linux/drivers/accel/ivpu/ivpu_ms.c
Tomasz Rusinowicz cdfad4db77 accel/ivpu: Add NPU profiling support
Implement time based Metric Streamer profiling UAPI.

This is a generic mechanism allowing user mode tools to sample
NPU metrics. These metrics are defined by the FW and transparent to
the driver.

The user space can check for this feature by checking
DRM_IVPU_CAP_METRIC_STREAMER driver capability.

Signed-off-by: Tomasz Rusinowicz <tomasz.rusinowicz@intel.com>
Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240513120431.3187212-9-jacek.lawrynowicz@linux.intel.com
2024-05-15 07:42:23 +02:00

310 lines
7.7 KiB
C

// SPDX-License-Identifier: GPL-2.0-only OR MIT
/*
* Copyright (C) 2020-2024 Intel Corporation
*/
#include <drm/drm_file.h>
#include "ivpu_drv.h"
#include "ivpu_gem.h"
#include "ivpu_jsm_msg.h"
#include "ivpu_ms.h"
#include "ivpu_pm.h"
#define MS_INFO_BUFFER_SIZE SZ_16K
#define MS_NUM_BUFFERS 2
#define MS_READ_PERIOD_MULTIPLIER 2
#define MS_MIN_SAMPLE_PERIOD_NS 1000000
static struct ivpu_ms_instance *
get_instance_by_mask(struct ivpu_file_priv *file_priv, u64 metric_mask)
{
struct ivpu_ms_instance *ms;
lockdep_assert_held(&file_priv->ms_lock);
list_for_each_entry(ms, &file_priv->ms_instance_list, ms_instance_node)
if (ms->mask == metric_mask)
return ms;
return NULL;
}
int ivpu_ms_start_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
{
struct ivpu_file_priv *file_priv = file->driver_priv;
struct drm_ivpu_metric_streamer_start *args = data;
struct ivpu_device *vdev = file_priv->vdev;
struct ivpu_ms_instance *ms;
u64 single_buff_size;
u32 sample_size;
int ret;
if (!args->metric_group_mask || !args->read_period_samples ||
args->sampling_period_ns < MS_MIN_SAMPLE_PERIOD_NS)
return -EINVAL;
mutex_lock(&file_priv->ms_lock);
if (get_instance_by_mask(file_priv, args->metric_group_mask)) {
ivpu_err(vdev, "Instance already exists (mask %#llx)\n", args->metric_group_mask);
ret = -EALREADY;
goto unlock;
}
ms = kzalloc(sizeof(*ms), GFP_KERNEL);
if (!ms) {
ret = -ENOMEM;
goto unlock;
}
ms->mask = args->metric_group_mask;
ret = ivpu_jsm_metric_streamer_info(vdev, ms->mask, 0, 0, &sample_size, NULL);
if (ret)
goto err_free_ms;
single_buff_size = sample_size *
((u64)args->read_period_samples * MS_READ_PERIOD_MULTIPLIER);
ms->bo = ivpu_bo_create_global(vdev, PAGE_ALIGN(single_buff_size * MS_NUM_BUFFERS),
DRM_IVPU_BO_CACHED | DRM_IVPU_BO_MAPPABLE);
if (!ms->bo) {
ivpu_err(vdev, "Failed to allocate MS buffer (size %llu)\n", single_buff_size);
ret = -ENOMEM;
goto err_free_ms;
}
ms->buff_size = ivpu_bo_size(ms->bo) / MS_NUM_BUFFERS;
ms->active_buff_vpu_addr = ms->bo->vpu_addr;
ms->inactive_buff_vpu_addr = ms->bo->vpu_addr + ms->buff_size;
ms->active_buff_ptr = ivpu_bo_vaddr(ms->bo);
ms->inactive_buff_ptr = ivpu_bo_vaddr(ms->bo) + ms->buff_size;
ret = ivpu_jsm_metric_streamer_start(vdev, ms->mask, args->sampling_period_ns,
ms->active_buff_vpu_addr, ms->buff_size);
if (ret)
goto err_free_bo;
args->sample_size = sample_size;
args->max_data_size = ivpu_bo_size(ms->bo);
list_add_tail(&ms->ms_instance_node, &file_priv->ms_instance_list);
goto unlock;
err_free_bo:
ivpu_bo_free(ms->bo);
err_free_ms:
kfree(ms);
unlock:
mutex_unlock(&file_priv->ms_lock);
return ret;
}
static int
copy_leftover_bytes(struct ivpu_ms_instance *ms,
void __user *user_ptr, u64 user_size, u64 *user_bytes_copied)
{
u64 copy_bytes;
if (ms->leftover_bytes) {
copy_bytes = min(user_size - *user_bytes_copied, ms->leftover_bytes);
if (copy_to_user(user_ptr + *user_bytes_copied, ms->leftover_addr, copy_bytes))
return -EFAULT;
ms->leftover_bytes -= copy_bytes;
ms->leftover_addr += copy_bytes;
*user_bytes_copied += copy_bytes;
}
return 0;
}
static int
copy_samples_to_user(struct ivpu_device *vdev, struct ivpu_ms_instance *ms,
void __user *user_ptr, u64 user_size, u64 *user_bytes_copied)
{
u64 bytes_written;
int ret;
*user_bytes_copied = 0;
ret = copy_leftover_bytes(ms, user_ptr, user_size, user_bytes_copied);
if (ret)
return ret;
if (*user_bytes_copied == user_size)
return 0;
ret = ivpu_jsm_metric_streamer_update(vdev, ms->mask, ms->inactive_buff_vpu_addr,
ms->buff_size, &bytes_written);
if (ret)
return ret;
swap(ms->active_buff_vpu_addr, ms->inactive_buff_vpu_addr);
swap(ms->active_buff_ptr, ms->inactive_buff_ptr);
ms->leftover_bytes = bytes_written;
ms->leftover_addr = ms->inactive_buff_ptr;
return copy_leftover_bytes(ms, user_ptr, user_size, user_bytes_copied);
}
int ivpu_ms_get_data_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
{
struct drm_ivpu_metric_streamer_get_data *args = data;
struct ivpu_file_priv *file_priv = file->driver_priv;
struct ivpu_device *vdev = file_priv->vdev;
struct ivpu_ms_instance *ms;
u64 bytes_written;
int ret;
if (!args->metric_group_mask)
return -EINVAL;
mutex_lock(&file_priv->ms_lock);
ms = get_instance_by_mask(file_priv, args->metric_group_mask);
if (!ms) {
ivpu_err(vdev, "Instance doesn't exist for mask: %#llx\n", args->metric_group_mask);
ret = -EINVAL;
goto unlock;
}
if (!args->buffer_size) {
ret = ivpu_jsm_metric_streamer_update(vdev, ms->mask, 0, 0, &bytes_written);
if (ret)
goto unlock;
args->data_size = bytes_written + ms->leftover_bytes;
goto unlock;
}
if (!args->buffer_ptr) {
ret = -EINVAL;
goto unlock;
}
ret = copy_samples_to_user(vdev, ms, u64_to_user_ptr(args->buffer_ptr),
args->buffer_size, &args->data_size);
unlock:
mutex_unlock(&file_priv->ms_lock);
return ret;
}
static void free_instance(struct ivpu_file_priv *file_priv, struct ivpu_ms_instance *ms)
{
lockdep_assert_held(&file_priv->ms_lock);
list_del(&ms->ms_instance_node);
ivpu_jsm_metric_streamer_stop(file_priv->vdev, ms->mask);
ivpu_bo_free(ms->bo);
kfree(ms);
}
int ivpu_ms_stop_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
{
struct ivpu_file_priv *file_priv = file->driver_priv;
struct drm_ivpu_metric_streamer_stop *args = data;
struct ivpu_ms_instance *ms;
if (!args->metric_group_mask)
return -EINVAL;
mutex_lock(&file_priv->ms_lock);
ms = get_instance_by_mask(file_priv, args->metric_group_mask);
if (ms)
free_instance(file_priv, ms);
mutex_unlock(&file_priv->ms_lock);
return ms ? 0 : -EINVAL;
}
static inline struct ivpu_bo *get_ms_info_bo(struct ivpu_file_priv *file_priv)
{
lockdep_assert_held(&file_priv->ms_lock);
if (file_priv->ms_info_bo)
return file_priv->ms_info_bo;
file_priv->ms_info_bo = ivpu_bo_create_global(file_priv->vdev, MS_INFO_BUFFER_SIZE,
DRM_IVPU_BO_CACHED | DRM_IVPU_BO_MAPPABLE);
return file_priv->ms_info_bo;
}
int ivpu_ms_get_info_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
{
struct drm_ivpu_metric_streamer_get_data *args = data;
struct ivpu_file_priv *file_priv = file->driver_priv;
struct ivpu_device *vdev = file_priv->vdev;
struct ivpu_bo *bo;
u64 info_size;
int ret;
if (!args->metric_group_mask)
return -EINVAL;
if (!args->buffer_size)
return ivpu_jsm_metric_streamer_info(vdev, args->metric_group_mask,
0, 0, NULL, &args->data_size);
if (!args->buffer_ptr)
return -EINVAL;
mutex_lock(&file_priv->ms_lock);
bo = get_ms_info_bo(file_priv);
if (!bo) {
ret = -ENOMEM;
goto unlock;
}
ret = ivpu_jsm_metric_streamer_info(vdev, args->metric_group_mask, bo->vpu_addr,
ivpu_bo_size(bo), NULL, &info_size);
if (ret)
goto unlock;
if (args->buffer_size < info_size) {
ret = -ENOSPC;
goto unlock;
}
if (copy_to_user(u64_to_user_ptr(args->buffer_ptr), ivpu_bo_vaddr(bo), info_size))
ret = -EFAULT;
args->data_size = info_size;
unlock:
mutex_unlock(&file_priv->ms_lock);
return ret;
}
void ivpu_ms_cleanup(struct ivpu_file_priv *file_priv)
{
struct ivpu_ms_instance *ms, *tmp;
mutex_lock(&file_priv->ms_lock);
if (file_priv->ms_info_bo) {
ivpu_bo_free(file_priv->ms_info_bo);
file_priv->ms_info_bo = NULL;
}
list_for_each_entry_safe(ms, tmp, &file_priv->ms_instance_list, ms_instance_node)
free_instance(file_priv, ms);
mutex_unlock(&file_priv->ms_lock);
}
void ivpu_ms_cleanup_all(struct ivpu_device *vdev)
{
struct ivpu_file_priv *file_priv;
unsigned long ctx_id;
mutex_lock(&vdev->context_list_lock);
xa_for_each(&vdev->context_xa, ctx_id, file_priv)
ivpu_ms_cleanup(file_priv);
mutex_unlock(&vdev->context_list_lock);
}