diff mbox series

[v1,3/4] net/ixgbe: implement good checksum flag for NEON vector

Message ID 20201118062929.21345-4-feifei.wang2@arm.com (mailing list archive)
State Superseded, archived
Headers show
Series Enable Checksum Offloading for NEON vector | expand

Checks

Context Check Description
ci/checkpatch warning coding style issues

Commit Message

Feifei Wang Nov. 18, 2020, 6:29 a.m. UTC
From: feiwan02 <feifei.wang2@arm.com>

Add CKSUM_GOOD flag to distinguish a good checksum from an unknown one
in neon vertor RX function.

Test Results:
NICs: 82599(igb)
Dirver: ixgbe(vector)
Packet: IPv4_checksum = correct value && UDP_checksum = correct value

$:./app/dpdk-testpmd -c 0x3 -w 0002:f9:00.0 -- -i --port-topology=chained
test-pmd> set fwd rxonly
test-pmd> set verbose 1
test-pmd> start

With this patch:
testpmd> port 0/queue 0: received 1 packets
src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 - length=70 - nb_segs=1
ol_flags: PKT_RX_L4_CKSUM_GOOD PKT_RX_IP_CKSUM_GOOD

Without this patch:
testpmd> port 0/queue 0: received 1 packets
src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 - length=70 - nb_segs=1
ol_flags: PKT_RX_L4_CKSUM_UNKNOWN PKT_RX_IP_CKSUM_UNKNOWN

Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
---
 drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c | 37 +++++++++++++++++--------
 1 file changed, 25 insertions(+), 12 deletions(-)

Comments

Wang, Haiyue Nov. 18, 2020, 7:21 a.m. UTC | #1
> -----Original Message-----
> From: Feifei Wang <feifei.wang2@arm.com>
> Sent: Wednesday, November 18, 2020 14:29
> To: Jerin Jacob <jerinj@marvell.com>; Ruifeng Wang <ruifeng.wang@arm.com>; Guo, Jia
> <jia.guo@intel.com>; Wang, Haiyue <haiyue.wang@intel.com>
> Cc: dev@dpdk.org; nd@arm.com; feiwan02 <feifei.wang2@arm.com>
> Subject: [PATCH v1 3/4] net/ixgbe: implement good checksum flag for NEON vector
> 
> From: feiwan02 <feifei.wang2@arm.com>
> 
> Add CKSUM_GOOD flag to distinguish a good checksum from an unknown one
> in neon vertor RX function.
> 
> Test Results:
> NICs: 82599(igb)
> Dirver: ixgbe(vector)
> Packet: IPv4_checksum = correct value && UDP_checksum = correct value
> 
> $:./app/dpdk-testpmd -c 0x3 -w 0002:f9:00.0 -- -i --port-topology=chained
> test-pmd> set fwd rxonly
> test-pmd> set verbose 1
> test-pmd> start
> 
> With this patch:
> testpmd> port 0/queue 0: received 1 packets
> src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 - length=70 - nb_segs=1
> ol_flags: PKT_RX_L4_CKSUM_GOOD PKT_RX_IP_CKSUM_GOOD
> 
> Without this patch:
> testpmd> port 0/queue 0: received 1 packets
> src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 - length=70 - nb_segs=1
> ol_flags: PKT_RX_L4_CKSUM_UNKNOWN PKT_RX_IP_CKSUM_UNKNOWN
> 

How about to merge "1 & 2" and "3 & 4"? Then you will have the above test result. ;-)

> Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> ---
> 2.17.1
Feifei Wang Nov. 18, 2020, 7:55 a.m. UTC | #2
Hi, Haiyue

> -----邮件原件-----
> 发件人: Wang, Haiyue <haiyue.wang@intel.com>
> 发送时间: 2020年11月18日 15:22
> 收件人: Feifei Wang <Feifei.Wang2@arm.com>; jerinj@marvell.com; Ruifeng
> Wang <Ruifeng.Wang@arm.com>; Guo, Jia <jia.guo@intel.com>
> 抄送: dev@dpdk.org; nd <nd@arm.com>
> 主题: RE: [PATCH v1 3/4] net/ixgbe: implement good checksum flag for NEON
> vector
> 
> > -----Original Message-----
> > From: Feifei Wang <feifei.wang2@arm.com>
> > Sent: Wednesday, November 18, 2020 14:29
> > To: Jerin Jacob <jerinj@marvell.com>; Ruifeng Wang
> > <ruifeng.wang@arm.com>; Guo, Jia <jia.guo@intel.com>; Wang, Haiyue
> > <haiyue.wang@intel.com>
> > Cc: dev@dpdk.org; nd@arm.com; feiwan02 <feifei.wang2@arm.com>
> > Subject: [PATCH v1 3/4] net/ixgbe: implement good checksum flag for
> > NEON vector
> >
> > From: feiwan02 <feifei.wang2@arm.com>
> >
> > Add CKSUM_GOOD flag to distinguish a good checksum from an unknown
> one
> > in neon vertor RX function.
> >
> > Test Results:
> > NICs: 82599(igb)
> > Dirver: ixgbe(vector)
> > Packet: IPv4_checksum = correct value && UDP_checksum = correct value
> >
> > $:./app/dpdk-testpmd -c 0x3 -w 0002:f9:00.0 -- -i
> > --port-topology=chained
> > test-pmd> set fwd rxonly
> > test-pmd> set verbose 1
> > test-pmd> start
> >
> > With this patch:
> > testpmd> port 0/queue 0: received 1 packets
> > src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 -
> > length=70 - nb_segs=1
> > ol_flags: PKT_RX_L4_CKSUM_GOOD PKT_RX_IP_CKSUM_GOOD
> >
> > Without this patch:
> > testpmd> port 0/queue 0: received 1 packets
> > src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 -
> > length=70 - nb_segs=1
> > ol_flags: PKT_RX_L4_CKSUM_UNKNOWN PKT_RX_IP_CKSUM_UNKNOWN
> >
> 
> How about to merge "1 & 2" and "3 & 4"? Then you will have the above test
> result. ;-)
Do you mean that combine patch 1 and patch 2 into one patch to support vlan stripping and bad checksum flag, 
Then also do the same work for patch 3 and patch 4 to support good checksum flag and enable IXGBE neon vector when DEV_RX_OFFLOAD_CHECKSUM is set?

Actually, the test results are achieved without patch 4, due to that testpmd does not check rxmode->offloads, it will call IXGBE NEON vector function directly. And this change
is for the default config of l3fwd example.

Best Regards
Feifei
> 
> > Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
> > Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> > ---
> > 2.17.1
Wang, Haiyue Nov. 18, 2020, 8:15 a.m. UTC | #3
> -----Original Message-----
> From: Feifei Wang <Feifei.Wang2@arm.com>
> Sent: Wednesday, November 18, 2020 15:56
> To: Wang, Haiyue <haiyue.wang@intel.com>; jerinj@marvell.com; Ruifeng Wang <Ruifeng.Wang@arm.com>; Guo,
> Jia <jia.guo@intel.com>
> Cc: dev@dpdk.org; nd <nd@arm.com>; nd <nd@arm.com>
> Subject: 回复: [PATCH v1 3/4] net/ixgbe: implement good checksum flag for NEON vector
> 
> Hi, Haiyue
> 
> > -----邮件原件-----
> > 发件人: Wang, Haiyue <haiyue.wang@intel.com>
> > 发送时间: 2020年11月18日 15:22
> > 收件人: Feifei Wang <Feifei.Wang2@arm.com>; jerinj@marvell.com; Ruifeng
> > Wang <Ruifeng.Wang@arm.com>; Guo, Jia <jia.guo@intel.com>
> > 抄送: dev@dpdk.org; nd <nd@arm.com>
> > 主题: RE: [PATCH v1 3/4] net/ixgbe: implement good checksum flag for NEON
> > vector
> >
> > > -----Original Message-----
> > > From: Feifei Wang <feifei.wang2@arm.com>
> > > Sent: Wednesday, November 18, 2020 14:29
> > > To: Jerin Jacob <jerinj@marvell.com>; Ruifeng Wang
> > > <ruifeng.wang@arm.com>; Guo, Jia <jia.guo@intel.com>; Wang, Haiyue
> > > <haiyue.wang@intel.com>
> > > Cc: dev@dpdk.org; nd@arm.com; feiwan02 <feifei.wang2@arm.com>
> > > Subject: [PATCH v1 3/4] net/ixgbe: implement good checksum flag for
> > > NEON vector
> > >
> > > From: feiwan02 <feifei.wang2@arm.com>
> > >
> > > Add CKSUM_GOOD flag to distinguish a good checksum from an unknown
> > one
> > > in neon vertor RX function.
> > >
> > > Test Results:
> > > NICs: 82599(igb)
> > > Dirver: ixgbe(vector)
> > > Packet: IPv4_checksum = correct value && UDP_checksum = correct value
> > >
> > > $:./app/dpdk-testpmd -c 0x3 -w 0002:f9:00.0 -- -i
> > > --port-topology=chained
> > > test-pmd> set fwd rxonly
> > > test-pmd> set verbose 1
> > > test-pmd> start
> > >
> > > With this patch:
> > > testpmd> port 0/queue 0: received 1 packets
> > > src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 -
> > > length=70 - nb_segs=1
> > > ol_flags: PKT_RX_L4_CKSUM_GOOD PKT_RX_IP_CKSUM_GOOD
> > >
> > > Without this patch:
> > > testpmd> port 0/queue 0: received 1 packets
> > > src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 -
> > > length=70 - nb_segs=1
> > > ol_flags: PKT_RX_L4_CKSUM_UNKNOWN PKT_RX_IP_CKSUM_UNKNOWN
> > >
> >
> > How about to merge "1 & 2" and "3 & 4"? Then you will have the above test
> > result. ;-)
> Do you mean that combine patch 1 and patch 2 into one patch to support vlan stripping and bad checksum
> flag,
> Then also do the same work for patch 3 and patch 4 to support good checksum flag and enable IXGBE neon
> vector when DEV_RX_OFFLOAD_CHECKSUM is set?
> 
> Actually, the test results are achieved without patch 4, due to that testpmd does not check rxmode-
> >offloads, it will call IXGBE NEON vector function directly. And this change
> is for the default config of l3fwd example.
> 

I was messed by the vector code, sorry. Patch b is for bad checksum,
update the commit message to align your design. Then the patch set
is OF for me, thanks!

Feifei Wang (4):
  a). net/ixgbe: add VLAN stripping support for Arm
  b). net/ixgbe: support checksum flags for NEON vector
  c). net/ixgbe: implement good checksum flag for NEON vector
  d). net/ixgbe: enable IXGBE NEON vector PMD when CHECKSUM is set

> Best Regards
> Feifei
> >
> > > Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
> > > Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> > > ---
> > > 2.17.1
Feifei Wang Nov. 18, 2020, 8:29 a.m. UTC | #4
> -----邮件原件-----
> 发件人: Wang, Haiyue <haiyue.wang@intel.com>
> 发送时间: 2020年11月18日 16:15
> 收件人: Feifei Wang <Feifei.Wang2@arm.com>; jerinj@marvell.com; Ruifeng
> Wang <Ruifeng.Wang@arm.com>; Guo, Jia <jia.guo@intel.com>
> 抄送: dev@dpdk.org; nd <nd@arm.com>; nd <nd@arm.com>
> 主题: RE: [PATCH v1 3/4] net/ixgbe: implement good checksum flag for NEON
> vector
> 
> > -----Original Message-----
> > From: Feifei Wang <Feifei.Wang2@arm.com>
> > Sent: Wednesday, November 18, 2020 15:56
> > To: Wang, Haiyue <haiyue.wang@intel.com>; jerinj@marvell.com; Ruifeng
> > Wang <Ruifeng.Wang@arm.com>; Guo, Jia <jia.guo@intel.com>
> > Cc: dev@dpdk.org; nd <nd@arm.com>; nd <nd@arm.com>
> > Subject: 回复: [PATCH v1 3/4] net/ixgbe: implement good checksum flag
> > for NEON vector
> >
> > Hi, Haiyue
> >
> > > -----邮件原件-----
> > > 发件人: Wang, Haiyue <haiyue.wang@intel.com>
> > > 发送时间: 2020年11月18日 15:22
> > > 收件人: Feifei Wang <Feifei.Wang2@arm.com>; jerinj@marvell.com;
> Ruifeng
> > > Wang <Ruifeng.Wang@arm.com>; Guo, Jia <jia.guo@intel.com>
> > > 抄送: dev@dpdk.org; nd <nd@arm.com>
> > > 主题: RE: [PATCH v1 3/4] net/ixgbe: implement good checksum flag for
> > > NEON vector
> > >
> > > > -----Original Message-----
> > > > From: Feifei Wang <feifei.wang2@arm.com>
> > > > Sent: Wednesday, November 18, 2020 14:29
> > > > To: Jerin Jacob <jerinj@marvell.com>; Ruifeng Wang
> > > > <ruifeng.wang@arm.com>; Guo, Jia <jia.guo@intel.com>; Wang, Haiyue
> > > > <haiyue.wang@intel.com>
> > > > Cc: dev@dpdk.org; nd@arm.com; feiwan02 <feifei.wang2@arm.com>
> > > > Subject: [PATCH v1 3/4] net/ixgbe: implement good checksum flag
> > > > for NEON vector
> > > >
> > > > From: feiwan02 <feifei.wang2@arm.com>
> > > >
> > > > Add CKSUM_GOOD flag to distinguish a good checksum from an unknown
> > > one
> > > > in neon vertor RX function.
> > > >
> > > > Test Results:
> > > > NICs: 82599(igb)
> > > > Dirver: ixgbe(vector)
> > > > Packet: IPv4_checksum = correct value && UDP_checksum = correct
> > > > value
> > > >
> > > > $:./app/dpdk-testpmd -c 0x3 -w 0002:f9:00.0 -- -i
> > > > --port-topology=chained
> > > > test-pmd> set fwd rxonly
> > > > test-pmd> set verbose 1
> > > > test-pmd> start
> > > >
> > > > With this patch:
> > > > testpmd> port 0/queue 0: received 1 packets
> > > > src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 -
> > > > length=70 - nb_segs=1
> > > > ol_flags: PKT_RX_L4_CKSUM_GOOD PKT_RX_IP_CKSUM_GOOD
> > > >
> > > > Without this patch:
> > > > testpmd> port 0/queue 0: received 1 packets
> > > > src=00:00:00:00:00:02 - dst=00:00:00:00:00:01 - type=0x0800 -
> > > > length=70 - nb_segs=1
> > > > ol_flags: PKT_RX_L4_CKSUM_UNKNOWN PKT_RX_IP_CKSUM_UNKNOWN
> > > >
> > >
> > > How about to merge "1 & 2" and "3 & 4"? Then you will have the above
> > > test result. ;-)
> > Do you mean that combine patch 1 and patch 2 into one patch to support
> > vlan stripping and bad checksum flag, Then also do the same work for
> > patch 3 and patch 4 to support good checksum flag and enable IXGBE
> > neon vector when DEV_RX_OFFLOAD_CHECKSUM is set?
> >
> > Actually, the test results are achieved without patch 4, due to that
> > testpmd does not check rxmode-
> > >offloads, it will call IXGBE NEON vector function directly. And this
> > >change
> > is for the default config of l3fwd example.
> >
> 
> I was messed by the vector code, sorry. Patch b is for bad checksum, update the
> commit message to align your design. Then the patch set is OF for me, thanks!
> 
> Feifei Wang (4):
>   a). net/ixgbe: add VLAN stripping support for Arm
>   b). net/ixgbe: support checksum flags for NEON vector
>   c). net/ixgbe: implement good checksum flag for NEON vector
>   d). net/ixgbe: enable IXGBE NEON vector PMD when CHECKSUM is set
> 
That's all right. I will update the commit message and then upload the new version.
Thanks for your attention.
> > Best Regards
> > Feifei
> > >
> > > > Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
> > > > Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> > > > ---
> > > > 2.17.1
diff mbox series

Patch

diff --git a/drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c b/drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c
index 4d6f057e7..b2bee2228 100644
--- a/drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c
+++ b/drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c
@@ -86,13 +86,13 @@  desc_to_olflags_v(uint8x16x2_t sterr_tmp1, uint8x16x2_t sterr_tmp2,
 		  uint8x16_t staterr, uint8_t vlan_flags, struct rte_mbuf **rx_pkts)
 {
 	uint8x16_t ptype;
-	uint8x16_t vtag;
+	uint8x16_t vtag_lo, vtag_hi, vtag;
 	uint8x16_t temp_csum;
 	uint32x4_t csum = {0, 0, 0, 0};
 
 	union {
-		uint8_t e[4];
-		uint32_t word;
+		uint16_t e[4];
+		uint64_t word;
 	} vol;
 
 	const uint8x16_t rsstype_msk = {
@@ -119,18 +119,26 @@  desc_to_olflags_v(uint8x16x2_t sterr_tmp1, uint8x16x2_t sterr_tmp2,
 			(IXGBE_RXDADV_ERR_TCPE | IXGBE_RXDADV_ERR_IPE) >> 24};
 
 	/* map vlan present (0x8), IPE (0x2), L4E (0x1) to ol_flags */
-	const uint8x16_t vlan_csum_map = {
-			0,
-			PKT_RX_L4_CKSUM_BAD,
+	const uint8x16_t vlan_csum_map_lo = {
+			PKT_RX_IP_CKSUM_GOOD,
+			PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_BAD,
 			PKT_RX_IP_CKSUM_BAD,
 			PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD,
 			0, 0, 0, 0,
-			vlan_flags,
-			vlan_flags | PKT_RX_L4_CKSUM_BAD,
+			vlan_flags | PKT_RX_IP_CKSUM_GOOD,
+			vlan_flags | PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_BAD,
 			vlan_flags | PKT_RX_IP_CKSUM_BAD,
 			vlan_flags | PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD,
 			0, 0, 0, 0};
 
+	const uint8x16_t vlan_csum_map_hi = {
+			PKT_RX_L4_CKSUM_GOOD >> sizeof(uint8_t), 0,
+			PKT_RX_L4_CKSUM_GOOD >> sizeof(uint8_t), 0,
+			0, 0, 0, 0,
+			PKT_RX_L4_CKSUM_GOOD >> sizeof(uint8_t), 0,
+			PKT_RX_L4_CKSUM_GOOD >> sizeof(uint8_t), 0,
+			0, 0, 0, 0};
+
 	ptype = vzipq_u8(sterr_tmp1.val[0], sterr_tmp2.val[0]).val[0];
 	ptype = vandq_u8(ptype, rsstype_msk);
 	ptype = vqtbl1q_u8(rss_flags, ptype);
@@ -150,11 +158,16 @@  desc_to_olflags_v(uint8x16x2_t sterr_tmp1, uint8x16x2_t sterr_tmp2,
 	csum = vsetq_lane_u32(vgetq_lane_u32(vreinterpretq_u32_u8(temp_csum), 3), csum, 0);
 	vtag = vorrq_u8(vreinterpretq_u8_u32(csum), vtag);
 
-	/* convert VP, IPE, L4E to ol_flags */
-	vtag = vqtbl1q_u8(vlan_csum_map, vtag);
-	vtag = vorrq_u8(ptype, vtag);
+	/* convert L4 checksum correct type to vtag_hi */
+	vtag_hi = vqtbl1q_u8(vlan_csum_map_hi, vtag);
+	vtag_hi = vshrq_n_u8(vtag_hi, 7);
+
+	/* convert VP, IPE, L4E to vtag_lo */
+	vtag_lo = vqtbl1q_u8(vlan_csum_map_lo, vtag);
+	vtag_lo = vorrq_u8(ptype, vtag_lo);
 
-	vol.word = vgetq_lane_u32(vreinterpretq_u32_u8(vtag), 0);
+	vtag = vzipq_u8(vtag_lo, vtag_hi).val[0];
+	vol.word = vgetq_lane_u64(vreinterpretq_u64_u8(vtag), 0);
 
 	rx_pkts[0]->ol_flags = vol.e[0];
 	rx_pkts[1]->ol_flags = vol.e[1];