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. 09 5月, 2014 1 次提交
  3. 25 4月, 2014 2 次提交
    • M
      bonding: Add tlb_dynamic_lb parameter for tlb mode · e9f0fb88
      由 Mahesh Bandewar 提交于 
      The aggresive load balancing causes packet re-ordering as active
flows are moved from a slave to another within the group. Sometime
this aggresive lb is not necessary if the preference is for less
re-ordering. This parameter if used with value "0" disables
this dynamic flow shuffling minimizing packet re-ordering. Of course
the side effect is that it has to live with the static load balancing
that the hashing distribution provides. This impact is less severe if
the correct xmit-hashing-policy is used for the tlb setup.

The default value of the parameter is set to "1" mimicing the earlier
behavior.

Ran the netperf test with 200 stream for 1 min between two hosts with
4x1G trunk (xmit-lb mode with xmit-policy L3+4) before and after these
changes. Following was the command used for those 200 instances -

    netperf -t TCP_RR -l 60 -s 5 -H <host> -- -r81920,81920

Transactions per second:
    Before change: 1,367.11
    After  change: 1,470.65

Change-Id: Ie3f75c77282cf602e83a6e833c6eb164e72a0990
Signed-off-by: NMahesh Bandewar <maheshb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      e9f0fb88
    • M
      bonding: Changed hashing function to just provide hash · ee62e868
      由 Mahesh Bandewar 提交于 
      Modified the hash function to return just hash separating from the
modulo operation that can be performed by the caller. This is to
make way for the tlb mode to use the same hashing policies that
are used in the 802.3ad and Xor mode.

Change-Id: I276609e87e0ca213c4d1b17b79c5e0b0f3d0dd6f
Signed-off-by: NMahesh Bandewar <maheshb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      ee62e868
  5. 27 3月, 2014 1 次提交
    • D
      bonding: support QinQ for bond arp interval · fbd929f2
      由 dingtianhong 提交于 
      The bond send arp request to indicate that the slave is active, and if the bond dev
is a vlan dev, it will set the vlan tag in skb to notice the vlan group, but the
bond could only send a skb with 802.1q proto, not support for QinQ.

So add outer tag for lower vlan tag and inner tag for upper vlan tag to support QinQ,
The new skb will be consist of two vlan tag just like this:

dst mac | src mac | outer vlan tag | inner vlan tag | data | .....

If We don't need QinQ, the inner vlan tag could be set to 0 and use outer vlan tag
 as a normal vlan group.

Using "ip link" to configure the bond for QinQ and add test log:

ip link add link bond0  bond0.20 type vlan proto 802.1ad id 20
ip link add link bond0.20  bond0.20.200 type vlan proto 802.1q id 200

ifconfig bond0.20 11.11.20.36/24
ifconfig bond0.20.200 11.11.200.36/24

echo +11.11.200.37 > /sys/class/net/bond0/bonding/arp_ip_target

90:e2:ba:07:4a:5c (oui Unknown) > Broadcast, ethertype 802.1Q-QinQ (0x88a8),length 50: vlan 20, p 0,ethertype 802.1Q, vlan 200, p 0, ethertype ARP, Ethernet (len 6), IPv4 (len 4), Request who-has 11.11.200.37 tell 11.11.200.36, length 28

90:e2:ba:06:f9:86 (oui Unknown) > 90:e2:ba:07:4a:5c (oui Unknown), ethertype 802.1Q-QinQ (0x88a8), length 50: vlan 20, p 0, ethertype 802.1Q, vlan 200, p 0, ethertype ARP, Ethernet (len 6), IPv4 (len 4), Reply 11.11.200.37 is-at 90:e2:ba:06:f9:86 (oui Unknown), length 28

v1->v2: remove the comment "TODO: QinQ?".

Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      fbd929f2
  7. 07 3月, 2014 3 次提交
  9. 27 2月, 2014 2 次提交
    • D
      bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for ab arp monitor · b0929915
      由 dingtianhong 提交于 
      Veaceslav has reported and fix this problem by commit f2ebd477
(bonding: restructure locking of bond_ab_arp_probe()). According Jay's
opinion, the current solution is not very well, because the notification
is to indicate that the interface has actually changed state in a meaningful
way, but these calls in the ab ARP monitor are internal settings of the flags
to allow the ARP monitor to search for a slave to become active when there are
no active slaves. The flag setting to active or backup is to permit the ARP
monitor's response logic to do the right thing when deciding if the test
slave (current_arp_slave) is up or not.

So the best way to fix the problem is that we should not send a notification
when the slave is in testing state, and check the state at the end of the
monitor, if the slave's state recover, avoid to send pointless notification
twice. And RTNL is really a big lock, hold it regardless the slave's state
changed or not when the current_active_slave is null will loss performance
(every 100ms), so we should hold it only when the slave's state changed and
need to notify.

I revert the old commit and add new modifications.

Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      b0929915
    • D
      bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode · 5e5b0665
      由 dingtianhong 提交于 
      The problem was introduced by the commit 1d3ee88a
(bonding: add netlink attributes to slave link dev).
The bond_set_active_slave() and bond_set_backup_slave()
will use rtmsg_ifinfo to send slave's states, so these
two functions should be called in RTNL.

In 802.3ad mode, acquiring RTNL for the __enable_port and
__disable_port cases is difficult, as those calls generally
already hold the state machine lock, and cannot unconditionally
call rtnl_lock because either they already hold RTNL (for calls
via bond_3ad_unbind_slave) or due to the potential for deadlock
with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed,
bond_3ad_link_change, or bond_3ad_update_lacp_rate.  All four of
those are called with RTNL held, and acquire the state machine lock
second.  The calling contexts for __enable_port and __disable_port
already hold the state machine lock, and may or may not need RTNL.

According to the Jay's opinion, I don't think it is a problem that
the slave don't send notify message synchronously when the status
changed, normally the state machine is running every 100 ms, send
the notify message at the end of the state machine if the slave's
state changed should be better.

I fix the problem through these steps:

1). add a new function bond_set_slave_state() which could change
    the slave's state and call rtmsg_ifinfo() according to the input
    parameters called notify.

2). Add a new slave parameter which called should_notify, if the slave's state
    changed and don't notify yet, the parameter will be set to 1, and then if
    the slave's state changed again, the param will be set to 0, it indicate that
    the slave's state has been restored, no need to notify any one.

3). the __enable_port and __disable_port should not call rtmsg_ifinfo
    in the state machine lock, any change in the state of slave could
    set a flag in the slave, it will indicated that an rtmsg_ifinfo
    should be called at the end of the state machine.

Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      5e5b0665
  11. 19 2月, 2014 4 次提交
  13. 29 1月, 2014 1 次提交
    • D
      bonding: fix locking in bond_loadbalance_arp_mon() · 6fde8f03
      由 Ding Tianhong 提交于 
      The commit 1d3ee88a
(bonding: add netlink attributes to slave link dev)
has add rtmsg_ifinfo() in bond_set_active_slave() and
bond_set_backup_slave(), so the two function need to
called in RTNL lock, but bond_loadbalance_arp_mon()
only calling these functions in RCU, warning message
will occurs.

fix this by add a new function bond_slave_state_change(),
which will reset the slave's state after slave link check,
so remove the bond_set_xxx_slave() from the cycle and only
record the slave_state_changed, this will call the new
function to set all slaves to new state in RTNL later.

Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      6fde8f03
  15. 23 1月, 2014 24 次提交
  17. 22 1月, 2014 1 次提交
    • H
      reciprocal_divide: update/correction of the algorithm · 809fa972
      由 Hannes Frederic Sowa 提交于 
      Jakub Zawadzki noticed that some divisions by reciprocal_divide()
were not correct [1][2], which he could also show with BPF code
after divisions are transformed into reciprocal_value() for runtime
invariance which can be passed to reciprocal_divide() later on;
reverse in BPF dump ended up with a different, off-by-one K in
some situations.

This has been fixed by Eric Dumazet in commit aee636c4
("bpf: do not use reciprocal divide"). This follow-up patch
improves reciprocal_value() and reciprocal_divide() to work in
all cases by using Granlund and Montgomery method, so that also
future use is safe and without any non-obvious side-effects.
Known problems with the old implementation were that division by 1
always returned 0 and some off-by-ones when the dividend and divisor
where very large. This seemed to not be problematic with its
current users, as far as we can tell. Eric Dumazet checked for
the slab usage, we cannot surely say so in the case of flex_array.
Still, in order to fix that, we propose an extension from the
original implementation from commit 6a2d7a95 resp. [3][4],
by using the algorithm proposed in "Division by Invariant Integers
Using Multiplication" [5], Torbjörn Granlund and Peter L.
Montgomery, that is, pseudocode for q = n/d where q, n, d is in
u32 universe:

1) Initialization:

  int l = ceil(log_2 d)
  uword m' = floor((1<<32)*((1<<l)-d)/d)+1
  int sh_1 = min(l,1)
  int sh_2 = max(l-1,0)

2) For q = n/d, all uword:

  uword t = (n*m')>>32
  q = (t+((n-t)>>sh_1))>>sh_2

The assembler implementation from Agner Fog [6] also helped a lot
while implementing. We have tested the implementation on x86_64,
ppc64, i686, s390x; on x86_64/haswell we're still half the latency
compared to normal divide.

Joint work with Daniel Borkmann.

  [1] http://www.wireshark.org/~darkjames/reciprocal-buggy.c
  [2] http://www.wireshark.org/~darkjames/set-and-dump-filter-k-bug.c
  [3] https://gmplib.org/~tege/division-paper.pdf
  [4] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html
  [5] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.1.2556
  [6] http://www.agner.org/optimize/asmlib.zipReported-by: NJakub Zawadzki <darkjames-ws@darkjames.pl>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Austin S Hemmelgarn <ahferroin7@gmail.com>
Cc: linux-kernel@vger.kernel.org
Cc: Jesse Gross <jesse@nicira.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Matt Mackall <mpm@selenic.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Andy Gospodarek <andy@greyhouse.net>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl>
Signed-off-by: NDaniel Borkmann <dborkman@redhat.com>
Signed-off-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      809fa972
  19. 18 1月, 2014 1 次提交