Quoting r. David Miller <[EMAIL PROTECTED]>:
> Subject: Re: Dropping NETIF_F_SG since no checksum feature.
>
> From: "Michael S. Tsirkin" <[EMAIL PROTECTED]>
> Date: Thu, 12 Oct 2006 21:12:06 +0200
>
> > Quoting r. David Miller <[EMAIL PROTECTED]>:
> > > Subject: Re: Dropping NETIF_F_SG since no
Sean Hefty <[EMAIL PROTECTED]> wrote on 10/12/2006 11:43:34 PM:
> > + if (!device->node_guid)
> > + return;
>
> I'm not sure that we even need this check anymore. All devices should
have a
> node_guid set. Maybe we can just remove it, rather than moving it up.
OK.
> Btw, were version
Hi Sean,
> Most of these seem to be a style issue. Should error handling be placed
at the
> end of the function, or within an if (error) type check? Keeping it at
the end
> of the function tends to make maintenance a little easier, otherwise, we
end up
> either duplicating the error handli
> There's lot of code doing :
To give examples on the above statement, the patches :
#2, #8, #10, #12 (all are one line, and one "go to" case patches)
#6, #9, #11, #13 (these are 2 line error handling, not 1 line)
#3 cleans up error handling to remove multiple goto err's.
#7 is an optimization in
Roland Dreier <[EMAIL PROTECTED]> wrote on 10/13/2006 12:08:25 AM:
> What's the motivation here?
There's lot of code doing :
ret = fn()
if (ret)
goto err;
return 0;
err:
one_line_cleanup;
return ret;
which could be easily made easier to c
From: "Michael S. Tsirkin" <[EMAIL PROTECTED]>
Date: Thu, 12 Oct 2006 21:12:06 +0200
> Quoting r. David Miller <[EMAIL PROTECTED]>:
> > Subject: Re: Dropping NETIF_F_SG since no checksum feature.
> >
> > Numbers?
>
> I created two subnets on top of the same pair infiniband HCAs:
I was asking fo
hi,
I am trying to use the APM support in openib kernel stack and facing
some problems.
here are the steps I follow:
1. first resolve both the path, primary and alternate path.
2. send REQ using:
active_param.primary_path = path;
active_param.alternate_path = alt_path;
ib_send_cm_req( cm_id, &
This improves switch's mcast tables dumping and eliminates multiple file
open/seek/close sequences. In one word - cleanup.
Signed-off-by: Sasha Khapyorsky <[EMAIL PROTECTED]>
---
osm/opensm/osm_mcast_mgr.c | 108 +---
1 files changed, 52 insertions(+), 56
Here is a patch to remove uDAPL scm provider from the build since it is no
longer needed nor
supported. This provider was merely a stop gap until uCMA was pushed into
kernel.
Tziporet, can you get this change into OFED 1.1?
Signed-off by: Arlin Davis [EMAIL PROTECTED]
Index: doc/dat.conf
>I'm not sure -- who are the customers for this? What do they want to
>do really?
The main customers I'm aware of are MPI developers. For direct IB multicast
usage, the national labs are converting some of their MPI algorithms to use IB
multicast. Their current approach is to create a multicast
What's the motivation here? Lots of these patches look like a step
backwards, for example this:
> -if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
> -goto out;
> -
> -if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
> -cma_exch(id_pr
Sean> On this same thought, do you have an idea of an interface
Sean> that you'd accept to export raw IB multicast support up to
Sean> userspace?
I'm not sure -- who are the customers for this? What do they want to
do really?
- R.
___
ope
Quoting r. Sean Hefty <[EMAIL PROTECTED]>:
> Subject: RE: [openib-general] [RFC] [PATCH 2/7] ib_multicast 2.6.20: add
> ib_multicast module to track join requests from the same port
>
> >Another comment that I'd like not to get in the noise is that we need
> >to handle the full set of SA queries,
>Another comment that I'd like not to get in the noise is that we need
>to handle the full set of SA queries, not just EQ.
I think that functionality can be added separately, but ib_multicast is only
intended to handle Set / Delete methods. Get and GetTable methods would still
go through the ib_s
Roland Dreier wrote:
> We want a way for unprivileged userspace to be able to use multicast.
> Usually I say "just use a privileged daemon in userspace" but I think
> in this case we actually need coordination between the kernel and
> userspace to track _all_ multicast joins, so it does make sense
Quoting r. David Miller <[EMAIL PROTECTED]>:
> Subject: Re: Dropping NETIF_F_SG since no checksum feature.
>
> From: "Michael S. Tsirkin" <[EMAIL PROTECTED]>
> Date: Wed, 11 Oct 2006 23:23:39 +0200
>
> > With my patch, there is a huge performance gain by increasing MTU to 64K.
> > And it seems th
Quoting r. Sean Hefty <[EMAIL PROTECTED]>:
> Subject: Re: [openib-general] [RFC] [PATCH 2/7] ib_multicast 2.6.20: add
> ib_multicast module to track join requests from the same port
>
> Michael S. Tsirkin wrote:
> > Ah. I get it.
> > If my callback will return error, I must make sure I won't dest
>But unlike the sa races which were unfixable without API changes,
>here users can synchronize the removal of the mc object.
>So I think what you describe is a user error.
The user can ensure that an id is only destroyed once. What they cannot ensure
is whether their callback is still running.
>
Krishna Kumar wrote:
> Re-write all cma error cases to simplify code. Splitting
> it as multiple patches (one per routine) in case some are
> found not required (in which case, later ones may apply
> with a fuzz).
Most of these seem to be a style issue. Should error handling be placed at the
end
Michael S. Tsirkin wrote:
> Ah. I get it.
> If my callback will return error, I must make sure I won't destroy the cm_id.
> But this means that I don't get the protection on destroy that was
> checking that callbacks have all gone.
>
>
> So, let's solve it in the same way we did for sa?
Yes - I
Quoting r. Sean Hefty <[EMAIL PROTECTED]>:
> Subject: RE: [openib-general] [RFC] [PATCH 2/7] ib_multicast 2.6.20: add
> ib_multicast module to track join requests from the same port
>
> >But unlike the sa races which were unfixable without API changes,
> >here users can synchronize the removal of
Krishna Kumar wrote:
> @@ -2288,14 +2288,15 @@ static void cma_add_one(struct ib_device
> struct cma_device *cma_dev;
> struct rdma_id_private *id_priv;
>
> + if (!device->node_guid)
> + return;
I'm not sure that we even need this check anymore. All devices should ha
>But what is the issue? some kind of race?
If we look at just the ib_multicast patches as an example...
Calling ib_join_multicast allocates a struct ib_multicast that must be freed.
Here's the relevant portion of ipoib's join callback:
@@ -325,11 +328,10 @@ ipoib_mcast_sendonly_join_complete(int
Quoting r. Sean Hefty <[EMAIL PROTECTED]>:
> Subject: RE: [openib-general] [RFC] [PATCH 2/7] ib_multicast 2.6.20: add
> ib_multicast module to track join requests from the same port
>
> >But what is the issue? some kind of race?
>
> If we look at just the ib_multicast patches as an example...
>
Eitan Zahavi wrote:
> I disagree. If you sniff at the MAD level you can simply react to the
> lower level messages.
First, when designing this, I did consider using the MAD snooping ability, and
changing what could be done with snooping. However, the multicast handling is
not simply sniffing M
Quoting r. Sean Hefty <[EMAIL PROTECTED]>:
> The ib_cm and rdma_cm have the issue if a client uses the return value from
> the callback to destroy their cm_id's.
But what is the issue? some kind of race?
--
MST
___
openib-general mailing list
openib-
Michael S. Tsirkin wrote:
> Hmm, sorry, I forgot.
> Could you restate what the ib_cm/rdma_cm problem is, please?
> Shouldn't we solve that, too?
The ib_multicast API needs register/unregister calls to prevent module unload
races.
The ib_cm and rdma_cm have the issue if a client uses the return v
Hi.
What should happen if one opens the IB device, allocate resources and
close the device?
for example, if a user do the following operations in a loop:
ibv_get_device_list
in a loop:
ibv_open_device
ibv_alloc_pd
ibv_create_cq
ibv_close_device?
should the ib
john> Hi This is with respect to Mellanox HCAs. Is there a way to
john> bypass/disable MPT (Memory Protection Table) and MTT (Memory
john> Translation Table) lookups done by HCA? If yes which
john> function should be called from a module to achieve this.
ib_get_dma_mr() gives you a
http://openib.org/bugzilla/show_bug.cgi?id=263
--- Comment #9 from [EMAIL PROTECTED] 2006-10-12 07:54 ---
Created an attachment (id=62)
--> (http://openib.org/bugzilla/attachment.cgi?id=62&action=view)
Please test this patch - does the crash happen with it?
Interesting. As a test, S
Hi all,
I am back from vacation and found you waited with the release for me :-)
From a quick look at status mails I think we can do the official
release next week.
Please reply if there are still any blocking issues you have.
Also - please update all documents till end of Monday next week.
http://openib.org/bugzilla/show_bug.cgi?id=263
--- Comment #8 from [EMAIL PROTECTED] 2006-10-12 07:17 ---
ipoib_start_xmit() only checks at entry to see if the queue is stopped.
ipoib_start_xmit() could still unicast_arp_send() after a netif_stop_queue().
in ipoib_stop(), i guess th
http://openib.org/bugzilla/show_bug.cgi?id=263
--- Comment #7 from [EMAIL PROTECTED] 2006-10-12 06:07 ---
why is it necessary to ensure that ipoib_start_xmit()
isnt in the middle of some work?
--- You are receiving this mail because: ---
You are the assignee for the bug, o
http://openib.org/bugzilla/show_bug.cgi?id=263
--- Comment #6 from [EMAIL PROTECTED] 2006-10-12 06:00 ---
calling netif_stop_queue() doesnt immediately stop the transmit queue. it
might be necessary to take priv->tx_lock when calling netif_stop_queue() from
ipoib_stop() to ensure tha
OpenSM/osmtest.c: Fix float calculation in osmtest_stress_large_rmpp_pr
Signed-off-by: Hal Rosenstock <[EMAIL PROTECTED]>
---
Index: osmtest/osmtest.c
===
--- osmtest/osmtest.c (revision 9795)
+++ osmtest/osmtest.c (working copy)
Hi
This is with respect to Mellanox HCAs. Is there a way to bypass/disable MPT (Memory Protection Table) and MTT (Memory Translation Table) lookups done by HCA? If yes which function should be called from a module to achieve this.
Regards,
John T.
__
OpenSM/osm_link_mgr.c: Use leaf HOQLife for IB router ports
Signed-off-by: Hal Rosenstock <[EMAIL PROTECTED]>
---
Index: opensm/osm_link_mgr.c
===
--- opensm/osm_link_mgr.c (revision 9771)
+++ opensm/osm_link_mgr.c (workin
Eitan Zahavi wrote:
> Roland Dreier wrote:
>> Eitan> 1 that you know about. Others did not make it into the
>> Eitan> kernel but are quite productive to those running them.
>>
>> What are those others?
> CFS is an example.
Lustre o2ibnld is using RC only and from what i know is coded over
Sean Hefty wrote:
> Eitan Zahavi wrote:
>
>> So if it is then there is no problem sniffing it and refcounting.
>>
>
> The MADs cannot simply be sniffed and counted. MADs which affect the same
> multicast group should not always be sent. Join operations must be
> serialized
> against le
Roland Dreier wrote:
> Eitan> 1 that you know about. Others did not make it into the
> Eitan> kernel but are quite productive to those running them.
>
> What are those others?
>
CFS is an example.
> Eitan> Changing top API for ULPs and Clients is simpler to
> Eitan> implement but
Roland Dreier wrote:
> Eitan> If the tracking (ref counting) was done at the MAD level -
> Eitan> no change to IPoIB would have been required ...
>
> It doesn't seem very feasible to implement a complete local copy of
> the SA (in the kernel no less) so that we can allow unprivileged
> proc
Arlin Davis wrote:
> Or Gerlitz wrote:
>> I see now that the uDAPL CMA provider code uses the MTU 1:1 as
>> returned by the SM in the path, so if the env is made of the Mellanox
>> PCI-X HCA there can be big BW drop, etc... we have discussed that.
>> I wonder how are you overcoming this when run
Roland Dreier wrote:
> Or> Its not a rush its a move for enabling user space code that
> Or> can offload IP Multicast. We have a library doing that which
> Or> is coded over the gen1 stack and is now in porting for the
> Or> gen2 stack.
>
> OK -- I would like to hear your experienc
43 matches
Mail list logo