@@ -214,6 +214,9 @@ struct cn10k_ml_dev {
/* ML Firmware */
struct cn10k_ml_fw fw;
+
+ /* Number of models loaded */
+ uint16_t nb_models_loaded;
};
uint64_t cn10k_ml_fw_flags_get(struct cn10k_ml_fw *fw);
new file mode 100644
@@ -0,0 +1,5 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2022 Marvell.
+ */
+
+#include "cn10k_ml_model.h"
new file mode 100644
@@ -0,0 +1,43 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2022 Marvell.
+ */
+
+#ifndef _CN10K_ML_MODEL_H_
+#define _CN10K_ML_MODEL_H_
+
+#include <rte_mldev.h>
+
+#include <roc_api.h>
+
+#include "cn10k_ml_dev.h"
+
+/* Maximum number of models per device */
+#define ML_CN10K_MAX_MODELS 16
+
+/* Model state */
+enum cn10k_ml_model_state {
+ ML_CN10K_MODEL_STATE_LOADED,
+ ML_CN10K_MODEL_STATE_JOB_ACTIVE,
+ ML_CN10K_MODEL_STATE_STARTED,
+ ML_CN10K_MODEL_STATE_UNKNOWN,
+};
+
+/* ML Model Object */
+struct cn10k_ml_model {
+ /* Device reference */
+ struct cn10k_ml_dev *mldev;
+
+ /* Model name */
+ char name[RTE_ML_STR_MAX];
+
+ /* Model ID */
+ int16_t model_id;
+
+ /* Model lock, used to update model state */
+ plt_spinlock_t lock;
+
+ /* Model state */
+ enum cn10k_ml_model_state state;
+};
+
+#endif /* _CN10K_ML_MODEL_H_ */
@@ -6,8 +6,12 @@
#include <rte_mldev_pmd.h>
#include "cn10k_ml_dev.h"
+#include "cn10k_ml_model.h"
#include "cn10k_ml_ops.h"
+/* ML model macros */
+#define CN10K_ML_MODEL_MEMZONE_NAME "ml_cn10k_model_mz"
+
static void
qp_memzone_name_get(char *name, int size, int dev_id, int qp_id)
{
@@ -120,9 +124,11 @@ static int
cn10k_ml_dev_configure(struct rte_ml_dev *dev, const struct rte_ml_dev_config *conf)
{
struct rte_ml_dev_info dev_info;
+ struct cn10k_ml_model *model;
struct cn10k_ml_dev *mldev;
struct cn10k_ml_qp *qp;
uint32_t mz_size;
+ int16_t model_id;
uint16_t qp_id;
int ret;
@@ -203,6 +209,48 @@ cn10k_ml_dev_configure(struct rte_ml_dev *dev, const struct rte_ml_dev_config *c
}
dev->data->nb_queue_pairs = conf->nb_queue_pairs;
+ /* Allocate ML models */
+ if (dev->data->models == NULL) {
+ mz_size = sizeof(dev->data->models[0]) * conf->nb_models;
+ dev->data->models = rte_zmalloc("cn10k_mldev_models", mz_size, RTE_CACHE_LINE_SIZE);
+ if (dev->data->models == NULL) {
+ dev->data->nb_models = 0;
+ plt_err("Failed to get memory for ml_models, nb_models %u",
+ conf->nb_models);
+ ret = -ENOMEM;
+ goto error;
+ }
+ } else {
+ /* Re-configure */
+ void **models;
+
+ /* Unload all models */
+ for (model_id = 0; model_id < dev->data->nb_models; model_id++) {
+ model = dev->data->models[model_id];
+ if (model != NULL) {
+ if (model->state == ML_CN10K_MODEL_STATE_LOADED) {
+ if (cn10k_ml_model_unload(dev, model_id) != 0)
+ plt_err("Could not unload model %d", model_id);
+ }
+ dev->data->models[model_id] = NULL;
+ }
+ }
+
+ models = dev->data->models;
+ models = rte_realloc(models, sizeof(models[0]) * conf->nb_models,
+ RTE_CACHE_LINE_SIZE);
+ if (models == NULL) {
+ dev->data->nb_models = 0;
+ plt_err("Failed to realloc ml_models, nb_models = %u", conf->nb_models);
+ ret = -ENOMEM;
+ goto error;
+ }
+ memset(models, 0, sizeof(models[0]) * conf->nb_models);
+ dev->data->models = models;
+ }
+ dev->data->nb_models = conf->nb_models;
+
+ mldev->nb_models_loaded = 0;
mldev->state = ML_CN10K_DEV_STATE_CONFIGURED;
return 0;
@@ -211,14 +259,19 @@ cn10k_ml_dev_configure(struct rte_ml_dev *dev, const struct rte_ml_dev_config *c
if (dev->data->queue_pairs != NULL)
rte_free(dev->data->queue_pairs);
+ if (dev->data->models != NULL)
+ rte_free(dev->data->models);
+
return ret;
}
static int
cn10k_ml_dev_close(struct rte_ml_dev *dev)
{
+ struct cn10k_ml_model *model;
struct cn10k_ml_dev *mldev;
struct cn10k_ml_qp *qp;
+ int16_t model_id;
uint16_t qp_id;
if (dev == NULL)
@@ -226,6 +279,21 @@ cn10k_ml_dev_close(struct rte_ml_dev *dev)
mldev = dev->data->dev_private;
+ /* Unload all models */
+ for (model_id = 0; model_id < dev->data->nb_models; model_id++) {
+ model = dev->data->models[model_id];
+ if (model != NULL) {
+ if (model->state == ML_CN10K_MODEL_STATE_LOADED) {
+ if (cn10k_ml_model_unload(dev, model_id) != 0)
+ plt_err("Could not unload model %d", model_id);
+ }
+ dev->data->models[model_id] = NULL;
+ }
+ }
+
+ if (dev->data->models)
+ rte_free(dev->data->models);
+
/* Destroy all queue pairs */
for (qp_id = 0; qp_id < dev->data->nb_queue_pairs; qp_id++) {
qp = dev->data->queue_pairs[qp_id];
@@ -337,6 +405,88 @@ cn10k_ml_dev_queue_pair_setup(struct rte_ml_dev *dev, uint16_t queue_pair_id,
return 0;
}
+int
+cn10k_ml_model_load(struct rte_ml_dev *dev, struct rte_ml_model_params *params, int16_t *model_id)
+{
+ struct cn10k_ml_model *model;
+ struct cn10k_ml_dev *mldev;
+
+ char str[RTE_MEMZONE_NAMESIZE];
+ const struct plt_memzone *mz;
+ uint64_t mz_size;
+ uint16_t idx;
+ bool found;
+
+ PLT_SET_USED(params);
+
+ mldev = dev->data->dev_private;
+
+ /* Find model ID */
+ found = false;
+ for (idx = 0; idx < dev->data->nb_models; idx++) {
+ if (dev->data->models[idx] == NULL) {
+ found = true;
+ break;
+ }
+ }
+
+ if (!found) {
+ plt_err("No slots available to load new model");
+ return -ENOMEM;
+ }
+
+ /* Get MZ size */
+ mz_size = PLT_ALIGN_CEIL(sizeof(struct cn10k_ml_model), ML_CN10K_ALIGN_SIZE);
+
+ /* Allocate memzone for model object and model data */
+ snprintf(str, RTE_MEMZONE_NAMESIZE, "%s_%u", CN10K_ML_MODEL_MEMZONE_NAME, idx);
+ mz = plt_memzone_reserve_aligned(str, mz_size, 0, ML_CN10K_ALIGN_SIZE);
+ if (!mz) {
+ plt_err("plt_memzone_reserve failed : %s", str);
+ return -ENOMEM;
+ }
+
+ model = mz->addr;
+ model->mldev = mldev;
+ model->model_id = idx;
+
+ plt_spinlock_init(&model->lock);
+ model->state = ML_CN10K_MODEL_STATE_LOADED;
+ dev->data->models[idx] = model;
+ mldev->nb_models_loaded++;
+
+ *model_id = idx;
+
+ return 0;
+}
+
+int
+cn10k_ml_model_unload(struct rte_ml_dev *dev, int16_t model_id)
+{
+ char str[RTE_MEMZONE_NAMESIZE];
+ struct cn10k_ml_model *model;
+ struct cn10k_ml_dev *mldev;
+
+ mldev = dev->data->dev_private;
+ model = dev->data->models[model_id];
+
+ if (model == NULL) {
+ plt_err("Invalid model_id = %d", model_id);
+ return -EINVAL;
+ }
+
+ if (model->state != ML_CN10K_MODEL_STATE_LOADED) {
+ plt_err("Cannot unload. Model in use.");
+ return -EBUSY;
+ }
+
+ dev->data->models[model_id] = NULL;
+ mldev->nb_models_loaded--;
+
+ snprintf(str, RTE_MEMZONE_NAMESIZE, "%s_%d", CN10K_ML_MODEL_MEMZONE_NAME, model_id);
+ return plt_memzone_free(plt_memzone_lookup(str));
+}
+
struct rte_ml_dev_ops cn10k_ml_ops = {
/* Device control ops */
.dev_info_get = cn10k_ml_dev_info_get,
@@ -348,4 +498,8 @@ struct rte_ml_dev_ops cn10k_ml_ops = {
/* Queue-pair handling ops */
.dev_queue_pair_setup = cn10k_ml_dev_queue_pair_setup,
.dev_queue_pair_release = cn10k_ml_dev_queue_pair_release,
+
+ /* Model ops */
+ .model_load = cn10k_ml_model_load,
+ .model_unload = cn10k_ml_model_unload,
};
@@ -53,4 +53,9 @@ struct cn10k_ml_qp {
/* CN10K device ops */
extern struct rte_ml_dev_ops cn10k_ml_ops;
+/* Slow-path ops */
+int cn10k_ml_model_load(struct rte_ml_dev *dev, struct rte_ml_model_params *params,
+ int16_t *model_id);
+int cn10k_ml_model_unload(struct rte_ml_dev *dev, int16_t model_id);
+
#endif /* _CN10K_ML_OPS_H_ */
@@ -10,11 +10,13 @@ endif
sources = files(
'cn10k_ml_dev.c',
'cn10k_ml_ops.c',
+ 'cn10k_ml_model.c',
)
headers = files(
'cn10k_ml_dev.h',
'cn10k_ml_ops.h',
+ 'cn10k_ml_model.h',
)
deps += ['mldev', 'common_ml', 'common_cnxk', 'kvargs']