[dpdk-dev] [PATCH v2] ixgbe: fix ixgbe access endian issue on bigendian arch

Ananyev, Konstantin konstantin.ananyev at intel.com
Fri Mar 27 12:37:34 CET 2015



> 
> Hi,
> 
> Please see my comments inline.
> 
> Thanks,
> Xuelin
> 
> > -----Original Message-----
> > From: Ananyev, Konstantin [mailto:konstantin.ananyev at intel.com]
> > Sent: Friday, March 27, 2015 09:04
> > To: Shi Xuelin-B29237; thomas.monjalon at 6wind.com
> > Cc: dev at dpdk.org
> > Subject: RE: [dpdk-dev] [PATCH v2] ixgbe: fix ixgbe access endian issue
> > on bigendian arch
> >
> > Hi,
> >
> > > -----Original Message-----
> > > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of
> > > xuelin.shi at freescale.com
> > > Sent: Tuesday, March 24, 2015 6:34 AM
> > > To: thomas.monjalon at 6wind.com
> > > Cc: dev at dpdk.org; Xuelin Shi
> > > Subject: [dpdk-dev] [PATCH v2] ixgbe: fix ixgbe access endian issue on
> > > bigendian arch
> > >
> > > From: Xuelin Shi <xuelin.shi at freescale.com>
> > >
> > > enforce rules of the cpu and ixgbe exchange data.
> > > 1. cpu use data owned by ixgbe must use rte_le_to_cpu_xx(...) 2. cpu
> > > fill data to ixgbe must use rte_cpu_to_le_xx(...)
> > >
> > > Signed-off-by: Xuelin Shi <xuelin.shi at freescale.com>
> > > ---
> > > changes for v2:
> > >   rebased on latest head.
> > >   fix some style issue detected by checkpatch.pl
> > >
> > >  lib/librte_pmd_ixgbe/ixgbe_rxtx.c | 104
> > > ++++++++++++++++++++++++--------------
> > >  1 file changed, 66 insertions(+), 38 deletions(-)
> > >
> > > diff --git a/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> > > b/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> > > index 9da2c7e..961ac08 100644
> > > --- a/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> > > +++ b/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> > > @@ -128,7 +128,7 @@ ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
> > >  	int i;
> > >
> > >  	/* check DD bit on threshold descriptor */
> > > -	status = txq->tx_ring[txq->tx_next_dd].wb.status;
> > > +	status = rte_le_to_cpu_32(txq->tx_ring[txq->tx_next_dd].wb.status);
> > >  	if (! (status & IXGBE_ADVTXD_STAT_DD))
> > >  		return 0;
> > >
> > > @@ -174,11 +174,14 @@ tx4(volatile union ixgbe_adv_tx_desc *txdp,
> > struct rte_mbuf **pkts)
> > >  		pkt_len = (*pkts)->data_len;
> > >
> > >  		/* write data to descriptor */
> > > -		txdp->read.buffer_addr = buf_dma_addr;
> > > +		txdp->read.buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
> > > +
> > >  		txdp->read.cmd_type_len =
> > > -				((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
> > > +			rte_cpu_to_le_32((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
> > > +
> > >  		txdp->read.olinfo_status =
> > > -				(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
> > > +			rte_cpu_to_le_32(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
> > > +
> > >  		rte_prefetch0(&(*pkts)->pool);
> > >  	}
> > >  }
> > > @@ -194,11 +197,14 @@ tx1(volatile union ixgbe_adv_tx_desc *txdp,
> > struct rte_mbuf **pkts)
> > >  	pkt_len = (*pkts)->data_len;
> > >
> > >  	/* write data to descriptor */
> > > -	txdp->read.buffer_addr = buf_dma_addr;
> > > +	txdp->read.buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
> > > +
> > >  	txdp->read.cmd_type_len =
> > > -			((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
> > > +			rte_cpu_to_le_32((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
> > > +
> > >  	txdp->read.olinfo_status =
> > > -			(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
> > > +			rte_cpu_to_le_32(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
> > > +
> > >  	rte_prefetch0(&(*pkts)->pool);
> > >  }
> > >
> > > @@ -285,7 +291,7 @@ tx_xmit_pkts(void *tx_queue, struct rte_mbuf
> > **tx_pkts,
> > >  		 * a divisor of the ring size
> > >  		 */
> > >  		tx_r[txq->tx_next_rs].read.cmd_type_len |=
> > > -			rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
> > > +				rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
> > >  		txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
> > >
> > >  		txq->tx_tail = 0;
> > > @@ -304,7 +310,7 @@ tx_xmit_pkts(void *tx_queue, struct rte_mbuf
> > **tx_pkts,
> > >  	 */
> > >  	if (txq->tx_tail > txq->tx_next_rs) {
> > >  		tx_r[txq->tx_next_rs].read.cmd_type_len |=
> > > -			rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
> > > +				rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
> > >  		txq->tx_next_rs = (uint16_t)(txq->tx_next_rs +
> > >  						txq->tx_rs_thresh);
> > >  		if (txq->tx_next_rs >= txq->nb_tx_desc) @@ -505,6 +511,7 @@
> > > ixgbe_xmit_cleanup(struct ixgbe_tx_queue *txq)
> > >  	uint16_t nb_tx_desc = txq->nb_tx_desc;
> > >  	uint16_t desc_to_clean_to;
> > >  	uint16_t nb_tx_to_clean;
> > > +	uint32_t stat;
> > >
> > >  	/* Determine the last descriptor needing to be cleaned */
> > >  	desc_to_clean_to = (uint16_t)(last_desc_cleaned +
> > > txq->tx_rs_thresh); @@ -513,7 +520,9 @@ ixgbe_xmit_cleanup(struct
> > > ixgbe_tx_queue *txq)
> > >
> > >  	/* Check to make sure the last descriptor to clean is done */
> > >  	desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
> > > -	if (! (txr[desc_to_clean_to].wb.status & IXGBE_TXD_STAT_DD))
> > > +
> > > +	stat = rte_le_to_cpu_32(txr[desc_to_clean_to].wb.status);
> > > +	if (!(stat & IXGBE_TXD_STAT_DD))
> > >  	{
> > >  		PMD_TX_FREE_LOG(DEBUG,
> > >  				"TX descriptor %4u is not done"
> > > @@ -544,7 +553,7 @@ ixgbe_xmit_cleanup(struct ixgbe_tx_queue *txq)
> > >  	 * up to the last descriptor with the RS bit set
> > >  	 * are done. Only reset the threshold descriptor.
> > >  	 */
> > > -	txr[desc_to_clean_to].wb.status = 0;
> > > +	txr[desc_to_clean_to].wb.status = rte_cpu_to_le_32(0);
> >
> > Don't think you need swap bytes here.
> 
> Yes, it's to accentuate the endian rule. I'm expecting compiler will optimizing it.

It probably would, though still don't see any reason for doing it.

> 
> >
> > >
> > >  	/* Update the txq to reflect the last descriptor that was cleaned
> > */
> > >  	txq->last_desc_cleaned = desc_to_clean_to; @@ -801,12 +810,14 @@
> > > ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
> > >  			 */
> > >  			slen = m_seg->data_len;
> > >  			buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
> > > +
> > >  			txd->read.buffer_addr =
> > > -				rte_cpu_to_le_64(buf_dma_addr);
> > > +					rte_cpu_to_le_64(buf_dma_addr);
> > >  			txd->read.cmd_type_len =
> > > -				rte_cpu_to_le_32(cmd_type_len | slen);
> > > +					rte_cpu_to_le_32(cmd_type_len | slen);
> > >  			txd->read.olinfo_status =
> > > -				rte_cpu_to_le_32(olinfo_status);
> > > +					rte_cpu_to_le_32(olinfo_status);
> > > +
> > >  			txe->last_id = tx_last;
> > >  			tx_id = txe->next_id;
> > >  			txe = txn;
> > > @@ -946,14 +957,16 @@ ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
> > >  	uint64_t pkt_flags;
> > >  	int s[LOOK_AHEAD], nb_dd;
> > >  	int i, j, nb_rx = 0;
> > > +	uint32_t stat;
> > >
> > >
> > >  	/* get references to current descriptor and S/W ring entry */
> > >  	rxdp = &rxq->rx_ring[rxq->rx_tail];
> > >  	rxep = &rxq->sw_ring[rxq->rx_tail];
> > >
> > > +	stat = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
> > >  	/* check to make sure there is at least 1 packet to receive */
> > > -	if (! (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD))
> > > +	if (!(stat & IXGBE_RXDADV_STAT_DD))
> > >  		return 0;
> > >
> > >  	/*
> > > @@ -965,7 +978,7 @@ ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
> > >  	{
> > >  		/* Read desc statuses backwards to avoid race condition */
> > >  		for (j = LOOK_AHEAD-1; j >= 0; --j)
> > > -			s[j] = rxdp[j].wb.upper.status_error;
> > > +			s[j] = rte_le_to_cpu_32(rxdp[j].wb.upper.status_error);
> > >
> > >  		/* Compute how many status bits were set */
> > >  		nb_dd = 0;
> > > @@ -976,28 +989,36 @@ ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue
> > > *rxq)
> > >
> > >  		/* Translate descriptor info to mbuf format */
> > >  		for (j = 0; j < nb_dd; ++j) {
> > > +			uint16_t tmp16;
> > > +			uint32_t tmp32;
> > > +
> > >  			mb = rxep[j].mbuf;
> > > -			pkt_len = (uint16_t)(rxdp[j].wb.upper.length - rxq-
> > >crc_len);
> > > +			tmp16 = rte_le_to_cpu_16(rxdp[j].wb.upper.length);
> > > +			pkt_len = tmp16 - rxq->crc_len;
> >
> > Here and in other places, wonder why do you introduce temporary variables?
> > Why not just:
> > pkt_len = rte_le_to_cpu_16( (rxdp[j].wb.upper.length) - rxq->crc_len; ?
> 
> Comply with 80 char width limit style


Ok, but what prevents you from:
 pkt_len = rte_le_to_cpu_16( (rxdp[j].wb.upper.length) -
	rxq->crc_len;

?
BTW, Nice to see that people try to follow 80-char rule :)

> 
> >
> > >  			mb->data_len = pkt_len;
> > >  			mb->pkt_len = pkt_len;
> > > -			mb->vlan_tci = rxdp[j].wb.upper.vlan;
> > >  			mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
> > >
> > >  			/* convert descriptor fields to rte mbuf flags */
> > > -			pkt_flags  = rx_desc_hlen_type_rss_to_pkt_flags(
> > > +			tmp32 = rte_le_to_cpu_32(
> > >  					rxdp[j].wb.lower.lo_dword.data);
> >
> >
> > Same here, why just not:
> > pkt_flags  =
> > rx_desc_hlen_type_rss_to_pkt_flags(rte_le_to_cpu_32(rxdp[j].wb.lower.lo_d
> > word.data));
> > ?
> >
> > > +			pkt_flags  = rx_desc_hlen_type_rss_to_pkt_flags(tmp32);
> > > +
> > >  			/* reuse status field from scan list */
> > >  			pkt_flags |= rx_desc_status_to_pkt_flags(s[j]);
> > >  			pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
> > >  			mb->ol_flags = pkt_flags;
> > >
> > >  			if (likely(pkt_flags & PKT_RX_RSS_HASH))
> > > -				mb->hash.rss = rxdp[j].wb.lower.hi_dword.rss;
> > > +				mb->hash.rss = rte_le_to_cpu_32(
> > > +						rxdp[j].wb.lower.hi_dword.rss);
> > >  			else if (pkt_flags & PKT_RX_FDIR) {
> > > -				mb->hash.fdir.hash =
> > > -
> > 	(uint16_t)((rxdp[j].wb.lower.hi_dword.csum_ip.csum)
> > > -						& IXGBE_ATR_HASH_MASK);
> > > -				mb->hash.fdir.id =
> > rxdp[j].wb.lower.hi_dword.csum_ip.ip_id;
> > > +				tmp16 = rxdp[j].wb.lower.hi_dword.csum_ip.csum;
> > > +				mb->hash.fdir.hash = rte_le_to_cpu_16(
> > > +						tmp16 & IXGBE_ATR_HASH_MASK);
> >
> > Shouldn't it be:
> > rte_le_to_cpu_16(tmp16) & IXGBE_ATR_HASH_MASK; ?
> 
> Because the value of IXGBE_ATR_HASH_MASK is assumed as little endian.
Didn't get you here, IXGBE_ATR_HASH_MASK is just a constant value:

lib/librte_pmd_ixgbe/ixgbe/ixgbe_type.h:3059:#define IXGBE_ATR_HASH_MASK       0x7fff

How it could be 'assumed as little endian'?

> 
> >
> > > +
> > > +				tmp16 = rxdp[j].wb.lower.hi_dword.csum_ip.ip_id;
> > > +				mb->hash.fdir.id = rte_le_to_cpu_16(tmp16);
> > >  			}
> > >  		}
> > >
> > > @@ -1051,8 +1072,8 @@ ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq)
> > >
> > >  		/* populate the descriptors */
> > >  		dma_addr =
> > rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
> > > -		rxdp[i].read.hdr_addr = dma_addr;
> > > -		rxdp[i].read.pkt_addr = dma_addr;
> > > +		rxdp[i].read.hdr_addr = rte_cpu_to_le_64(dma_addr);
> > > +		rxdp[i].read.pkt_addr = rte_cpu_to_le_64(dma_addr);
> >
> > No need these 2 swaps here.
> > dma_addr already converted to LE, just a line above.
> 
> OK, these two should be removed.
> 
> >
> > >  	}
> > >
> > >  	/* update tail pointer */
> > > @@ -1220,8 +1241,8 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf
> > **rx_pkts,
> > >  		 * using invalid descriptor fields when read from rxd.
> > >  		 */
> > >  		rxdp = &rx_ring[rx_id];
> > > -		staterr = rxdp->wb.upper.status_error;
> > > -		if (! (staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
> > > +		staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
> > > +		if (!(staterr & IXGBE_RXDADV_STAT_DD))
> >
> > Why do you need that change here?
> > Original code seems also ok to me.
> 
> I think it is more clear. We read some value from PCIe and change it to cpu format.
> I accentuate the rules of read/write between cpu and PCIe.

It is probably a bit clearer, but from other side it seems a bit slower.
As you pointed above, rte_constant_bswap32(constant_value) will most likely be optimized by the compiler,
so no real byte swap would happen here.  

> 
> >
> >
> > >  			break;
> > >  		rxd = *rxdp;
> > >
> > > @@ -1325,12 +1346,17 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf
> > **rx_pkts,
> > >  		rxm->ol_flags = pkt_flags;
> > >
> > >  		if (likely(pkt_flags & PKT_RX_RSS_HASH))
> > > -			rxm->hash.rss = rxd.wb.lower.hi_dword.rss;
> > > +			rxm->hash.rss = rte_le_to_cpu_32(
> > > +						rxd.wb.lower.hi_dword.rss);
> > >  		else if (pkt_flags & PKT_RX_FDIR) {
> > > -			rxm->hash.fdir.hash =
> > > -				(uint16_t)((rxd.wb.lower.hi_dword.csum_ip.csum)
> > > -					   & IXGBE_ATR_HASH_MASK);
> > > -			rxm->hash.fdir.id = rxd.wb.lower.hi_dword.csum_ip.ip_id;
> > > +			uint16_t tmp16;
> > > +
> > > +			tmp16 = rxd.wb.lower.hi_dword.csum_ip.csum;
> > > +			rxm->hash.fdir.hash = rte_le_to_cpu_16(
> > > +						tmp16 & IXGBE_ATR_HASH_MASK);
> >
> >
> > Same thing here, seems should be: rte_le_to_cpu_16(tmp16) &
> > IXGBE_ATR_HASH_MASK;
> >
> > > +
> > > +			rxm->hash.fdir.id = rte_le_to_cpu_16(
> > > +					rxd.wb.lower.hi_dword.csum_ip.ip_id);
> > >  		}
> > >  		/*
> > >  		 * Store the mbuf address into the next entry of the array @@
> > > -1412,8 +1438,8 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct
> > rte_mbuf **rx_pkts,
> > >  		 * using invalid descriptor fields when read from rxd.
> > >  		 */
> > >  		rxdp = &rx_ring[rx_id];
> > > -		staterr = rxdp->wb.upper.status_error;
> > > -		if (! (staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
> > > +		staterr = rte_cpu_to_le_32(rxdp->wb.upper.status_error);
> > > +		if (!(staterr & IXGBE_RXDADV_STAT_DD))
> >
> >
> > Again, seems unnecessary.
> >
> > >  			break;
> > >  		rxd = *rxdp;
> > >
> > > @@ -1742,7 +1768,7 @@ ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
> > >  	prev = (uint16_t) (txq->nb_tx_desc - 1);
> > >  	for (i = 0; i < txq->nb_tx_desc; i++) {
> > >  		volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
> > > -		txd->wb.status = IXGBE_TXD_STAT_DD;
> > > +		txd->wb.status = rte_cpu_to_le_32(IXGBE_TXD_STAT_DD);
> > >  		txe[i].mbuf = NULL;
> > >  		txe[i].last_id = i;
> > >  		txe[prev].next_id = i;
> > > @@ -2293,7 +2319,8 @@ ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev,
> > uint16_t rx_queue_id)
> > >  	rxdp = &(rxq->rx_ring[rxq->rx_tail]);
> > >
> > >  	while ((desc < rxq->nb_rx_desc) &&
> > > -		(rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD)) {
> > > +		(rte_le_to_cpu_32(rxdp->wb.upper.status_error) &
> > > +				  IXGBE_RXDADV_STAT_DD)) {
> > >  		desc += IXGBE_RXQ_SCAN_INTERVAL;
> > >  		rxdp += IXGBE_RXQ_SCAN_INTERVAL;
> > >  		if (rxq->rx_tail + desc >= rxq->nb_rx_desc) @@ -2318,7
> > +2345,8 @@
> > > ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
> > >  		desc -= rxq->nb_rx_desc;
> > >
> > >  	rxdp = &rxq->rx_ring[desc];
> > > -	return !!(rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD);
> > > +	return !!(rte_le_to_cpu_32(rxdp->wb.upper.status_error) &
> > > +			IXGBE_RXDADV_STAT_DD);
> > >  }
> > >
> > >  void
> > > --
> > > 1.9.1



More information about the dev mailing list