Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 27 1月, 2012 2 次提交
  3. 09 12月, 2011 1 次提交
  5. 30 11月, 2011 1 次提交
  7. 17 11月, 2011 1 次提交
  9. 15 11月, 2011 1 次提交
    • E
      bnx2x: uses build_skb() in receive path · e52fcb24
      由 Eric Dumazet 提交于 
      bnx2x uses following formula to compute its rx_buf_sz :

dev->mtu + 2*L1_CACHE_BYTES + 14 + 8 + 8 + 2

Then core network adds NET_SKB_PAD and SKB_DATA_ALIGN(sizeof(struct
skb_shared_info))

Final allocated size for skb head on x86_64 (L1_CACHE_BYTES = 64,
MTU=1500) : 2112 bytes : SLUB/SLAB round this to 4096 bytes.

Since skb truesize is then bigger than SK_MEM_QUANTUM, we have lot of
false sharing because of mem_reclaim in UDP stack.

One possible way to half truesize is to reduce the need by 64 bytes
(2112 -> 2048 bytes)

Instead of allocating a full cache line at the end of packet for
alignment, we can use the fact that skb_shared_info sits at the end of
skb->head, and we can use this room, if we convert bnx2x to new
build_skb() infrastructure.

skb_shared_info will be initialized after hardware finished its
transfert, so we can eventually overwrite the final padding.

Using build_skb() also reduces cache line misses in the driver, since we
use cache hot skb instead of cold ones. Number of in-flight sk_buff
structures is lower, they are recycled while still hot.

Performance results :

(820.000 pps on a rx UDP monothread benchmark, instead of 720.000 pps)
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Acked-by: NEilon Greenstein <eilong@broadcom.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      e52fcb24
  11. 14 11月, 2011 4 次提交
  13. 30 8月, 2011 1 次提交
  15. 19 8月, 2011 1 次提交
  17. 18 8月, 2011 1 次提交
  19. 11 8月, 2011 1 次提交
    • J
      broadcom: Move the Broadcom drivers · adfc5217
      由 Jeff Kirsher 提交于 
      Moves the drivers for Broadcom devices into
drivers/net/ethernet/broadcom/ and the necessary Kconfig and Makefile
changes.

CC: Eilon Greenstein <eilong@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Matt Carlson <mcarlson@broadcom.com>
CC: Gary Zambrano <zambrano@broadcom.com>
CC: "Maciej W. Rozycki" <macro@linux-mips.org>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
      adfc5217
  21. 22 7月, 2011 1 次提交
  23. 20 7月, 2011 1 次提交
  25. 15 7月, 2011 1 次提交
    • A
      bnx2x: Multiple concurrent l2 traffic classes · 6383c0b3
      由 Ariel Elior 提交于 
      Overview:
 Support mapping of priorities to traffic classes and
 traffic classes to transmission queues ranges in the net device.
 The queue ranges are (count, offset) pairs relating to the txq
 array.
 This can be done via DCBX negotiation or by kernel.
 As a result Enhanced Transmission Selection (ETS) and Priority Flow
 Control (PFC) are supported between L2 network traffic classes.

 Mapping:
 This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and
 netdev_set_tc_queue functions to map priorities to traffic classes
 and traffic classes to transmission queue ranges.
 This mapping is performed by bnx2x_setup_tc function which is
 connected to the ndo_setup_tc.
 This function is always called at nic load where by default it
 maps all priorities to tc 0, and it may also be called by the
 kernel or by the bnx2x upon DCBX negotiation to modify the mapping.

 rtnl lock:
 When the ndo_setup_tc is called at nic load or by kernel the rtnl
 lock is already taken. However, when DCBX negotiation takes place
 the lock is not taken. The work is therefore scheduled to be
 handled by the sp_rtnl task.

 Fastpath:
 The fastpath structure of the bnx2x which was previously used
 to hold the information of one tx queue and one rx queue was
 redesigned to represent multiple tx queues, one for each traffic
 class.
 The transmission queue supplied in the skb by the kernel can no
 longer be interpreted as a straightforward index into the fastpath
 structure array, but it must rather be decoded to the appropriate
 fastpath index and the tc within that fastpath.

 Slowpath:
 The bnx2x's queue object was redesigned to accommodate multiple
 transmission queues. The queue object's state machine was enhanced
 to allow opening multiple transmission-only connections on top of
 the regular tx-rx connection.

 Firmware:
 This feature relies on the tx-only queue feature introduced in the
 bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi
 cos support.
Signed-off-by: NAriel Elior <ariele@broadcom.com>
Signed-off-by: NEilon Greenstein <eilong@broadcom.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      6383c0b3
  27. 15 6月, 2011 5 次提交
  29. 02 6月, 2011 1 次提交
  31. 23 5月, 2011 1 次提交
  33. 06 5月, 2011 4 次提交
  35. 13 4月, 2011 1 次提交
    • M
      net: bnx2x: convert to hw_features · 66371c44
      由 Michał Mirosław 提交于 
      Since ndo_fix_features callback is postponing features change when
bp->recovery_state != BNX2X_RECOVERY_DONE, netdev_update_features()
has to be called again when this condition changes. Previously,
ethtool_ops->set_flags callback returned -EBUSY in that case
(it's not possible in the new model).
Signed-off-by: NMichał Mirosław <mirq-linux@rere.qmqm.pl>

v5: - don't delay set_features, as it's rtnl_locked - same as recovery process
v4: - complete bp->rx_csum -> NETIF_F_RXCSUM conversion
    - add check for failed ndo_set_features in ndo_open callback
v3: - include NETIF_F_LRO in hw_features
    - don't call netdev_update_features() if bnx2x_nic_load() failed
v2: - comment in ndo_fix_features callback
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      66371c44
  37. 01 4月, 2011 1 次提交
  39. 07 3月, 2011 1 次提交
  41. 01 3月, 2011 1 次提交
  43. 07 2月, 2011 1 次提交
  45. 17 12月, 2010 3 次提交
  47. 02 12月, 2010 1 次提交
    • D
      bnx2x: Add Nic partitioning mode (57712 devices) · 0793f83f
      由 Dmitry Kravkov 提交于 
      NIC partitioning is another flavor of multi function - having few
PCI functions share the same physical port. Unlike the currently
supported mode of multi-function which depends on the switch
configuration and uses outer-VLAN, the NPAR mode is switch independent
and uses the MAC addresses to distribute incoming packets to the different
functions. This patch adds the specific HW setting of the NPAR mode
and some distinctions between switch dependent (SD) and
switch independent (SI) multi-function (MF) modes where the configuration
is not the same.

Advance driver version to 1.60.00-6
Signed-off-by: NDmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: NEilon Greenstein <eilong@broadcom.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      0793f83f
  49. 25 10月, 2010 1 次提交
  51. 18 10月, 2010 1 次提交
  53. 07 10月, 2010 1 次提交