Re: [dpdk-dev] [PATCH] app/testpmd: fix random number of Tx segments

["Li, Xiaoyun" <xiaoyun.li@intel.com>]

> -----Original Message-----
> From: Zhang, AlvinX <alvinx.zhang@intel.com>
> Sent: Tuesday, September 7, 2021 10:25
> To: Li, Xiaoyun <xiaoyun.li@intel.com>; Ananyev, Konstantin
> <konstantin.ananyev@intel.com>
> Cc: dev@dpdk.org; stable@dpdk.org
> Subject: RE: [PATCH] app/testpmd: fix random number of Tx segments
> 
> > -----Original Message-----
> > From: Li, Xiaoyun <xiaoyun.li@intel.com>
> > Sent: Monday, September 6, 2021 6:55 PM
> > To: Zhang, AlvinX <alvinx.zhang@intel.com>; Ananyev, Konstantin
> > <konstantin.ananyev@intel.com>
> > Cc: dev@dpdk.org; stable@dpdk.org
> > Subject: RE: [PATCH] app/testpmd: fix random number of Tx segments
> >
> >
> >
> > > -----Original Message-----
> > > From: Zhang, AlvinX <alvinx.zhang@intel.com>
> > > Sent: Monday, September 6, 2021 18:04
> > > To: Li, Xiaoyun <xiaoyun.li@intel.com>; Ananyev, Konstantin
> > > <konstantin.ananyev@intel.com>
> > > Cc: dev@dpdk.org; stable@dpdk.org
> > > Subject: RE: [PATCH] app/testpmd: fix random number of Tx segments
> > >
> > > > -----Original Message-----
> > > > From: Li, Xiaoyun <xiaoyun.li@intel.com>
> > > > Sent: Monday, September 6, 2021 4:59 PM
> > > > To: Zhang, AlvinX <alvinx.zhang@intel.com>; Ananyev, Konstantin
> > > > <konstantin.ananyev@intel.com>
> > > > Cc: dev@dpdk.org; stable@dpdk.org
> > > > Subject: RE: [PATCH] app/testpmd: fix random number of Tx segments
> > > >
> > > > Hi
> > > >
> > > > > -----Original Message-----
> > > > > From: Zhang, AlvinX <alvinx.zhang@intel.com>
> > > > > Sent: Thursday, September 2, 2021 16:20
> > > > > To: Li, Xiaoyun <xiaoyun.li@intel.com>; Ananyev, Konstantin
> > > > > <konstantin.ananyev@intel.com>
> > > > > Cc: dev@dpdk.org; Zhang, AlvinX <alvinx.zhang@intel.com>;
> > > > > stable@dpdk.org
> > > > > Subject: [PATCH] app/testpmd: fix random number of Tx segments
> > > > >
> > > > > When random number of segments in Tx packets is enabled, the
> > > > > total data space length of all segments must be greater or equal
> > > > > than the size of an Eth/IP/UDP/timestamp packet, that's total 14
> > > > > + 20 +
> > > > > 8 +
> > > > > 16 bytes. Otherwise the Tx engine may cause the application to crash.
> > > > >
> > > > > Bugzilla ID: 797
> > > > > Fixes: 79bec05b32b7 ("app/testpmd: add ability to split outgoing
> > > > > packets")
> > > > > Cc: stable@dpdk.org
> > > > >
> > > > > Signed-off-by: Alvin Zhang <alvinx.zhang@intel.com>
> > > > > ---
> > > > >  app/test-pmd/config.c  | 16 +++++++++++-----
> > > > > app/test-pmd/testpmd.c
> > > > > |  5
> > > > > +++++  app/test-pmd/testpmd.h |  5 +++++  app/test-pmd/txonly.c
> > > > > +++++ |
> > > > > +++++ 7
> > > > > +++++ +++++--
> > > > >  4 files changed, 26 insertions(+), 7 deletions(-)
> > > > >
> > > > > diff --git a/app/test-pmd/config.c b/app/test-pmd/config.c index
> > > > > 31d8ba1..5105b3b 100644
> > > > > --- a/app/test-pmd/config.c
> > > > > +++ b/app/test-pmd/config.c
> > > > > @@ -3837,10 +3837,11 @@ struct igb_ring_desc_16_bytes {
> > > > >  	 * Check that each segment length is greater or equal than
> > > > >  	 * the mbuf data size.
> > > > >  	 * Check also that the total packet length is greater or equal than the
> > > > > -	 * size of an empty UDP/IP packet (sizeof(struct rte_ether_hdr) +
> > > > > -	 * 20 + 8).
> > > > > +	 * size of an Eth/IP/UDP + timestamp packet
> > > > > +	 * (sizeof(struct rte_ether_hdr) + 20 + 8 + 16).
> > > >
> > > > I don't really agree on this. Most of the time, txonly generate
> > > > packets with Eth/IP/UDP. It's not fair to limit the hdr length to
> > > > include
> > > timestamp in all cases.
> > > > And to be honest, I don't see why you need to add
> > > > "tx_pkt_nb_min_segs". It's only used in txonly when
> > > > "TX_PKT_SPLIT_RND". So this issue is because when
> > > > "TX_PKT_SPLIT_RND", the
> > > random nb_segs is not enough for the hdr.
> > > >
> > > > But if you read txonly carefully, if "TX_PKT_SPLIT_RND", the first
> > > > segment length should be equal or greater than 42 (14+20+8).
> > > > Because when "TX_PKT_SPLIT_RND", update_pkt_header() should be
> > > > called. And that function doesn't deal with header in multi-segments.
> > > > I think there's bug here.
> > > >
> > > > So I think you should only add a check in pkt_burst_prepare() in txonly().
> > > > 	if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND) ||
> > > > txonly_multi_flow)
> > > > +                           if (tx_pkt_seg_lengths[0] < 42) {
> > > > +			err_log;
> > > > +			return false;
> > > > +		}
> > > > 		update_pkt_header(pkt, pkt_len);
> 
> 
> As above, If user have below configuration, there will no one packet be sent out
> and have lots and lots of repeated annoying logs.
> testpmd> set fwd txonly
> Set txonly packet forwarding mode
> testpmd> set txpkts 40,64
> testpmd> set txsplit rand
> testpmd> start
> txonly packet forwarding - ports=1 - cores=1 - streams=4 - NUMA support
> enabled, MP allocation mode: native Logical Core 2 (socket 0) forwards packets
> on 4 streams:
>   RX P=0/Q=0 (socket 0) -> TX P=0/Q=0 (socket 0) peer=02:00:00:00:00:00
>   RX P=0/Q=1 (socket 0) -> TX P=0/Q=1 (socket 0) peer=02:00:00:00:00:00
>   RX P=0/Q=2 (socket 0) -> TX P=0/Q=2 (socket 0) peer=02:00:00:00:00:00
>   RX P=0/Q=3 (socket 0) -> TX P=0/Q=3 (socket 0) peer=02:00:00:00:00:00
> 
>   txonly packet forwarding packets/burst=32
>   packet len=104 - nb packet segments=2
>   nb forwarding cores=1 - nb forwarding ports=1
>   port 0: RX queue number: 4 Tx queue number: 4
>     Rx offloads=0x0 Tx offloads=0x10000
>     RX queue: 0
>       RX desc=1024 - RX free threshold=32
>       RX threshold registers: pthresh=0 hthresh=0  wthresh=0
>       RX Offloads=0x0
>     TX queue: 0
>       TX desc=1024 - TX free threshold=32
>       TX threshold registers: pthresh=32 hthresh=0  wthresh=0
>       TX offloads=0x10000 - TX RS bit threshold=32 tx_pkt_seg_lengths[0] must
> bigger than 41 bytes tetx_pkt_seg_lengths[0] must bigger than 41 bytes
> stpmd> tx_pkt_seg_lengths[0] must bigger than 41 bytes
> tx_pkt_seg_lengths[0] must bigger than 41 bytes tx_pkt_seg_lengths[0] must
> bigger than 41 bytes tx_pkt_seg_lengths[0] must bigger than 41 bytes
> tx_pkt_seg_lengths[0] must bigger than 41 bytes tx_pkt_seg_lengths[0] must
> bigger than 41 bytes tx_pkt_seg_lengths[0] must bigger than 41 bytes
> tx_pkt_seg_lengths[0] must bigger than 41 bytes tx_pkt_seg_lengths[0] must
> bigger than 41 bytes tx_pkt_seg_lengths[0] must bigger than 41 bytes
> tx_pkt_seg_lengths[0] must bigger than 41 bytes tx_pkt_seg_lengths[0] must
> bigger than 41 bytes tx_pkt_seg_lengths[0] must bigger than 41 bytes
> tx_pkt_seg_lengths[0] must bigger than 41 bytes tx_pkt_seg_lengths[0] must
> bigger than 41 bytes tx_pkt_seg_lengths[0] must bigger than 41 bytes
> tx_pkt_seg_lengths[0] must bigger than 41 bytes tx_pkt_seg_lengths[0] must
> bigger than 41 bytes tx_pkt_seg_lengths[0] must bigger than 41 bytes
> tx_pkt_seg_lengths[0] must bigger than 41 bytes tx_pkt_seg_lengths[0] must
> bigger than 41 bytes tx_pkt_seg_lengths[0] must bigger than 41 bytes
> tx_pkt_seg_lengths[0] must bigger than 41 bytes ...
> ...
> testpmd> stop
> Telling cores to stop...
> Waiting for lcores to finish...
> 
>   ---------------------- Forward statistics for port 0  ----------------------
>   RX-packets: 0              RX-dropped: 0             RX-total: 0
>   TX-packets: 0              TX-dropped: 0             TX-total: 0
>   ----------------------------------------------------------------------------
> 
>   +++++++++++++++ Accumulated forward statistics for all
> ports+++++++++++++++
>   RX-packets: 0              RX-dropped: 0             RX-total: 0
>   TX-packets: 0              TX-dropped: 0             TX-total: 0
> 
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> +++++++++++
> 
> By the way, if multi flow was enabled, the ip header also will be updated, so we
> should put the check before below codes.
> 
> 	if (txonly_multi_flow) {
> 		uint8_t  ip_var = RTE_PER_LCORE(_ip_var);
> 		struct rte_ipv4_hdr *ip_hdr;
> 		uint32_t addr;
> 
> 		ip_hdr = rte_pktmbuf_mtod_offset(pkt,
> 				struct rte_ipv4_hdr *,
> 				sizeof(struct rte_ether_hdr));
> 		/*
> 		 * Generate multiple flows by varying IP src addr. This
> 		 * enables packets are well distributed by RSS in
> 		 * receiver side if any and txonly mode can be a decent
> 		 * packet generator for developer's quick performance
> 		 * regression test.
> 		 */
> 		addr = (tx_ip_dst_addr | (ip_var++ << 8)) + rte_lcore_id();
> 		ip_hdr->src_addr = rte_cpu_to_be_32(addr);
> 		RTE_PER_LCORE(_ip_var) = ip_var;
> 	}
> 

Yes. No matter where you put the check in txonly, there'll be annoying print.
Tricky.
I can only think of 2 solutions:
1. If needs to update hdr (ip or udp) and first seg len < 42, copy pkt_ip_hdr and pkt_udp_hdr to a temp value, update the temp value and use copy_buf_to_pkt() to copy the tmp hdr to segs.
And other cases keep the original code in case perf drop. Also, This one needs to keep your min_segs design.

The shortcoming is it's quite complicated change.

2. Force the first segment length is over 42 bytes in set_tx_pkt_segments()

This way is very simple but the shortcoming is the limit usage and you probably need to add a notice in doc to tell users this change.

> > >
> > > Yes, I didn't notice the updating for the UDP header, but the bug
> > > first occurs in this function copy_buf_to_pkt(&pkt_udp_hdr,
> > sizeof(pkt_udp_hdr), pkt,
> > > 			sizeof(struct rte_ether_hdr) +
> > > 			sizeof(struct rte_ipv4_hdr));
> > > not in update_pkt_header.
> > >
> > > Here we expecting users should set minimum 42 byte for first segment
> > > seems ok, But I think we putting the check in configuring the data
> > > space length of first segment is more graceful.
> >
> > No. You didn't get my point. It's not graceful at all.
> > The segment fault will only happen when "TX_PKT_SPLIT_RND". Because
> > the hdr may take 2 segs while random nb_segs is only 1.
> > But if it's not "TX_PKT_SPLIT_RND", the current set_tx_pkt_segments()
> > already make sure pkt_len is enough for 42.
> >
> > So the only case you need to deal with is "TX_PKT_SPLIT_RND". And
> > since "TX_PKT_SPLIT_RND" actually needs the first segment to be enough
> > to contain
> > 42 bytes.
> > And in cmd_set_txpkts_parsed, you may not get the configuration to
> > know if it will be configured as "TX_PKT_SPLIT_RND". That's why you
> > should check before update_pkt_header().
> >
> > In this way, when it's NOT "TX_PKT_SPLIT_RND", it will maintain the
> > old behavior that hdrs may cross several segs which makes more sense.
> >
> > >
> > > >
> > > > As for timestamp, maybe refer to "pkt_copy_split" in csumonly.c is
> > > > better? Copy the extra to the last segment if it's not enough. Not
> > > > sure how to deal with this issue better.
> > > >
> > > > >  	 */
> > > > >  	tx_pkt_len = 0;
> > > > > +	tx_pkt_nb_min_segs = 0;
> > > > >  	for (i = 0; i < nb_segs; i++) {
> > > > >  		if (seg_lengths[i] > mbuf_data_size[0]) {
> > > > >  			fprintf(stderr,
> > > > > @@ -3849,11 +3850,16 @@ struct igb_ring_desc_16_bytes {
> > > > >  			return;
> > > > >  		}
> > > > >  		tx_pkt_len = (uint16_t)(tx_pkt_len + seg_lengths[i]);
> > > > > +
> > > > > +		if (!tx_pkt_nb_min_segs &&
> > > > > +		    tx_pkt_len >= (sizeof(struct rte_ether_hdr) + 20 + 8 +
> 16))
> > > > > +			tx_pkt_nb_min_segs = i + 1;
> > > > >  	}
> > > > > -	if (tx_pkt_len < (sizeof(struct rte_ether_hdr) + 20 + 8)) {
> > > > > +
> > > > > +	if (!tx_pkt_nb_min_segs) {
> > > > >  		fprintf(stderr, "total packet length=%u < %d - give up\n",
> > > > > -				(unsigned) tx_pkt_len,
> > > > > -				(int)(sizeof(struct rte_ether_hdr) + 20 + 8));
> > > > > +			(unsigned int) tx_pkt_len,
> > > > > +			(int)(sizeof(struct rte_ether_hdr) + 20 + 8 + 16));
> > > > >  		return;
> > > > >  	}
> > > > >
> > > > > diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> > > > > index
> > > > > 6cbe9ba..c496e59 100644
> > > > > --- a/app/test-pmd/testpmd.c
> > > > > +++ b/app/test-pmd/testpmd.c
> > > > > @@ -232,6 +232,11 @@ struct fwd_engine * fwd_engines[] = {  };
> > > > > uint8_t tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY
> > > > > packets */
> > > > >
> > > > > +/**< Minimum number of segments in TXONLY packets to
> > > > > +accommodate all packet
> > > > > + * headers.
> > > > > + */
> > > > > +uint8_t  tx_pkt_nb_min_segs = 1;
> > > > > +
> > > > >  enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;  /**< Split
> > > > > policy for packets to TX. */
> > > > >
> > > > > diff --git a/app/test-pmd/testpmd.h b/app/test-pmd/testpmd.h
> > > > > index
> > > > > 16a3598..f5bc427 100644
> > > > > --- a/app/test-pmd/testpmd.h
> > > > > +++ b/app/test-pmd/testpmd.h
> > > > > @@ -464,6 +464,11 @@ enum dcb_mode_enable  extern uint16_t
> > > > > tx_pkt_length; /**< Length of TXONLY packet */  extern uint16_t
> > > > > tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT]; /**< Seg. lengths */
> > > > > extern uint8_t  tx_pkt_nb_segs; /**< Number of segments in TX
> > > > > packets */
> > > > > +
> > > > > +/**< Minimum number of segments in TXONLY packets to
> > > > > +accommodate all packet
> > > > > + * headers.
> > > > > + */
> > > > > +extern uint8_t  tx_pkt_nb_min_segs;
> > > > >  extern uint32_t tx_pkt_times_intra;  extern uint32_t
> > > > > tx_pkt_times_inter;
> > > > >
> > > > > diff --git a/app/test-pmd/txonly.c b/app/test-pmd/txonly.c index
> > > > > aed820f..27e4458 100644
> > > > > --- a/app/test-pmd/txonly.c
> > > > > +++ b/app/test-pmd/txonly.c
> > > > > @@ -195,8 +195,11 @@
> > > > >  	uint32_t nb_segs, pkt_len;
> > > > >  	uint8_t i;
> > > > >
> > > > > -	if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND))
> > > > > -		nb_segs = rte_rand() % tx_pkt_nb_segs + 1;
> > > > > +	if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND) &&
> > > > > +	    tx_pkt_nb_segs > tx_pkt_nb_min_segs)
> > > > > +		nb_segs = rte_rand() %
> > > > > +			  (tx_pkt_nb_segs - tx_pkt_nb_min_segs + 1) +
> > > > > +			  tx_pkt_nb_min_segs;
> > > > >  	else
> > > > >  		nb_segs = tx_pkt_nb_segs;
> > > > >
> > > > > --
> > > > > 1.8.3.1