gpu: nvgpu: Move TSG IOCTL code to Linux module

Move TSG IOCTL specific code to Linux module. This clears most
Linux dependencies from tsg_gk20a.c.

Move also remaining file_operations declarations from channel_gk20a.h
to ioctl_channel.h.

JIRA NVGPU-32

Change-Id: Idcc2a525ebe12b30db46c3893a2735509c41ff39
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/1330805
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This commit is contained in:
Terje Bergstrom
2017-03-23 14:19:01 -07:00
committed by mobile promotions
parent a07e10f494
commit 06fe28567d
10 changed files with 523 additions and 464 deletions

View File

@@ -29,6 +29,7 @@ nvgpu-y := \
common/linux/ioctl_ctrl.o \
common/linux/ioctl_as.o \
common/linux/ioctl_channel.o \
common/linux/ioctl_tsg.o \
common/mm/nvgpu_allocator.o \
common/mm/bitmap_allocator.o \
common/mm/buddy_allocator.o \

View File

@@ -23,10 +23,10 @@
#include "gk20a/gk20a.h"
#include "gk20a/dbg_gpu_gk20a.h"
#include "gk20a/ctxsw_trace_gk20a.h"
#include "gk20a/tsg_gk20a.h"
#include "ioctl_channel.h"
#include "ioctl_ctrl.h"
#include "ioctl_as.h"
#include "ioctl_tsg.h"
#define GK20A_NUM_CDEVS 7
@@ -89,12 +89,12 @@ static const struct file_operations gk20a_prof_ops = {
static const struct file_operations gk20a_tsg_ops = {
.owner = THIS_MODULE,
.release = gk20a_tsg_dev_release,
.open = gk20a_tsg_dev_open,
.release = nvgpu_ioctl_tsg_dev_release,
.open = nvgpu_ioctl_tsg_dev_open,
#ifdef CONFIG_COMPAT
.compat_ioctl = gk20a_tsg_dev_ioctl,
.compat_ioctl = nvgpu_ioctl_tsg_dev_ioctl,
#endif
.unlocked_ioctl = gk20a_tsg_dev_ioctl,
.unlocked_ioctl = nvgpu_ioctl_tsg_dev_ioctl,
};
static const struct file_operations gk20a_ctxsw_ops = {

View File

@@ -31,6 +31,7 @@
#include "gk20a/ctxsw_trace_gk20a.h"
#include "gk20a/dbg_gpu_gk20a.h"
#include "gk20a/fence_gk20a.h"
#include "ioctl_channel.h"
/*
* Although channels do have pointers back to the gk20a struct that they were

View File

@@ -18,4 +18,7 @@ int gk20a_channel_release(struct inode *inode, struct file *filp);
long gk20a_channel_ioctl(struct file *filp,
unsigned int cmd, unsigned long arg);
extern const struct file_operations gk20a_event_id_ops;
extern const struct file_operations gk20a_channel_ops;
#endif

View File

@@ -24,6 +24,7 @@
#include <nvgpu/kmem.h>
#include "ioctl_ctrl.h"
#include "ioctl_tsg.h"
#include "gk20a/gk20a.h"
#include "gk20a/fence_gk20a.h"
@@ -250,7 +251,7 @@ static int gk20a_ctrl_open_tsg(struct gk20a *g,
goto clean_up;
}
err = gk20a_tsg_open(g, file);
err = nvgpu_ioctl_tsg_open(g, file);
if (err)
goto clean_up_file;

View File

@@ -0,0 +1,470 @@
/*
* Copyright (c) 2014-2017, NVIDIA CORPORATION. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
* version 2, as published by the Free Software Foundation.
*
* This program is distributed in the hope it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/cdev.h>
#include <linux/uaccess.h>
#include <linux/nvhost.h>
#include <uapi/linux/nvgpu.h>
#include <linux/anon_inodes.h>
#include <nvgpu/kmem.h>
#include "gk20a/gk20a.h"
#include "gk20a/tsg_gk20a.h"
#include "ioctl_channel.h"
struct tsg_private {
struct gk20a *g;
struct tsg_gk20a *tsg;
};
static int gk20a_tsg_bind_channel_fd(struct tsg_gk20a *tsg, int ch_fd)
{
struct channel_gk20a *ch;
int err;
ch = gk20a_get_channel_from_file(ch_fd);
if (!ch)
return -EINVAL;
err = ch->g->ops.fifo.tsg_bind_channel(tsg, ch);
return err;
}
static int gk20a_tsg_get_event_data_from_id(struct tsg_gk20a *tsg,
unsigned int event_id,
struct gk20a_event_id_data **event_id_data)
{
struct gk20a_event_id_data *local_event_id_data;
bool event_found = false;
nvgpu_mutex_acquire(&tsg->event_id_list_lock);
list_for_each_entry(local_event_id_data, &tsg->event_id_list,
event_id_node) {
if (local_event_id_data->event_id == event_id) {
event_found = true;
break;
}
}
nvgpu_mutex_release(&tsg->event_id_list_lock);
if (event_found) {
*event_id_data = local_event_id_data;
return 0;
} else {
return -1;
}
}
void gk20a_tsg_event_id_post_event(struct tsg_gk20a *tsg,
int event_id)
{
struct gk20a_event_id_data *event_id_data;
int err = 0;
err = gk20a_tsg_get_event_data_from_id(tsg, event_id,
&event_id_data);
if (err)
return;
nvgpu_mutex_acquire(&event_id_data->lock);
gk20a_dbg_info(
"posting event for event_id=%d on tsg=%d\n",
event_id, tsg->tsgid);
event_id_data->event_posted = true;
wake_up_interruptible_all(&event_id_data->event_id_wq);
nvgpu_mutex_release(&event_id_data->lock);
}
static int gk20a_tsg_event_id_enable(struct tsg_gk20a *tsg,
int event_id,
int *fd)
{
int err = 0;
int local_fd;
struct file *file;
char name[64];
struct gk20a_event_id_data *event_id_data;
struct gk20a *g;
g = gk20a_get(tsg->g);
if (!g)
return -ENODEV;
err = gk20a_tsg_get_event_data_from_id(tsg,
event_id, &event_id_data);
if (err == 0) {
/* We already have event enabled */
err = -EINVAL;
goto free_ref;
}
err = get_unused_fd_flags(O_RDWR);
if (err < 0)
goto free_ref;
local_fd = err;
snprintf(name, sizeof(name), "nvgpu-event%d-fd%d",
event_id, local_fd);
file = anon_inode_getfile(name, &gk20a_event_id_ops,
NULL, O_RDWR);
if (IS_ERR(file)) {
err = PTR_ERR(file);
goto clean_up;
}
event_id_data = nvgpu_kzalloc(tsg->g, sizeof(*event_id_data));
if (!event_id_data) {
err = -ENOMEM;
goto clean_up_file;
}
event_id_data->g = g;
event_id_data->id = tsg->tsgid;
event_id_data->is_tsg = true;
event_id_data->event_id = event_id;
init_waitqueue_head(&event_id_data->event_id_wq);
err = nvgpu_mutex_init(&event_id_data->lock);
if (err)
goto clean_up_free;
INIT_LIST_HEAD(&event_id_data->event_id_node);
nvgpu_mutex_acquire(&tsg->event_id_list_lock);
list_add_tail(&event_id_data->event_id_node, &tsg->event_id_list);
nvgpu_mutex_release(&tsg->event_id_list_lock);
fd_install(local_fd, file);
file->private_data = event_id_data;
*fd = local_fd;
return 0;
clean_up_free:
kfree(event_id_data);
clean_up_file:
fput(file);
clean_up:
put_unused_fd(local_fd);
free_ref:
gk20a_put(g);
return err;
}
static int gk20a_tsg_event_id_ctrl(struct gk20a *g, struct tsg_gk20a *tsg,
struct nvgpu_event_id_ctrl_args *args)
{
int err = 0;
int fd = -1;
if (args->event_id >= NVGPU_IOCTL_CHANNEL_EVENT_ID_MAX)
return -EINVAL;
switch (args->cmd) {
case NVGPU_IOCTL_CHANNEL_EVENT_ID_CMD_ENABLE:
err = gk20a_tsg_event_id_enable(tsg, args->event_id, &fd);
if (!err)
args->event_fd = fd;
break;
default:
gk20a_err(dev_from_gk20a(tsg->g),
"unrecognized tsg event id cmd: 0x%x",
args->cmd);
err = -EINVAL;
break;
}
return err;
}
int nvgpu_ioctl_tsg_open(struct gk20a *g, struct file *filp)
{
struct tsg_private *priv;
struct tsg_gk20a *tsg;
struct device *dev;
int err;
g = gk20a_get(g);
if (!g)
return -ENODEV;
dev = dev_from_gk20a(g);
gk20a_dbg(gpu_dbg_fn, "tsg: %s", dev_name(dev));
priv = nvgpu_kmalloc(g, sizeof(*priv));
if (!priv) {
err = -ENOMEM;
goto free_ref;
}
tsg = gk20a_tsg_open(g);
if (!tsg) {
nvgpu_kfree(g, priv);
err = -ENOMEM;
goto free_ref;
}
priv->g = g;
priv->tsg = tsg;
filp->private_data = priv;
return 0;
free_ref:
gk20a_put(g);
return err;
}
int nvgpu_ioctl_tsg_dev_open(struct inode *inode, struct file *filp)
{
struct gk20a *g;
int ret;
g = container_of(inode->i_cdev,
struct gk20a, tsg.cdev);
gk20a_dbg_fn("");
ret = nvgpu_ioctl_tsg_open(g, filp);
gk20a_dbg_fn("done");
return ret;
}
int nvgpu_ioctl_tsg_dev_release(struct inode *inode, struct file *filp)
{
struct tsg_private *priv = filp->private_data;
struct tsg_gk20a *tsg = priv->tsg;
kref_put(&tsg->refcount, gk20a_tsg_release);
nvgpu_kfree(tsg->g, priv);
return 0;
}
static int gk20a_tsg_ioctl_set_priority(struct gk20a *g,
struct tsg_gk20a *tsg, struct nvgpu_set_priority_args *arg)
{
struct gk20a_sched_ctrl *sched = &g->sched_ctrl;
int err;
nvgpu_mutex_acquire(&sched->control_lock);
if (sched->control_locked) {
err = -EPERM;
goto done;
}
err = gk20a_busy(g);
if (err) {
gk20a_err(dev_from_gk20a(g), "failed to power on gpu");
goto done;
}
err = gk20a_tsg_set_priority(g, tsg, arg->priority);
gk20a_idle(g);
done:
nvgpu_mutex_release(&sched->control_lock);
return err;
}
static int gk20a_tsg_ioctl_set_runlist_interleave(struct gk20a *g,
struct tsg_gk20a *tsg, struct nvgpu_runlist_interleave_args *arg)
{
struct gk20a_sched_ctrl *sched = &g->sched_ctrl;
int err;
gk20a_dbg(gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsg->tsgid);
nvgpu_mutex_acquire(&sched->control_lock);
if (sched->control_locked) {
err = -EPERM;
goto done;
}
err = gk20a_busy(g);
if (err) {
gk20a_err(dev_from_gk20a(g), "failed to power on gpu");
goto done;
}
err = gk20a_tsg_set_runlist_interleave(tsg, arg->level);
gk20a_idle(g);
done:
nvgpu_mutex_release(&sched->control_lock);
return err;
}
static int gk20a_tsg_ioctl_set_timeslice(struct gk20a *g,
struct tsg_gk20a *tsg, struct nvgpu_timeslice_args *arg)
{
struct gk20a_sched_ctrl *sched = &g->sched_ctrl;
int err;
gk20a_dbg(gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsg->tsgid);
nvgpu_mutex_acquire(&sched->control_lock);
if (sched->control_locked) {
err = -EPERM;
goto done;
}
err = gk20a_busy(g);
if (err) {
gk20a_err(dev_from_gk20a(g), "failed to power on gpu");
goto done;
}
err = gk20a_tsg_set_timeslice(tsg, arg->timeslice_us);
gk20a_idle(g);
done:
nvgpu_mutex_release(&sched->control_lock);
return err;
}
long nvgpu_ioctl_tsg_dev_ioctl(struct file *filp, unsigned int cmd,
unsigned long arg)
{
struct tsg_private *priv = filp->private_data;
struct tsg_gk20a *tsg = priv->tsg;
struct gk20a *g = tsg->g;
u8 __maybe_unused buf[NVGPU_TSG_IOCTL_MAX_ARG_SIZE];
int err = 0;
gk20a_dbg(gpu_dbg_fn, "");
if ((_IOC_TYPE(cmd) != NVGPU_TSG_IOCTL_MAGIC) ||
(_IOC_NR(cmd) == 0) ||
(_IOC_NR(cmd) > NVGPU_TSG_IOCTL_LAST) ||
(_IOC_SIZE(cmd) > NVGPU_TSG_IOCTL_MAX_ARG_SIZE))
return -EINVAL;
memset(buf, 0, sizeof(buf));
if (_IOC_DIR(cmd) & _IOC_WRITE) {
if (copy_from_user(buf, (void __user *)arg, _IOC_SIZE(cmd)))
return -EFAULT;
}
if (!g->gr.sw_ready) {
err = gk20a_busy(g);
if (err)
return err;
gk20a_idle(g);
}
switch (cmd) {
case NVGPU_TSG_IOCTL_BIND_CHANNEL:
{
int ch_fd = *(int *)buf;
if (ch_fd < 0) {
err = -EINVAL;
break;
}
err = gk20a_tsg_bind_channel_fd(tsg, ch_fd);
break;
}
case NVGPU_TSG_IOCTL_UNBIND_CHANNEL:
/* We do not support explicitly unbinding channel from TSG.
* Channel will be unbounded from TSG when it is closed.
*/
break;
case NVGPU_IOCTL_TSG_ENABLE:
{
err = gk20a_busy(g);
if (err) {
gk20a_err(g->dev,
"failed to host gk20a for ioctl cmd: 0x%x", cmd);
return err;
}
gk20a_enable_tsg(tsg);
gk20a_idle(g);
break;
}
case NVGPU_IOCTL_TSG_DISABLE:
{
err = gk20a_busy(g);
if (err) {
gk20a_err(g->dev,
"failed to host gk20a for ioctl cmd: 0x%x", cmd);
return err;
}
gk20a_disable_tsg(tsg);
gk20a_idle(g);
break;
}
case NVGPU_IOCTL_TSG_PREEMPT:
{
err = gk20a_busy(g);
if (err) {
gk20a_err(g->dev,
"failed to host gk20a for ioctl cmd: 0x%x", cmd);
return err;
}
/* preempt TSG */
err = g->ops.fifo.preempt_tsg(g, tsg->tsgid);
gk20a_idle(g);
break;
}
case NVGPU_IOCTL_TSG_SET_PRIORITY:
{
err = gk20a_tsg_ioctl_set_priority(g, tsg,
(struct nvgpu_set_priority_args *)buf);
break;
}
case NVGPU_IOCTL_TSG_EVENT_ID_CTRL:
{
err = gk20a_tsg_event_id_ctrl(g, tsg,
(struct nvgpu_event_id_ctrl_args *)buf);
break;
}
case NVGPU_IOCTL_TSG_SET_RUNLIST_INTERLEAVE:
err = gk20a_tsg_ioctl_set_runlist_interleave(g, tsg,
(struct nvgpu_runlist_interleave_args *)buf);
break;
case NVGPU_IOCTL_TSG_SET_TIMESLICE:
{
err = gk20a_tsg_ioctl_set_timeslice(g, tsg,
(struct nvgpu_timeslice_args *)buf);
break;
}
default:
gk20a_err(dev_from_gk20a(g),
"unrecognized tsg gpu ioctl cmd: 0x%x",
cmd);
err = -ENOTTY;
break;
}
if ((err == 0) && (_IOC_DIR(cmd) & _IOC_READ))
err = copy_to_user((void __user *)arg,
buf, _IOC_SIZE(cmd));
return err;
}

View File

@@ -0,0 +1,26 @@
/*
* Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
* version 2, as published by the Free Software Foundation.
*
* This program is distributed in the hope it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*/
#ifndef NVGPU_IOCTL_TSG_H
#define NVGPU_IOCTL_TSG_H
struct inode;
struct file;
struct gk20a;
int nvgpu_ioctl_tsg_dev_release(struct inode *inode, struct file *filp);
int nvgpu_ioctl_tsg_dev_open(struct inode *inode, struct file *filp);
int nvgpu_ioctl_tsg_open(struct gk20a *g, struct file *filp);
long nvgpu_ioctl_tsg_dev_ioctl(struct file *filp,
unsigned int cmd, unsigned long arg);
#endif

View File

@@ -37,9 +37,6 @@ struct fifo_profile_gk20a;
#include "gr_gk20a.h"
#include "fence_gk20a.h"
extern const struct file_operations gk20a_event_id_ops;
extern const struct file_operations gk20a_channel_ops;
struct notification {
struct {
u32 nanoseconds[2];

View File

@@ -14,24 +14,10 @@
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/cdev.h>
#include <linux/uaccess.h>
#include <linux/nvhost.h>
#include <uapi/linux/nvgpu.h>
#include <linux/anon_inodes.h>
#include <nvgpu/kmem.h>
#include "gk20a.h"
#include <nvgpu/hw/gk20a/hw_ccsr_gk20a.h>
struct tsg_private {
struct gk20a *g;
struct tsg_gk20a *tsg;
};
#include "tsg_gk20a.h"
bool gk20a_is_channel_marked_as_tsg(struct channel_gk20a *ch)
{
@@ -81,19 +67,6 @@ static bool gk20a_is_channel_active(struct gk20a *g, struct channel_gk20a *ch)
return false;
}
static int gk20a_tsg_bind_channel_fd(struct tsg_gk20a *tsg, int ch_fd)
{
struct channel_gk20a *ch;
int err;
ch = gk20a_get_channel_from_file(ch_fd);
if (!ch)
return -EINVAL;
err = ch->g->ops.fifo.tsg_bind_channel(tsg, ch);
return err;
}
/*
* API to mark channel as part of TSG
*
@@ -181,7 +154,7 @@ int gk20a_init_tsg_support(struct gk20a *g, u32 tsgid)
return 0;
}
static int gk20a_tsg_set_priority(struct gk20a *g, struct tsg_gk20a *tsg,
int gk20a_tsg_set_priority(struct gk20a *g, struct tsg_gk20a *tsg,
u32 priority)
{
u32 timeslice_us;
@@ -204,158 +177,6 @@ static int gk20a_tsg_set_priority(struct gk20a *g, struct tsg_gk20a *tsg,
return gk20a_tsg_set_timeslice(tsg, timeslice_us);
}
static int gk20a_tsg_get_event_data_from_id(struct tsg_gk20a *tsg,
unsigned int event_id,
struct gk20a_event_id_data **event_id_data)
{
struct gk20a_event_id_data *local_event_id_data;
bool event_found = false;
nvgpu_mutex_acquire(&tsg->event_id_list_lock);
list_for_each_entry(local_event_id_data, &tsg->event_id_list,
event_id_node) {
if (local_event_id_data->event_id == event_id) {
event_found = true;
break;
}
}
nvgpu_mutex_release(&tsg->event_id_list_lock);
if (event_found) {
*event_id_data = local_event_id_data;
return 0;
} else {
return -1;
}
}
void gk20a_tsg_event_id_post_event(struct tsg_gk20a *tsg,
int event_id)
{
struct gk20a_event_id_data *event_id_data;
int err = 0;
err = gk20a_tsg_get_event_data_from_id(tsg, event_id,
&event_id_data);
if (err)
return;
nvgpu_mutex_acquire(&event_id_data->lock);
gk20a_dbg_info(
"posting event for event_id=%d on tsg=%d\n",
event_id, tsg->tsgid);
event_id_data->event_posted = true;
wake_up_interruptible_all(&event_id_data->event_id_wq);
nvgpu_mutex_release(&event_id_data->lock);
}
static int gk20a_tsg_event_id_enable(struct tsg_gk20a *tsg,
int event_id,
int *fd)
{
int err = 0;
int local_fd;
struct file *file;
char name[64];
struct gk20a_event_id_data *event_id_data;
struct gk20a *g;
g = gk20a_get(tsg->g);
if (!g)
return -ENODEV;
err = gk20a_tsg_get_event_data_from_id(tsg,
event_id, &event_id_data);
if (err == 0) {
/* We already have event enabled */
err = -EINVAL;
goto free_ref;
}
err = get_unused_fd_flags(O_RDWR);
if (err < 0)
goto free_ref;
local_fd = err;
snprintf(name, sizeof(name), "nvgpu-event%d-fd%d",
event_id, local_fd);
file = anon_inode_getfile(name, &gk20a_event_id_ops,
NULL, O_RDWR);
if (IS_ERR(file)) {
err = PTR_ERR(file);
goto clean_up;
}
event_id_data = nvgpu_kzalloc(tsg->g, sizeof(*event_id_data));
if (!event_id_data) {
err = -ENOMEM;
goto clean_up_file;
}
event_id_data->g = g;
event_id_data->id = tsg->tsgid;
event_id_data->is_tsg = true;
event_id_data->event_id = event_id;
init_waitqueue_head(&event_id_data->event_id_wq);
err = nvgpu_mutex_init(&event_id_data->lock);
if (err)
goto clean_up_free;
INIT_LIST_HEAD(&event_id_data->event_id_node);
nvgpu_mutex_acquire(&tsg->event_id_list_lock);
list_add_tail(&event_id_data->event_id_node, &tsg->event_id_list);
nvgpu_mutex_release(&tsg->event_id_list_lock);
fd_install(local_fd, file);
file->private_data = event_id_data;
*fd = local_fd;
return 0;
clean_up_free:
kfree(event_id_data);
clean_up_file:
fput(file);
clean_up:
put_unused_fd(local_fd);
free_ref:
gk20a_put(g);
return err;
}
static int gk20a_tsg_event_id_ctrl(struct gk20a *g, struct tsg_gk20a *tsg,
struct nvgpu_event_id_ctrl_args *args)
{
int err = 0;
int fd = -1;
if (args->event_id >= NVGPU_IOCTL_CHANNEL_EVENT_ID_MAX)
return -EINVAL;
switch (args->cmd) {
case NVGPU_IOCTL_CHANNEL_EVENT_ID_CMD_ENABLE:
err = gk20a_tsg_event_id_enable(tsg, args->event_id, &fd);
if (!err)
args->event_fd = fd;
break;
default:
gk20a_err(dev_from_gk20a(tsg->g),
"unrecognized tsg event id cmd: 0x%x",
args->cmd);
err = -EINVAL;
break;
}
return err;
}
int gk20a_tsg_set_runlist_interleave(struct tsg_gk20a *tsg, u32 level)
{
struct gk20a *g = tsg->g;
@@ -396,7 +217,7 @@ static void release_used_tsg(struct fifo_gk20a *f, struct tsg_gk20a *tsg)
nvgpu_mutex_release(&f->tsg_inuse_mutex);
}
static struct tsg_gk20a *acquire_unused_tsg(struct fifo_gk20a *f)
static struct tsg_gk20a *gk20a_tsg_acquire_unused_tsg(struct fifo_gk20a *f)
{
struct tsg_gk20a *tsg = NULL;
unsigned int tsgid;
@@ -414,33 +235,14 @@ static struct tsg_gk20a *acquire_unused_tsg(struct fifo_gk20a *f)
return tsg;
}
int gk20a_tsg_open(struct gk20a *g, struct file *filp)
struct tsg_gk20a *gk20a_tsg_open(struct gk20a *g)
{
struct tsg_private *priv;
struct tsg_gk20a *tsg;
struct device *dev;
int err;
g = gk20a_get(g);
if (!g)
return -ENODEV;
dev = dev_from_gk20a(g);
gk20a_dbg(gpu_dbg_fn, "tsg: %s", g->name);
priv = nvgpu_kmalloc(g, sizeof(*priv));
if (!priv) {
err = -ENOMEM;
goto free_ref;
}
tsg = acquire_unused_tsg(&g->fifo);
if (!tsg) {
nvgpu_kfree(g, priv);
err = -ENOMEM;
goto free_ref;
}
tsg = gk20a_tsg_acquire_unused_tsg(&g->fifo);
if (!tsg)
return NULL;
tsg->g = g;
tsg->num_active_channels = 0;
@@ -455,14 +257,11 @@ int gk20a_tsg_open(struct gk20a *g, struct file *filp)
tsg->runlist_id = ~0;
tsg->tgid = current->tgid;
priv->g = g;
priv->tsg = tsg;
filp->private_data = priv;
if (g->ops.fifo.tsg_open) {
err = g->ops.fifo.tsg_open(tsg);
if (err) {
gk20a_err(dev, "tsg %d fifo open failed %d",
gk20a_err(dev_from_gk20a(g),
"tsg %d fifo open failed %d",
tsg->tsgid, err);
goto clean_up;
}
@@ -472,26 +271,11 @@ int gk20a_tsg_open(struct gk20a *g, struct file *filp)
gk20a_sched_ctrl_tsg_added(g, tsg);
return 0;
return tsg;
clean_up:
kref_put(&tsg->refcount, gk20a_tsg_release);
free_ref:
gk20a_put(g);
return err;
}
int gk20a_tsg_dev_open(struct inode *inode, struct file *filp)
{
struct gk20a *g;
int ret;
g = container_of(inode->i_cdev,
struct gk20a, tsg.cdev);
gk20a_dbg_fn("");
ret = gk20a_tsg_open(g, filp);
gk20a_dbg_fn("done");
return ret;
return NULL;
}
void gk20a_tsg_release(struct kref *ref)
@@ -528,225 +312,6 @@ void gk20a_tsg_release(struct kref *ref)
gk20a_put(g);
}
int gk20a_tsg_dev_release(struct inode *inode, struct file *filp)
{
struct tsg_private *priv = filp->private_data;
struct tsg_gk20a *tsg = priv->tsg;
kref_put(&tsg->refcount, gk20a_tsg_release);
nvgpu_kfree(tsg->g, priv);
return 0;
}
static int gk20a_tsg_ioctl_set_priority(struct gk20a *g,
struct tsg_gk20a *tsg, struct nvgpu_set_priority_args *arg)
{
struct gk20a_sched_ctrl *sched = &g->sched_ctrl;
int err;
nvgpu_mutex_acquire(&sched->control_lock);
if (sched->control_locked) {
err = -EPERM;
goto done;
}
err = gk20a_busy(g);
if (err) {
gk20a_err(dev_from_gk20a(g), "failed to power on gpu");
goto done;
}
err = gk20a_tsg_set_priority(g, tsg, arg->priority);
gk20a_idle(g);
done:
nvgpu_mutex_release(&sched->control_lock);
return err;
}
static int gk20a_tsg_ioctl_set_runlist_interleave(struct gk20a *g,
struct tsg_gk20a *tsg, struct nvgpu_runlist_interleave_args *arg)
{
struct gk20a_sched_ctrl *sched = &g->sched_ctrl;
int err;
gk20a_dbg(gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsg->tsgid);
nvgpu_mutex_acquire(&sched->control_lock);
if (sched->control_locked) {
err = -EPERM;
goto done;
}
err = gk20a_busy(g);
if (err) {
gk20a_err(dev_from_gk20a(g), "failed to power on gpu");
goto done;
}
err = gk20a_tsg_set_runlist_interleave(tsg, arg->level);
gk20a_idle(g);
done:
nvgpu_mutex_release(&sched->control_lock);
return err;
}
static int gk20a_tsg_ioctl_set_timeslice(struct gk20a *g,
struct tsg_gk20a *tsg, struct nvgpu_timeslice_args *arg)
{
struct gk20a_sched_ctrl *sched = &g->sched_ctrl;
int err;
gk20a_dbg(gpu_dbg_fn | gpu_dbg_sched, "tsgid=%u", tsg->tsgid);
nvgpu_mutex_acquire(&sched->control_lock);
if (sched->control_locked) {
err = -EPERM;
goto done;
}
err = gk20a_busy(g);
if (err) {
gk20a_err(dev_from_gk20a(g), "failed to power on gpu");
goto done;
}
err = gk20a_tsg_set_timeslice(tsg, arg->timeslice_us);
gk20a_idle(g);
done:
nvgpu_mutex_release(&sched->control_lock);
return err;
}
long gk20a_tsg_dev_ioctl(struct file *filp, unsigned int cmd,
unsigned long arg)
{
struct tsg_private *priv = filp->private_data;
struct tsg_gk20a *tsg = priv->tsg;
struct gk20a *g = tsg->g;
u8 __maybe_unused buf[NVGPU_TSG_IOCTL_MAX_ARG_SIZE];
int err = 0;
gk20a_dbg(gpu_dbg_fn, "");
if ((_IOC_TYPE(cmd) != NVGPU_TSG_IOCTL_MAGIC) ||
(_IOC_NR(cmd) == 0) ||
(_IOC_NR(cmd) > NVGPU_TSG_IOCTL_LAST) ||
(_IOC_SIZE(cmd) > NVGPU_TSG_IOCTL_MAX_ARG_SIZE))
return -EINVAL;
memset(buf, 0, sizeof(buf));
if (_IOC_DIR(cmd) & _IOC_WRITE) {
if (copy_from_user(buf, (void __user *)arg, _IOC_SIZE(cmd)))
return -EFAULT;
}
if (!g->gr.sw_ready) {
err = gk20a_busy(g);
if (err)
return err;
gk20a_idle(g);
}
switch (cmd) {
case NVGPU_TSG_IOCTL_BIND_CHANNEL:
{
int ch_fd = *(int *)buf;
if (ch_fd < 0) {
err = -EINVAL;
break;
}
err = gk20a_tsg_bind_channel_fd(tsg, ch_fd);
break;
}
case NVGPU_TSG_IOCTL_UNBIND_CHANNEL:
/* We do not support explicitly unbinding channel from TSG.
* Channel will be unbounded from TSG when it is closed.
*/
break;
case NVGPU_IOCTL_TSG_ENABLE:
{
err = gk20a_busy(g);
if (err) {
gk20a_err(g->dev,
"failed to host gk20a for ioctl cmd: 0x%x", cmd);
return err;
}
gk20a_enable_tsg(tsg);
gk20a_idle(g);
break;
}
case NVGPU_IOCTL_TSG_DISABLE:
{
err = gk20a_busy(g);
if (err) {
gk20a_err(g->dev,
"failed to host gk20a for ioctl cmd: 0x%x", cmd);
return err;
}
gk20a_disable_tsg(tsg);
gk20a_idle(g);
break;
}
case NVGPU_IOCTL_TSG_PREEMPT:
{
err = gk20a_busy(g);
if (err) {
gk20a_err(g->dev,
"failed to host gk20a for ioctl cmd: 0x%x", cmd);
return err;
}
/* preempt TSG */
err = g->ops.fifo.preempt_tsg(g, tsg->tsgid);
gk20a_idle(g);
break;
}
case NVGPU_IOCTL_TSG_SET_PRIORITY:
{
err = gk20a_tsg_ioctl_set_priority(g, tsg,
(struct nvgpu_set_priority_args *)buf);
break;
}
case NVGPU_IOCTL_TSG_EVENT_ID_CTRL:
{
err = gk20a_tsg_event_id_ctrl(g, tsg,
(struct nvgpu_event_id_ctrl_args *)buf);
break;
}
case NVGPU_IOCTL_TSG_SET_RUNLIST_INTERLEAVE:
err = gk20a_tsg_ioctl_set_runlist_interleave(g, tsg,
(struct nvgpu_runlist_interleave_args *)buf);
break;
case NVGPU_IOCTL_TSG_SET_TIMESLICE:
{
err = gk20a_tsg_ioctl_set_timeslice(g, tsg,
(struct nvgpu_timeslice_args *)buf);
break;
}
default:
gk20a_err(dev_from_gk20a(g),
"unrecognized tsg gpu ioctl cmd: 0x%x",
cmd);
err = -ENOTTY;
break;
}
if ((err == 0) && (_IOC_DIR(cmd) & _IOC_READ))
err = copy_to_user((void __user *)arg,
buf, _IOC_SIZE(cmd));
return err;
}
void gk20a_init_tsg_ops(struct gpu_ops *gops)
{
gops->fifo.tsg_bind_channel = gk20a_tsg_bind_channel;

View File

@@ -22,17 +22,10 @@
struct channel_gk20a;
struct gpu_ops;
struct file;
bool gk20a_is_channel_marked_as_tsg(struct channel_gk20a *ch);
int gk20a_tsg_dev_release(struct inode *inode, struct file *filp);
int gk20a_tsg_dev_open(struct inode *inode, struct file *filp);
struct tsg_gk20a *gk20a_tsg_open(struct gk20a *g);
void gk20a_tsg_release(struct kref *ref);
int gk20a_tsg_open(struct gk20a *g, struct file *filp);
long gk20a_tsg_dev_ioctl(struct file *filp,
unsigned int cmd, unsigned long arg);
int gk20a_init_tsg_support(struct gk20a *g, u32 tsgid);
void gk20a_init_tsg_ops(struct gpu_ops *gops);
@@ -76,6 +69,8 @@ void gk20a_tsg_event_id_post_event(struct tsg_gk20a *tsg,
int event_id);
int gk20a_tsg_set_runlist_interleave(struct tsg_gk20a *tsg, u32 level);
int gk20a_tsg_set_timeslice(struct tsg_gk20a *tsg, u32 timeslice);
int gk20a_tsg_set_priority(struct gk20a *g, struct tsg_gk20a *tsg,
u32 priority);
#endif /* __TSG_GK20A_H_ */