Skip to content

cloud-kernel
cloud-kernel

openanolis / cloud-kernel
1 年多 前同步成功

通知 160
Star 36
Fork 7
cloud-kernel
cloud-kernel
切换分支/标签
  1. 13 3月, 2011 1 次提交
  3. 05 3月, 2011 1 次提交
  5. 03 3月, 2011 1 次提交
  7. 02 3月, 2011 1 次提交
  9. 24 2月, 2011 1 次提交
  11. 23 2月, 2011 2 次提交
  13. 27 1月, 2011 1 次提交
    • D
      net: Implement read-only protection and COW'ing of metrics. · 62fa8a84
      由 David S. Miller 提交于 
      Routing metrics are now copy-on-write.

Initially a route entry points it's metrics at a read-only location.
If a routing table entry exists, it will point there.  Else it will
point at the all zero metric place-holder called 'dst_default_metrics'.

The writeability state of the metrics is stored in the low bits of the
metrics pointer, we have two bits left to spare if we want to store
more states.

For the initial implementation, COW is implemented simply via kmalloc.
However future enhancements will change this to place the writable
metrics somewhere else, in order to increase sharing.  Very likely
this "somewhere else" will be the inetpeer cache.

Note also that this means that metrics updates may transiently fail
if we cannot COW the metrics successfully.

But even by itself, this patch should decrease memory usage and
increase cache locality especially for routing workloads.  In those
cases the read-only metric copies stay in place and never get written
to.

TCP workloads where metrics get updated, and those rare cases where
PMTU triggers occur, will take a very slight performance hit.  But
that hit will be alleviated when the long-term writable metrics
move to a more sharable location.

Since the metrics storage went from a u32 array of RTAX_MAX entries to
what is essentially a pointer, some retooling of the dst_entry layout
was necessary.

Most importantly, we need to preserve the alignment of the reference
count so that it doesn't share cache lines with the read-mostly state,
as per Eric Dumazet's alignment assertion checks.

The only non-trivial bit here is the move of the 'flags' member into
the writeable cacheline.  This is OK since we are always accessing the
flags around the same moment when we made a modification to the
reference count.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      62fa8a84
  15. 18 11月, 2010 1 次提交
  17. 16 11月, 2010 1 次提交
  19. 12 11月, 2010 1 次提交
    • E
      net: get rid of rtable->idev · 72cdd1d9
      由 Eric Dumazet 提交于 
      It seems idev field in struct rtable has no special purpose, but adding
extra atomic ops.

We hold refcounts on the device itself (using percpu data, so pretty
cheap in current kernel).

infiniband case is solved using dst.dev instead of idev->dev

Removal of this field means routing without route cache is now using
shared data, percpu data, and only potential contention is a pair of
atomic ops on struct neighbour per forwarded packet.

About 5% speedup on routing test.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Roland Dreier <rolandd@cisco.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      72cdd1d9
  21. 12 10月, 2010 1 次提交
    • E
      net dst: use a percpu_counter to track entries · fc66f95c
      由 Eric Dumazet 提交于 
      struct dst_ops tracks number of allocated dst in an atomic_t field,
subject to high cache line contention in stress workload.

Switch to a percpu_counter, to reduce number of time we need to dirty a
central location. Place it on a separate cache line to avoid dirtying
read only fields.

Stress test :

(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE, SLUB/NUMA)

Before:

real    0m51.179s
user    0m15.329s
sys     10m15.942s

After:

real	0m45.570s
user	0m15.525s
sys	9m56.669s

With a small reordering of struct neighbour fields, subject of a
following patch, (to separate refcnt from other read mostly fields)

real	0m41.841s
user	0m15.261s
sys	8m45.949s
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      fc66f95c
  23. 23 9月, 2010 1 次提交
  25. 05 7月, 2010 1 次提交
  27. 11 6月, 2010 1 次提交
  29. 07 4月, 2010 1 次提交
    • T
      xfrm: cache bundles instead of policies for outgoing flows · 80c802f3
      由 Timo Teräs 提交于 
      __xfrm_lookup() is called for each packet transmitted out of
system. The xfrm_find_bundle() does a linear search which can
kill system performance depending on how many bundles are
required per policy.

This modifies __xfrm_lookup() to store bundles directly in
the flow cache. If we did not get a hit, we just create a new
bundle instead of doing slow search. This means that we can now
get multiple xfrm_dst's for same flow (on per-cpu basis).
Signed-off-by: NTimo Teras <timo.teras@iki.fi>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      80c802f3
  31. 03 3月, 2010 1 次提交
  33. 25 1月, 2010 1 次提交
  35. 12 11月, 2009 1 次提交
    • E
      sysctl net: Remove unused binary sysctl code · f8572d8f
      由 Eric W. Biederman 提交于 
      Now that sys_sysctl is a compatiblity wrapper around /proc/sys
all sysctl strategy routines, and all ctl_name and strategy
entries in the sysctl tables are unused, and can be
revmoed.

In addition neigh_sysctl_register has been modified to no longer
take a strategy argument and it's callers have been modified not
to pass one.

Cc: "David Miller" <davem@davemloft.net>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Cc: netdev@vger.kernel.org
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>
      f8572d8f
  37. 05 8月, 2009 1 次提交
  39. 31 7月, 2009 1 次提交
    • N
      xfrm: select sane defaults for xfrm[4|6] gc_thresh · a33bc5c1
      由 Neil Horman 提交于 
      Choose saner defaults for xfrm[4|6] gc_thresh values on init

Currently, the xfrm[4|6] code has hard-coded initial gc_thresh values
(set to 1024).  Given that the ipv4 and ipv6 routing caches are sized
dynamically at boot time, the static selections can be non-sensical.
This patch dynamically selects an appropriate gc threshold based on
the corresponding main routing table size, using the assumption that
we should in the worst case be able to handle as many connections as
the routing table can.

For ipv4, the maximum route cache size is 16 * the number of hash
buckets in the route cache.  Given that xfrm4 starts garbage
collection at the gc_thresh and prevents new allocations at 2 *
gc_thresh, we set gc_thresh to half the maximum route cache size.

For ipv6, its a bit trickier.  there is no maximum route cache size,
but the ipv6 dst_ops gc_thresh is statically set to 1024.  It seems
sane to select a simmilar gc_thresh for the xfrm6 code that is half
the number of hash buckets in the v6 route cache times 16 (like the v4
code does).
Signed-off-by: NNeil Horman <nhorman@tuxdriver.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      a33bc5c1
  41. 28 7月, 2009 1 次提交
    • N
      xfrm: export xfrm garbage collector thresholds via sysctl · a44a4a00
      由 Neil Horman 提交于 
      Export garbage collector thresholds for xfrm[4|6]_dst_ops

Had a problem reported to me recently in which a high volume of ipsec
connections on a system began reporting ENOBUFS for new connections
eventually.

It seemed that after about 2000 connections we started being unable to
create more.  A quick look revealed that the xfrm code used a dst_ops
structure that limited the gc_thresh value to 1024, and always
dropped route cache entries after 2x the gc_thresh.

It seems the most direct solution is to export the gc_thresh values in
the xfrm[4|6] dst_ops as sysctls, like the main routing table does, so
that higher volumes of connections can be supported.  This patch has
been tested and allows the reporter to increase their ipsec connection
volume successfully.
Reported-by: NJoe Nall <joe@nall.com>
Signed-off-by: NNeil Horman <nhorman@tuxdriver.com>

ipv4/xfrm4_policy.c |   18 ++++++++++++++++++
ipv6/xfrm6_policy.c |   18 ++++++++++++++++++
2 files changed, 36 insertions(+)
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      a44a4a00
  43. 04 7月, 2009 1 次提交
  45. 01 2月, 2009 1 次提交
  47. 26 11月, 2008 3 次提交
  49. 12 11月, 2008 1 次提交
  51. 03 11月, 2008 1 次提交
  53. 26 3月, 2008 1 次提交
  55. 01 2月, 2008 1 次提交
  57. 29 1月, 2008 9 次提交