCPLL_COEFF_N_WIDTH)))
return 0;
- /* setup */
- nvkm_mask(device, GPCPLL_CFG2, 0xff << GPCPLL_CFG2_PLL_STEPA_SHIFT,
- 0x2b << GPCPLL_CFG2_PLL_STEPA_SHIFT);
- nvkm_mask(device, GPCPLL_CFG3, 0xff << GPCPLL_CFG3_PLL_STEPB_SHIFT,
- 0xb << GPCPLL_CFG3_PLL_STEPB_SHIFT);
-
/* pll slowdown mode */
nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT),
@@ -581,6 +575,42 @@ gk20a_clk_tidy(struct nvkm_clk *base)
{
}
+int
+gk20a_clk_setup_slide(struct gk20a_clk *clk)
+{
+ struct nvkm_subdev *subdev = &clk->base.subdev;
+ struct nvkm_device *device = subdev->device;
+ u32 step_a, step_b;
+
+ switch (clk->parent_rate) {
+ case 12000000:
+ case 12800000:
+ case 13000000:
+ step_a = 0x2b;
+ step_b = 0x0b;
+ break;
+ case 19200000:
+ step_a = 0x12;
+ step_b = 0x08;
+ break;
+ case 38400000:
+ step_a = 0x04;
+ step_b = 0x05;
+ break;
+ default:
+ nvkm_error(subdev, "invalid parent clock rate %u KHz",
+ clk->parent_rate / KHZ);
+ return -EINVAL;
+ }
+
+ nvkm_mask(device, GPCPLL_CFG2, 0xff << GPCPLL_CFG2_PLL_STEPA_SHIFT,
+ step_a << GPCPLL_CFG2_PLL_STEPA_SHIFT);
+ nvkm_mask(device, GPCPLL_CFG3, 0xff << GPCPLL_CFG3_PLL_STEPB_SHIFT,
+ step_b << GPCPLL_CFG3_PLL_STEPB_SHIFT);
+
+ return 0;
+}
+
void
gk20a_clk_fini(struct nvkm_clk *base)
{
@@ -617,6 +647,10 @@ gk20a_clk_init(struct nvkm_clk *base)
nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_INIT_MASK,
GPC2CLK_OUT_INIT_VAL);
+ ret = gk20a_clk_setup_slide(clk);
+ if (ret)
+ return ret;
+
/* Start with lowest frequency */
base->func->calc(base, &base->func->pstates[0].base);
ret = base->func->prog(&clk->base);
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.h
b/drm/nouveau/nvkm/subdev/clk/gk20a.h
index 13c46740197d..028c27237947 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.h
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.h
@@ -62,4 +62,6 @@ int gk20a_clk_calc(struct nvkm_clk *, struct nvkm_cstate *);
int gk20a_clk_prog(struct nvkm_clk *);
void gk20a_clk_tidy(struct nvkm_clk *);
+int gk20a_clk_setup_slide(struct gk20a_clk *);
+
#endif
diff --git a/drm/nouveau/nvkm/subdev/clk/gm20b.c
b/drm/nouveau/nvkm/subdev/clk/gm20b.c
index 274a74c14fe7..a34acafe37ee 100644
--- a/drm/nouveau/nvkm/subdev/clk/gm20b.c
+++ b/drm/nouveau/nvkm/subdev/clk/gm20b.c
@@ -144,6 +144,10 @@ gm20b_clk_init(struct nvkm_clk *base)
struct nvkm_device *device = subdev->device;
int ret;
+ ret = gk20a_clk_setup_slide(clk);
+ if (ret)
+ return ret;
+
/* Set the global bypass control to VCO */
nvkm_mask(device, BYPASSCTRL_SYS,
MASK(BYPASSCTRL_SYS_GPCPLL_WIDTH) << BYPASSCTRL_SYS_GPCPLL_SHIFT,
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 08/15] clk/gk20a: reorganize MNP calculation a bit
Move variables declarations to their actual scope of use, and simplify
code a bit.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 13 +++++--------
1 file changed, 5 insertions(+), 8 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index 166f51cc5621..4b9e7ec11412 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -163,16 +163,13 @@ gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long
rate)
target_clk_f = rate * 2 / KHZ;
ref_clk_f = clk->parent_rate / KHZ;
- max_vco_f = clk->params->max_vco;
+ target_vco_f = target_clk_f + target_clk_f / 50;
+ max_vco_f = max(clk->params->max_vco, target_vco_f);
min_vco_f = clk->params->min_vco;
best_m = clk->params->max_m;
best_n = clk->params->min_n;
best_pl = clk->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, clk->params->max_pl);
@@ -195,9 +192,7 @@ gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long
rate)
target_vco_f = target_clk_f * clk->pl_to_div(pl);
for (m = clk->params->min_m; m <= clk->params->max_m; m++) {
- u32 u_f, vco_f;
-
- u_f = ref_clk_f / m;
+ u32 u_f = ref_clk_f / m;
if (u_f < clk->params->min_u)
break;
@@ -211,6 +206,8 @@ gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long
rate)
break;
for (; n <= n2; n++) {
+ u32 vco_f;
+
if (n < clk->params->min_n)
continue;
if (n > clk->params->max_n)
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 09/15] clk/gk20a: use nvkm_ functions in slide()
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 26 ++++++++++----------------
1 file changed, 10 insertions(+), 16 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index 4b9e7ec11412..f07cf5b44fc6 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -263,7 +263,7 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
struct nvkm_subdev *subdev = &clk->base.subdev;
struct nvkm_device *device = subdev->device;
u32 val;
- int ramp_timeout;
+ int ret = 0;
/* get old coefficients */
val = nvkm_rd32(device, GPCPLL_COEFF);
@@ -284,17 +284,16 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
nvkm_wr32(device, GPCPLL_COEFF, val);
/* dynamic ramp to new ndiv */
- val = nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
- val |= 0x1 << GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT;
udelay(1);
- nvkm_wr32(device, GPCPLL_NDIV_SLOWDOWN, val);
+ nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
+ BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT),
+ BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT));
- for (ramp_timeout = 500; ramp_timeout > 0; ramp_timeout--) {
- udelay(1);
- val = nvkm_rd32(device, GPC_BCAST_NDIV_SLOWDOWN_DEBUG);
- if (val & GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK)
- break;
- }
+ /* wait for ramping to complete */
+ if (nvkm_wait_usec(device, 500, GPC_BCAST_NDIV_SLOWDOWN_DEBUG,
+ GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK,
+ GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK) < 0)
+ ret = -ETIMEDOUT;
/* exit slowdown mode */
nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
@@ -302,12 +301,7 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT), 0);
nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
- if (ramp_timeout <= 0) {
- nvkm_error(subdev, "gpcpll dynamic ramp timeout\n");
- return -ETIMEDOUT;
- }
-
- return 0;
+ return ret;
}
static void
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 10/15] clk/gk20a: add and use MNP programming functions
Add relevant functions to work with the gk20a_pll structure and use them
where they ought to be instead of directly manipulating registers.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 35 ++++++++++++++++++++++-------------
1 file changed, 22 insertions(+), 13 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index f07cf5b44fc6..80031c0f9e3f 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -136,6 +136,18 @@ gk20a_pllg_read_mnp(struct gk20a_clk *clk, struct gk20a_pll
*pll)
pll->pl = (val >> GPCPLL_COEFF_P_SHIFT) &
MASK(GPCPLL_COEFF_P_WIDTH);
}
+static void
+gk20a_pllg_write_mnp(struct gk20a_clk *clk, const struct gk20a_pll *pll)
+{
+ struct nvkm_device *device = clk->base.subdev.device;
+ u32 val;
+
+ val = (pll->m & MASK(GPCPLL_COEFF_M_WIDTH)) <<
GPCPLL_COEFF_M_SHIFT;
+ val |= (pll->n & MASK(GPCPLL_COEFF_N_WIDTH)) <<
GPCPLL_COEFF_N_SHIFT;
+ val |= (pll->pl & MASK(GPCPLL_COEFF_P_WIDTH)) <<
GPCPLL_COEFF_P_SHIFT;
+ nvkm_wr32(device, GPCPLL_COEFF, val);
+}
+
static u32
gk20a_pllg_calc_rate(struct gk20a_clk *clk)
{
@@ -262,13 +274,13 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
{
struct nvkm_subdev *subdev = &clk->base.subdev;
struct nvkm_device *device = subdev->device;
- u32 val;
+ struct gk20a_pll pll;
int ret = 0;
/* get old coefficients */
- val = nvkm_rd32(device, GPCPLL_COEFF);
+ gk20a_pllg_read_mnp(clk, &pll);
/* do nothing if NDIV is the same */
- if (n == ((val >> GPCPLL_COEFF_N_SHIFT) &
MASK(GPCPLL_COEFF_N_WIDTH)))
+ if (n == pll.n)
return 0;
/* pll slowdown mode */
@@ -277,11 +289,9 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT));
/* new ndiv ready for ramp */
- val = nvkm_rd32(device, GPCPLL_COEFF);
- val &= ~(MASK(GPCPLL_COEFF_N_WIDTH) << GPCPLL_COEFF_N_SHIFT);
- val |= (n & MASK(GPCPLL_COEFF_N_WIDTH)) << GPCPLL_COEFF_N_SHIFT;
+ pll.n = n;
udelay(1);
- nvkm_wr32(device, GPCPLL_COEFF, val);
+ gk20a_pllg_write_mnp(clk, &pll);
/* dynamic ramp to new ndiv */
udelay(1);
@@ -377,12 +387,11 @@ _gk20a_pllg_program_mnp(struct gk20a_clk *clk, bool
allow_slide)
nvkm_debug(subdev, "%s: m=%d n=%d pl=%d\n", __func__,
clk->pll.m, clk->pll.n, clk->pll.pl);
- n_lo = DIV_ROUND_UP(clk->pll.m * clk->params->min_vco,
- clk->parent_rate / KHZ);
- val = clk->pll.m << GPCPLL_COEFF_M_SHIFT;
- val |= (allow_slide ? n_lo : clk->pll.n) << GPCPLL_COEFF_N_SHIFT;
- val |= clk->pll.pl << GPCPLL_COEFF_P_SHIFT;
- nvkm_wr32(device, GPCPLL_COEFF, val);
+ old_pll = clk->pll;
+ if (allow_slide)
+ old_pll.n = DIV_ROUND_UP(clk->pll.m * clk->params->min_vco,
+ clk->parent_rate / KHZ);
+ gk20a_pllg_write_mnp(clk, &old_pll);
gk20a_pllg_enable(clk);
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 11/15] clk/gk20a: parameterize PLL settings
Make functions manipulating PLL settings take them as an argument,
instead of assuming we want to work on the copy in the gk20a_clk
structure. This makes these functions more flexible, which we will need
in GM20B.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 51 ++++++++++++++++++++-----------------
1 file changed, 27 insertions(+), 24 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index 80031c0f9e3f..a95eda7c5df4 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -149,19 +149,20 @@ gk20a_pllg_write_mnp(struct gk20a_clk *clk, const struct
gk20a_pll *pll)
}
static u32
-gk20a_pllg_calc_rate(struct gk20a_clk *clk)
+gk20a_pllg_calc_rate(struct gk20a_clk *clk, struct gk20a_pll *pll)
{
u32 rate;
u32 divider;
- rate = clk->parent_rate * clk->pll.n;
- divider = clk->pll.m * clk->pl_to_div(clk->pll.pl);
+ rate = clk->parent_rate * pll->n;
+ divider = pll->m * clk->pl_to_div(pll->pl);
return rate / divider / 2;
}
static int
-gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate)
+gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate,
+ struct gk20a_pll *pll)
{
struct nvkm_subdev *subdev = &clk->base.subdev;
u32 target_clk_f, ref_clk_f, target_freq;
@@ -256,16 +257,16 @@ found_match:
"no best match for target @ %dMHz on gpc_pll",
target_clk_f / KHZ);
- clk->pll.m = best_m;
- clk->pll.n = best_n;
- clk->pll.pl = best_pl;
+ pll->m = best_m;
+ pll->n = best_n;
+ pll->pl = best_pl;
- target_freq = gk20a_pllg_calc_rate(clk);
+ target_freq = gk20a_pllg_calc_rate(clk, pll);
nvkm_debug(subdev,
- "actual target freq %d MHz, M %d, N %d, PL %d(div%d)\n",
- target_freq / MHZ, clk->pll.m, clk->pll.n, clk->pll.pl,
- clk->pl_to_div(clk->pll.pl));
+ "actual target freq %d KHz, M %d, N %d, PL %d(div%d)\n",
+ target_freq / KHZ, pll->m, pll->n, pll->pl,
+ clk->pl_to_div(pll->pl));
return 0;
}
@@ -333,7 +334,8 @@ gk20a_pllg_disable(struct gk20a_clk *clk)
}
static int
-_gk20a_pllg_program_mnp(struct gk20a_clk *clk, bool allow_slide)
+_gk20a_pllg_program_mnp(struct gk20a_clk *clk, struct gk20a_pll *pll,
+ bool allow_slide)
{
struct nvkm_subdev *subdev = &clk->base.subdev;
struct nvkm_device *device = subdev->device;
@@ -346,9 +348,9 @@ _gk20a_pllg_program_mnp(struct gk20a_clk *clk, bool
allow_slide)
/* do NDIV slide if there is no change in M and PL */
cfg = nvkm_rd32(device, GPCPLL_CFG);
- if (allow_slide && clk->pll.m == old_pll.m &&
- clk->pll.pl == old_pll.pl && (cfg & GPCPLL_CFG_ENABLE)) {
- return gk20a_pllg_slide(clk, clk->pll.n);
+ if (allow_slide && pll->m == old_pll.m &&
+ pll->pl == old_pll.pl && (cfg & GPCPLL_CFG_ENABLE)) {
+ return gk20a_pllg_slide(clk, pll->n);
}
/* slide down to NDIV_LO */
@@ -385,11 +387,11 @@ _gk20a_pllg_program_mnp(struct gk20a_clk *clk, bool
allow_slide)
gk20a_pllg_disable(clk);
nvkm_debug(subdev, "%s: m=%d n=%d pl=%d\n", __func__,
- clk->pll.m, clk->pll.n, clk->pll.pl);
+ pll->m, pll->n, pll->pl);
- old_pll = clk->pll;
+ old_pll = *pll;
if (allow_slide)
- old_pll.n = DIV_ROUND_UP(clk->pll.m * clk->params->min_vco,
+ old_pll.n = DIV_ROUND_UP(pll->m * clk->params->min_vco,
clk->parent_rate / KHZ);
gk20a_pllg_write_mnp(clk, &old_pll);
@@ -425,7 +427,7 @@ _gk20a_pllg_program_mnp(struct gk20a_clk *clk, bool
allow_slide)
}
/* slide up to new NDIV */
- return allow_slide ? gk20a_pllg_slide(clk, clk->pll.n) : 0;
+ return allow_slide ? gk20a_pllg_slide(clk, pll->n) : 0;
}
static int
@@ -433,9 +435,9 @@ gk20a_pllg_program_mnp(struct gk20a_clk *clk)
{
int err;
- err = _gk20a_pllg_program_mnp(clk, true);
+ err = _gk20a_pllg_program_mnp(clk, &clk->pll, true);
if (err)
- err = _gk20a_pllg_program_mnp(clk, false);
+ err = _gk20a_pllg_program_mnp(clk, &clk->pll, false);
return err;
}
@@ -540,13 +542,14 @@ gk20a_clk_read(struct nvkm_clk *base, enum nv_clk_src src)
struct gk20a_clk *clk = gk20a_clk(base);
struct nvkm_subdev *subdev = &clk->base.subdev;
struct nvkm_device *device = subdev->device;
+ struct gk20a_pll pll;
switch (src) {
case nv_clk_src_crystal:
return device->crystal;
case nv_clk_src_gpc:
- gk20a_pllg_read_mnp(clk, &clk->pll);
- return gk20a_pllg_calc_rate(clk) / GK20A_CLK_GPC_MDIV;
+ gk20a_pllg_read_mnp(clk, &pll);
+ return gk20a_pllg_calc_rate(clk, &pll) / GK20A_CLK_GPC_MDIV;
default:
nvkm_error(subdev, "invalid clock source %d\n", src);
return -EINVAL;
@@ -559,7 +562,7 @@ gk20a_clk_calc(struct nvkm_clk *base, struct nvkm_cstate
*cstate)
struct gk20a_clk *clk = gk20a_clk(base);
return gk20a_pllg_calc_mnp(clk, cstate->domain[nv_clk_src_gpc] *
- GK20A_CLK_GPC_MDIV);
+ GK20A_CLK_GPC_MDIV, &clk->pll);
}
int
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 12/15] clk/gk20a: factorize n_lo computation code
Use a dedicated function instead of always calculating n_lo on the fly.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 19 ++++++++++---------
1 file changed, 10 insertions(+), 9 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index a95eda7c5df4..cc46e3abac97 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -160,6 +160,13 @@ gk20a_pllg_calc_rate(struct gk20a_clk *clk, struct
gk20a_pll *pll)
return rate / divider / 2;
}
+static u32
+gk20a_pllg_n_lo(struct gk20a_clk *clk, struct gk20a_pll *pll)
+{
+ return DIV_ROUND_UP(pll->m * clk->params->min_vco,
+ clk->parent_rate / KHZ);
+}
+
static int
gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate,
struct gk20a_pll *pll)
@@ -341,7 +348,6 @@ _gk20a_pllg_program_mnp(struct gk20a_clk *clk, struct
gk20a_pll *pll,
struct nvkm_device *device = subdev->device;
u32 val, cfg;
struct gk20a_pll old_pll;
- u32 n_lo;
/* get old coefficients */
gk20a_pllg_read_mnp(clk, &old_pll);
@@ -357,10 +363,7 @@ _gk20a_pllg_program_mnp(struct gk20a_clk *clk, struct
gk20a_pll *pll,
if (allow_slide && (cfg & GPCPLL_CFG_ENABLE)) {
int ret;
- n_lo = DIV_ROUND_UP(old_pll.m * clk->params->min_vco,
- clk->parent_rate / KHZ);
- ret = gk20a_pllg_slide(clk, n_lo);
-
+ ret = gk20a_pllg_slide(clk, gk20a_pllg_n_lo(clk, &old_pll));
if (ret)
return ret;
}
@@ -391,8 +394,7 @@ _gk20a_pllg_program_mnp(struct gk20a_clk *clk, struct
gk20a_pll *pll,
old_pll = *pll;
if (allow_slide)
- old_pll.n = DIV_ROUND_UP(pll->m * clk->params->min_vco,
- clk->parent_rate / KHZ);
+ old_pll.n = gk20a_pllg_n_lo(clk, pll);
gk20a_pllg_write_mnp(clk, &old_pll);
gk20a_pllg_enable(clk);
@@ -628,8 +630,7 @@ gk20a_clk_fini(struct nvkm_clk *base)
u32 n_lo;
gk20a_pllg_read_mnp(clk, &pll);
- n_lo = DIV_ROUND_UP(pll.m * clk->params->min_vco,
- clk->parent_rate / KHZ);
+ n_lo = gk20a_pllg_n_lo(clk, &pll);
gk20a_pllg_slide(clk, n_lo);
}
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 13/15] clk/gk20a: improve MNP programming
Split the MNP programming function into two functions for the cases
where we allow sliding or not, instead of making it take a parameter for
this. This results in less conditionals in the code and makes it easier
to read.
Also make the MNP programming functions take the PLL parameters as
arguments, and move bits of code to more relevant places (previous
programming tended to be just-in-time, which added more conditionnals in
the code).
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 185 +++++++++++++++++++-----------------
1 file changed, 96 insertions(+), 89 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index cc46e3abac97..513d112e8268 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -72,6 +72,7 @@
#define GPC2CLK_OUT_VCODIV_WIDTH 6
#define GPC2CLK_OUT_VCODIV_SHIFT 8
#define GPC2CLK_OUT_VCODIV1 0
+#define GPC2CLK_OUT_VCODIV2 2
#define GPC2CLK_OUT_VCODIV_MASK (MASK(GPC2CLK_OUT_VCODIV_WIDTH) << \
GPC2CLK_OUT_VCODIV_SHIFT)
#define GPC2CLK_OUT_BYPDIV_WIDTH 6
@@ -322,13 +323,42 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
return ret;
}
-static void
+static bool
+gk20a_pllg_is_enabled(struct gk20a_clk *clk)
+{
+ struct nvkm_device *device = clk->base.subdev.device;
+ u32 val;
+
+ val = nvkm_rd32(device, GPCPLL_CFG);
+ return val & GPCPLL_CFG_ENABLE;
+}
+
+static int
gk20a_pllg_enable(struct gk20a_clk *clk)
{
struct nvkm_device *device = clk->base.subdev.device;
+ u32 val;
nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, GPCPLL_CFG_ENABLE);
nvkm_rd32(device, GPCPLL_CFG);
+
+ /* enable lock detection */
+ val = nvkm_rd32(device, GPCPLL_CFG);
+ if (val & GPCPLL_CFG_LOCK_DET_OFF) {
+ val &= ~GPCPLL_CFG_LOCK_DET_OFF;
+ nvkm_wr32(device, GPCPLL_CFG, val);
+ }
+
+ /* wait for lock */
+ if (nvkm_wait_usec(device, 300, GPCPLL_CFG, GPCPLL_CFG_LOCK,
+ GPCPLL_CFG_LOCK) < 0)
+ return -ETIMEDOUT;
+
+ /* switch to VCO mode */
+ nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT),
+ BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+
+ return 0;
}
static void
@@ -336,112 +366,81 @@ gk20a_pllg_disable(struct gk20a_clk *clk)
{
struct nvkm_device *device = clk->base.subdev.device;
+ /* put PLL in bypass before disabling it */
+ nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
+
nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, 0);
nvkm_rd32(device, GPCPLL_CFG);
}
static int
-_gk20a_pllg_program_mnp(struct gk20a_clk *clk, struct gk20a_pll *pll,
- bool allow_slide)
+gk20a_pllg_program_mnp(struct gk20a_clk *clk, const struct gk20a_pll *pll)
{
struct nvkm_subdev *subdev = &clk->base.subdev;
struct nvkm_device *device = subdev->device;
- u32 val, cfg;
- struct gk20a_pll old_pll;
-
- /* get old coefficients */
- gk20a_pllg_read_mnp(clk, &old_pll);
-
- /* do NDIV slide if there is no change in M and PL */
- cfg = nvkm_rd32(device, GPCPLL_CFG);
- if (allow_slide && pll->m == old_pll.m &&
- pll->pl == old_pll.pl && (cfg & GPCPLL_CFG_ENABLE)) {
- return gk20a_pllg_slide(clk, pll->n);
- }
+ struct gk20a_pll cur_pll;
+ int ret;
- /* slide down to NDIV_LO */
- if (allow_slide && (cfg & GPCPLL_CFG_ENABLE)) {
- int ret;
+ gk20a_pllg_read_mnp(clk, &cur_pll);
- ret = gk20a_pllg_slide(clk, gk20a_pllg_n_lo(clk, &old_pll));
- if (ret)
- return ret;
- }
-
- /* split FO-to-bypass jump in halfs by setting out divider 1:2 */
+ /* split VCO-to-bypass jump in half by setting out divider 1:2 */
nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
- 0x2 << GPC2CLK_OUT_VCODIV_SHIFT);
-
- /* put PLL in bypass before programming it */
- val = nvkm_rd32(device, SEL_VCO);
- val &= ~(BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+ GPC2CLK_OUT_VCODIV2 << GPC2CLK_OUT_VCODIV_SHIFT);
+ /* Intentional 2nd write to assure linear divider operation */
+ nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+ GPC2CLK_OUT_VCODIV2 << GPC2CLK_OUT_VCODIV_SHIFT);
+ nvkm_rd32(device, GPC2CLK_OUT);
udelay(2);
- nvkm_wr32(device, SEL_VCO, val);
-
- /* get out from IDDQ */
- val = nvkm_rd32(device, GPCPLL_CFG);
- if (val & GPCPLL_CFG_IDDQ) {
- val &= ~GPCPLL_CFG_IDDQ;
- nvkm_wr32(device, GPCPLL_CFG, val);
- nvkm_rd32(device, GPCPLL_CFG);
- udelay(2);
- }
gk20a_pllg_disable(clk);
- nvkm_debug(subdev, "%s: m=%d n=%d pl=%d\n", __func__,
- pll->m, pll->n, pll->pl);
-
- old_pll = *pll;
- if (allow_slide)
- old_pll.n = gk20a_pllg_n_lo(clk, pll);
- gk20a_pllg_write_mnp(clk, &old_pll);
+ gk20a_pllg_write_mnp(clk, pll);
- gk20a_pllg_enable(clk);
-
- val = nvkm_rd32(device, GPCPLL_CFG);
- if (val & GPCPLL_CFG_LOCK_DET_OFF) {
- val &= ~GPCPLL_CFG_LOCK_DET_OFF;
- nvkm_wr32(device, GPCPLL_CFG, val);
- }
-
- if (nvkm_usec(device, 300,
- if (nvkm_rd32(device, GPCPLL_CFG) & GPCPLL_CFG_LOCK)
- break;
- ) < 0)
- return -ETIMEDOUT;
-
- /* switch to VCO mode */
- nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT),
- BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+ ret = gk20a_pllg_enable(clk);
+ if (ret)
+ return ret;
/* restore out divider 1:1 */
- val = nvkm_rd32(device, GPC2CLK_OUT);
- if ((val & GPC2CLK_OUT_VCODIV_MASK) !- (GPC2CLK_OUT_VCODIV1 <<
GPC2CLK_OUT_VCODIV_SHIFT)) {
- val &= ~GPC2CLK_OUT_VCODIV_MASK;
- val |= GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT;
- udelay(2);
- nvkm_wr32(device, GPC2CLK_OUT, val);
- /* Intentional 2nd write to assure linear divider operation */
- nvkm_wr32(device, GPC2CLK_OUT, val);
- nvkm_rd32(device, GPC2CLK_OUT);
- }
+ udelay(2);
+ nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+ GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT);
+ /* Intentional 2nd write to assure linear divider operation */
+ nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+ GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT);
+ nvkm_rd32(device, GPC2CLK_OUT);
- /* slide up to new NDIV */
- return allow_slide ? gk20a_pllg_slide(clk, pll->n) : 0;
+ return 0;
}
static int
-gk20a_pllg_program_mnp(struct gk20a_clk *clk)
+gk20a_pllg_program_mnp_slide(struct gk20a_clk *clk, const struct gk20a_pll
*pll)
{
- int err;
+ struct gk20a_pll cur_pll;
+ int ret;
+
+ if (gk20a_pllg_is_enabled(clk)) {
+ gk20a_pllg_read_mnp(clk, &cur_pll);
- err = _gk20a_pllg_program_mnp(clk, &clk->pll, true);
- if (err)
- err = _gk20a_pllg_program_mnp(clk, &clk->pll, false);
+ /* just do NDIV slide if there is no change to M and PL */
+ if (pll->m == cur_pll.m && pll->pl == cur_pll.pl)
+ return gk20a_pllg_slide(clk, pll->n);
- return err;
+ /* slide down to current NDIV_LO */
+ cur_pll.n = gk20a_pllg_n_lo(clk, &cur_pll);
+ ret = gk20a_pllg_slide(clk, cur_pll.n);
+ if (ret)
+ return ret;
+ }
+
+ /* program MNP with the new clock parameters and new NDIV_LO */
+ cur_pll = *pll;
+ cur_pll.n = gk20a_pllg_n_lo(clk, &cur_pll);
+ ret = gk20a_pllg_program_mnp(clk, &cur_pll);
+ if (ret)
+ return ret;
+
+ /* slide up to new NDIV */
+ return gk20a_pllg_slide(clk, pll->n);
}
static struct nvkm_pstate
@@ -571,8 +570,13 @@ int
gk20a_clk_prog(struct nvkm_clk *base)
{
struct gk20a_clk *clk = gk20a_clk(base);
+ int ret;
+
+ ret = gk20a_pllg_program_mnp_slide(clk, &clk->pll);
+ if (ret)
+ ret = gk20a_pllg_program_mnp(clk, &clk->pll);
- return gk20a_pllg_program_mnp(clk);
+ return ret;
}
void
@@ -621,11 +625,9 @@ gk20a_clk_fini(struct nvkm_clk *base)
{
struct nvkm_device *device = base->subdev.device;
struct gk20a_clk *clk = gk20a_clk(base);
- u32 val;
/* slide to VCO min */
- val = nvkm_rd32(device, GPCPLL_CFG);
- if (val & GPCPLL_CFG_ENABLE) {
+ if (gk20a_pllg_is_enabled(clk)) {
struct gk20a_pll pll;
u32 n_lo;
@@ -634,10 +636,10 @@ gk20a_clk_fini(struct nvkm_clk *base)
gk20a_pllg_slide(clk, n_lo);
}
- /* put PLL in bypass before disabling it */
- nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
-
gk20a_pllg_disable(clk);
+
+ /* set IDDQ */
+ nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_IDDQ, 1);
}
static int
@@ -648,6 +650,11 @@ gk20a_clk_init(struct nvkm_clk *base)
struct nvkm_device *device = subdev->device;
int ret;
+ /* get out from IDDQ */
+ nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_IDDQ, 0);
+ nvkm_rd32(device, GPCPLL_CFG);
+ udelay(5);
+
nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_INIT_MASK,
GPC2CLK_OUT_INIT_VAL);
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 14/15] clk/gk20a: rename constructor
Strip the _ prefix off the gk20a clock constructor.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 4 ++--
drm/nouveau/nvkm/subdev/clk/gk20a.h | 2 +-
drm/nouveau/nvkm/subdev/clk/gm20b.c | 4 ++--
3 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index 513d112e8268..607d74a81c5e 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -691,7 +691,7 @@ gk20a_clk = {
};
int
-_gk20a_clk_ctor(struct nvkm_device *device, int index,
+gk20a_clk_ctor(struct nvkm_device *device, int index,
const struct nvkm_clk_func *func,
const struct gk20a_clk_pllg_params *params,
struct gk20a_clk *clk)
@@ -730,7 +730,7 @@ gk20a_clk_new(struct nvkm_device *device, int index, struct
nvkm_clk **pclk)
return -ENOMEM;
*pclk = &clk->base;
- ret = _gk20a_clk_ctor(device, index, &gk20a_clk, &gk20a_pllg_params,
+ ret = gk20a_clk_ctor(device, index, &gk20a_clk, &gk20a_pllg_params,
clk);
clk->pl_to_div = pl_to_div;
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.h
b/drm/nouveau/nvkm/subdev/clk/gk20a.h
index 028c27237947..73551f8209fa 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.h
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.h
@@ -54,7 +54,7 @@ struct gk20a_clk {
};
#define gk20a_clk(p) container_of((p), struct gk20a_clk, base)
-int _gk20a_clk_ctor(struct nvkm_device *, int, const struct nvkm_clk_func *,
+int gk20a_clk_ctor(struct nvkm_device *, int, const struct nvkm_clk_func *,
const struct gk20a_clk_pllg_params *, struct gk20a_clk *);
void gk20a_clk_fini(struct nvkm_clk *);
int gk20a_clk_read(struct nvkm_clk *, enum nv_clk_src);
diff --git a/drm/nouveau/nvkm/subdev/clk/gm20b.c
b/drm/nouveau/nvkm/subdev/clk/gm20b.c
index a34acafe37ee..68749bbbe623 100644
--- a/drm/nouveau/nvkm/subdev/clk/gm20b.c
+++ b/drm/nouveau/nvkm/subdev/clk/gm20b.c
@@ -192,8 +192,8 @@ gm20b_clk_new(struct nvkm_device *device, int index, struct
nvkm_clk **pclk)
return -ENOMEM;
*pclk = &clk->base;
- ret = _gk20a_clk_ctor(device, index, &gm20b_clk_speedo0,
- &gm20b_pllg_params, clk);
+ ret = gk20a_clk_ctor(device, index, &gm20b_clk_speedo0,
+ &gm20b_pllg_params, clk);
clk->pl_to_div = pl_to_div;
clk->div_to_pl = div_to_pl;
--
2.8.3
Alexandre Courbot
2016-Jun-01 08:39 UTC
[Nouveau] [PATCH 15/15] clk/gm20b: add glitchless and DFS support
This patch adds support for advanced features supported by the
Noise-Aware PLL of Maxwell. Glitchless switch allows the PL field to be
updated without disabling the PLL first if the SYNC_MODE bit of the CFG
register is set.
More significantly, DFS allows the PLL to monitor the actual input
voltage and to dynamically lower the output frequency accordingly. This
allows the clock to be more tolerant of lower voltages.
These improvements are only supported for Tegra speedos >= 1.
Also add the voltage table that is suitable for GM20B's NAPLL. This
change needs to be done atomically for the right voltages to be used by
the clock driver.
Signed-off-by: Alexandre Courbot <acourbot at nvidia.com>
---
drm/nouveau/nvkm/subdev/clk/gk20a.c | 89 +---
drm/nouveau/nvkm/subdev/clk/gk20a.h | 92 ++++
drm/nouveau/nvkm/subdev/clk/gm20b.c | 892 ++++++++++++++++++++++++++++++++++-
drm/nouveau/nvkm/subdev/volt/gm20b.c | 25 +-
4 files changed, 1000 insertions(+), 98 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.c
b/drm/nouveau/nvkm/subdev/clk/gk20a.c
index 607d74a81c5e..218893e3e5f9 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -28,70 +28,6 @@
#include <core/tegra.h>
#include <subdev/timer.h>
-#define KHZ (1000)
-#define MHZ (KHZ * 1000)
-
-#define MASK(w) ((1 << (w)) - 1)
-
-#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 GPC_BCASE_GPCPLL_CFG_BASE 0x00132800
-#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_VCODIV2 2
-#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)
-
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,
@@ -125,7 +61,7 @@ static const struct gk20a_clk_pllg_params gk20a_pllg_params =
{
.min_pl = 1, .max_pl = 32,
};
-static void
+void
gk20a_pllg_read_mnp(struct gk20a_clk *clk, struct gk20a_pll *pll)
{
struct nvkm_device *device = clk->base.subdev.device;
@@ -137,7 +73,7 @@ gk20a_pllg_read_mnp(struct gk20a_clk *clk, struct gk20a_pll
*pll)
pll->pl = (val >> GPCPLL_COEFF_P_SHIFT) &
MASK(GPCPLL_COEFF_P_WIDTH);
}
-static void
+void
gk20a_pllg_write_mnp(struct gk20a_clk *clk, const struct gk20a_pll *pll)
{
struct nvkm_device *device = clk->base.subdev.device;
@@ -149,7 +85,7 @@ gk20a_pllg_write_mnp(struct gk20a_clk *clk, const struct
gk20a_pll *pll)
nvkm_wr32(device, GPCPLL_COEFF, val);
}
-static u32
+u32
gk20a_pllg_calc_rate(struct gk20a_clk *clk, struct gk20a_pll *pll)
{
u32 rate;
@@ -161,14 +97,7 @@ gk20a_pllg_calc_rate(struct gk20a_clk *clk, struct gk20a_pll
*pll)
return rate / divider / 2;
}
-static u32
-gk20a_pllg_n_lo(struct gk20a_clk *clk, struct gk20a_pll *pll)
-{
- return DIV_ROUND_UP(pll->m * clk->params->min_vco,
- clk->parent_rate / KHZ);
-}
-
-static int
+int
gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate,
struct gk20a_pll *pll)
{
@@ -323,16 +252,6 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
return ret;
}
-static bool
-gk20a_pllg_is_enabled(struct gk20a_clk *clk)
-{
- struct nvkm_device *device = clk->base.subdev.device;
- u32 val;
-
- val = nvkm_rd32(device, GPCPLL_CFG);
- return val & GPCPLL_CFG_ENABLE;
-}
-
static int
gk20a_pllg_enable(struct gk20a_clk *clk)
{
diff --git a/drm/nouveau/nvkm/subdev/clk/gk20a.h
b/drm/nouveau/nvkm/subdev/clk/gk20a.h
index 73551f8209fa..0d1450972162 100644
--- a/drm/nouveau/nvkm/subdev/clk/gk20a.h
+++ b/drm/nouveau/nvkm/subdev/clk/gk20a.h
@@ -24,9 +24,79 @@
#ifndef __NVKM_CLK_GK20A_H__
#define __NVKM_CLK_GK20A_H__
+#define KHZ (1000)
+#define MHZ (KHZ * 1000)
+
+#define MASK(w) ((1 << (w)) - 1)
+
#define GK20A_CLK_GPC_MDIV 1000
#define SYS_GPCPLL_CFG_BASE 0x00137000
+#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_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_VCO_CTRL_SHIFT 0
+#define GPCPLL_CFG3_VCO_CTRL_WIDTH 9
+#define GPCPLL_CFG3_VCO_CTRL_MASK \
+ (MASK(GPCPLL_CFG3_VCO_CTRL_WIDTH) << GPCPLL_CFG3_VCO_CTRL_SHIFT)
+#define GPCPLL_CFG3_PLL_STEPB_SHIFT 16
+#define GPCPLL_CFG3_PLL_STEPB_WIDTH 8
+
+#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_N_MASK \
+ (MASK(GPCPLL_COEFF_N_WIDTH) << GPCPLL_COEFF_N_SHIFT)
+#define GPCPLL_COEFF_P_SHIFT 16
+#define GPCPLL_COEFF_P_WIDTH 6
+
+#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 GPC_BCAST_GPCPLL_CFG_BASE 0x00132800
+#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG (GPC_BCAST_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)
+
+#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_VCODIV2 2
+#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))
/* All frequencies in Khz */
struct gk20a_clk_pllg_params {
@@ -54,6 +124,28 @@ struct gk20a_clk {
};
#define gk20a_clk(p) container_of((p), struct gk20a_clk, base)
+u32 gk20a_pllg_calc_rate(struct gk20a_clk *, struct gk20a_pll *);
+int gk20a_pllg_calc_mnp(struct gk20a_clk *, unsigned long, struct gk20a_pll *);
+void gk20a_pllg_read_mnp(struct gk20a_clk *, struct gk20a_pll *);
+void gk20a_pllg_write_mnp(struct gk20a_clk *, const struct gk20a_pll *);
+
+static inline bool
+gk20a_pllg_is_enabled(struct gk20a_clk *clk)
+{
+ struct nvkm_device *device = clk->base.subdev.device;
+ u32 val;
+
+ val = nvkm_rd32(device, GPCPLL_CFG);
+ return val & GPCPLL_CFG_ENABLE;
+}
+
+static inline u32
+gk20a_pllg_n_lo(struct gk20a_clk *clk, struct gk20a_pll *pll)
+{
+ return DIV_ROUND_UP(pll->m * clk->params->min_vco,
+ clk->parent_rate / KHZ);
+}
+
int gk20a_clk_ctor(struct nvkm_device *, int, const struct nvkm_clk_func *,
const struct gk20a_clk_pllg_params *, struct gk20a_clk *);
void gk20a_clk_fini(struct nvkm_clk *);
diff --git a/drm/nouveau/nvkm/subdev/clk/gm20b.c
b/drm/nouveau/nvkm/subdev/clk/gm20b.c
index 68749bbbe623..3089e4bc5e54 100644
--- a/drm/nouveau/nvkm/subdev/clk/gm20b.c
+++ b/drm/nouveau/nvkm/subdev/clk/gm20b.c
@@ -21,20 +21,123 @@
*/
#include <subdev/clk.h>
+#include <subdev/volt.h>
+#include <subdev/timer.h>
#include <core/device.h>
+#include <core/tegra.h>
#include "priv.h"
#include "gk20a.h"
-#define KHZ (1000)
-#define MHZ (KHZ * 1000)
-
-#define MASK(w) ((1 << (w)) - 1)
+#define GPCPLL_CFG_SYNC_MODE BIT(2)
#define BYPASSCTRL_SYS (SYS_GPCPLL_CFG_BASE + 0x340)
#define BYPASSCTRL_SYS_GPCPLL_SHIFT 0
#define BYPASSCTRL_SYS_GPCPLL_WIDTH 1
+#define GPCPLL_CFG2_SDM_DIN_SHIFT 0
+#define GPCPLL_CFG2_SDM_DIN_WIDTH 8
+#define GPCPLL_CFG2_SDM_DIN_MASK \
+ (MASK(GPCPLL_CFG2_SDM_DIN_WIDTH) << GPCPLL_CFG2_SDM_DIN_SHIFT)
+#define GPCPLL_CFG2_SDM_DIN_NEW_SHIFT 8
+#define GPCPLL_CFG2_SDM_DIN_NEW_WIDTH 15
+#define GPCPLL_CFG2_SDM_DIN_NEW_MASK \
+ (MASK(GPCPLL_CFG2_SDM_DIN_NEW_WIDTH) << GPCPLL_CFG2_SDM_DIN_NEW_SHIFT)
+#define GPCPLL_CFG2_SETUP2_SHIFT 16
+#define GPCPLL_CFG2_PLL_STEPA_SHIFT 24
+
+#define GPCPLL_DVFS0 (SYS_GPCPLL_CFG_BASE + 0x10)
+#define GPCPLL_DVFS0_DFS_COEFF_SHIFT 0
+#define GPCPLL_DVFS0_DFS_COEFF_WIDTH 7
+#define GPCPLL_DVFS0_DFS_COEFF_MASK \
+ (MASK(GPCPLL_DVFS0_DFS_COEFF_WIDTH) << GPCPLL_DVFS0_DFS_COEFF_SHIFT)
+#define GPCPLL_DVFS0_DFS_DET_MAX_SHIFT 8
+#define GPCPLL_DVFS0_DFS_DET_MAX_WIDTH 7
+#define GPCPLL_DVFS0_DFS_DET_MAX_MASK \
+ (MASK(GPCPLL_DVFS0_DFS_DET_MAX_WIDTH) << GPCPLL_DVFS0_DFS_DET_MAX_SHIFT)
+
+#define GPCPLL_DVFS1 (SYS_GPCPLL_CFG_BASE + 0x14)
+#define GPCPLL_DVFS1_DFS_EXT_DET_SHIFT 0
+#define GPCPLL_DVFS1_DFS_EXT_DET_WIDTH 7
+#define GPCPLL_DVFS1_DFS_EXT_STRB_SHIFT 7
+#define GPCPLL_DVFS1_DFS_EXT_STRB_WIDTH 1
+#define GPCPLL_DVFS1_DFS_EXT_CAL_SHIFT 8
+#define GPCPLL_DVFS1_DFS_EXT_CAL_WIDTH 7
+#define GPCPLL_DVFS1_DFS_EXT_SEL_SHIFT 15
+#define GPCPLL_DVFS1_DFS_EXT_SEL_WIDTH 1
+#define GPCPLL_DVFS1_DFS_CTRL_SHIFT 16
+#define GPCPLL_DVFS1_DFS_CTRL_WIDTH 12
+#define GPCPLL_DVFS1_EN_SDM_SHIFT 28
+#define GPCPLL_DVFS1_EN_SDM_WIDTH 1
+#define GPCPLL_DVFS1_EN_SDM_BIT BIT(28)
+#define GPCPLL_DVFS1_EN_DFS_SHIFT 29
+#define GPCPLL_DVFS1_EN_DFS_WIDTH 1
+#define GPCPLL_DVFS1_EN_DFS_BIT BIT(29)
+#define GPCPLL_DVFS1_EN_DFS_CAL_SHIFT 30
+#define GPCPLL_DVFS1_EN_DFS_CAL_WIDTH 1
+#define GPCPLL_DVFS1_EN_DFS_CAL_BIT BIT(30)
+#define GPCPLL_DVFS1_DFS_CAL_DONE_SHIFT 31
+#define GPCPLL_DVFS1_DFS_CAL_DONE_WIDTH 1
+#define GPCPLL_DVFS1_DFS_CAL_DONE_BIT BIT(31)
+
+#define GPC_BCAST_GPCPLL_DVFS2 (GPC_BCAST_GPCPLL_CFG_BASE + 0x20)
+#define GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT BIT(16)
+
+#define GPCPLL_CFG3_PLL_DFS_TESTOUT_SHIFT 24
+#define GPCPLL_CFG3_PLL_DFS_TESTOUT_WIDTH 7
+
+#define DFS_DET_RANGE 6 /* -2^6 ... 2^6-1 */
+#define SDM_DIN_RANGE 12 /* -2^12 ... 2^12-1 */
+
+struct gm20b_clk_dvfs_params {
+ s32 coeff_slope;
+ s32 coeff_offs;
+ u32 vco_ctrl;
+};
+
+static const struct gm20b_clk_dvfs_params gm20b_dvfs_params = {
+ .coeff_slope = -165230,
+ .coeff_offs = 214007,
+ .vco_ctrl = 0x7 << 3,
+};
+
+/*
+ * base.n is now the *integer* part of the N factor.
+ * sdm_din contains n's decimal part.
+ */
+struct gm20b_pll {
+ struct gk20a_pll base;
+ u32 sdm_din;
+};
+
+struct gm20b_clk_dvfs {
+ u32 dfs_coeff;
+ s32 dfs_det_max;
+ s32 dfs_ext_cal;
+};
+
+struct gm20b_clk {
+ /* currently applied parameters */
+ struct gk20a_clk base;
+ struct gm20b_clk_dvfs dvfs;
+ u32 uv;
+
+ /* new parameters to apply */
+ struct gk20a_pll new_pll;
+ struct gm20b_clk_dvfs new_dvfs;
+ u32 new_uv;
+
+ const struct gm20b_clk_dvfs_params *dvfs_params;
+
+ /* fused parameters */
+ s32 uvdet_slope;
+ s32 uvdet_offs;
+
+ /* safe frequency we can use at minimum voltage */
+ u32 safe_fmax_vmin;
+};
+#define gm20b_clk(p) container_of((gk20a_clk(p)), struct gm20b_clk, base)
+
static u32 pl_to_div(u32 pl)
{
return pl;
@@ -53,6 +156,484 @@ static const struct gk20a_clk_pllg_params gm20b_pllg_params
= {
.min_pl = 1, .max_pl = 31,
};
+static void
+gm20b_pllg_read_mnp(struct gm20b_clk *clk, struct gm20b_pll *pll)
+{
+ struct nvkm_subdev *subdev = &clk->base.base.subdev;
+ struct nvkm_device *device = subdev->device;
+ u32 val;
+
+ gk20a_pllg_read_mnp(&clk->base, &pll->base);
+ val = nvkm_rd32(device, GPCPLL_CFG2);
+ pll->sdm_din = (val >> GPCPLL_CFG2_SDM_DIN_SHIFT) &
+ MASK(GPCPLL_CFG2_SDM_DIN_WIDTH);
+}
+
+static void
+gm20b_pllg_write_mnp(struct gm20b_clk *clk, const struct gm20b_pll *pll)
+{
+ struct nvkm_device *device = clk->base.base.subdev.device;
+
+ nvkm_mask(device, GPCPLL_CFG2, GPCPLL_CFG2_SDM_DIN_MASK,
+ pll->sdm_din << GPCPLL_CFG2_SDM_DIN_SHIFT);
+ gk20a_pllg_write_mnp(&clk->base, &pll->base);
+}
+
+/*
+ * Determine DFS_COEFF for the requested voltage. Always select external
+ * calibration override equal to the voltage, and set maximum detection
+ * limit "0" (to make sure that PLL output remains under F/V curve
when
+ * voltage increases).
+ */
+static void
+gm20b_dvfs_calc_det_coeff(struct gm20b_clk *clk, s32 uv,
+ struct gm20b_clk_dvfs *dvfs)
+{
+ struct nvkm_subdev *subdev = &clk->base.base.subdev;
+ const struct gm20b_clk_dvfs_params *p = clk->dvfs_params;
+ u32 coeff;
+ /* Work with mv as uv would likely trigger an overflow */
+ s32 mv = DIV_ROUND_CLOSEST(uv, 1000);
+
+ /* coeff = slope * voltage + offset */
+ coeff = DIV_ROUND_CLOSEST(mv * p->coeff_slope, 1000) + p->coeff_offs;
+ coeff = DIV_ROUND_CLOSEST(coeff, 1000);
+ dvfs->dfs_coeff = min_t(u32, coeff, MASK(GPCPLL_DVFS0_DFS_COEFF_WIDTH));
+
+ dvfs->dfs_ext_cal = DIV_ROUND_CLOSEST(uv - clk->uvdet_offs,
+ clk->uvdet_slope);
+ /* should never happen */
+ if (abs(dvfs->dfs_ext_cal) >= BIT(DFS_DET_RANGE))
+ nvkm_error(subdev, "dfs_ext_cal overflow!\n");
+
+ dvfs->dfs_det_max = 0;
+
+ nvkm_debug(subdev, "%s uv: %d coeff: %x, ext_cal: %d, det_max:
%d\n",
+ __func__, uv, dvfs->dfs_coeff, dvfs->dfs_ext_cal,
+ dvfs->dfs_det_max);
+}
+
+/*
+ * Solve equation for integer and fractional part of the effective NDIV:
+ *
+ * n_eff = n_int + 1/2 + (SDM_DIN / 2^(SDM_DIN_RANGE + 1)) +
+ * (DVFS_COEFF * DVFS_DET_DELTA) / 2^DFS_DET_RANGE
+ *
+ * The SDM_DIN LSB is finally shifted out, since it is not accessible by sw.
+ */
+static void
+gm20b_dvfs_calc_ndiv(struct gm20b_clk *clk, u32 n_eff, u32 *n_int, u32
*sdm_din)
+{
+ struct nvkm_subdev *subdev = &clk->base.base.subdev;
+ const struct gk20a_clk_pllg_params *p = clk->base.params;
+ u32 n;
+ s32 det_delta;
+ u32 rem, rem_range;
+
+ /* calculate current ext_cal and subtract previous one */
+ det_delta = DIV_ROUND_CLOSEST(((s32)clk->uv) - clk->uvdet_offs,
+ clk->uvdet_slope);
+ det_delta -= clk->dvfs.dfs_ext_cal;
+ det_delta = min(det_delta, clk->dvfs.dfs_det_max);
+ det_delta *= clk->dvfs.dfs_coeff;
+
+ /* integer part of n */
+ n = (n_eff << DFS_DET_RANGE) - det_delta;
+ /* should never happen! */
+ if (n <= 0) {
+ nvkm_error(subdev, "ndiv <= 0 - setting to 1...\n");
+ n = 1 << DFS_DET_RANGE;
+ }
+ if (n >> DFS_DET_RANGE > p->max_n) {
+ nvkm_error(subdev, "ndiv > max_n - setting to max_n...\n");
+ n = p->max_n << DFS_DET_RANGE;
+ }
+ *n_int = n >> DFS_DET_RANGE;
+
+ /* fractional part of n */
+ rem = ((u32)n) & MASK(DFS_DET_RANGE);
+ rem_range = SDM_DIN_RANGE + 1 - DFS_DET_RANGE;
+ /* subtract 2^SDM_DIN_RANGE to account for the 1/2 of the equation */
+ rem = (rem << rem_range) - BIT(SDM_DIN_RANGE);
+ /* lose 8 LSB and clip - sdm_din only keeps the most significant byte */
+ *sdm_din = (rem >> BITS_PER_BYTE) & MASK(GPCPLL_CFG2_SDM_DIN_WIDTH);
+
+ nvkm_debug(subdev, "%s n_eff: %d, n_int: %d, sdm_din: %d\n",
__func__,
+ n_eff, *n_int, *sdm_din);
+}
+
+static int
+gm20b_pllg_slide(struct gm20b_clk *clk, u32 n)
+{
+ struct nvkm_subdev *subdev = &clk->base.base.subdev;
+ struct nvkm_device *device = subdev->device;
+ struct gm20b_pll pll;
+ u32 n_int, sdm_din;
+ int ret = 0;
+
+ /* calculate the new n_int/sdm_din for this n/uv */
+ gm20b_dvfs_calc_ndiv(clk, n, &n_int, &sdm_din);
+
+ /* get old coefficients */
+ gm20b_pllg_read_mnp(clk, &pll);
+ /* do nothing if NDIV is the same */
+ if (n_int == pll.base.n && sdm_din == pll.sdm_din)
+ return 0;
+
+ /* pll slowdown mode */
+ nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
+ BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT),
+ BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT));
+
+ /* new ndiv ready for ramp */
+ /* in DVFS mode SDM is updated via "new" field */
+ nvkm_mask(device, GPCPLL_CFG2, GPCPLL_CFG2_SDM_DIN_NEW_MASK,
+ sdm_din << GPCPLL_CFG2_SDM_DIN_NEW_SHIFT);
+ pll.base.n = n_int;
+ udelay(1);
+ gk20a_pllg_write_mnp(&clk->base, &pll.base);
+
+ /* dynamic ramp to new ndiv */
+ udelay(1);
+ nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
+ BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT),
+ BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT));
+
+ /* wait for ramping to complete */
+ if (nvkm_wait_usec(device, 500, GPC_BCAST_NDIV_SLOWDOWN_DEBUG,
+ GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK,
+ GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK) < 0)
+ ret = -ETIMEDOUT;
+
+ /* in DVFS mode complete SDM update */
+ nvkm_mask(device, GPCPLL_CFG2, GPCPLL_CFG2_SDM_DIN_MASK,
+ sdm_din << GPCPLL_CFG2_SDM_DIN_SHIFT);
+
+ /* exit slowdown mode */
+ nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
+ BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT) |
+ BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT), 0);
+ nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
+
+ return ret;
+}
+
+static int
+gm20b_pllg_enable(struct gm20b_clk *clk)
+{
+ struct nvkm_device *device = clk->base.base.subdev.device;
+
+ nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, GPCPLL_CFG_ENABLE);
+ nvkm_rd32(device, GPCPLL_CFG);
+
+ /* In DVFS mode lock cannot be used - so just delay */
+ udelay(40);
+
+ /* set SYNC_MODE for glitchless switch out of bypass */
+ nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_SYNC_MODE,
+ GPCPLL_CFG_SYNC_MODE);
+ nvkm_rd32(device, GPCPLL_CFG);
+
+ /* switch to VCO mode */
+ nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT),
+ BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+
+ return 0;
+}
+
+static void
+gm20b_pllg_disable(struct gm20b_clk *clk)
+{
+ struct nvkm_device *device = clk->base.base.subdev.device;
+
+ /* put PLL in bypass before disabling it */
+ nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
+
+ /* clear SYNC_MODE before disabling PLL */
+ nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_SYNC_MODE, 0);
+
+ nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, 0);
+ nvkm_rd32(device, GPCPLL_CFG);
+}
+
+static int
+gm20b_pllg_program_mnp(struct gm20b_clk *clk, const struct gk20a_pll *pll)
+{
+ struct nvkm_subdev *subdev = &clk->base.base.subdev;
+ struct nvkm_device *device = subdev->device;
+ struct gm20b_pll cur_pll;
+ u32 n_int, sdm_din;
+ /* if we only change pdiv, we can do a glitchless transition */
+ bool pdiv_only;
+ int ret;
+
+ gm20b_dvfs_calc_ndiv(clk, pll->n, &n_int, &sdm_din);
+ gm20b_pllg_read_mnp(clk, &cur_pll);
+ pdiv_only = cur_pll.base.n == n_int && cur_pll.sdm_din == sdm_din
&&
+ cur_pll.base.m == pll->m;
+
+ /* need full sequence if clock not enabled yet */
+ if (!gk20a_pllg_is_enabled(&clk->base))
+ pdiv_only = false;
+
+ /* split VCO-to-bypass jump in half by setting out divider 1:2 */
+ nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+ GPC2CLK_OUT_VCODIV2 << GPC2CLK_OUT_VCODIV_SHIFT);
+ /* Intentional 2nd write to assure linear divider operation */
+ nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+ GPC2CLK_OUT_VCODIV2 << GPC2CLK_OUT_VCODIV_SHIFT);
+ nvkm_rd32(device, GPC2CLK_OUT);
+ udelay(2);
+
+ if (pdiv_only) {
+ u32 old = cur_pll.base.pl;
+ u32 new = pll->pl;
+
+ /*
+ * we can do a glitchless transition only if the old and new PL
+ * parameters share at least one bit set to 1. If this is not
+ * the case, calculate and program an interim PL that will allow
+ * us to respect that rule.
+ */
+ if ((old & new) == 0) {
+ cur_pll.base.pl = min(old | BIT(ffs(new) - 1),
+ new | BIT(ffs(old) - 1));
+ gk20a_pllg_write_mnp(&clk->base, &cur_pll.base);
+ }
+
+ cur_pll.base.pl = new;
+ gk20a_pllg_write_mnp(&clk->base, &cur_pll.base);
+ } else {
+ /* disable before programming if more than pdiv changes */
+ gm20b_pllg_disable(clk);
+
+ cur_pll.base = *pll;
+ cur_pll.base.n = n_int;
+ cur_pll.sdm_din = sdm_din;
+ gm20b_pllg_write_mnp(clk, &cur_pll);
+
+ ret = gm20b_pllg_enable(clk);
+ if (ret)
+ return ret;
+ }
+
+ /* restore out divider 1:1 */
+ udelay(2);
+ nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+ GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT);
+ /* Intentional 2nd write to assure linear divider operation */
+ nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+ GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT);
+ nvkm_rd32(device, GPC2CLK_OUT);
+
+ return 0;
+}
+
+static int
+gm20b_pllg_program_mnp_slide(struct gm20b_clk *clk, const struct gk20a_pll
*pll)
+{
+ struct gk20a_pll cur_pll;
+ int ret;
+
+ if (gk20a_pllg_is_enabled(&clk->base)) {
+ gk20a_pllg_read_mnp(&clk->base, &cur_pll);
+
+ /* just do NDIV slide if there is no change to M and PL */
+ if (pll->m == cur_pll.m && pll->pl == cur_pll.pl)
+ return gm20b_pllg_slide(clk, pll->n);
+
+ /* slide down to current NDIV_LO */
+ cur_pll.n = gk20a_pllg_n_lo(&clk->base, &cur_pll);
+ ret = gm20b_pllg_slide(clk, cur_pll.n);
+ if (ret)
+ return ret;
+ }
+
+ /* program MNP with the new clock parameters and new NDIV_LO */
+ cur_pll = *pll;
+ cur_pll.n = gk20a_pllg_n_lo(&clk->base, &cur_pll);
+ ret = gm20b_pllg_program_mnp(clk, &cur_pll);
+ if (ret)
+ return ret;
+
+ /* slide up to new NDIV */
+ return gm20b_pllg_slide(clk, pll->n);
+}
+
+static int
+gm20b_clk_calc(struct nvkm_clk *base, struct nvkm_cstate *cstate)
+{
+ struct gm20b_clk *clk = gm20b_clk(base);
+ struct nvkm_subdev *subdev = &base->subdev;
+ struct nvkm_volt *volt = base->subdev.device->volt;
+ int ret;
+
+ ret = gk20a_pllg_calc_mnp(&clk->base, cstate->domain[nv_clk_src_gpc]
*
+ GK20A_CLK_GPC_MDIV, &clk->new_pll);
+ if (ret)
+ return ret;
+
+ clk->new_uv = volt->vid[cstate->voltage].uv;
+ gm20b_dvfs_calc_det_coeff(clk, clk->new_uv, &clk->new_dvfs);
+
+ nvkm_debug(subdev, "%s uv: %d uv\n", __func__, clk->new_uv);
+
+ return 0;
+}
+
+/*
+ * Compute PLL parameters that are always safe for the current voltage
+ */
+static void
+gm20b_dvfs_calc_safe_pll(struct gm20b_clk *clk, struct gk20a_pll *pll)
+{
+ u32 rate = gk20a_pllg_calc_rate(&clk->base, pll) / KHZ;
+ u32 parent_rate = clk->base.parent_rate / KHZ;
+ u32 nmin, nsafe;
+
+ /* remove a safe margin of 10% */
+ if (rate > clk->safe_fmax_vmin)
+ rate = rate * (100 - 10) / 100;
+
+ /* gpc2clk */
+ rate *= 2;
+
+ nmin = DIV_ROUND_UP(pll->m * clk->base.params->min_vco, parent_rate);
+ nsafe = pll->m * rate / (clk->base.parent_rate);
+
+ if (nsafe < nmin) {
+ pll->pl = DIV_ROUND_UP(nmin * parent_rate, pll->m * rate);
+ nsafe = nmin;
+ }
+
+ pll->n = nsafe;
+}
+
+static void
+gm20b_dvfs_program_coeff(struct gm20b_clk *clk, u32 coeff)
+{
+ struct nvkm_device *device = clk->base.base.subdev.device;
+
+ /* strobe to read external DFS coefficient */
+ nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2,
+ GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT,
+ GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT);
+
+ nvkm_mask(device, GPCPLL_DVFS0, GPCPLL_DVFS0_DFS_COEFF_MASK,
+ coeff << GPCPLL_DVFS0_DFS_COEFF_SHIFT);
+
+ udelay(1);
+ nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2,
+ GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT, 0);
+}
+
+static void
+gm20b_dvfs_program_ext_cal(struct gm20b_clk *clk, u32 dfs_det_cal)
+{
+ struct nvkm_device *device = clk->base.base.subdev.device;
+ u32 val;
+
+ nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2, MASK(DFS_DET_RANGE + 1),
+ dfs_det_cal);
+ udelay(1);
+
+ val = nvkm_rd32(device, GPCPLL_DVFS1);
+ if (!(val & BIT(25))) {
+ /* Use external value to overwrite calibration value */
+ val |= BIT(25) | BIT(16);
+ nvkm_wr32(device, GPCPLL_DVFS1, val);
+ }
+}
+
+static void
+gm20b_dvfs_program_dfs_detection(struct gm20b_clk *clk,
+ struct gm20b_clk_dvfs *dvfs)
+{
+ struct nvkm_device *device = clk->base.base.subdev.device;
+
+ /* strobe to read external DFS coefficient */
+ nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2,
+ GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT,
+ GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT);
+
+ nvkm_mask(device, GPCPLL_DVFS0,
+ GPCPLL_DVFS0_DFS_COEFF_MASK | GPCPLL_DVFS0_DFS_DET_MAX_MASK,
+ dvfs->dfs_coeff << GPCPLL_DVFS0_DFS_COEFF_SHIFT |
+ dvfs->dfs_det_max << GPCPLL_DVFS0_DFS_DET_MAX_SHIFT);
+
+ udelay(1);
+ nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2,
+ GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT, 0);
+
+ gm20b_dvfs_program_ext_cal(clk, dvfs->dfs_ext_cal);
+}
+
+static int
+gm20b_clk_prog(struct nvkm_clk *base)
+{
+ struct gm20b_clk *clk = gm20b_clk(base);
+ u32 cur_freq;
+ int ret;
+
+ /* No change in DVFS settings? */
+ if (clk->uv == clk->new_uv)
+ goto prog;
+
+ /*
+ * Interim step for changing DVFS detection settings: low enough
+ * frequency to be safe at at DVFS coeff = 0.
+ *
+ * 1. If voltage is increasing:
+ * - safe frequency target matches the lowest - old - frequency
+ * - DVFS settings are still old
+ * - Voltage already increased to new level by volt, but maximum
+ * detection limit assures PLL output remains under F/V curve
+ *
+ * 2. If voltage is decreasing:
+ * - safe frequency target matches the lowest - new - frequency
+ * - DVFS settings are still old
+ * - Voltage is also old, it will be lowered by volt afterwards
+ *
+ * Interim step can be skipped if old frequency is below safe minimum,
+ * i.e., it is low enough to be safe at any voltage in operating range
+ * with zero DVFS coefficient.
+ */
+ cur_freq = nvkm_clk_read(&clk->base.base, nv_clk_src_gpc);
+ if (cur_freq > clk->safe_fmax_vmin) {
+ struct gk20a_pll pll_safe;
+
+ if (clk->uv < clk->new_uv)
+ /* voltage will raise: safe frequency is current one */
+ pll_safe = clk->base.pll;
+ else
+ /* voltage will drop: safe frequency is new one */
+ pll_safe = clk->new_pll;
+
+ gm20b_dvfs_calc_safe_pll(clk, &pll_safe);
+ ret = gm20b_pllg_program_mnp_slide(clk, &pll_safe);
+ if (ret)
+ return ret;
+ }
+
+ /*
+ * DVFS detection settings transition:
+ * - Set DVFS coefficient zero
+ * - Set calibration level to new voltage
+ * - Set DVFS coefficient to match new voltage
+ */
+ gm20b_dvfs_program_coeff(clk, 0);
+ gm20b_dvfs_program_ext_cal(clk, clk->new_dvfs.dfs_ext_cal);
+ gm20b_dvfs_program_coeff(clk, clk->new_dvfs.dfs_coeff);
+ gm20b_dvfs_program_dfs_detection(clk, &clk->new_dvfs);
+
+prog:
+ clk->uv = clk->new_uv;
+ clk->dvfs = clk->new_dvfs;
+ clk->base.pll = clk->new_pll;
+
+ return gm20b_pllg_program_mnp_slide(clk, &clk->base.pll);
+}
+
static struct nvkm_pstate
gm20b_pstates[] = {
{
@@ -133,9 +714,99 @@ gm20b_pstates[] = {
.voltage = 12,
},
},
-
};
+static void
+gm20b_clk_fini(struct nvkm_clk *base)
+{
+ struct nvkm_device *device = base->subdev.device;
+ struct gm20b_clk *clk = gm20b_clk(base);
+
+ /* slide to VCO min */
+ if (gk20a_pllg_is_enabled(&clk->base)) {
+ struct gk20a_pll pll;
+ u32 n_lo;
+
+ gk20a_pllg_read_mnp(&clk->base, &pll);
+ n_lo = gk20a_pllg_n_lo(&clk->base, &pll);
+ gm20b_pllg_slide(clk, n_lo);
+ }
+
+ gm20b_pllg_disable(clk);
+
+ /* set IDDQ */
+ nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_IDDQ, 1);
+}
+
+static int
+gm20b_clk_init_dvfs(struct gm20b_clk *clk)
+{
+ struct nvkm_subdev *subdev = &clk->base.base.subdev;
+ struct nvkm_device *device = subdev->device;
+ bool fused = clk->uvdet_offs && clk->uvdet_slope;
+ static const s32 ADC_SLOPE_UV = 10000; /* default ADC detection slope */
+ u32 data;
+ int ret;
+
+ /* Enable NA DVFS */
+ nvkm_mask(device, GPCPLL_DVFS1, GPCPLL_DVFS1_EN_DFS_BIT,
+ GPCPLL_DVFS1_EN_DFS_BIT);
+
+ /* Set VCO_CTRL */
+ if (clk->dvfs_params->vco_ctrl)
+ nvkm_mask(device, GPCPLL_CFG3, GPCPLL_CFG3_VCO_CTRL_MASK,
+ clk->dvfs_params->vco_ctrl << GPCPLL_CFG3_VCO_CTRL_SHIFT);
+
+ if (fused) {
+ /* Start internal calibration, but ignore results */
+ nvkm_mask(device, GPCPLL_DVFS1, GPCPLL_DVFS1_EN_DFS_CAL_BIT,
+ GPCPLL_DVFS1_EN_DFS_CAL_BIT);
+
+ /* got uvdev parameters from fuse, skip calibration */
+ goto calibrated;
+ }
+
+ /*
+ * If calibration parameters are not fused, start internal calibration,
+ * wait for completion, and use results along with default slope to
+ * calculate ADC offset during boot.
+ */
+ nvkm_mask(device, GPCPLL_DVFS1, GPCPLL_DVFS1_EN_DFS_CAL_BIT,
+ GPCPLL_DVFS1_EN_DFS_CAL_BIT);
+
+ /* Wait for internal calibration done (spec < 2us). */
+ ret = nvkm_wait_usec(device, 10, GPCPLL_DVFS1,
+ GPCPLL_DVFS1_DFS_CAL_DONE_BIT,
+ GPCPLL_DVFS1_DFS_CAL_DONE_BIT);
+ if (ret < 0) {
+ nvkm_error(subdev, "GPCPLL calibration timeout\n");
+ return -ETIMEDOUT;
+ }
+
+ data = nvkm_rd32(device, GPCPLL_CFG3) >>
+ GPCPLL_CFG3_PLL_DFS_TESTOUT_SHIFT;
+ data &= MASK(GPCPLL_CFG3_PLL_DFS_TESTOUT_WIDTH);
+
+ clk->uvdet_slope = ADC_SLOPE_UV;
+ clk->uvdet_offs = ((s32)clk->uv) - data * ADC_SLOPE_UV;
+
+ nvkm_debug(subdev, "calibrated DVFS parameters: offs %d, slope
%d\n",
+ clk->uvdet_offs, clk->uvdet_slope);
+
+calibrated:
+ /* Compute and apply initial DVFS parameters */
+ gm20b_dvfs_calc_det_coeff(clk, clk->uv, &clk->dvfs);
+ gm20b_dvfs_program_coeff(clk, 0);
+ gm20b_dvfs_program_ext_cal(clk, clk->dvfs.dfs_ext_cal);
+ gm20b_dvfs_program_coeff(clk, clk->dvfs.dfs_coeff);
+ gm20b_dvfs_program_dfs_detection(clk, &clk->new_dvfs);
+
+ return 0;
+}
+
+/* Forward declaration to detect speedo >=1 in gm20b_clk_init() */
+static const struct nvkm_clk_func gm20b_clk;
+
static int
gm20b_clk_init(struct nvkm_clk *base)
{
@@ -143,19 +814,56 @@ gm20b_clk_init(struct nvkm_clk *base)
struct nvkm_subdev *subdev = &clk->base.subdev;
struct nvkm_device *device = subdev->device;
int ret;
+ u32 data;
- ret = gk20a_clk_setup_slide(clk);
- if (ret)
- return ret;
+ /* get out from IDDQ */
+ nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_IDDQ, 0);
+ nvkm_rd32(device, GPCPLL_CFG);
+ udelay(5);
+
+ nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_INIT_MASK,
+ GPC2CLK_OUT_INIT_VAL);
/* Set the global bypass control to VCO */
nvkm_mask(device, BYPASSCTRL_SYS,
MASK(BYPASSCTRL_SYS_GPCPLL_WIDTH) << BYPASSCTRL_SYS_GPCPLL_SHIFT,
0);
+ ret = gk20a_clk_setup_slide(clk);
+ if (ret)
+ return ret;
+
+ /* If not fused, set RAM SVOP PDP data 0x2, and enable fuse override */
+ data = nvkm_rd32(device, 0x021944);
+ if (!(data & 0x3)) {
+ data |= 0x2;
+ nvkm_wr32(device, 0x021944, data);
+
+ data = nvkm_rd32(device, 0x021948);
+ data |= 0x1;
+ nvkm_wr32(device, 0x021948, data);
+ }
+
+ /* Disable idle slow down */
+ nvkm_mask(device, 0x20160, 0x003f0000, 0x0);
+
+ /* speedo >= 1? */
+ if (clk->base.func == &gm20b_clk) {
+ struct gm20b_clk *_clk = gm20b_clk(base);
+ struct nvkm_volt *volt = device->volt;
+
+ /* Get current voltage */
+ _clk->uv = nvkm_volt_get(volt);
+
+ /* Initialize DVFS */
+ ret = gm20b_clk_init_dvfs(_clk);
+ if (ret)
+ return ret;
+ }
+
/* Start with lowest frequency */
base->func->calc(base, &base->func->pstates[0].base);
- ret = base->func->prog(&clk->base);
+ ret = base->func->prog(base);
if (ret) {
nvkm_error(subdev, "cannot initialize clock\n");
return ret;
@@ -173,6 +881,7 @@ gm20b_clk_speedo0 = {
.prog = gk20a_clk_prog,
.tidy = gk20a_clk_tidy,
.pstates = gm20b_pstates,
+ /* Speedo 0 only supports 12 voltages */
.nr_pstates = ARRAY_SIZE(gm20b_pstates) - 1,
.domains = {
{ nv_clk_src_crystal, 0xff },
@@ -181,8 +890,26 @@ gm20b_clk_speedo0 = {
},
};
-int
-gm20b_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
+static const struct nvkm_clk_func
+gm20b_clk = {
+ .init = gm20b_clk_init,
+ .fini = gm20b_clk_fini,
+ .read = gk20a_clk_read,
+ .calc = gm20b_clk_calc,
+ .prog = gm20b_clk_prog,
+ .tidy = gk20a_clk_tidy,
+ .pstates = gm20b_pstates,
+ .nr_pstates = ARRAY_SIZE(gm20b_pstates),
+ .domains = {
+ { nv_clk_src_crystal, 0xff },
+ { nv_clk_src_gpc, 0xff, 0, "core", GK20A_CLK_GPC_MDIV },
+ { nv_clk_src_max },
+ },
+};
+
+static int
+gm20b_clk_new_speedo0(struct nvkm_device *device, int index,
+ struct nvkm_clk **pclk)
{
struct gk20a_clk *clk;
int ret;
@@ -200,3 +927,146 @@ gm20b_clk_new(struct nvkm_device *device, int index,
struct nvkm_clk **pclk)
return ret;
}
+
+/* FUSE register */
+#define FUSE_RESERVED_CALIB0 0x204
+#define FUSE_RESERVED_CALIB0_INTERCEPT_FRAC_SHIFT 0
+#define FUSE_RESERVED_CALIB0_INTERCEPT_FRAC_WIDTH 4
+#define FUSE_RESERVED_CALIB0_INTERCEPT_INT_SHIFT 4
+#define FUSE_RESERVED_CALIB0_INTERCEPT_INT_WIDTH 10
+#define FUSE_RESERVED_CALIB0_SLOPE_FRAC_SHIFT 14
+#define FUSE_RESERVED_CALIB0_SLOPE_FRAC_WIDTH 10
+#define FUSE_RESERVED_CALIB0_SLOPE_INT_SHIFT 24
+#define FUSE_RESERVED_CALIB0_SLOPE_INT_WIDTH 6
+#define FUSE_RESERVED_CALIB0_FUSE_REV_SHIFT 30
+#define FUSE_RESERVED_CALIB0_FUSE_REV_WIDTH 2
+
+static int
+gm20b_clk_init_fused_params(struct gm20b_clk *clk)
+{
+ struct nvkm_subdev *subdev = &clk->base.base.subdev;
+ u32 val;
+ u32 rev;
+
+ tegra_fuse_readl(FUSE_RESERVED_CALIB0, &val);
+ rev = (val >> FUSE_RESERVED_CALIB0_FUSE_REV_SHIFT) &
+ MASK(FUSE_RESERVED_CALIB0_FUSE_REV_WIDTH);
+
+ /* No fused parameters, we will calibrate later */
+ if (rev == 0)
+ return -EINVAL;
+
+ /* Integer part in mV + fractional part in uV */
+ clk->uvdet_slope = ((val >> FUSE_RESERVED_CALIB0_SLOPE_INT_SHIFT)
&
+ MASK(FUSE_RESERVED_CALIB0_SLOPE_INT_WIDTH)) * 1000 +
+ ((val >> FUSE_RESERVED_CALIB0_SLOPE_FRAC_SHIFT) &
+ MASK(FUSE_RESERVED_CALIB0_SLOPE_FRAC_WIDTH));
+
+ /* Integer part in mV + fractional part in 100uV */
+ clk->uvdet_offs = ((val >> FUSE_RESERVED_CALIB0_INTERCEPT_INT_SHIFT)
&
+ MASK(FUSE_RESERVED_CALIB0_INTERCEPT_INT_WIDTH)) * 1000 +
+ ((val >> FUSE_RESERVED_CALIB0_INTERCEPT_FRAC_SHIFT) &
+ MASK(FUSE_RESERVED_CALIB0_INTERCEPT_FRAC_WIDTH)) * 100;
+
+ nvkm_debug(subdev, "fused calibration data: slope %d, offs %d\n",
+ clk->uvdet_slope, clk->uvdet_offs);
+ return 0;
+}
+
+static int
+gm20b_clk_init_safe_fmax(struct gm20b_clk *clk)
+{
+ struct nvkm_subdev *subdev = &clk->base.base.subdev;
+ struct nvkm_volt *volt = subdev->device->volt;
+ struct nvkm_pstate *pstates = clk->base.base.func->pstates;
+ int nr_pstates = clk->base.base.func->nr_pstates;
+ int vmin, id = 0;
+ u32 fmax = 0;
+ int i;
+
+ /* find lowest voltage we can use */
+ vmin = volt->vid[0].uv;
+ for (i = 1; i < volt->vid_nr; i++) {
+ if (volt->vid[i].uv <= vmin) {
+ vmin = volt->vid[i].uv;
+ id = volt->vid[i].vid;
+ }
+ }
+
+ /* find max frequency at this voltage */
+ for (i = 0; i < nr_pstates; i++)
+ if (pstates[i].base.voltage == id)
+ fmax = max(fmax,
+ pstates[i].base.domain[nv_clk_src_gpc]);
+
+ if (!fmax) {
+ nvkm_error(subdev, "failed to evaluate safe fmax\n");
+ return -EINVAL;
+ }
+
+ /* we are safe at 90% of the max frequency */
+ clk->safe_fmax_vmin = fmax * (100 - 10) / 100;
+ nvkm_debug(subdev, "safe fmax @ vmin = %u Khz\n",
clk->safe_fmax_vmin);
+
+ return 0;
+}
+
+int
+gm20b_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
+{
+ struct nvkm_device_tegra *tdev = device->func->tegra(device);
+ struct gm20b_clk *clk;
+ struct nvkm_subdev *subdev;
+ struct gk20a_clk_pllg_params *clk_params;
+ int ret;
+
+ /* Speedo 0 GPUs cannot use noise-aware PLL */
+ if (tdev->gpu_speedo_id == 0)
+ return gm20b_clk_new_speedo0(device, index, pclk);
+
+ /* Speedo >= 1, use NAPLL */
+ clk = kzalloc(sizeof(*clk) + sizeof(*clk_params), GFP_KERNEL);
+ if (!clk)
+ return -ENOMEM;
+ *pclk = &clk->base.base;
+ subdev = &clk->base.base.subdev;
+
+ /* duplicate the clock parameters since we will patch them below */
+ clk_params = (void *) (clk + 1);
+ *clk_params = gm20b_pllg_params;
+ ret = gk20a_clk_ctor(device, index, &gm20b_clk, clk_params,
+ &clk->base);
+ if (ret)
+ return ret;
+
+ /*
+ * NAPLL can only work with max_u, clamp the m range so
+ * gk20a_pllg_calc_mnp always uses it
+ */
+ clk_params->max_m = clk_params->min_m =
DIV_ROUND_UP(clk_params->max_u,
+ (clk->base.parent_rate / KHZ));
+ if (clk_params->max_m == 0) {
+ nvkm_warn(subdev, "cannot use NAPLL, using legacy clock...\n");
+ kfree(clk);
+ return gm20b_clk_new_speedo0(device, index, pclk);
+ }
+
+ clk->base.pl_to_div = pl_to_div;
+ clk->base.div_to_pl = div_to_pl;
+
+ clk->dvfs_params = &gm20b_dvfs_params;
+
+ ret = gm20b_clk_init_fused_params(clk);
+ /*
+ * we will calibrate during init - should never happen on
+ * prod parts
+ */
+ if (ret)
+ nvkm_warn(subdev, "no fused calibration parameters\n");
+
+ ret = gm20b_clk_init_safe_fmax(clk);
+ if (ret)
+ return ret;
+
+ return 0;
+}
diff --git a/drm/nouveau/nvkm/subdev/volt/gm20b.c
b/drm/nouveau/nvkm/subdev/volt/gm20b.c
index d2d39ab74b8e..74db4d28930f 100644
--- a/drm/nouveau/nvkm/subdev/volt/gm20b.c
+++ b/drm/nouveau/nvkm/subdev/volt/gm20b.c
@@ -41,6 +41,23 @@ const struct cvb_coef gm20b_cvb_coef[] = {
/* 921600 */ { 2647676, -106455, 1632 },
};
+static const struct cvb_coef gm20b_na_cvb_coef[] = {
+ /* KHz, c0, c1, c2, c3, c4, c5 */
+ /* 76800 */ { 814294, 8144, -940, 808, -21583, 226 },
+ /* 153600 */ { 856185, 8144, -940, 808, -21583, 226 },
+ /* 230400 */ { 898077, 8144, -940, 808, -21583, 226 },
+ /* 307200 */ { 939968, 8144, -940, 808, -21583, 226 },
+ /* 384000 */ { 981860, 8144, -940, 808, -21583, 226 },
+ /* 460800 */ { 1023751, 8144, -940, 808, -21583, 226 },
+ /* 537600 */ { 1065642, 8144, -940, 808, -21583, 226 },
+ /* 614400 */ { 1107534, 8144, -940, 808, -21583, 226 },
+ /* 691200 */ { 1149425, 8144, -940, 808, -21583, 226 },
+ /* 768000 */ { 1191317, 8144, -940, 808, -21583, 226 },
+ /* 844800 */ { 1233208, 8144, -940, 808, -21583, 226 },
+ /* 921600 */ { 1275100, 8144, -940, 808, -21583, 226 },
+ /* 998400 */ { 1316991, 8144, -940, 808, -21583, 226 },
+};
+
const u32 speedo_to_vmin[] = {
/* 0, 1, 2, 3, 4, */
950000, 840000, 818750, 840000, 810000,
@@ -66,6 +83,10 @@ gm20b_volt_new(struct nvkm_device *device, int index, struct
nvkm_volt **pvolt)
vmin = speedo_to_vmin[tdev->gpu_speedo_id];
- return gk20a_volt_ctor(device, index, gm20b_cvb_coef,
- ARRAY_SIZE(gm20b_cvb_coef), vmin, volt);
+ if (tdev->gpu_speedo_id >= 1)
+ return gk20a_volt_ctor(device, index, gm20b_na_cvb_coef,
+ ARRAY_SIZE(gm20b_na_cvb_coef), vmin, volt);
+ else
+ return gk20a_volt_ctor(device, index, gm20b_cvb_coef,
+ ARRAY_SIZE(gm20b_cvb_coef), vmin, volt);
}
--
2.8.3