linux/drivers/net/ethernet
David S. Miller 20eb08b2b0 mlx5-updates-2019-04-22
This series includes updates to mlx5e driver RX data path and some
 significant XDP RX/TX improvements to overcome/mitigate HW and PCIE
 bottlenecks.
 
 From Tariq:
 1) Some Enhancements in rq->flags
 2) Stabilize RX packet rate (on Striding RQ) with
 multiple outstanding UMR posts
 In this patch, we add support for multiple outstanding UMR posts,
  to allow faster gap closure between consuming MPWQEs and reposting
 them back into the WQ.
 
 Performance test:
 As expected, huge improvement in large-scale (48 cores).
 
 xdp_redirect_map, 64B UDP multi-stream.
 Redirect from ConnectX-5 100Gbps to ConnectX-6 100Gbps.
 CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz.
 
 Before: Unstable, 7 to 30 Mpps
 After:  Stable,   at 70.5 Mpps
 
 From Shay:
 3) XDP, Inline small packets into the TX MPWQE in XDP xmit flow
 
 Upon high packet rate with multiple CPUs TX workloads, much of the HCA's
 resources are spent on prefetching TX descriptors, thus affecting
 transmission rates.
 This patch comes to mitigate this problem by moving some workload to the
 CPU and reducing the HW data prefetch overhead for small packets (<= 256B).
 
 When forwarding packets with XDP, a packet that is smaller
 than a certain size (set to ~256 bytes) would be sent inline within
 its WQE TX descrptor (mem-copied), when the hardware tx queue is congested
 beyond a pre-defined water-mark.
 
 Performance:
     Tested packet rate for UDP 64Byte multi-stream
     over two dual port ConnectX-5 100Gbps NICs.
     CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz
 
     * Tested with hyper-threading disabled
 
     XDP_TX:
 
     |          | before | after   |       |
     | 24 rings | 51Mpps | 116Mpps | +126% |
     | 1 ring   | 12Mpps | 12Mpps  | same  |
 
     XDP_REDIRECT:
 
     ** Below is the transmit rate, not the redirection rate
     which might be larger, and is not affected by this patch.
 
     |          | before  | after   |      |
     | 32 rings | 64Mpps  | 92Mpps  | +43% |
     | 1 ring   | 6.4Mpps | 6.4Mpps | same |
 
 As we can see, feature significantly improves scaling, without
 hurting single ring performance.
 
 From Maxim:
 4) Some trivial refactoring and code improvements prior to a larger series
 to support AF_XDP.
 
 -Saeed.
 -----BEGIN PGP SIGNATURE-----
 
 iQEcBAABAgAGBQJcv2LjAAoJEEg/ir3gV/o+90gIAI8+4lwkXZAVk4mxf9PMjxuB
 bQiKd80e++26sgrNHCyuWZnIzTQqYAnUJ3WRC+Kk1pFTo1O23A+fvweT8m1dqAvP
 Z/5ktfbAeF3fwOVu7aGu9vh4zJEWJj8oO+I+G+OaOe2iV7FVTTFnWHxiiCfungAW
 oUnXozq4vERSQLechqqgz6nACxOPgEOCJrp4T9lDYSbqZizHgFttmInMQguq/7KS
 LvITcNu3EF5l4y2LxwCFiKRgGc2y/belU63AK+2pQUXhH46kQPEHdncdLg5d9QYA
 xJwthn697qxS0PIP5oHPHNVN+qJXfuUHVonXqVOAJebGQnV82of6+sPweRxwh1s=
 =MfAR
 -----END PGP SIGNATURE-----

Merge tag 'mlx5-updates-2019-04-22' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2019-04-22

This series includes updates to mlx5e driver RX data path and some
significant XDP RX/TX improvements to overcome/mitigate HW and PCIE
bottlenecks.

From Tariq:
1) Some Enhancements in rq->flags
2) Stabilize RX packet rate (on Striding RQ) with
multiple outstanding UMR posts
In this patch, we add support for multiple outstanding UMR posts,
 to allow faster gap closure between consuming MPWQEs and reposting
them back into the WQ.

Performance test:
As expected, huge improvement in large-scale (48 cores).

xdp_redirect_map, 64B UDP multi-stream.
Redirect from ConnectX-5 100Gbps to ConnectX-6 100Gbps.
CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz.

Before: Unstable, 7 to 30 Mpps
After:  Stable,   at 70.5 Mpps

From Shay:
3) XDP, Inline small packets into the TX MPWQE in XDP xmit flow

Upon high packet rate with multiple CPUs TX workloads, much of the HCA's
resources are spent on prefetching TX descriptors, thus affecting
transmission rates.
This patch comes to mitigate this problem by moving some workload to the
CPU and reducing the HW data prefetch overhead for small packets (<= 256B).

When forwarding packets with XDP, a packet that is smaller
than a certain size (set to ~256 bytes) would be sent inline within
its WQE TX descrptor (mem-copied), when the hardware tx queue is congested
beyond a pre-defined water-mark.

Performance:
    Tested packet rate for UDP 64Byte multi-stream
    over two dual port ConnectX-5 100Gbps NICs.
    CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz

    * Tested with hyper-threading disabled

    XDP_TX:

    |          | before | after   |       |
    | 24 rings | 51Mpps | 116Mpps | +126% |
    | 1 ring   | 12Mpps | 12Mpps  | same  |

    XDP_REDIRECT:

    ** Below is the transmit rate, not the redirection rate
    which might be larger, and is not affected by this patch.

    |          | before  | after   |      |
    | 32 rings | 64Mpps  | 92Mpps  | +43% |
    | 1 ring   | 6.4Mpps | 6.4Mpps | same |

As we can see, feature significantly improves scaling, without
hurting single ring performance.

From Maxim:
4) Some trivial refactoring and code improvements prior to a larger series
to support AF_XDP.
====================

Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2019-04-23 17:03:40 -07:00
..
3com 3c515: fix integer overflow warning 2019-03-19 15:07:49 -07:00
8390 mac8390: Fix mmio access size probe 2019-03-18 16:56:57 -07:00
adaptec net: adaptec: starfire: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-14 12:43:15 -05:00
aeroflex
agere
alacritech
allwinner
alteon
altera
amazon net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
amd net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
apm
apple net: apple: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-13 20:50:42 -08:00
aquantia net: aquantia: enable driver build for arm64 or compile_test 2019-03-23 22:16:53 -04:00
arc net: arc_emac: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-14 12:43:15 -05:00
atheros Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-02-24 12:06:19 -08:00
aurora drivers: net: aurora: use netdev_xmit_more helper 2019-04-01 18:35:02 -07:00
broadcom net: systemport: Remove need for DMA descriptor 2019-04-22 22:20:15 -07:00
brocade
cadence Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-04-08 23:39:36 -07:00
calxeda
cavium Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-04-17 11:26:25 -07:00
chelsio cxgb4: Don't return EAGAIN when TCAM is full. 2019-04-08 10:01:50 -07:00
cirrus
cisco net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
cortina net: gemini: remove unnecessary assert 2019-04-18 17:06:15 -07:00
davicom
dec Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-02-08 15:00:17 -08:00
dlink net:dl2k: Replace dev_kfree_skb_irq by dev_consume_skb_irq 2019-02-14 12:25:24 -05:00
emulex net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
ezchip
faraday drivers: net: Remove unnecessary semicolon 2019-03-01 23:13:49 -08:00
freescale Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-04-17 11:26:25 -07:00
fujitsu net: fujitsu: fix a potential NULL pointer dereference 2019-03-12 14:48:35 -07:00
hisilicon net: phy: marvell: add new default led configure for m88e151x 2019-04-23 10:40:32 -07:00
hp
huawei net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
i825xx net: i825xx: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-14 12:43:14 -05:00
ibm Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-04-17 11:26:25 -07:00
intel ice: Calculate ITR increment based on direct calculation 2019-04-18 08:38:48 -07:00
marvell net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
mediatek net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
mellanox mlx5-updates-2019-04-22 2019-04-23 17:03:40 -07:00
micrel net: ks8851: Deduplicate register macros 2019-03-20 12:43:52 -07:00
microchip enc28j60: Convert to use SPDX identifier 2019-03-19 14:59:32 -07:00
moxa net: moxa: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-13 20:50:42 -08:00
mscc ocelot: Clean up stats update deferred work 2019-04-16 21:43:53 -07:00
myricom net: myri10ge: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-14 11:56:36 -05:00
natsemi net: ns83820: code cleanup for ns83820_probe_phy() 2019-02-19 14:08:19 -08:00
neterion vxge: fix return of a free'd memblock on a failed dma mapping 2019-04-12 16:54:41 -07:00
netronome net: devlink: Add extack to shared buffer operations 2019-04-22 22:09:32 -07:00
ni net: nixge: Add support for fixed-link configurations 2019-02-05 10:34:34 -08:00
nuvoton net: nuvoton: w90p910_ether: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-14 11:56:36 -05:00
nvidia
nxp
oki-semi pci-v5.1-changes 2019-03-09 14:57:08 -08:00
packetengines net: packetengines: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-14 12:43:15 -05:00
pasemi net: pasemi: remove set but not used variable 'cpyhdr' 2019-04-04 17:36:23 -07:00
qlogic Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-04-17 11:26:25 -07:00
qualcomm net: qualcomm: emac: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-13 20:50:42 -08:00
rdc net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
realtek r8169: change irq handler to always trigger NAPI polling 2019-04-14 13:58:15 -07:00
renesas net: remove 'fallback' argument from dev->ndo_select_queue() 2019-03-20 11:18:55 -07:00
rocker ipv4: Flag fib_info with a fib_nh using IPv6 gateway 2019-04-08 15:22:41 -07:00
samsung
seeq
sfc drivers: net: sfc: use netdev_xmit_more helper 2019-04-01 18:35:02 -07:00
sgi net: sgi: use GFP_ATOMIC under spin lock 2019-02-17 15:51:46 -08:00
silan
sis net: sis900: fix indentation issues, remove some spaces 2019-03-15 11:59:00 -07:00
smsc smc911x: pass struct device to DMA API functions 2019-02-12 12:09:24 -05:00
socionext
stmicro net: stmmac: Set Flow Control to automatic mode in the driver 2019-04-17 10:14:28 -07:00
sun net: remove 'fallback' argument from dev->ndo_select_queue() 2019-03-20 11:18:55 -07:00
synopsys net: move skb->xmit_more hint to softnet data 2019-04-01 18:35:02 -07:00
tehuti net: tehuti: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-06 17:38:15 -08:00
ti net: ethernet: ti: davinci_mdio: switch to readl/writel() 2019-04-01 15:02:50 -07:00
toshiba
tundra
via net: via-velocity: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-06 17:38:15 -08:00
wiznet
xilinx net: xilinx: emaclite: add minimal ndo_do_ioctl hook 2019-04-08 10:00:10 -07:00
xircom
xscale net: ixp4xx_eth: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-13 20:50:54 -08:00
dnet.c
dnet.h
ec_bhf.c
ethoc.c
fealnx.c net: fealnx: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles 2019-02-13 20:50:42 -08:00
jme.c
jme.h
Kconfig
korina.c
lantiq_etop.c lantiq_etop: pass struct device to DMA API functions 2019-02-12 12:09:23 -05:00
lantiq_xrx200.c net: lantiq: Do not use eth_change_mtu() 2019-02-24 17:35:42 -08:00
Makefile
netx-eth.c