From: Ethan Jackson
Previously, we had a separate flow_dumper thread that fetched flows from
the datapath to distribute to revalidator threads. This patch takes the
logic for dumping and pushes it into the revalidator threads, resulting
in simpler code with similar performance to the current code
Improve random distribution for an hmap with a small number of nodes
with the expense of the increased cpu cost.
It would be a fair trade-off because the situation is rather common
for bond, which is currently the only consumer of this API in tree.
Signed-off-by: YAMAMOTO Takashi
---
lib/hmap.c
Applied, thx very much for the review.
On Mon, Apr 21, 2014 at 9:10 PM, Ethan Jackson wrote:
> LGTM as well
>
> On Mon, Apr 21, 2014 at 9:09 PM, Alex Wang wrote:
> > Commit 1f8675481e (ofproto-dpif-upcall: Fix ovs-vswitchd crash.)
> > directly copied the udpif_set_threads() logic to udpif_stop
LGTM as well
On Mon, Apr 21, 2014 at 9:09 PM, Alex Wang wrote:
> Commit 1f8675481e (ofproto-dpif-upcall: Fix ovs-vswitchd crash.)
> directly copied the udpif_set_threads() logic to udpif_stop_threads()
> and udpif_start_threads(). In fact, this was erroneous and caused
> unittest failures.
>
> T
Commit 1f8675481e (ofproto-dpif-upcall: Fix ovs-vswitchd crash.)
directly copied the udpif_set_threads() logic to udpif_stop_threads()
and udpif_start_threads(). In fact, this was erroneous and caused
unittest failures.
This commit fixes the above issue by correcting the checks in
udpif_stop_thre
Thanks for patching this. I'll hold off on rebasing until this is pushed.
On 22 April 2014 15:28, Ethan Jackson wrote:
> Acked-by: Ethan Jackson
>
>
> On Mon, Apr 21, 2014 at 8:25 PM, Alex Wang wrote:
> > Commit 1f8675481e (ofproto-dpif-upcall: Fix ovs-vswitchd crash.)
> > directly copied the
Acked-by: Ethan Jackson
On Mon, Apr 21, 2014 at 8:25 PM, Alex Wang wrote:
> Commit 1f8675481e (ofproto-dpif-upcall: Fix ovs-vswitchd crash.)
> directly copied the udpif_set_threads() logic to udpif_stop_threads()
> and udpif_start_threads(). In fact, this was erroneous and caused
> unittest fa
Commit 1f8675481e (ofproto-dpif-upcall: Fix ovs-vswitchd crash.)
directly copied the udpif_set_threads() logic to udpif_stop_threads()
and udpif_start_threads(). In fact, this was erroneous and caused
unittest failures.
This commit fixes the above issue by correcting the checks in
udpif_stop_thre
Sorry, seemed my patch caused more issues,
There is a typo in the udpif_stop_threads() function,
There are also unit test failures,
I'm addressing it,
On Mon, Apr 21, 2014 at 6:19 PM, Ethan Jackson wrote:
> Acked-by: Ethan Jackson
>
>
> On Mon, Apr 21, 2014 at 6:13 PM, Alex Wang wrote:
>
Acked-by: Ethan Jackson
On Mon, Apr 21, 2014 at 6:13 PM, Alex Wang wrote:
> On current master, caller of udpif_set_threads() can pass 0 value
> on n_handlers and n_revalidators to delete all handler and revalidator
> threads.
>
> After commit 9a159f748866 (ofproto-dpif-upcall: Remove the dispat
On current master, caller of udpif_set_threads() can pass 0 value
on n_handlers and n_revalidators to delete all handler and revalidator
threads.
After commit 9a159f748866 (ofproto-dpif-upcall: Remove the dispatcher
thread.), udpif_set_threads() also calls the dpif_handlers_set() with
the 0 value
>
> On Apr 20, 2014, at 7:49 PM, YAMAMOTO Takashi wrote:
>
>>> +hash = mhash_finish(hash, 13); /* No need to match byte length
>>> here. */
>>
>> is it worth being special?
>>
>
> I知 not exactly sure what you ask here, but in general, the length parameter
> is used to distinguish be
Thanks!
On Apr 21, 2014 5:08 PM, "Alex Wang" wrote:
> Hey Ben,
>
> Yeah, this is introduced by my changes. I'm working on a fix.
>
> Alex Wang,
> Thx,
>
>
> On Mon, Apr 21, 2014 at 4:48 PM, Ben Pfaff wrote:
>
>> Yes, I'm seeing these too on current master. I bisected it to the
>> following com
Hey Ben,
Yeah, this is introduced by my changes. I'm working on a fix.
Alex Wang,
Thx,
On Mon, Apr 21, 2014 at 4:48 PM, Ben Pfaff wrote:
> Yes, I'm seeing these too on current master. I bisected it to the
> following commit:
>
> commit 9a159f748866136a126474137604b2b020bd3574
> Auth
Yes, I'm seeing these too on current master. I bisected it to the
following commit:
commit 9a159f748866136a126474137604b2b020bd3574
Author: Alex Wang
Date: Wed Feb 26 23:03:24 2014 -0800
ofproto-dpif-upcall: Remove the dispatcher thread.
With the foundation laid i
From: Ethan Jackson
Previously, we had a separate flow_dumper thread that fetched flows from
the datapath to distribute to revalidator threads. This patch takes the
logic for dumping and pushes it into the revalidator threads, resulting
in simpler code with similar performance to the current code
Commits 1579cf677fc (dpif-linux: Implement the API functions to allow
multiple handler threads read upcall.) causes the ovs-vswitchd crash in
the following scenario:
- when udpif_flush() is called, it sets both n-handlers and n-revalidators
to zero. and it calls the dpif_handlers_set() to refre
Thx Andy for providing the core file. It is very helpful!
The root cause is that:
1. when udpif_flush() is called, it sets both n-handlers and n-revalidators
to 0
2. in refresh_channels() (dpif-linux.c), though the new 'n_handlers = 0',
we still try
dpif->handlers = xzalloc(n_handlers * size
Thanks a lot for great reviews from Jesse and Pravin. Pushed both
patches with the typos fixed.
On Mon, Apr 21, 2014 at 9:13 AM, Jesse Gross wrote:
> On Sun, Apr 20, 2014 at 11:05 PM, Andy Zhou wrote:
>> diff --git a/include/linux/openvswitch.h b/include/linux/openvswitch.h
>> index 1aa9158..dd2
Hey Andy,
Thx for the fix. Could you provide more info on how to reproduce it?
maybe the backtrace?
I expect vport_del_channels() to be called only when 'port_idx >=
dpif->uc_array_size' or dpif->handlers[0].channels is non-NULL.
I want to know more about the context.
Thx
Alex Wang,
On Sun,
On Sat, Apr 19, 2014 at 8:30 PM, YAMAMOTO Takashi
wrote:
>> ovsdb-server on windows crashes without it.
>>
>> Signed-off-by: Gurucharan Shetty
>
> why is it necessary on windows?
I see why this patch is not correct. There is a dynamic initialization
of the mutex in netdev_initialize().
For the
On Sun, Apr 20, 2014 at 2:19 AM, Rami Rosen wrote:
> This patch removes the new_vport() forward declaration in datapath.c
> as it is not needed.
>
> Signed-off-by: Rami Rosen
Applied.
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mail
On Sun, Apr 20, 2014 at 11:06 PM, Andy Zhou wrote:
> Recirculation implementation for Linux kernel data path.
>
> Signed-off-by: Andy Zhou
Acked-by: Jesse Gross
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On Sun, Apr 20, 2014 at 11:05 PM, Andy Zhou wrote:
> diff --git a/include/linux/openvswitch.h b/include/linux/openvswitch.h
> index 1aa9158..dd2cceb 100644
> --- a/include/linux/openvswitch.h
> +++ b/include/linux/openvswitch.h
> @@ -312,7 +312,8 @@ enum ovs_key_attr {
> OVS_KEY_ATTR_TUNNE
On Apr 20, 2014, at 7:54 PM, YAMAMOTO Takashi wrote:
> do these inlining yield measurable improvements?
>
I have not measured this recently, but I’ll report back with later on this.
> YAMAMOTO Takashi
Jarno
___
dev mailing list
dev@openvswitch.o
On Apr 20, 2014, at 7:49 PM, YAMAMOTO Takashi wrote:
>> +hash = mhash_finish(hash, 13); /* No need to match byte length
>> here. */
>
> is it worth being special?
>
I’m not exactly sure what you ask here, but in general, the length parameter is
used to distinguish between inputs tha
Hi David,
On Thu, Mar 27, 2014 at 03:29:56PM -0400, David Miller wrote:
> From: Zoltan Kiss
> Date: Wed, 26 Mar 2014 22:37:45 +
>
> > skb_zerocopy can copy elements of the frags array between skbs, but it
> > doesn't
> > orphan them. Also, it doesn't handle errors, so this patch takes care
JOB ID: EU/DCL/00867-13
ATTN:
Complements of the season; We want to inform you that we are hiring candidates that will work with Disney Cruise Line [Australia] under the job Identification Number EU/DCL/00867-13. The total recruitment will be 205 applicants. Minimum Age requirements is 18-y
28 matches
Mail list logo
