Roland Dreier <[EMAIL PROTECTED]> wrote on 11/14/2005 11:47:13 AM:
> Pradeep> I am trying to use copy_from_user()/copy_to_user of data
> Pradeep> structures that contains pointers.
>
> If you are defining a new interface, then the simplest thing is not to
> do that: always put pointers i
On Mon, Nov 14, 2005 at 09:54:28PM +0200, Eitan Zahavi wrote:
> Hi Troy
>
> Try to move aside your /lib/tls directory and see if you still get these
> crashes.
> We have issues with TLS pthread and glibc
We still have issues with -maxsmps=8. And no, running with maxsmps=1 is
not an option on thi
I am trying to compile some of the userspace utilities as 64-bit apps on a ppc64 machine with a
sles9sp2 distribution.
I am getting the following compile errors (appended below). Yes, I am using some older bits, but I
do not think that is the issue here.
I have exported LDFLAGS=-L /lib64 -m64 a
I have initial RPM support for both of these releases available for
use/testing. For Fedora Core 4, I didn't compile a new kernel since the
current FC4 kernel is 2.6.14 based and includes the upstream Infiniband
support. For RHEL4 I obviously compiled a new kernel, but it used the
code pulled
Eitan> Hi Roland, Now I get it ! In the port info record there is
Eitan> a field named LocalPortNumber. This field is NOT the port
Eitan> number the data is about. It is the port number the packet
Eitan> of the query came from. (see table 145 p823 l-38). When
Eitan> OpenSM ob
In absence of any protocol level
ack (and regardless of protocol level ack), it is the application which has
to implement its own reliability. RDS becomes a passive channel passing
packet back and forth including duplicate packets. The responsibility then
shifts to the a
Hi Roland,
Now I get it !
In the port info record there is a field named LocalPortNumber.
This field is NOT the port number the data is about.
It is the port number the packet of the query came from.
(see table 145 p823 l-38).
When OpenSM obtains the PortInfo associated with that particular port
[EMAIL PROTECTED] wrote:
> Kanevsky, Arkady wrote:
>> Which entity is responsible to "use" the proposed protocol is an
>> interesting one. I was assuming that this will be CM. After all the
>> proposed protocol is CM extension protocol. But it can be another
>> entity module between CM and ULP.
>
Are you referring to SCinet ? It is definitely running off HCA port 1.
-- Hal
From: [EMAIL PROTECTED] on behalf of Roland Dreier
Sent: Tue 11/15/2005 11:04 AM
To: Eitan Zahavi
Cc: openib-general@openib.org
Subject: Re: [openib-general] SRP device management clie
Kanevsky, Arkady wrote:
Which entity is responsible to "use" the proposed protocol is an
interesting one.
I was assuming that this will be CM. After all the proposed protocol is
CM extension
protocol.
But it can be another entity module between CM and ULP.
The use of a reserved bit in the CM me
I just noticed that the host port that the SM is running on is
connected to switch port 2.
What seems to be happening is that all of the switch's ports (except
port 0) are seen as having local port number 2 in the actual PortInfo
attribute information, even though the PortNum field in the SA recor
Eitan> Could you dump out the content of the PortRecords that you
Eitan> get as response?
They look like valid records for switch ports, except the local port
number field doesn't match the port number field in the SA record
identifier wrapper.
- R.
__
Eitan> You should not get more then one SA header. I assumed you
Eitan> are doing GetTable of PortInfoRecord. If this is correct
Eitan> you should only get one SA header in the resulting RMPP
Eitan> (reassembled MAD).
Yes, I only have one SA header. I just meant the SA wrapper of
You should not get more then one SA header.
I assumed you are doing GetTable of PortInfoRecord. If this is correct
you should only get one SA header in the resulting RMPP (reassembled
MAD).
EZ
Eitan Zahavi
Design Technology Director
Mellanox Technologies LTD
Tel:+972-4-9097208
Fax:+972-4-9593245
Hi Roland,
If you only got single 24port switch you should only see 1 record with
base lid = 0 and port num = 2. But maybe we have a bug not comparing
port num.
On our test today we have seen only one record for port 2 from each
switch (we had two switches so got 2 recodrs).
Could you dump out
Hal> Are those the only component fields which were zeroed out ?
No, for example the capability mask is all 0 as well.
It seems to be something a little bit more complicated. In my test
fabric, which has a single 24-port switch with hosts connected to
both port 1 and port 2 (of the switch),
Yael> Hello Roland, When turning on only the comp_mask for the
Yael> local_port_num you will get all relevant PortInfo records
Yael> from the switches. These records do have many fields zeroed
Yael> out (e.g subnet_prefix), but they are still valid records.
Yael> Is this what y
Quoting r. Michael S. Tsirkin <[EMAIL PROTECTED]>:
> Subject: [PATCH] mthca: fix qp max_send/recv_sge calculation
>
> Roland, I think I see a problem in mthca, where qp capability values
> we return arent safe.
> How does the following look (compile tested only)?
This is tested now, please review
Date: Mon, 14 Nov 2005 15:16:51 -0800
To: [EMAIL PROTECTED], openib-general@openib.org,
[EMAIL PROTECTED], Eric Lantz, [EMAIL PROTECTED]
From: Bill Boas <[EMAIL PROTECTED]>
Subject: Invitation to OpenIB BOF at SC05 Wednesday 11-12 Room 205
in Convention Center
Cc: [EMAIL PROTECTED], [EMAIL PR
The goal that this proposal is to provide underpinning for common RDMA
transport CM.
Thus, the API ULP (both user space and kernel space) use socket
addressing.
For ULP addressing this means 5 tuple: protocol, src IP addr, src port,
dst IP addr,
and dst port. Port is 16 bit entity.
The proposal j
Roland,
Just to close the loop on this:
Are those the only component fields which were zeroed out ?
Thanks.
-- Hal
From: Eitan Zahavi [mailto:[EMAIL PROTECTED]
Sent: Tue 11/15/2005 9:41 AM
To: Hal Rosenstock; Roland Dreier
Cc: openib-general@openib.org
Sub
At 12:49 PM 11/14/2005, Nitin Hande wrote:
Michael Krause wrote:
At 01:01 PM 11/11/2005, Nitin
Hande wrote:
Michael Krause wrote:
At 10:28 AM 11/9/2005, Rick
Frank wrote:
Yes, the application is
responsible for detecting lost msgs at the application level - the
transport can not do this.
RDS do
At 12:49 PM 11/14/2005, Nitin Hande wrote:
Michael Krause wrote:
At 01:02 PM 11/11/2005, Ranjit
Pandit wrote:
On 11/11/05, Michael Krause
<[EMAIL PROTECTED]> wrote:
> Please clarify the following which was in the document provided by
Oracle.
>
> On page 3 of the RDS document, under the section "R
I think Yael figured it out:
Looking at Roland's code it seems like it will not filter out the
PortRecords coming from switch physical ports. So actually he gets many
records that all have base lid = 0 and gid = 0 from these ports...
I assume this is the case. There is no trivial way to know from
Hi,
It's not necessarily an RMPP bug. A lot of the port 2s on SCinet are not
plugged in.
-- Hal
From: [EMAIL PROTECTED] on behalf of Roland Dreier
Sent: Tue 11/15/2005 3:27 AM
To: Eitan Zahavi
Cc: openib-general@openib.org
Subject: Re: [openib-general] SRP de
Quoting r. Roland Dreier <[EMAIL PROTECTED]>:
> Subject: Re: ipoib oops
>
> Sorry I haven't been able to look at this immediately, since I've been
> busy with SC05-related stuff.
>
> I hope to sit down and think about this in detail tomorrow...
>
> - R.
>
OK.
Meanwhile, there's another possib
Hello Roland,
When turning on only the comp_mask for the local_port_num you will get
all
relevant PortInfo records from the switches.
These records do have many fields zeroed out (e.g subnet_prefix), but
they are
still valid records.
Is this what you are seeing?
Thanks,
Yael
-Original Messag
Thanks. We will try and reproduce it here.
Eitan Zahavi
Design Technology Director
Mellanox Technologies LTD
Tel:+972-4-9097208
Fax:+972-4-9593245
P.O. Box 586 Yokneam 20692 ISRAEL
> -Original Message-
> From: Roland Dreier [mailto:[EMAIL PROTECTED]
> Sent: Tuesday, November 15, 2005 10:
Have __srp_get_tx_iu() fail if the target port's request limit will
not allow the initiator to post a send. This avoids continuing on and
posting a receive, and then failing to post a corresponding send. If
that happens, then the initiator will end up with an extra receive
posted, and if this hap
Increase SRP max_luns to 512 to match the kernel's default, since SRP
storage targets can have lots of LUNs and the SRP initiator itself
doesn't have any particular limit.
Signed-off-by: Roland Dreier <[EMAIL PROTECTED]>
---
drivers/infiniband/ulp/srp/ib_srp.c |2 ++
drivers/infiniband/ulp/
Responder resources are only required to handle RDMA reads and atomic
operations, not RDMA writes. So the driver should allow RDMA writes
even if responder resources are set to 0. This is especially
important for the UC transport -- with the old code, it was impossible
to enable RDMA writes for U
Sorry I haven't been able to look at this immediately, since I've been
busy with SC05-related stuff.
I hope to sit down and think about this in detail tomorrow...
- R.
___
openib-general mailing list
openib-general@openib.org
http://openib.org/mailman/
Roland> Quite easy in my setup -- it seems to happen every time on
Roland> my fabric when I do a get table for PortInfoRecords with
Roland> local port num 2.
And running ibsrpdm on the scinet fabric at SC'05 I see hundreds of
PortInfoRecords with a base LID of 0...
- R.
_
Eitan> Yes this is correct we never got requested for that query.
Eitan> If you are only interested in obtaining the guid of the
Eitan> port you can simply use NodeInfoRecord and you get the guid
Eitan> in the NodeInfo. But you probably know that. Is there
Eitan> anything more
Hi!
1. Did you strip it?
# ls -l /usr/local/bin/opensm
-rwxr-xr-x 1 root root 2124734 Nov 15 10:07 /usr/local/bin/opensm
# strip /usr/local/bin/opensm
# ls -l /usr/local/bin/opensm
-rwxr-xr-x 1 root root 333024 Nov 15 10:22 /usr/local/bin/opensm
2. Compile with -Os: edit Makefile in
/usr/src/o
Roland Dreier wrote:
The opensm issues I saw were:
- GUIDInfoRecord SA queries are not implemented (I think), so by
default my code does a (non-compliant) SM class query to get ports'
GUIDs.
Yes this is correct we never got requested for that query.
If you are only interested in obtaini
36 matches
Mail list logo