hollalinux
/
linux-sg2042
mirror of https://github.com/sophgo/linux-riscv.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

drm/nouveau/pwr: add support for GK20A 

This patch adds PWR support for GK20A. But instead of adding the PWR
features like firmware loading and communication with PMU firmware, we
add the DVFS (Dynamic Voltage and Frequency Scaling), which is one of
the PMU firmware's jobs on dGPUs, in this patch. This refers to the
idle signals provided by the NVIDIA hardware and tries to adjust the
performance level based on the calculated target. The reclocking policy
can be fine-tuned later when we have more real use cases.

Signed-off-by: Vince Hsu <vinceh@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This commit is contained in:
Vince Hsu 2014-12-22 17:11:39 +08:00 committed by Ben Skeggs
parent 47e7df39d9
commit 9f7fd620af
4 changed files with 235 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
drivers/gpu/drm/nouveau/Makefile
View File

@ -201,6 +201,7 @@ nouveau-y += core/subdev/pwr/nvc0.o
nouveau-y += core/subdev/pwr/nvd0.o
nouveau-y += core/subdev/pwr/gk104.o
nouveau-y += core/subdev/pwr/nv108.o
nouveau-y += core/subdev/pwr/gk20a.o
nouveau-y += core/subdev/therm/base.o
nouveau-y += core/subdev/therm/fan.o
nouveau-y += core/subdev/therm/fannil.o

1
drivers/gpu/drm/nouveau/core/engine/device/nve0.c
View File

@ -180,6 +180,7 @@ nve0_identify(struct nouveau_device *device)
device->oclass[NVDEV_ENGINE_COPY2 ] = &nve0_copy2_oclass;
device->oclass[NVDEV_ENGINE_PERFMON] = &nve0_perfmon_oclass;
device->oclass[NVDEV_SUBDEV_VOLT ] = &gk20a_volt_oclass;
device->oclass[NVDEV_SUBDEV_PWR ] = gk20a_pwr_oclass;
break;
case 0xf0:
device->cname = "GK110";

1
drivers/gpu/drm/nouveau/core/include/subdev/pwr.h
View File

@ -38,6 +38,7 @@ extern struct nouveau_oclass *nvc0_pwr_oclass;
extern struct nouveau_oclass *nvd0_pwr_oclass;
extern struct nouveau_oclass *gk104_pwr_oclass;
extern struct nouveau_oclass *nv108_pwr_oclass;
extern struct nouveau_oclass *gk20a_pwr_oclass;
/* interface to MEMX process running on PPWR */
struct nouveau_memx;

232
drivers/gpu/drm/nouveau/core/subdev/pwr/gk20a.c Normal file
View File

@ -0,0 +1,232 @@
/*
* Copyright (c) 2014, NVIDIA CORPORATION. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*/
#include "priv.h"
#include <subdev/clock.h>
#include <subdev/timer.h>
#include <subdev/volt.h>
#define BUSY_SLOT 0
#define CLK_SLOT 7
struct gk20a_pwr_dvfs_data {
int p_load_target;
int p_load_max;
int p_smooth;
unsigned int avg_load;
};
struct gk20a_pwr_priv {
struct nouveau_pwr base;
struct nouveau_alarm alarm;
struct gk20a_pwr_dvfs_data *data;
};
struct gk20a_pwr_dvfs_dev_status {
unsigned long total;
unsigned long busy;
int cur_state;
};
static int
gk20a_pwr_dvfs_target(struct gk20a_pwr_priv *priv, int *state)
{
struct nouveau_clock *clk = nouveau_clock(priv);
return nouveau_clock_astate(clk, *state, 0, false);
}
static int
gk20a_pwr_dvfs_get_cur_state(struct gk20a_pwr_priv *priv, int *state)
{
struct nouveau_clock *clk = nouveau_clock(priv);
*state = clk->pstate;
return 0;
}
static int
gk20a_pwr_dvfs_get_target_state(struct gk20a_pwr_priv *priv,
int *state, int load)
{
struct gk20a_pwr_dvfs_data *data = priv->data;
struct nouveau_clock *clk = nouveau_clock(priv);
int cur_level, level;
/* For GK20A, the performance level is directly mapped to pstate */
level = cur_level = clk->pstate;
if (load > data->p_load_max) {
level = min(clk->state_nr - 1, level + (clk->state_nr / 3));
} else {
level += ((load - data->p_load_target) * 10 /
data->p_load_target) / 2;
level = max(0, level);
level = min(clk->state_nr - 1, level);
}
nv_trace(priv, "cur level = %d, new level = %d\n", cur_level, level);
*state = level;
if (level == cur_level)
return 0;
else
return 1;
}
static int
gk20a_pwr_dvfs_get_dev_status(struct gk20a_pwr_priv *priv,
struct gk20a_pwr_dvfs_dev_status *status)
{
status->busy = nv_rd32(priv, 0x10a508 + (BUSY_SLOT * 0x10));
status->total= nv_rd32(priv, 0x10a508 + (CLK_SLOT * 0x10));
return 0;
}
static void
gk20a_pwr_dvfs_reset_dev_status(struct gk20a_pwr_priv *priv)
{
nv_wr32(priv, 0x10a508 + (BUSY_SLOT * 0x10), 0x80000000);
nv_wr32(priv, 0x10a508 + (CLK_SLOT * 0x10), 0x80000000);
}
static void
gk20a_pwr_dvfs_work(struct nouveau_alarm *alarm)
{
struct gk20a_pwr_priv *priv = container_of(alarm,
struct gk20a_pwr_priv, alarm);
struct gk20a_pwr_dvfs_data *data = priv->data;
struct gk20a_pwr_dvfs_dev_status status;
struct nouveau_clock *clk = nouveau_clock(priv);
struct nouveau_volt *volt = nouveau_volt(priv);
u32 utilization = 0;
int state, ret;
/*
* The PWR is initialized before CLK and VOLT, so we have to make sure the
* CLK and VOLT are ready here.
*/
if (!clk || !volt)
goto resched;
ret = gk20a_pwr_dvfs_get_dev_status(priv, &status);
if (ret) {
nv_warn(priv, "failed to get device status\n");
goto resched;
}
if (status.total)
utilization = div_u64((u64)status.busy * 100, status.total);
data->avg_load = (data->p_smooth * data->avg_load) + utilization;
data->avg_load /= data->p_smooth + 1;
nv_trace(priv, "utilization = %d %%, avg_load = %d %%\n",
utilization, data->avg_load);
ret = gk20a_pwr_dvfs_get_cur_state(priv, &state);
if (ret) {
nv_warn(priv, "failed to get current state\n");
goto resched;
}
if (gk20a_pwr_dvfs_get_target_state(priv, &state, data->avg_load)) {
nv_trace(priv, "set new state to %d\n", state);
gk20a_pwr_dvfs_target(priv, &state);
}
resched:
gk20a_pwr_dvfs_reset_dev_status(priv);
nouveau_timer_alarm(priv, 100000000, alarm);
}
int
gk20a_pwr_fini(struct nouveau_object *object, bool suspend)
{
struct nouveau_pwr *ppwr = (void *)object;
struct gk20a_pwr_priv *priv = (void *)ppwr;
nouveau_timer_alarm_cancel(priv, &priv->alarm);
return nouveau_subdev_fini(&ppwr->base, suspend);
}
int
gk20a_pwr_init(struct nouveau_object *object)
{
struct nouveau_pwr *ppwr = (void *)object;
struct gk20a_pwr_priv *priv = (void *)ppwr;
int ret;
ret = nouveau_subdev_init(&ppwr->base);
if (ret)
return ret;
ppwr->pgob = nouveau_pwr_pgob;
/* init pwr perf counter */
nv_wr32(ppwr, 0x10a504 + (BUSY_SLOT * 0x10), 0x00200001);
nv_wr32(ppwr, 0x10a50c + (BUSY_SLOT * 0x10), 0x00000002);
nv_wr32(ppwr, 0x10a50c + (CLK_SLOT * 0x10), 0x00000003);
nouveau_timer_alarm(ppwr, 2000000000, &priv->alarm);
return ret;
}
struct gk20a_pwr_dvfs_data gk20a_dvfs_data= {
.p_load_target = 70,
.p_load_max = 90,
.p_smooth = 1,
};
static int
gk20a_pwr_ctor(struct nouveau_object *parent,
struct nouveau_object *engine,
struct nouveau_oclass *oclass, void *data, u32 size,
struct nouveau_object **pobject)
{
struct gk20a_pwr_priv *priv;
int ret;
ret = nouveau_pwr_create(parent, engine, oclass, &priv);
*pobject = nv_object(priv);
if (ret)
return ret;
priv->data = &gk20a_dvfs_data;
nouveau_alarm_init(&priv->alarm, gk20a_pwr_dvfs_work);
return 0;
}
struct nouveau_oclass *
gk20a_pwr_oclass = &(struct nvkm_pwr_impl) {
.base.handle = NV_SUBDEV(PWR, 0xea),
.base.ofuncs = &(struct nouveau_ofuncs) {
.ctor = gk20a_pwr_ctor,
.dtor = _nouveau_pwr_dtor,
.init = gk20a_pwr_init,
.fini = gk20a_pwr_fini,
},
}.base;