This is a issue involve RS bit set rule in ixgbe. Let us take function i40e_xmit_pkts_vec () as an example, in this function RS bit will be set for descriptor with index txq->tx_next_rs, and also descriptor free function i40e_tx_free_bufs() also check RS bit for descriptor with index txq->tx_next_rs, This is perfect ok. Let us take an example, if app set tx_rs_thresh = 32 and nb_desc = 512, then i40e PMD code will init txq->tx_next_rs = 31 in function i40e_reset_tx_queue when tx queue setup. And also txq->tx_next_rs will be update as 63, 95 and so on. But, in the function i40e_dev_tx_descriptor_status(), the RS bit to check is " desc = ((desc + txq->tx_rs_thresh - 1) / txq->tx_rs_thresh) * txq-tx_rs_thresh", which is 32 ,64, 96 and so on. So, they are all wrong! In tx function of i40e_xmit_pkts_simple, the RS bit rule is also the same, it also set index 31 ,64, 95. we need to correct it.
Fixes: 68a43d1bb023 ("net/i40e: implement descriptor status API") Signed-off-by: Wei Zhao <wei.zh...@intel.com> --- drivers/net/i40e/i40e_rxtx.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/drivers/net/i40e/i40e_rxtx.c b/drivers/net/i40e/i40e_rxtx.c index 6032d55..7e37d8d 100644 --- a/drivers/net/i40e/i40e_rxtx.c +++ b/drivers/net/i40e/i40e_rxtx.c @@ -2032,15 +2032,15 @@ i40e_dev_tx_descriptor_status(void *tx_queue, uint16_t offset) return -EINVAL; desc = txq->tx_tail + offset; + if (desc >= txq->nb_tx_desc) + desc -= txq->nb_tx_desc; /* go to next desc that has the RS bit */ - desc = ((desc + txq->tx_rs_thresh - 1) / txq->tx_rs_thresh) * - txq->tx_rs_thresh; - if (desc >= txq->nb_tx_desc) { + desc = (desc / txq->tx_rs_thresh + 1) * + txq->tx_rs_thresh - 1; + if (desc >= txq->nb_tx_desc) desc -= txq->nb_tx_desc; - if (desc >= txq->nb_tx_desc) - desc -= txq->nb_tx_desc; - } + desc = txq->sw_ring[desc].last_id; status = &txq->tx_ring[desc].cmd_type_offset_bsz; mask = rte_le_to_cpu_64(I40E_TXD_QW1_DTYPE_MASK); expect = rte_cpu_to_le_64( -- 2.7.5