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. 22 5月, 2014 5 次提交
    • J
      wimax/i2400m: make return of 0 explicit · aff4b974
      由 Julia Lawall 提交于 
      Delete unnecessary local variable whose value is always 0 and that hides
the fact that the result is always 0.

A simplified version of the semantic patch that fixes this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
@r exists@
local idexpression ret;
expression e;
position p;
@@

-ret = 0;
... when != ret = e
return
- ret
+ 0
  ;
// </smpl>
Signed-off-by: NJulia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      aff4b974
    • S
      enic: Add support for adaptive interrupt coalescing · 7c2ce6e6
      由 Sujith Sankar 提交于 
      This patch adds support for adaptive interrupt coalescing.

For small pkts with low pkt rate, we can decrease the coalescing interrupt
dynamically which decreases the latency. This however increases the cpu
utilization. Based on testing with different coal intr and pkt rate we came up
with a table(mod_table) with rx_rate and coalescing interrupt value where we
get low latency without significant increase in cpu. mod_table table stores
the coalescing timer percentage value for different throughputs.

Function enic_calc_int_moderation() calculates the desired coalescing intr timer
value. This function is called in driver rx napi_poll. The actual value is set
by enic_set_int_moderation() which is called when napi_poll is complete. i.e
when we unmask the rx intr.

Adaptive coal intr is support only when driver is using msix intr. Because
intr is not shared.

Struct mod_range is used to store only the default adaptive coalescing intr
value.

Adaptive coal intr calue is calculated by

timer = range_start + ((rx_coal->range_end - range_start) *
		       mod_table[index].range_percent / 100);

rx_coal->range_end is the rx-usecs-high value set using ethtool.
range_start is rx-usecs-low, set using ethtool, if rx_small_pkt_bytes_cnt is
greater than 2 * rx_large_pkt_bytes_cnt. i.e small pkts are dominant. Else its
rx-usecs-low + 3.

Cc: Christian Benvenuti <benve@cisco.com>
Cc: Neel Patel <neepatel@cisco.com>
Signed-off-by: NSujith Sankar <ssujith@cisco.com>
Signed-off-by: NGovindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      7c2ce6e6
    • M
      vxge: Use time_before() · f6e92d10
      由 Manuel Schölling 提交于 
      To be future-proof and for better readability the time comparisons are modified
to use time_before() instead of plain, error-prone math.
Signed-off-by: NManuel Schölling <manuel.schoelling@gmx.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      f6e92d10
    • H
      ieee802154: Introduce the use of the managed version of kzalloc · 12b5c38f
      由 Himangi Saraogi 提交于 
      This patch moves data allocated using kzalloc to managed data allocated
using devm_kzalloc and cleans now unnecessary kfrees in probe and remove
functions. An explicit linux/device.h include is added to make sure
the devm_*() routine declarations are unambiguously available.

The following Coccinelle semantic patch was used for making the change:

@platform@
identifier p, probefn, removefn;
@@
struct platform_driver p = {
  .probe = probefn,
  .remove = removefn,
};

@prb@
identifier platform.probefn, pdev;
expression e, e1, e2;
@@
probefn(struct platform_device *pdev, ...) {
  <+...
- e = kzalloc(e1, e2)
+ e = devm_kzalloc(&pdev->dev, e1, e2)
  ...
?-kfree(e);
  ...+>
}

@rem depends on prb@
identifier platform.removefn;
expression e;
@@
removefn(...) {
  <...
- kfree(e);
  ...>
}
Signed-off-by: NHimangi Saraogi <himangi774@gmail.com>
Acked-by: NJulia Lawall <julia.lawall@lip6.fr>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      12b5c38f
    • X
      net-tun: restructure tun_do_read for better sleep/wakeup efficiency · 9e641bdc
      由 Xi Wang 提交于 
      tun_do_read always adds current thread to wait queue, even if a packet
is ready to read. This is inefficient because both sleeper and waker
want to acquire the wait queue spin lock when packet rate is high.

We restructure the read function and use common kernel networking
routines to handle receive, sleep and wakeup. With the change
available packets are checked first before the reading thread is added
to the wait queue.

Ran performance tests with the following configuration:

 - my packet generator -> tap1 -> br0 -> tap0 -> my packet consumer
 - sender pinned to one core and receiver pinned to another core
 - sender send small UDP packets (64 bytes total) as fast as it can
 - sandy bridge cores
 - throughput are receiver side goodput numbers

The results are

baseline: 731k pkts/sec, cpu utilization at 1.50 cpus
 changed: 783k pkts/sec, cpu utilization at 1.53 cpus

The performance difference is largely determined by packet rate and
inter-cpu communication cost. For example, if the sender and
receiver are pinned to different cpu sockets, the results are

baseline: 558k pkts/sec, cpu utilization at 1.71 cpus
 changed: 690k pkts/sec, cpu utilization at 1.67 cpus
Co-authored-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NXi Wang <xii@google.com>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      9e641bdc
  3. 21 5月, 2014 16 次提交
  5. 19 5月, 2014 12 次提交
  7. 17 5月, 2014 7 次提交