Using model_id from op structure as layer index for TVM models with multiple MRVL layers is incorrect. This patch fixes the issue by reading the layer index from the job descriptor structure.
Fixes: 31cc30b5f088 ("ml/cnxk: update model start/stop functions") Signed-off-by: Srikanth Yalavarthi <syalavar...@marvell.com> --- v2: - restart upstream CI drivers/ml/cnxk/cn10k_ml_ops.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/drivers/ml/cnxk/cn10k_ml_ops.c b/drivers/ml/cnxk/cn10k_ml_ops.c index b9d30278c61..7f7e5efceac 100644 --- a/drivers/ml/cnxk/cn10k_ml_ops.c +++ b/drivers/ml/cnxk/cn10k_ml_ops.c @@ -1191,13 +1191,15 @@ cn10k_ml_result_update(struct cnxk_ml_dev *cnxk_mldev, int qp_id, void *request) uint64_t fw_latency; uint16_t model_id; uint16_t layer_id; + uint16_t idx; req = (struct cnxk_ml_req *)request; result = &req->cn10k_req.result; op = req->op; if (likely(result->error_code == 0)) { - model_id = cnxk_mldev->index_map[op->model_id].model_id; - layer_id = cnxk_mldev->index_map[op->model_id].layer_id; + idx = req->cn10k_req.jd.hdr.model_id; + model_id = cnxk_mldev->index_map[idx].model_id; + layer_id = cnxk_mldev->index_map[idx].layer_id; model = cnxk_mldev->mldev->data->models[model_id]; layer = &model->layer[layer_id]; if (likely(qp_id >= 0)) { -- 2.42.0