net/netvsc: patches
cover net/netvsc: patches
A couple of fixes for netvsc PMD.
The first is more serious than the second.
Stephen Hemminger (2):
net/netvsc: handle receive packets during multi-channel setup net/netvsc: avoid mixing buffered and direct packets
drivers/net/netvsc/hn_ethdev.c | 1 + drivers/net/netvsc/hn_nvs.c | 40 ++++++++++++++++++++++++++++++++-- drivers/net/netvsc/hn_rxtx.c | 15 ++++++------- drivers/net/netvsc/hn_var.h | 1 + 4 files changed, 47 insertions(+), 10 deletions(-)
Signature · 1 line
2.20.1
PATCH 1/2 net/netvsc: handle receive packets during multi-channel setup
It is possible for a packet to arrive during the configuration process when setting up multiple queue mode. This would cause configure to fail; fix by just ignoring receive packets while waiting for control commands.
Add a spinlock to ensure that two control operations are not attempted at once.
--- drivers/net/netvsc/hn_ethdev.c | 1 + drivers/net/netvsc/hn_nvs.c | 40 ++++++++++++++++++++++++++++++++-- drivers/net/netvsc/hn_var.h | 1 + 3 files changed, 40 insertions(+), 2 deletions(-)
drivers/net/netvsc/hn_ethdev.c +1 -0
@@ -951,6 +951,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev) hv->latency =HN_CHAN_LATENCY_NS ; hv->max_queues =1 ;rte_spinlock_init (&hv->vf_lock); +rte_spinlock_init (&hv->nvs_lock); hv->vf_port =HN_INVALID_PORT ; err =hn_parse_args (eth_dev);
drivers/net/netvsc/hn_nvs.c +38 -2
@@ -54,7 +54,7 @@ static int hn_nvs_req_send(struct hn_data *hv, }static int -hn_nvs_execute (struct hn_data *hv, +__hn_nvs_execute (struct hn_data *hv,void *req,uint32_t reqlen,void *resp,uint32_t resplen,uint32_t type)
@@ -62,6 +62,7 @@ hn_nvs_execute(struct hn_data *hv,struct vmbus_channel *chan =hn_primary_chan (hv);char buffer[NVS_RESPSIZE_MAX ];const struct hn_nvs_hdr *hdr; +uint64_t xactid;uint32_t len;int ret;
@@ -77,7 +78,7 @@ hn_nvs_execute(struct hn_data *hv, retry: len =sizeof (buffer); - ret =rte_vmbus_chan_recv (chan, buffer, &len,NULL ); + ret =rte_vmbus_chan_recv (chan, buffer, &len, &xactid);if (ret == -EAGAIN ) {rte_delay_us (HN_CHAN_INTERVAL_US ); goto retry;
@@ -88,7 +89,20 @@ hn_nvs_execute(struct hn_data *hv,return ret; } +if (len <sizeof (*hdr)) { +PMD_DRV_LOG (ERR ,"response missing NVS header" ); +return -EINVAL ; + } + hdr = (struct hn_nvs_hdr *)buffer; + +/* Silently drop received packets while waiting for response */ +if (hdr->type ==NVS_TYPE_RNDIS ) { +hn_nvs_ack_rxbuf (chan, xactid); + --hv->rxbuf_outstanding; + goto retry; + } +if (hdr->type != type) {PMD_DRV_LOG (ERR ,"unexpected NVS resp %#x, expect %#x" , hdr->type, type);
@@ -108,6 +122,28 @@ hn_nvs_execute(struct hn_data *hv,return 0 ; } + +/* +* Execute one control command and get the response. +* Only one command can be active on a channel at once +* Unlike BSD, DPDK does not have an interrupt context +* so the polling is required to wait for response. +*/ +static int +hn_nvs_execute (struct hn_data *hv, +void *req,uint32_t reqlen, +void *resp,uint32_t resplen, +uint32_t type) +{ +int ret; + +rte_spinlock_lock (&hv->nvs_lock); + ret =__hn_nvs_execute (hv, req, reqlen, resp, resplen, type); +rte_spinlock_unlock (&hv->nvs_lock); + +return ret; +} +static int hn_nvs_doinit (struct hn_data *hv,uint32_t nvs_ver) {
drivers/net/netvsc/hn_var.h +1 -0
@@ -97,6 +97,7 @@ struct hn_data {struct rte_vmbus_device *vmbus;struct hn_rx_queue *primary;rte_spinlock_t vf_lock; +rte_spinlock_t nvs_lock;uint16_t port_id;uint16_t vf_port;
Signature · 1 line
2.20.1
PATCH 2/2 net/netvsc: avoid mixing buffered and direct packets
In reply to Stephen HemmingerPutting buffered and direct packets together into one request is an unnecessary optimization. Better to split into two requests (buffered and direct).
--- drivers/net/netvsc/hn_rxtx.c | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-)
drivers/net/netvsc/hn_rxtx.c +7 -8
@@ -1362,15 +1362,14 @@ hn_xmit_pkts(void *ptxq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) }else {struct hn_txdesc *txd; -/* can send chimney data and large packet at once */ +/* Send any outstanding packets in buffer */ txd = txq->agg_txd; -if (txd) { -hn_reset_txagg (txq); - }else { - txd =hn_new_txd (hv, txq); -if (unlikely (!txd)) -break ; - } +if (txd &&hn_flush_txagg (txq, &need_sig)) + goto fail; + + txd =hn_new_txd (hv, txq); +if (unlikely (!txd)) +break ; pkt = txd->rndis_pkt; txd->m = m;
Signature · 1 line
2.20.1
Quoted 7 lines · Stephen Hemminger
A couple of fixes for netvsc PMD.
The first is more serious than the second.Stephen Hemminger (2):
net/netvsc: handle receive packets during multi-channel setup net/netvsc: avoid mixing buffered and direct packets
Series applied to dpdk-next-net/master, thanks.
Quoted 10 lines · On Wed, 18 Mar 2020 17:33:18 +0000 Ferruh Yigit <[email protected]> wrote:
Quoted 7 lines · Stephen Hemminger
A couple of fixes for netvsc PMD.
The first is more serious than the second.Stephen Hemminger (2):
net/netvsc: handle receive packets during multi-channel setup net/netvsc: avoid mixing buffered and direct packetsSeries applied to dpdk-next-net/master, thanks.
Could you back it out. I have v2 to send today.
Quoted 7 lines · Stephen Hemminger
Quoted 4 lines · On Wed, 18 Mar 2020 17:33:18 +0000 Ferruh Yigit <[email protected]> wrote:
Quoted 3 lines · Stephen Hemminger
A couple of fixes for netvsc PMD.
The first is more serious than the second.
Quoted 9 lines
Quoted 6 lines
Quoted 4 lines
Stephen Hemminger (2):
net/netvsc: handle receive packets during multi-channel setup net/netvsc: avoid mixing buffered and direct packetsSeries applied to dpdk-next-net/master, thanks.
Could you back it out. I have v2 to send today.
Done, dropped from next-net.
v2 cover net/netvsc patches
In reply to Stephen HemmingerThese are bug fixes for netvsc PMD mostly related to issues discovered by users of multi-queue.
v2 -- simplify the locking on multi-channel setup
add patches to handle single queue and tx descriptor pool
Stephen Hemminger (4):
net/netvsc: propogate descriptor limits from VF to netvsc net/netvsc: handle receive packets during multi-channel setup net/netvsc: split send buffers from transmit descriptors net/netvsc: don't enable RSS if only single receive queue
drivers/net/netvsc/hn_ethdev.c | 22 ++- drivers/net/netvsc/hn_nvs.c | 41 +++++- drivers/net/netvsc/hn_rxtx.c | 258 ++++++++++++++++++++------------- drivers/net/netvsc/hn_var.h | 10 +- drivers/net/netvsc/hn_vf.c | 13 ++ 5 files changed, 231 insertions(+), 113 deletions(-)
Signature · 1 line
2.20.1
PATCH v2 1/4 net/netvsc: propogate descriptor limits from VF to netvsc
If application cares about descriptor limits, the netvsc device values should reflect those of the VF as well.
--- drivers/net/netvsc/hn_vf.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
drivers/net/netvsc/hn_vf.c +13 -0
@@ -167,6 +167,17 @@ hn_nvs_handle_vfassoc(struct rte_eth_dev *dev,hn_vf_remove (hv); } +static void +hn_vf_merge_desc_lim (struct rte_eth_desc_lim *lim, +const struct rte_eth_desc_lim *vf_lim) +{ + lim->nb_max =RTE_MIN (vf_lim->nb_max, lim->nb_max); + lim->nb_min =RTE_MAX (vf_lim->nb_min, lim->nb_min); + lim->nb_align =RTE_MAX (vf_lim->nb_align, lim->nb_align); + lim->nb_seg_max =RTE_MIN (vf_lim->nb_seg_max, lim->nb_seg_max); + lim->nb_mtu_seg_max =RTE_MIN (vf_lim->nb_seg_max, lim->nb_seg_max); +} +/* * Merge the info from the VF and synthetic path. * use the default config of the VF
@@ -196,11 +207,13 @@ static int hn_vf_info_merge(struct rte_eth_dev *vf_dev, info->max_tx_queues); info->tx_offload_capa &= vf_info.tx_offload_capa; info->tx_queue_offload_capa &= vf_info.tx_queue_offload_capa; +hn_vf_merge_desc_lim (&info->tx_desc_lim, &vf_info.tx_desc_lim); info->min_rx_bufsize =RTE_MAX (vf_info.min_rx_bufsize, info->min_rx_bufsize); info->max_rx_pktlen =RTE_MAX (vf_info.max_rx_pktlen, info->max_rx_pktlen); +hn_vf_merge_desc_lim (&info->rx_desc_lim, &vf_info.rx_desc_lim);return 0 ; }
Signature · 1 line
2.20.1
PATCH v2 2/4 net/netvsc: handle receive packets during multi-channel setup
In reply to Stephen HemmingerIt is possible for a packet to arrive during the configuration process when setting up multiple queue mode. This would cause configure to fail; fix by just ignoring receive packets while waiting for control commands.
Use the receive ring lock to avoid possible races between oddly behaved applications doing rx_burst and control operations concurrently.
--- drivers/net/netvsc/hn_nvs.c | 41 +++++++++++++++++++++++++++++++++++-- 1 file changed, 39 insertions(+), 2 deletions(-)
drivers/net/netvsc/hn_nvs.c +39 -2
@@ -54,7 +54,7 @@ static int hn_nvs_req_send(struct hn_data *hv, }static int -hn_nvs_execute (struct hn_data *hv, +__hn_nvs_execute (struct hn_data *hv,void *req,uint32_t reqlen,void *resp,uint32_t resplen,uint32_t type)
@@ -62,6 +62,7 @@ hn_nvs_execute(struct hn_data *hv,struct vmbus_channel *chan =hn_primary_chan (hv);char buffer[NVS_RESPSIZE_MAX ];const struct hn_nvs_hdr *hdr; +uint64_t xactid;uint32_t len;int ret;
@@ -77,7 +78,7 @@ hn_nvs_execute(struct hn_data *hv, retry: len =sizeof (buffer); - ret =rte_vmbus_chan_recv (chan, buffer, &len,NULL ); + ret =rte_vmbus_chan_recv (chan, buffer, &len, &xactid);if (ret == -EAGAIN ) {rte_delay_us (HN_CHAN_INTERVAL_US ); goto retry;
@@ -88,7 +89,20 @@ hn_nvs_execute(struct hn_data *hv,return ret; } +if (len <sizeof (*hdr)) { +PMD_DRV_LOG (ERR ,"response missing NVS header" ); +return -EINVAL ; + } + hdr = (struct hn_nvs_hdr *)buffer; + +/* Silently drop received packets while waiting for response */ +if (hdr->type ==NVS_TYPE_RNDIS ) { +hn_nvs_ack_rxbuf (chan, xactid); + --hv->rxbuf_outstanding; + goto retry; + } +if (hdr->type != type) {PMD_DRV_LOG (ERR ,"unexpected NVS resp %#x, expect %#x" , hdr->type, type);
@@ -108,6 +122,29 @@ hn_nvs_execute(struct hn_data *hv,return 0 ; } + +/* +* Execute one control command and get the response. +* Only one command can be active on a channel at once +* Unlike BSD, DPDK does not have an interrupt context +* so the polling is required to wait for response. +*/ +static int +hn_nvs_execute (struct hn_data *hv, +void *req,uint32_t reqlen, +void *resp,uint32_t resplen, +uint32_t type) +{ +struct hn_rx_queue *rxq = hv->primary; +int ret; + +rte_spinlock_lock (&rxq->ring_lock); + ret =__hn_nvs_execute (hv, req, reqlen, resp, resplen, type); +rte_spinlock_unlock (&rxq->ring_lock); + +return ret; +} +static int hn_nvs_doinit (struct hn_data *hv,uint32_t nvs_ver) {
Signature · 1 line
2.20.1
PATCH v2 3/4 net/netvsc: split send buffers from transmit descriptors
In reply to Stephen HemmingerThe VMBus has reserved transmit area (per device) and transmit descriptors (per queue). The previous code was always having a 1:1 mapping between send buffers and descriptors. This can lead to one queue starving another and also buffer bloat.
Change to working more like FreeBSD where there is a pool of transmit descriptors per queue. If send buffer is not available then no aggregation happens but the queue can still drain.
--- drivers/net/netvsc/hn_ethdev.c | 9 +- drivers/net/netvsc/hn_rxtx.c | 258 ++++++++++++++++++++------------- drivers/net/netvsc/hn_var.h | 10 +- 3 files changed, 171 insertions(+), 106 deletions(-)
drivers/net/netvsc/hn_ethdev.c +6 -3
@@ -257,6 +257,9 @@ static int hn_dev_info_get(struct rte_eth_dev *dev, dev_info->max_rx_queues = hv->max_queues; dev_info->max_tx_queues = hv->max_queues; + dev_info->tx_desc_lim.nb_min =1 ; + dev_info->tx_desc_lim.nb_max =4096 ; +if (rte_eal_process_type () !=RTE_PROC_PRIMARY )return 0 ;
@@ -982,7 +985,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)if (err) goto failed; - err =hn_tx_pool_init (eth_dev); + err =hn_chim_init (eth_dev);if (err) goto failed;
@@ -1018,7 +1021,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev) failed:PMD_INIT_LOG (NOTICE ,"device init failed" ); -hn_tx_pool_uninit (eth_dev); +hn_chim_uninit (eth_dev);hn_detach (hv);return err; }
@@ -1042,7 +1045,7 @@ eth_hn_dev_uninit(struct rte_eth_dev *eth_dev) eth_dev->rx_pkt_burst =NULL ;hn_detach (hv); -hn_tx_pool_uninit (eth_dev); +hn_chim_uninit (eth_dev);rte_vmbus_chan_close (hv->primary->chan);rte_free (hv->primary); ret =rte_eth_dev_owner_delete (hv->owner.id);
drivers/net/netvsc/hn_rxtx.c +158 -100
408 lines not on this page · raw message
drivers/net/netvsc/hn_var.h +7 -3
@@ -52,6 +52,8 @@ struct hn_tx_queue {uint16_t port_id;uint16_t queue_id;uint32_t free_thresh; +struct rte_mempool *txdesc_pool; +void *tx_rndis;/* Applied packet transmission aggregation limits. */ uint32_t agg_szmax;
@@ -115,8 +117,10 @@ struct hn_data {uint16_t num_queues;uint64_t rss_offloads; +rte_spinlock_t chim_lock;struct rte_mem_resource *chim_res;/* UIO resource for Tx */ -struct rte_mempool *tx_pool;/* Tx descriptors */ +struct rte_bitmap *chim_bmap;/* Send buffer map */ +void *chim_bmem;uint32_t chim_szmax;/* Max size per buffer */ uint32_t chim_cnt;/* Max packets per buffer */
@@ -157,8 +161,8 @@ uint16_t hn_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,uint16_t hn_recv_pkts (void *rx_queue,struct rte_mbuf **rx_pkts,uint16_t nb_pkts); -int hn_tx_pool_init (struct rte_eth_dev *dev); -void hn_tx_pool_uninit (struct rte_eth_dev *dev); +int hn_chim_init (struct rte_eth_dev *dev); +void hn_chim_uninit (struct rte_eth_dev *dev);int hn_dev_link_update (struct rte_eth_dev *dev,int wait); int hn_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, uint16_t nb_desc, unsigned int socket_id,
Signature · 1 line
2.20.1
PATCH v2 4/4 net/netvsc: don't enable RSS if only single receive queue
In reply to Stephen HemmingerIf application has multiple transmit queues but only a single receive queue, then do not enable RSS.
--- drivers/net/netvsc/hn_ethdev.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)
drivers/net/netvsc/hn_ethdev.c +8 -5
@@ -597,11 +597,14 @@ static int hn_dev_configure(struct rte_eth_dev *dev)return err; } - err =hn_rndis_conf_rss (hv,0 ); -if (err) { -PMD_DRV_LOG (NOTICE , -"initial RSS config failed" ); -return err; +if ((dev_conf->rxmode.offloads &DEV_RX_OFFLOAD_RSS_HASH ) && + dev->data->nb_rx_queues >1 ) { + err =hn_rndis_conf_rss (hv,0 ); +if (err) { +PMD_DRV_LOG (NOTICE , +"initial RSS config failed" ); +return err; + } } }
Signature · 1 line
2.20.1
v3 cover net/netvsc patches
In reply to Stephen HemmingerThese are bug fixes for netvsc PMD mostly related to issues discovered by users of multi-queue.
v3 - fix checkpatch complaints
v2 - simplify the locking on multi-channel setup
add patches to handle single queue and tx descriptor pool
Stephen Hemminger (4):
net/netvsc: propagate descriptor limits from VF to netvsc net/netvsc: handle receive packets during multi-channel setup net/netvsc: split send buffers from transmit descriptors net/netvsc: don't enable RSS if only single receive queue
drivers/net/netvsc/hn_ethdev.c | 22 ++- drivers/net/netvsc/hn_nvs.c | 41 +++++- drivers/net/netvsc/hn_rxtx.c | 261 ++++++++++++++++++++------------- drivers/net/netvsc/hn_var.h | 10 +- drivers/net/netvsc/hn_vf.c | 13 ++ 5 files changed, 234 insertions(+), 113 deletions(-)
Signature · 1 line
2.20.1
PATCH v3 1/4 net/netvsc: propagate descriptor limits from VF to netvsc
If application cares about descriptor limits, the netvsc device values should reflect those of the VF as well.
--- drivers/net/netvsc/hn_vf.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
drivers/net/netvsc/hn_vf.c +13 -0
@@ -167,6 +167,17 @@ hn_nvs_handle_vfassoc(struct rte_eth_dev *dev,hn_vf_remove (hv); } +static void +hn_vf_merge_desc_lim (struct rte_eth_desc_lim *lim, +const struct rte_eth_desc_lim *vf_lim) +{ + lim->nb_max =RTE_MIN (vf_lim->nb_max, lim->nb_max); + lim->nb_min =RTE_MAX (vf_lim->nb_min, lim->nb_min); + lim->nb_align =RTE_MAX (vf_lim->nb_align, lim->nb_align); + lim->nb_seg_max =RTE_MIN (vf_lim->nb_seg_max, lim->nb_seg_max); + lim->nb_mtu_seg_max =RTE_MIN (vf_lim->nb_seg_max, lim->nb_seg_max); +} +/* * Merge the info from the VF and synthetic path. * use the default config of the VF
@@ -196,11 +207,13 @@ static int hn_vf_info_merge(struct rte_eth_dev *vf_dev, info->max_tx_queues); info->tx_offload_capa &= vf_info.tx_offload_capa; info->tx_queue_offload_capa &= vf_info.tx_queue_offload_capa; +hn_vf_merge_desc_lim (&info->tx_desc_lim, &vf_info.tx_desc_lim); info->min_rx_bufsize =RTE_MAX (vf_info.min_rx_bufsize, info->min_rx_bufsize); info->max_rx_pktlen =RTE_MAX (vf_info.max_rx_pktlen, info->max_rx_pktlen); +hn_vf_merge_desc_lim (&info->rx_desc_lim, &vf_info.rx_desc_lim);return 0 ; }
Signature · 1 line
2.20.1
PATCH v3 2/4 net/netvsc: handle receive packets during multi-channel setup
In reply to Stephen HemmingerIt is possible for a packet to arrive during the configuration process when setting up multiple queue mode. This would cause configure to fail; fix by just ignoring receive packets while waiting for control commands.
Use the receive ring lock to avoid possible races between oddly behaved applications doing rx_burst and control operations concurrently.
--- drivers/net/netvsc/hn_nvs.c | 41 +++++++++++++++++++++++++++++++++++-- 1 file changed, 39 insertions(+), 2 deletions(-)
drivers/net/netvsc/hn_nvs.c +39 -2
@@ -54,7 +54,7 @@ static int hn_nvs_req_send(struct hn_data *hv, }static int -hn_nvs_execute (struct hn_data *hv, +__hn_nvs_execute (struct hn_data *hv,void *req,uint32_t reqlen,void *resp,uint32_t resplen,uint32_t type)
@@ -62,6 +62,7 @@ hn_nvs_execute(struct hn_data *hv,struct vmbus_channel *chan =hn_primary_chan (hv);char buffer[NVS_RESPSIZE_MAX ];const struct hn_nvs_hdr *hdr; +uint64_t xactid;uint32_t len;int ret;
@@ -77,7 +78,7 @@ hn_nvs_execute(struct hn_data *hv, retry: len =sizeof (buffer); - ret =rte_vmbus_chan_recv (chan, buffer, &len,NULL ); + ret =rte_vmbus_chan_recv (chan, buffer, &len, &xactid);if (ret == -EAGAIN ) {rte_delay_us (HN_CHAN_INTERVAL_US ); goto retry;
@@ -88,7 +89,20 @@ hn_nvs_execute(struct hn_data *hv,return ret; } +if (len <sizeof (*hdr)) { +PMD_DRV_LOG (ERR ,"response missing NVS header" ); +return -EINVAL ; + } + hdr = (struct hn_nvs_hdr *)buffer; + +/* Silently drop received packets while waiting for response */ +if (hdr->type ==NVS_TYPE_RNDIS ) { +hn_nvs_ack_rxbuf (chan, xactid); + --hv->rxbuf_outstanding; + goto retry; + } +if (hdr->type != type) {PMD_DRV_LOG (ERR ,"unexpected NVS resp %#x, expect %#x" , hdr->type, type);
@@ -108,6 +122,29 @@ hn_nvs_execute(struct hn_data *hv,return 0 ; } + +/* +* Execute one control command and get the response. +* Only one command can be active on a channel at once +* Unlike BSD, DPDK does not have an interrupt context +* so the polling is required to wait for response. +*/ +static int +hn_nvs_execute (struct hn_data *hv, +void *req,uint32_t reqlen, +void *resp,uint32_t resplen, +uint32_t type) +{ +struct hn_rx_queue *rxq = hv->primary; +int ret; + +rte_spinlock_lock (&rxq->ring_lock); + ret =__hn_nvs_execute (hv, req, reqlen, resp, resplen, type); +rte_spinlock_unlock (&rxq->ring_lock); + +return ret; +} +static int hn_nvs_doinit (struct hn_data *hv,uint32_t nvs_ver) {
Signature · 1 line
2.20.1
PATCH v3 3/4 net/netvsc: split send buffers from transmit descriptors
In reply to Stephen HemmingerThe VMBus has reserved transmit area (per device) and transmit descriptors (per queue). The previous code was always having a 1:1 mapping between send buffers and descriptors. This can lead to one queue starving another and also buffer bloat.
Change to working more like FreeBSD where there is a pool of transmit descriptors per queue. If send buffer is not available then no aggregation happens but the queue can still drain.
--- drivers/net/netvsc/hn_ethdev.c | 9 +- drivers/net/netvsc/hn_rxtx.c | 261 ++++++++++++++++++++------------- drivers/net/netvsc/hn_var.h | 10 +- 3 files changed, 174 insertions(+), 106 deletions(-)
drivers/net/netvsc/hn_ethdev.c +6 -3
@@ -257,6 +257,9 @@ static int hn_dev_info_get(struct rte_eth_dev *dev, dev_info->max_rx_queues = hv->max_queues; dev_info->max_tx_queues = hv->max_queues; + dev_info->tx_desc_lim.nb_min =1 ; + dev_info->tx_desc_lim.nb_max =4096 ; +if (rte_eal_process_type () !=RTE_PROC_PRIMARY )return 0 ;
@@ -982,7 +985,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)if (err) goto failed; - err =hn_tx_pool_init (eth_dev); + err =hn_chim_init (eth_dev);if (err) goto failed;
@@ -1018,7 +1021,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev) failed:PMD_INIT_LOG (NOTICE ,"device init failed" ); -hn_tx_pool_uninit (eth_dev); +hn_chim_uninit (eth_dev);hn_detach (hv);return err; }
@@ -1042,7 +1045,7 @@ eth_hn_dev_uninit(struct rte_eth_dev *eth_dev) eth_dev->rx_pkt_burst =NULL ;hn_detach (hv); -hn_tx_pool_uninit (eth_dev); +hn_chim_uninit (eth_dev);rte_vmbus_chan_close (hv->primary->chan);rte_free (hv->primary); ret =rte_eth_dev_owner_delete (hv->owner.id);
drivers/net/netvsc/hn_rxtx.c +161 -100
411 lines not on this page · raw message
drivers/net/netvsc/hn_var.h +7 -3
@@ -52,6 +52,8 @@ struct hn_tx_queue {uint16_t port_id;uint16_t queue_id;uint32_t free_thresh; +struct rte_mempool *txdesc_pool; +void *tx_rndis;/* Applied packet transmission aggregation limits. */ uint32_t agg_szmax;
@@ -115,8 +117,10 @@ struct hn_data {uint16_t num_queues;uint64_t rss_offloads; +rte_spinlock_t chim_lock;struct rte_mem_resource *chim_res;/* UIO resource for Tx */ -struct rte_mempool *tx_pool;/* Tx descriptors */ +struct rte_bitmap *chim_bmap;/* Send buffer map */ +void *chim_bmem;uint32_t chim_szmax;/* Max size per buffer */ uint32_t chim_cnt;/* Max packets per buffer */
@@ -157,8 +161,8 @@ uint16_t hn_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,uint16_t hn_recv_pkts (void *rx_queue,struct rte_mbuf **rx_pkts,uint16_t nb_pkts); -int hn_tx_pool_init (struct rte_eth_dev *dev); -void hn_tx_pool_uninit (struct rte_eth_dev *dev); +int hn_chim_init (struct rte_eth_dev *dev); +void hn_chim_uninit (struct rte_eth_dev *dev);int hn_dev_link_update (struct rte_eth_dev *dev,int wait); int hn_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, uint16_t nb_desc, unsigned int socket_id,
Signature · 1 line
2.20.1
PATCH v3 4/4 net/netvsc: don't enable RSS if only single receive queue
In reply to Stephen HemmingerIf application has multiple transmit queues but only a single receive queue, then do not enable RSS.
--- drivers/net/netvsc/hn_ethdev.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)
drivers/net/netvsc/hn_ethdev.c +8 -5
@@ -597,11 +597,14 @@ static int hn_dev_configure(struct rte_eth_dev *dev)return err; } - err =hn_rndis_conf_rss (hv,0 ); -if (err) { -PMD_DRV_LOG (NOTICE , -"initial RSS config failed" ); -return err; +if ((dev_conf->rxmode.offloads &DEV_RX_OFFLOAD_RSS_HASH ) && + dev->data->nb_rx_queues >1 ) { + err =hn_rndis_conf_rss (hv,0 ); +if (err) { +PMD_DRV_LOG (NOTICE , +"initial RSS config failed" ); +return err; + } } }
Signature · 1 line
2.20.1
v4 cover net/netvsc: bug fixes
In reply to Stephen HemmingerThese are bug fixes for netvsc PMD mostly related to issues in the transmit ring discovered by users of multi-queue.
v4 - add more bug fixes
v3 - fix checkpatch complaints
v2 - simplify the locking on multi-channel setup
Stephen Hemminger (8):
net/netvsc: propagate descriptor limits from VF to netvsc net/netvsc: handle receive packets during multi-channel setup net/netvsc: split send buffers from transmit descriptors net/netvsc: fix invalid rte_free on dev_close net/netvsc: remove process event optimization net/netvsc: handle transmit completions based on burst size bus/vmbus: simplify args to need_signal
net/netvsc: avoid possible live lock
drivers/bus/vmbus/vmbus_bufring.c | 8 +- drivers/net/netvsc/hn_ethdev.c | 25 ++- drivers/net/netvsc/hn_nvs.c | 41 ++++- drivers/net/netvsc/hn_rxtx.c | 281 ++++++++++++++++++------------ drivers/net/netvsc/hn_var.h | 12 +- drivers/net/netvsc/hn_vf.c | 13 ++ 6 files changed, 248 insertions(+), 132 deletions(-)
Signature · 1 line
2.20.1
PATCH v4 1/8 net/netvsc: propagate descriptor limits from VF to netvsc
If application cares about descriptor limits, the netvsc device values should reflect those of the VF as well.
--- drivers/net/netvsc/hn_vf.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
drivers/net/netvsc/hn_vf.c +13 -0
@@ -167,6 +167,17 @@ hn_nvs_handle_vfassoc(struct rte_eth_dev *dev,hn_vf_remove (hv); } +static void +hn_vf_merge_desc_lim (struct rte_eth_desc_lim *lim, +const struct rte_eth_desc_lim *vf_lim) +{ + lim->nb_max =RTE_MIN (vf_lim->nb_max, lim->nb_max); + lim->nb_min =RTE_MAX (vf_lim->nb_min, lim->nb_min); + lim->nb_align =RTE_MAX (vf_lim->nb_align, lim->nb_align); + lim->nb_seg_max =RTE_MIN (vf_lim->nb_seg_max, lim->nb_seg_max); + lim->nb_mtu_seg_max =RTE_MIN (vf_lim->nb_seg_max, lim->nb_seg_max); +} +/* * Merge the info from the VF and synthetic path. * use the default config of the VF
@@ -196,11 +207,13 @@ static int hn_vf_info_merge(struct rte_eth_dev *vf_dev, info->max_tx_queues); info->tx_offload_capa &= vf_info.tx_offload_capa; info->tx_queue_offload_capa &= vf_info.tx_queue_offload_capa; +hn_vf_merge_desc_lim (&info->tx_desc_lim, &vf_info.tx_desc_lim); info->min_rx_bufsize =RTE_MAX (vf_info.min_rx_bufsize, info->min_rx_bufsize); info->max_rx_pktlen =RTE_MAX (vf_info.max_rx_pktlen, info->max_rx_pktlen); +hn_vf_merge_desc_lim (&info->rx_desc_lim, &vf_info.rx_desc_lim);return 0 ; }
Signature · 1 line
2.20.1
PATCH v4 2/8 net/netvsc: handle receive packets during multi-channel setup
In reply to Stephen HemmingerIt is possible for a packet to arrive during the configuration process when setting up multiple queue mode. This would cause configure to fail; fix by just ignoring receive packets while waiting for control commands.
Use the receive ring lock to avoid possible races between oddly behaved applications doing rx_burst and control operations concurrently.
--- drivers/net/netvsc/hn_nvs.c | 41 +++++++++++++++++++++++++++++++++++-- 1 file changed, 39 insertions(+), 2 deletions(-)
drivers/net/netvsc/hn_nvs.c +39 -2
@@ -54,7 +54,7 @@ static int hn_nvs_req_send(struct hn_data *hv, }static int -hn_nvs_execute (struct hn_data *hv, +__hn_nvs_execute (struct hn_data *hv,void *req,uint32_t reqlen,void *resp,uint32_t resplen,uint32_t type)
@@ -62,6 +62,7 @@ hn_nvs_execute(struct hn_data *hv,struct vmbus_channel *chan =hn_primary_chan (hv);char buffer[NVS_RESPSIZE_MAX ];const struct hn_nvs_hdr *hdr; +uint64_t xactid;uint32_t len;int ret;
@@ -77,7 +78,7 @@ hn_nvs_execute(struct hn_data *hv, retry: len =sizeof (buffer); - ret =rte_vmbus_chan_recv (chan, buffer, &len,NULL ); + ret =rte_vmbus_chan_recv (chan, buffer, &len, &xactid);if (ret == -EAGAIN ) {rte_delay_us (HN_CHAN_INTERVAL_US ); goto retry;
@@ -88,7 +89,20 @@ hn_nvs_execute(struct hn_data *hv,return ret; } +if (len <sizeof (*hdr)) { +PMD_DRV_LOG (ERR ,"response missing NVS header" ); +return -EINVAL ; + } + hdr = (struct hn_nvs_hdr *)buffer; + +/* Silently drop received packets while waiting for response */ +if (hdr->type ==NVS_TYPE_RNDIS ) { +hn_nvs_ack_rxbuf (chan, xactid); + --hv->rxbuf_outstanding; + goto retry; + } +if (hdr->type != type) {PMD_DRV_LOG (ERR ,"unexpected NVS resp %#x, expect %#x" , hdr->type, type);
@@ -108,6 +122,29 @@ hn_nvs_execute(struct hn_data *hv,return 0 ; } + +/* +* Execute one control command and get the response. +* Only one command can be active on a channel at once +* Unlike BSD, DPDK does not have an interrupt context +* so the polling is required to wait for response. +*/ +static int +hn_nvs_execute (struct hn_data *hv, +void *req,uint32_t reqlen, +void *resp,uint32_t resplen, +uint32_t type) +{ +struct hn_rx_queue *rxq = hv->primary; +int ret; + +rte_spinlock_lock (&rxq->ring_lock); + ret =__hn_nvs_execute (hv, req, reqlen, resp, resplen, type); +rte_spinlock_unlock (&rxq->ring_lock); + +return ret; +} +static int hn_nvs_doinit (struct hn_data *hv,uint32_t nvs_ver) {
Signature · 1 line
2.20.1
PATCH v4 3/8 net/netvsc: split send buffers from transmit descriptors
In reply to Stephen HemmingerThe VMBus has reserved transmit area (per device) and transmit descriptors (per queue). The previous code was always having a 1:1 mapping between send buffers and descriptors. This can lead to one queue starving another and also buffer bloat.
Change to working more like FreeBSD where there is a pool of transmit descriptors per queue. If send buffer is not available then no aggregation happens but the queue can still drain.
--- drivers/net/netvsc/hn_ethdev.c | 9 +- drivers/net/netvsc/hn_rxtx.c | 271 ++++++++++++++++++++------------- drivers/net/netvsc/hn_var.h | 10 +- 3 files changed, 180 insertions(+), 110 deletions(-)
drivers/net/netvsc/hn_ethdev.c +6 -3
@@ -257,6 +257,9 @@ static int hn_dev_info_get(struct rte_eth_dev *dev, dev_info->max_rx_queues = hv->max_queues; dev_info->max_tx_queues = hv->max_queues; + dev_info->tx_desc_lim.nb_min =1 ; + dev_info->tx_desc_lim.nb_max =4096 ; +if (rte_eal_process_type () !=RTE_PROC_PRIMARY )return 0 ;
@@ -982,7 +985,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)if (err) goto failed; - err =hn_tx_pool_init (eth_dev); + err =hn_chim_init (eth_dev);if (err) goto failed;
@@ -1018,7 +1021,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev) failed:PMD_INIT_LOG (NOTICE ,"device init failed" ); -hn_tx_pool_uninit (eth_dev); +hn_chim_uninit (eth_dev);hn_detach (hv);return err; }
@@ -1042,7 +1045,7 @@ eth_hn_dev_uninit(struct rte_eth_dev *eth_dev) eth_dev->rx_pkt_burst =NULL ;hn_detach (hv); -hn_tx_pool_uninit (eth_dev); +hn_chim_uninit (eth_dev);rte_vmbus_chan_close (hv->primary->chan);rte_free (hv->primary); ret =rte_eth_dev_owner_delete (hv->owner.id);
drivers/net/netvsc/hn_rxtx.c +167 -104
430 lines not on this page · raw message
drivers/net/netvsc/hn_var.h +7 -3
@@ -52,6 +52,8 @@ struct hn_tx_queue {uint16_t port_id;uint16_t queue_id;uint32_t free_thresh; +struct rte_mempool *txdesc_pool; +void *tx_rndis;/* Applied packet transmission aggregation limits. */ uint32_t agg_szmax;
@@ -115,8 +117,10 @@ struct hn_data {uint16_t num_queues;uint64_t rss_offloads; +rte_spinlock_t chim_lock;struct rte_mem_resource *chim_res;/* UIO resource for Tx */ -struct rte_mempool *tx_pool;/* Tx descriptors */ +struct rte_bitmap *chim_bmap;/* Send buffer map */ +void *chim_bmem;uint32_t chim_szmax;/* Max size per buffer */ uint32_t chim_cnt;/* Max packets per buffer */
@@ -157,8 +161,8 @@ uint16_t hn_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,uint16_t hn_recv_pkts (void *rx_queue,struct rte_mbuf **rx_pkts,uint16_t nb_pkts); -int hn_tx_pool_init (struct rte_eth_dev *dev); -void hn_tx_pool_uninit (struct rte_eth_dev *dev); +int hn_chim_init (struct rte_eth_dev *dev); +void hn_chim_uninit (struct rte_eth_dev *dev);int hn_dev_link_update (struct rte_eth_dev *dev,int wait); int hn_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, uint16_t nb_desc, unsigned int socket_id,
Signature · 1 line
2.20.1
PATCH v4 4/8 net/netvsc: fix invalid rte_free on dev_close
In reply to Stephen HemmingerThe netvsc PMD was putting the mac address in private data but the core rte_ethdev doesn't allow that it. It has to be in rte_malloc'd memory or the a message will be printed on shutdown/close.
EAL: Invalid memory
--- drivers/net/netvsc/hn_ethdev.c | 16 ++++++++++------ drivers/net/netvsc/hn_var.h | 2 -- 2 files changed, 10 insertions(+), 8 deletions(-)
drivers/net/netvsc/hn_ethdev.c +10 -6
@@ -134,8 +134,6 @@ eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)static void eth_dev_vmbus_release (struct rte_eth_dev *eth_dev) { -/* mac_addrs must not be freed alone because part of dev_private */ - eth_dev->data->mac_addrs =NULL ;/* free ether device */ rte_eth_dev_release_port (eth_dev);
@@ -937,9 +935,6 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev) eth_dev->tx_pkt_burst = &hn_xmit_pkts; eth_dev->rx_pkt_burst = &hn_recv_pkts; -/* Since Hyper-V only supports one MAC address, just use local data */ - eth_dev->data->mac_addrs = &hv->mac_addr; -/* * for secondary processes, we don't initialize any further as primary * has already done this work.
@@ -947,6 +942,15 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)if (rte_eal_process_type () !=RTE_PROC_PRIMARY )return 0 ; +/* Since Hyper-V only supports one MAC address */ + eth_dev->data->mac_addrs =rte_calloc ("hv_mac" ,HN_MAX_MAC_ADDRS , +sizeof (struct rte_ether_addr ),0 ); +if (eth_dev->data->mac_addrs ==NULL ) { +PMD_INIT_LOG (ERR , +"Failed to allocate memory store MAC addresses" ); +return -ENOMEM ; + } + hv->vmbus = vmbus; hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP ]; hv->chim_res = &vmbus->resource[HV_SEND_BUF_MAP ];
@@ -989,7 +993,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)if (err) goto failed; - err =hn_rndis_get_eaddr (hv, hv->mac_addr.addr_bytes); + err =hn_rndis_get_eaddr (hv, eth_dev->data->mac_addrs->addr_bytes);if (err) goto failed;
drivers/net/netvsc/hn_var.h +0 -2
@@ -139,8 +139,6 @@ struct hn_data {uint8_t rss_key[40 ];uint16_t rss_ind[128 ]; -struct rte_ether_addr mac_addr; -struct rte_eth_dev_owner owner;struct rte_intr_handle vf_intr;
Signature · 1 line
2.20.1
Re: [dpdk-dev] [dpdk-stable] [PATCH v4 4/8] net/netvsc: fix invalid rte_free on dev_close
Quoted 3 lines · Stephen Hemminger
The netvsc PMD was putting the mac address in private data but the core
rte_ethdevdoesn't allow that it. It has to be in rte_malloc'd memory or the a message will be printed on shutdown/close.
It doesn't have to be rte_malloc'ed. Setting 'eth_dev->data->mac_addrs' to null in 'hn_dev_close()' should fix it.
But dynamically allocating it also works if that is what preferred.
Quoted 72 lines
EAL: Invalid memory
Fixes f8279f47dd89 ("net/netvsc: fix crash in secondary process")Signed-off-by Stephen Hemminger <[email protected]>--- drivers/net/netvsc/hn_ethdev.c | 16 ++++++++++------ drivers/net/netvsc/hn_var.h | 2 -- 2 files changed, 10 insertions(+), 8 deletions(-)drivers/net/netvsc/hn_ethdev.c +10
-6@@ -134,8 +134,6 @@ eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)static void eth_dev_vmbus_release (struct rte_eth_dev *eth_dev) { -/* mac_addrs must not be freed alone because part of dev_private */ - eth_dev->data->mac_addrs =NULL ;/* free ether device */ rte_eth_dev_release_port (eth_dev);@@ -937,9 +935,6 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev) eth_dev->tx_pkt_burst = &hn_xmit_pkts; eth_dev->rx_pkt_burst = &hn_recv_pkts; -/* Since Hyper-V only supports one MAC address, just use local data */ - eth_dev->data->mac_addrs = &hv->mac_addr; -/* * for secondary processes, we don't initialize any further as primary * has already done this work. @@ -947,6 +942,15 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)if (rte_eal_process_type () !=RTE_PROC_PRIMARY )return 0 ; +/* Since Hyper-V only supports one MAC address */ + eth_dev->data->mac_addrs =rte_calloc ("hv_mac" ,HN_MAX_MAC_ADDRS , +sizeof (struct rte_ether_addr ),0 ); +if (eth_dev->data->mac_addrs ==NULL ) { +PMD_INIT_LOG (ERR , +"Failed to allocate memory store MAC addresses" ); +return -ENOMEM ; + } + hv->vmbus = vmbus; hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP ]; hv->chim_res = &vmbus->resource[HV_SEND_BUF_MAP ];@@ -989,7 +993,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)if (err) goto failed; - err =hn_rndis_get_eaddr (hv, hv->mac_addr.addr_bytes); + err =hn_rndis_get_eaddr (hv, eth_dev->data->mac_addrs->addr_bytes);if (err) goto failed;drivers/net/netvsc/hn_var.h +0
-2@@ -139,8 +139,6 @@ struct hn_data {uint8_t rss_key[40 ];uint16_t rss_ind[128 ]; -struct rte_ether_addr mac_addr; -struct rte_eth_dev_owner owner;struct rte_intr_handle vf_intr;
PATCH v4 5/8 net/netvsc: remove process event optimization
In reply to Stephen HemmingerRemove unlocked check for data in receive ring. This check is not safe because of missing barriers etc.
--- drivers/net/netvsc/hn_rxtx.c | 4 ---- 1 file changed, 4 deletions(-)
drivers/net/netvsc/hn_rxtx.c +0 -4
@@ -969,10 +969,6 @@ uint32_t hn_process_events(struct hn_data *hv, uint16_t queue_id, rxq = queue_id ==0 ? hv->primary : dev->data->rx_queues[queue_id]; -/* If no pending data then nothing to do */ -if (rte_vmbus_chan_rx_empty (rxq->chan)) -return 0 ; -/* * Since channel is shared between Rx and TX queue need to have a lock * since DPDK does not force same CPU to be used for Rx/Tx.
Signature · 1 line
2.20.1
PATCH v4 6/8 net/netvsc: handle transmit completions based on burst size
In reply to Stephen HemmingerIf tx_free_thresh is quite low, it is possible that we need to cleanup based on burst size.
--- drivers/net/netvsc/hn_rxtx.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/net/netvsc/hn_rxtx.c +3 -2
@@ -1375,7 +1375,7 @@ hn_xmit_pkts(void *ptxq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)struct hn_data *hv = txq->hv;struct rte_eth_dev *vf_dev;bool need_sig = false; -uint16_t nb_tx; +uint16_t nb_tx, avail;int ret;if (unlikely (hv->closed))
@@ -1390,7 +1390,8 @@ hn_xmit_pkts(void *ptxq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)return (*vf_dev->tx_pkt_burst)(sub_q, tx_pkts, nb_pkts); } -if (rte_mempool_avail_count (txq->txdesc_pool) <= txq->free_thresh) + avail =rte_mempool_avail_count (txq->txdesc_pool); +if (nb_pkts > avail || avail <= txq->free_thresh)hn_process_events (hv, txq->queue_id,0 );for (nb_tx =0 ; nb_tx < nb_pkts; nb_tx++) {
Signature · 1 line
2.20.1
PATCH v4 7/8 bus/vmbus: simplify args to need_signal
In reply to Stephen HemmingerThe transmit need signal function can avoid an unnecessary dereference by passing the right pointer. This also makes code better match FreeBSD driver.
--- drivers/bus/vmbus/vmbus_bufring.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
drivers/bus/vmbus/vmbus_bufring.c +4 -4
@@ -54,10 +54,10 @@ void vmbus_br_setup(struct vmbus_br *br, void *buf, unsigned int blen)* data have arrived. */ static inline bool -vmbus_txbr_need_signal (const struct vmbus_br *tbr,uint32_t old_windex) +vmbus_txbr_need_signal (const struct vmbus_bufring *vbr,uint32_t old_windex) {rte_smp_mb (); -if (tbr->vbr->imask) +if (vbr->imask)return false;rte_smp_rmb ();
@@ -66,7 +66,7 @@ vmbus_txbr_need_signal(const struct vmbus_br *tbr, uint32_t old_windex)* This is the only case we need to signal when the * ring transitions from being empty to non-empty. */ -return old_windex == tbr->vbr->rindex; +return old_windex == vbr->rindex; } static inline uint32_t
@@ -163,7 +163,7 @@ vmbus_txbr_write(struct vmbus_br *tbr, const struct iovec iov[], int iovlen,rte_pause ();/* If host had read all data before this, then need to signal */ - *need_sig |=vmbus_txbr_need_signal (tbr, old_windex); + *need_sig |=vmbus_txbr_need_signal (vbr, old_windex);return 0 ; }
Signature · 1 line
2.20.1
PATCH v4 8/8 net/netvsc: avoid possible live lock
In reply to Stephen HemmingerSince the ring buffer with host is shared for both transmit completions and receive packets, it is possible that transmitter could get starved if receive ring gets full.
Better to process all outstanding events which frees up transmit buffer slots, even if means dropping some packets.
--- drivers/net/netvsc/hn_rxtx.c | 3 --- 1 file changed, 3 deletions(-)
drivers/net/netvsc/hn_rxtx.c +0 -3
@@ -1032,9 +1032,6 @@ uint32_t hn_process_events(struct hn_data *hv, uint16_t queue_id,if (tx_limit && tx_done >= tx_limit)break ; - -if (rxq->rx_ring &&rte_ring_full (rxq->rx_ring)) -break ; }if (bytes_read >0 )
Signature · 1 line
2.20.1
Re: [dpdk-dev] [PATCH v4 0/8] net/netvsc: bug fixes
In reply to Stephen HemmingerQuoted 7 lines · Stephen Hemminger
These are bug fixes for netvsc PMD mostly related to issues in the transmit ring discovered by users of multi-queue.
v4 - add more bug fixes
v3 - fix checkpatch complaints
v2 - simplify the locking on multi-channel setup
Quoted 9 lines
Stephen Hemminger (8):
net/netvsc: propagate descriptor limits from VF to netvsc net/netvsc: handle receive packets during multi-channel setup net/netvsc: split send buffers from transmit descriptors net/netvsc: fix invalidrte_freeondev_closenet/netvsc: remove process event optimization net/netvsc: handle transmit completions based on burst size bus/vmbus: simplify args toneed_signal
net/netvsc: avoid possible live lock
Series applied to dpdk-next-net/master, thanks.