DPDK patches and discussions
 help / color / mirror / Atom feed
* [dpdk-dev] [PATCH] The VMXNET3 PMD can't receive packet suddenly after a lot of traffic coming in
@ 2015-07-23  8:51 Marco Lee
  2015-07-23 16:49 ` Stephen Hemminger
  0 siblings, 1 reply; 6+ messages in thread
From: Marco Lee @ 2015-07-23  8:51 UTC (permalink / raw)
  To: dev; +Cc: Marco Lee

The RX of VMXNET3 PMD will have deadlock when a lot of traffic coming in.
The root cause is due to mbuf allocation fail in vmxnet3_post_rx_bufs() and there is no error handling when it is called
from vmxnet3_recv_pkts(). The RXD will not have "free" mbuf for it but the counter still increment.
Finally, no packet can be received.

This fix is allocate the mbuf first, if the allocation is failed, then reuse the old mbuf
If the allocation is sucess, the vmxnet3_post_rx_bufs() will call vmxnet3_renew_desc()
and RXD will be renew inside.

Signed-off-by: Marco Lee <mac_leehk@yahoo.com.hk/marco.lee@ruckuswireless.com>
---
 drivers/net/vmxnet3/vmxnet3_rxtx.c |   36 +++++++++++++++++++++++++++++++++++-
 1 file changed, 35 insertions(+), 1 deletion(-)

diff --git a/drivers/net/vmxnet3/vmxnet3_rxtx.c b/drivers/net/vmxnet3/vmxnet3_rxtx.c
index 39ad6ef..3a15009 100644
--- a/drivers/net/vmxnet3/vmxnet3_rxtx.c
+++ b/drivers/net/vmxnet3/vmxnet3_rxtx.c
@@ -421,6 +421,34 @@ vmxnet3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	return nb_tx;
 }
 
+static inline void
+vmxnet3_renew_desc(vmxnet3_rx_queue_t *rxq, uint8_t ring_id,struct rte_mbuf *mbuf)
+{
+	uint32_t  val = 0;
+	struct vmxnet3_cmd_ring *ring = &rxq->cmd_ring[ring_id];
+
+	struct Vmxnet3_RxDesc *rxd;
+	vmxnet3_buf_info_t *buf_info = &ring->buf_info[ring->next2fill];
+
+	rxd = (struct Vmxnet3_RxDesc *)(ring->base + ring->next2fill);
+
+	if (ring->rid == 0) {
+		val = VMXNET3_RXD_BTYPE_HEAD;
+	} else {
+		val = VMXNET3_RXD_BTYPE_BODY;
+	}
+
+	buf_info->m = mbuf;
+	buf_info->len = (uint16_t)(mbuf->buf_len - RTE_PKTMBUF_HEADROOM);
+	buf_info->bufPA = RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf);
+
+	rxd->addr = buf_info->bufPA;
+	rxd->btype = val;
+	rxd->len = buf_info->len;
+	rxd->gen = ring->gen;
+
+	vmxnet3_cmd_ring_adv_next2fill(ring);
+}
 /*
  *  Allocates mbufs and clusters. Post rx descriptors with buffer details
  *  so that device can receive packets in those buffers.
@@ -578,6 +606,8 @@ vmxnet3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 		if (nb_rx >= nb_pkts)
 			break;
 
+		struct rte_mbuf *rep;
+		rep = rte_rxmbuf_alloc(rxq->mb_pool);
 		idx = rcd->rxdIdx;
 		ring_idx = (uint8_t)((rcd->rqID == rxq->qid1) ? 0 : 1);
 		rxd = (Vmxnet3_RxDesc *)rxq->cmd_ring[ring_idx].base + idx;
@@ -651,13 +681,17 @@ vmxnet3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
 		vmxnet3_rx_offload(rcd, rxm);
 
+		if (unlikely(rep == NULL)) {
+			rep = rxm;
+			goto rcd_done;
+		}
 		rx_pkts[nb_rx++] = rxm;
 rcd_done:
 		rxq->cmd_ring[ring_idx].next2comp = idx;
 		VMXNET3_INC_RING_IDX_ONLY(rxq->cmd_ring[ring_idx].next2comp, rxq->cmd_ring[ring_idx].size);
 
 		/* It's time to allocate some new buf and renew descriptors */
-		vmxnet3_post_rx_bufs(rxq, ring_idx);
+		vmxnet3_renew_desc(rxq, ring_idx,rep);
 		if (unlikely(rxq->shared->ctrl.updateRxProd)) {
 			VMXNET3_WRITE_BAR0_REG(hw, rxprod_reg[ring_idx] + (rxq->queue_id * VMXNET3_REG_ALIGN),
 					       rxq->cmd_ring[ring_idx].next2fill);
-- 
1.7.9.5

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [dpdk-dev] [PATCH] The VMXNET3 PMD can't receive packet suddenly after a lot of traffic coming in
  2015-07-23  8:51 [dpdk-dev] [PATCH] The VMXNET3 PMD can't receive packet suddenly after a lot of traffic coming in Marco Lee
@ 2015-07-23 16:49 ` Stephen Hemminger
  0 siblings, 0 replies; 6+ messages in thread
From: Stephen Hemminger @ 2015-07-23 16:49 UTC (permalink / raw)
  To: Marco Lee; +Cc: dev

On Thu, 23 Jul 2015 16:51:32 +0800
Marco Lee <mac_leehk@yahoo.com.hk> wrote:

> The RX of VMXNET3 PMD will have deadlock when a lot of traffic coming in.
> The root cause is due to mbuf allocation fail in vmxnet3_post_rx_bufs() and there is no error handling when it is called
> from vmxnet3_recv_pkts(). The RXD will not have "free" mbuf for it but the counter still increment.
> Finally, no packet can be received.
> 
> This fix is allocate the mbuf first, if the allocation is failed, then reuse the old mbuf
> If the allocation is sucess, the vmxnet3_post_rx_bufs() will call vmxnet3_renew_desc()
> and RXD will be renew inside.
> 
> Signed-off-by: Marco Lee <mac_leehk@yahoo.com.hk/marco.lee@ruckuswireless.com>

Much better. I ran this patch through checkpatch and the following
things should be fixed.

WARNING: Possible unwrapped commit description (prefer a maximum 75 chars per line)
#69: 
The root cause is due to mbuf allocation fail in vmxnet3_post_rx_bufs() and there is no error handling when it is called

WARNING: 'sucess' may be misspelled - perhaps 'success'?
#74: 
If the allocation is sucess, the vmxnet3_post_rx_bufs() will call vmxnet3_renew_desc()

WARNING: line over 80 characters
#91: FILE: drivers/net/vmxnet3/vmxnet3_rxtx.c:425:
+vmxnet3_renew_desc(vmxnet3_rx_queue_t *rxq, uint8_t ring_id,struct rte_mbuf *mbuf)

ERROR: space required after that ',' (ctx:VxV)
#91: FILE: drivers/net/vmxnet3/vmxnet3_rxtx.c:425:
+vmxnet3_renew_desc(vmxnet3_rx_queue_t *rxq, uint8_t ring_id,struct rte_mbuf *mbuf)
                                                            ^
WARNING: braces {} are not necessary for any arm of this statement
#101: FILE: drivers/net/vmxnet3/vmxnet3_rxtx.c:435:
+	if (ring->rid == 0) {
[...]
+	} else {
[...]

ERROR: space required after that ',' (ctx:VxV)
#145: FILE: drivers/net/vmxnet3/vmxnet3_rxtx.c:694:
+		vmxnet3_renew_desc(rxq, ring_idx,rep);
 		                                ^

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [dpdk-dev] [PATCH] The VMXNET3 PMD can't receive packet suddenly after a lot of traffic coming in
  2015-07-24  0:53 Marco Lee
  2015-07-24  7:58 ` Thomas Monjalon
@ 2015-12-22 21:36 ` Yong Wang
  1 sibling, 0 replies; 6+ messages in thread
From: Yong Wang @ 2015-12-22 21:36 UTC (permalink / raw)
  To: Marco Lee, dev

On 7/23/15, 5:53 PM, "dev on behalf of Marco Lee" <dev-bounces@dpdk.org on behalf of mac_leehk@yahoo.com.hk> wrote:


>The RX of VMXNET3 PMD will have deadlock when a lot of traffic coming in.
>The root cause is due to mbuf allocation fail in vmxnet3_post_rx_bufs()
>and there is no error handling when it is called from vmxnet3_recv_pkts().
>The RXD will not have "free" mbuf for it but the counter still increment.

Can you describe what counter this refers to?

>Finally, no packet can be received.
>
>This fix is allocate the mbuf first, if the allocation is failed,
>then reuse the old mbuf. If the allocation is success,
>the vmxnet3_post_rx_bufs() will call vmxnet3_renew_desc()
>and RXD will be renew inside.

I didn’t see this part of logic implemented.

>
>Signed-off-by: Marco Lee <mac_leehk@yahoo.com.hk/marco.lee@ruckuswireless.com>
>---
> drivers/net/vmxnet3/vmxnet3_rxtx.c |   37 +++++++++++++++++++++++++++++++++++-
> 1 file changed, 36 insertions(+), 1 deletion(-)
>
>diff --git a/drivers/net/vmxnet3/vmxnet3_rxtx.c b/drivers/net/vmxnet3/vmxnet3_rxtx.c
>index 39ad6ef..cbed438 100644
>--- a/drivers/net/vmxnet3/vmxnet3_rxtx.c
>+++ b/drivers/net/vmxnet3/vmxnet3_rxtx.c
>@@ -421,6 +421,35 @@ vmxnet3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
> 	return nb_tx;
> }
> 
>+static inline void
>+vmxnet3_renew_desc(vmxnet3_rx_queue_t *rxq, uint8_t ring_id,
>+		struct rte_mbuf *mbuf)
>+{
>+	uint32_t  val = 0;
>+	struct vmxnet3_cmd_ring *ring = &rxq->cmd_ring[ring_id];
>+

Remove this blank line.

>+	struct Vmxnet3_RxDesc *rxd;
>+	vmxnet3_buf_info_t *buf_info = &ring->buf_info[ring->next2fill];
>+
>+	rxd = (struct Vmxnet3_RxDesc *)(ring->base + ring->next2fill);
>+

nit: this can be merged with the above definition.

>+	if (ring->rid == 0)
>+		val = VMXNET3_RXD_BTYPE_HEAD;
>+	else
>+		val = VMXNET3_RXD_BTYPE_BODY;
>+	

Remove the trailing space here.
 
>+
>+	buf_info->m = mbuf;
>+	buf_info->len = (uint16_t)(mbuf->buf_len - RTE_PKTMBUF_HEADROOM);
>+	buf_info->bufPA = RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf);
>+
>+	rxd->addr = buf_info->bufPA;
>+	rxd->btype = val;
>+	rxd->len = buf_info->len;
>+	rxd->gen = ring->gen;
>+
>+	vmxnet3_cmd_ring_adv_next2fill(ring);
>+}
> /*
>  *  Allocates mbufs and clusters. Post rx descriptors with buffer details
>  *  so that device can receive packets in those buffers.
>@@ -578,6 +607,8 @@ vmxnet3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
> 		if (nb_rx >= nb_pkts)
> 			break;
> 
>+		struct rte_mbuf *rep;

What does rep mean? Can you rename it to something easier to understand
(say newm, or m2)?

Also, please move the definition to the top of this block.

>+		rep = rte_rxmbuf_alloc(rxq->mb_pool);
> 		idx = rcd->rxdIdx;
> 		ring_idx = (uint8_t)((rcd->rqID == rxq->qid1) ? 0 : 1);
> 		rxd = (Vmxnet3_RxDesc *)rxq->cmd_ring[ring_idx].base + idx;
>@@ -651,13 +682,17 @@ vmxnet3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
> 
> 		vmxnet3_rx_offload(rcd, rxm);
> 
>+		if (unlikely(rep == NULL)) {
>+			rep = rxm;
>+			goto rcd_done;
>+		}

Should this be moved earlier? Also need to update the rx_buf_alloc_failure
counter.

> 		rx_pkts[nb_rx++] = rxm;
> rcd_done:
> 		rxq->cmd_ring[ring_idx].next2comp = idx;
> 		VMXNET3_INC_RING_IDX_ONLY(rxq->cmd_ring[ring_idx].next2comp, rxq->cmd_ring[ring_idx].size);
> 
> 		/* It's time to allocate some new buf and renew descriptors */
>-		vmxnet3_post_rx_bufs(rxq, ring_idx);
>+		vmxnet3_renew_desc(rxq, ring_idx, rep);
> 		if (unlikely(rxq->shared->ctrl.updateRxProd)) {
> 			VMXNET3_WRITE_BAR0_REG(hw, rxprod_reg[ring_idx] + (rxq->queue_id * VMXNET3_REG_ALIGN),
> 					       rxq->cmd_ring[ring_idx].next2fill);
>-- 
>1.7.9.5
>

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [dpdk-dev] [PATCH] The VMXNET3 PMD can't receive packet suddenly after a lot of traffic coming in
  2015-07-24  7:58 ` Thomas Monjalon
@ 2015-10-21 16:39   ` Thomas Monjalon
  0 siblings, 0 replies; 6+ messages in thread
From: Thomas Monjalon @ 2015-10-21 16:39 UTC (permalink / raw)
  To: Marco Lee, Yong Wang; +Cc: dev

Any update?
Yong, have you looked at it?

2015-07-24 09:58, Thomas Monjalon:
> Hi Marco,
> 
> It seems it is a new version of the patch.
> Please, in order to ease patch management and review, could you add:
> 	- a version number
> 	- a changelog
> 	- threading with previous patches
> It can be done with:
> 	- subject-prefix option
> 	- annotate option
> 	- in-reply-to option
> as explained here:
> 	http://dpdk.org/dev#send
> 
> Thanks

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [dpdk-dev] [PATCH] The VMXNET3 PMD can't receive packet suddenly after a lot of traffic coming in
  2015-07-24  0:53 Marco Lee
@ 2015-07-24  7:58 ` Thomas Monjalon
  2015-10-21 16:39   ` Thomas Monjalon
  2015-12-22 21:36 ` Yong Wang
  1 sibling, 1 reply; 6+ messages in thread
From: Thomas Monjalon @ 2015-07-24  7:58 UTC (permalink / raw)
  To: Marco Lee; +Cc: dev

Hi Marco,

It seems it is a new version of the patch.
Please, in order to ease patch management and review, could you add:
	- a version number
	- a changelog
	- threading with previous patches
It can be done with:
	- subject-prefix option
	- annotate option
	- in-reply-to option
as explained here:
	http://dpdk.org/dev#send

Thanks

^ permalink raw reply	[flat|nested] 6+ messages in thread

* [dpdk-dev] [PATCH] The VMXNET3 PMD can't receive packet suddenly after a lot of traffic coming in
@ 2015-07-24  0:53 Marco Lee
  2015-07-24  7:58 ` Thomas Monjalon
  2015-12-22 21:36 ` Yong Wang
  0 siblings, 2 replies; 6+ messages in thread
From: Marco Lee @ 2015-07-24  0:53 UTC (permalink / raw)
  To: dev; +Cc: Marco Lee

The RX of VMXNET3 PMD will have deadlock when a lot of traffic coming in.
The root cause is due to mbuf allocation fail in vmxnet3_post_rx_bufs()
and there is no error handling when it is called from vmxnet3_recv_pkts().
The RXD will not have "free" mbuf for it but the counter still increment.
Finally, no packet can be received.

This fix is allocate the mbuf first, if the allocation is failed,
then reuse the old mbuf. If the allocation is success,
the vmxnet3_post_rx_bufs() will call vmxnet3_renew_desc()
and RXD will be renew inside.

Signed-off-by: Marco Lee <mac_leehk@yahoo.com.hk/marco.lee@ruckuswireless.com>
---
 drivers/net/vmxnet3/vmxnet3_rxtx.c |   37 +++++++++++++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)

diff --git a/drivers/net/vmxnet3/vmxnet3_rxtx.c b/drivers/net/vmxnet3/vmxnet3_rxtx.c
index 39ad6ef..cbed438 100644
--- a/drivers/net/vmxnet3/vmxnet3_rxtx.c
+++ b/drivers/net/vmxnet3/vmxnet3_rxtx.c
@@ -421,6 +421,35 @@ vmxnet3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	return nb_tx;
 }
 
+static inline void
+vmxnet3_renew_desc(vmxnet3_rx_queue_t *rxq, uint8_t ring_id,
+		struct rte_mbuf *mbuf)
+{
+	uint32_t  val = 0;
+	struct vmxnet3_cmd_ring *ring = &rxq->cmd_ring[ring_id];
+
+	struct Vmxnet3_RxDesc *rxd;
+	vmxnet3_buf_info_t *buf_info = &ring->buf_info[ring->next2fill];
+
+	rxd = (struct Vmxnet3_RxDesc *)(ring->base + ring->next2fill);
+
+	if (ring->rid == 0)
+		val = VMXNET3_RXD_BTYPE_HEAD;
+	else
+		val = VMXNET3_RXD_BTYPE_BODY;
+	
+
+	buf_info->m = mbuf;
+	buf_info->len = (uint16_t)(mbuf->buf_len - RTE_PKTMBUF_HEADROOM);
+	buf_info->bufPA = RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf);
+
+	rxd->addr = buf_info->bufPA;
+	rxd->btype = val;
+	rxd->len = buf_info->len;
+	rxd->gen = ring->gen;
+
+	vmxnet3_cmd_ring_adv_next2fill(ring);
+}
 /*
  *  Allocates mbufs and clusters. Post rx descriptors with buffer details
  *  so that device can receive packets in those buffers.
@@ -578,6 +607,8 @@ vmxnet3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 		if (nb_rx >= nb_pkts)
 			break;
 
+		struct rte_mbuf *rep;
+		rep = rte_rxmbuf_alloc(rxq->mb_pool);
 		idx = rcd->rxdIdx;
 		ring_idx = (uint8_t)((rcd->rqID == rxq->qid1) ? 0 : 1);
 		rxd = (Vmxnet3_RxDesc *)rxq->cmd_ring[ring_idx].base + idx;
@@ -651,13 +682,17 @@ vmxnet3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
 		vmxnet3_rx_offload(rcd, rxm);
 
+		if (unlikely(rep == NULL)) {
+			rep = rxm;
+			goto rcd_done;
+		}
 		rx_pkts[nb_rx++] = rxm;
 rcd_done:
 		rxq->cmd_ring[ring_idx].next2comp = idx;
 		VMXNET3_INC_RING_IDX_ONLY(rxq->cmd_ring[ring_idx].next2comp, rxq->cmd_ring[ring_idx].size);
 
 		/* It's time to allocate some new buf and renew descriptors */
-		vmxnet3_post_rx_bufs(rxq, ring_idx);
+		vmxnet3_renew_desc(rxq, ring_idx, rep);
 		if (unlikely(rxq->shared->ctrl.updateRxProd)) {
 			VMXNET3_WRITE_BAR0_REG(hw, rxprod_reg[ring_idx] + (rxq->queue_id * VMXNET3_REG_ALIGN),
 					       rxq->cmd_ring[ring_idx].next2fill);
-- 
1.7.9.5

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2015-12-22 21:36 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-07-23  8:51 [dpdk-dev] [PATCH] The VMXNET3 PMD can't receive packet suddenly after a lot of traffic coming in Marco Lee
2015-07-23 16:49 ` Stephen Hemminger
2015-07-24  0:53 Marco Lee
2015-07-24  7:58 ` Thomas Monjalon
2015-10-21 16:39   ` Thomas Monjalon
2015-12-22 21:36 ` Yong Wang

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).