forked from Minki/linux
29b4433d99
We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.
There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)
We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.
On x86, dev_hold(dev) code :
before
lock incl 0x280(%ebx)
after:
movl 0x260(%ebx),%eax
incl fs:(%eax)
Stress bench :
(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)
Before:
real 1m1.662s
user 0m14.373s
sys 12m55.960s
After:
real 0m51.179s
user 0m15.329s
sys 10m15.942s
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
||
|---|---|---|
| .. | ||
| datagram.c | ||
| dev_addr_lists.c | ||
| dev.c | ||
| drop_monitor.c | ||
| dst.c | ||
| ethtool.c | ||
| fib_rules.c | ||
| filter.c | ||
| flow.c | ||
| gen_estimator.c | ||
| gen_stats.c | ||
| iovec.c | ||
| kmap_skb.h | ||
| link_watch.c | ||
| Makefile | ||
| neighbour.c | ||
| net_namespace.c | ||
| net-sysfs.c | ||
| net-sysfs.h | ||
| net-traces.c | ||
| netevent.c | ||
| netpoll.c | ||
| pktgen.c | ||
| request_sock.c | ||
| rtnetlink.c | ||
| scm.c | ||
| skbuff.c | ||
| sock.c | ||
| stream.c | ||
| sysctl_net_core.c | ||
| timestamping.c | ||
| user_dma.c | ||
| utils.c | ||