From: Shijith Thotton Date: Sat, 25 Mar 2017 06:24:40 +0000 (+0530) Subject: net/liquidio: add API to flush IQ X-Git-Url: http://git.droids-corp.org/?a=commitdiff_plain;h=5ee7640f95a0bd48f6ae18a776d037ce1abeb451;p=dpdk.git net/liquidio: add API to flush IQ API to flush instruction queue checks how many packets reached device and frees associated host buffers using request list. Signed-off-by: Shijith Thotton Signed-off-by: Jerin Jacob Signed-off-by: Derek Chickles Signed-off-by: Venkat Koppula Signed-off-by: Srisivasubramanian S Signed-off-by: Mallesham Jatharakonda --- diff --git a/drivers/net/liquidio/lio_ethdev.c b/drivers/net/liquidio/lio_ethdev.c index 5c5aade5fd..b8baa4f41b 100644 --- a/drivers/net/liquidio/lio_ethdev.c +++ b/drivers/net/liquidio/lio_ethdev.c @@ -281,6 +281,7 @@ static int lio_dev_configure(struct rte_eth_dev *eth_dev) * response arrived or timed-out. */ while ((*sc->status_word == LIO_COMPLETION_WORD_INIT) && --timeout) { + lio_flush_iq(lio_dev, lio_dev->instr_queue[sc->iq_no]); lio_process_ordered_list(lio_dev); rte_delay_ms(1); } diff --git a/drivers/net/liquidio/lio_rxtx.c b/drivers/net/liquidio/lio_rxtx.c index 75ecbd563a..64c038510a 100644 --- a/drivers/net/liquidio/lio_rxtx.c +++ b/drivers/net/liquidio/lio_rxtx.c @@ -41,6 +41,9 @@ #include "lio_rxtx.h" #define LIO_MAX_SG 12 +/* Flush iq if available tx_desc fall below LIO_FLUSH_WM */ +#define LIO_FLUSH_WM(_iq) ((_iq)->max_count / 2) +#define LIO_PKT_IN_DONE_CNT_MASK 0x00000000FFFFFFFFULL static void lio_droq_compute_max_packet_bufs(struct lio_droq *droq) @@ -977,6 +980,146 @@ lio_add_to_request_list(struct lio_instr_queue *iq, iq->request_list[idx].reqtype = reqtype; } +static inline void +lio_free_netsgbuf(void *buf) +{ + struct lio_buf_free_info *finfo = buf; + struct lio_device *lio_dev = finfo->lio_dev; + struct rte_mbuf *m = finfo->mbuf; + struct lio_gather *g = finfo->g; + uint8_t iq = finfo->iq_no; + + /* This will take care of multiple segments also */ + rte_pktmbuf_free(m); + + rte_spinlock_lock(&lio_dev->glist_lock[iq]); + STAILQ_INSERT_TAIL(&lio_dev->glist_head[iq], &g->list, entries); + rte_spinlock_unlock(&lio_dev->glist_lock[iq]); + rte_free(finfo); +} + +/* Can only run in process context */ +static int +lio_process_iq_request_list(struct lio_device *lio_dev, + struct lio_instr_queue *iq) +{ + struct octeon_instr_irh *irh = NULL; + uint32_t old = iq->flush_index; + struct lio_soft_command *sc; + uint32_t inst_count = 0; + int reqtype; + void *buf; + + while (old != iq->lio_read_index) { + reqtype = iq->request_list[old].reqtype; + buf = iq->request_list[old].buf; + + if (reqtype == LIO_REQTYPE_NONE) + goto skip_this; + + switch (reqtype) { + case LIO_REQTYPE_NORESP_NET: + rte_pktmbuf_free((struct rte_mbuf *)buf); + break; + case LIO_REQTYPE_NORESP_NET_SG: + lio_free_netsgbuf(buf); + break; + case LIO_REQTYPE_SOFT_COMMAND: + sc = buf; + irh = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh; + if (irh->rflag) { + /* We're expecting a response from Octeon. + * It's up to lio_process_ordered_list() to + * process sc. Add sc to the ordered soft + * command response list because we expect + * a response from Octeon. + */ + rte_spinlock_lock(&lio_dev->response_list.lock); + rte_atomic64_inc( + &lio_dev->response_list.pending_req_count); + STAILQ_INSERT_TAIL( + &lio_dev->response_list.head, + &sc->node, entries); + rte_spinlock_unlock( + &lio_dev->response_list.lock); + } else { + if (sc->callback) { + /* This callback must not sleep */ + sc->callback(LIO_REQUEST_DONE, + sc->callback_arg); + } + } + break; + default: + lio_dev_err(lio_dev, + "Unknown reqtype: %d buf: %p at idx %d\n", + reqtype, buf, old); + } + + iq->request_list[old].buf = NULL; + iq->request_list[old].reqtype = 0; + +skip_this: + inst_count++; + old = lio_incr_index(old, 1, iq->max_count); + } + + iq->flush_index = old; + + return inst_count; +} + +static void +lio_update_read_index(struct lio_instr_queue *iq) +{ + uint32_t pkt_in_done = rte_read32(iq->inst_cnt_reg); + uint32_t last_done; + + last_done = pkt_in_done - iq->pkt_in_done; + iq->pkt_in_done = pkt_in_done; + + /* Add last_done and modulo with the IQ size to get new index */ + iq->lio_read_index = (iq->lio_read_index + + (uint32_t)(last_done & LIO_PKT_IN_DONE_CNT_MASK)) % + iq->max_count; +} + +int +lio_flush_iq(struct lio_device *lio_dev, struct lio_instr_queue *iq) +{ + uint32_t tot_inst_processed = 0; + uint32_t inst_processed = 0; + int tx_done = 1; + + if (rte_atomic64_test_and_set(&iq->iq_flush_running) == 0) + return tx_done; + + rte_spinlock_lock(&iq->lock); + + lio_update_read_index(iq); + + do { + /* Process any outstanding IQ packets. */ + if (iq->flush_index == iq->lio_read_index) + break; + + inst_processed = lio_process_iq_request_list(lio_dev, iq); + + if (inst_processed) + rte_atomic64_sub(&iq->instr_pending, inst_processed); + + tot_inst_processed += inst_processed; + inst_processed = 0; + + } while (1); + + rte_spinlock_unlock(&iq->lock); + + rte_atomic64_clear(&iq->iq_flush_running); + + return tx_done; +} + static int lio_send_command(struct lio_device *lio_dev, uint32_t iq_no, void *cmd, void *buf, uint32_t datasize __rte_unused, uint32_t reqtype) @@ -1385,6 +1528,35 @@ lio_delete_instruction_queue(struct lio_device *lio_dev, int iq_no) lio_dev->num_iqs--; } +static inline uint32_t +lio_iq_get_available(struct lio_device *lio_dev, uint32_t q_no) +{ + return ((lio_dev->instr_queue[q_no]->max_count - 1) - + (uint32_t)rte_atomic64_read( + &lio_dev->instr_queue[q_no]->instr_pending)); +} + +static inline int +lio_iq_is_full(struct lio_device *lio_dev, uint32_t q_no) +{ + return ((uint32_t)rte_atomic64_read( + &lio_dev->instr_queue[q_no]->instr_pending) >= + (lio_dev->instr_queue[q_no]->max_count - 2)); +} + +static int +lio_dev_cleanup_iq(struct lio_device *lio_dev, int iq_no) +{ + struct lio_instr_queue *iq = lio_dev->instr_queue[iq_no]; + uint32_t count = 10000; + + while ((lio_iq_get_available(lio_dev, iq_no) < LIO_FLUSH_WM(iq)) && + --count) + lio_flush_iq(lio_dev, iq); + + return count ? 0 : 1; +} + /** Send data packet to the device * @param lio_dev - lio device pointer * @param ndata - control structure with queueing, and buffer information @@ -1421,6 +1593,8 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts) goto xmit_failed; } + lio_dev_cleanup_iq(lio_dev, iq_no); + for (i = 0; i < nb_pkts; i++) { uint32_t pkt_len = 0; @@ -1432,6 +1606,14 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts) ndata.buf = m; ndata.q_no = iq_no; + if (lio_iq_is_full(lio_dev, ndata.q_no)) { + if (lio_dev_cleanup_iq(lio_dev, iq_no)) { + PMD_TX_LOG(lio_dev, ERR, + "Transmit failed iq:%d full\n", + ndata.q_no); + break; + } + } cmdsetup.cmd_setup64 = 0; cmdsetup.s.iq_no = iq_no; @@ -1524,8 +1706,11 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts) break; } - if (unlikely(status == LIO_IQ_SEND_STOP)) + if (unlikely(status == LIO_IQ_SEND_STOP)) { PMD_TX_LOG(lio_dev, DEBUG, "iq full\n"); + /* create space as iq is full */ + lio_dev_cleanup_iq(lio_dev, iq_no); + } processed++; } diff --git a/drivers/net/liquidio/lio_rxtx.h b/drivers/net/liquidio/lio_rxtx.h index b555bde9d0..0a4cc2b517 100644 --- a/drivers/net/liquidio/lio_rxtx.h +++ b/drivers/net/liquidio/lio_rxtx.h @@ -706,6 +706,7 @@ uint16_t lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, int lio_setup_iq(struct lio_device *lio_dev, int q_index, union octeon_txpciq iq_no, uint32_t num_descs, void *app_ctx, unsigned int socket_id); +int lio_flush_iq(struct lio_device *lio_dev, struct lio_instr_queue *iq); void lio_delete_instruction_queue(struct lio_device *lio_dev, int iq_no); /** Setup instruction queue zero for the device * @param lio_dev which lio device to setup