Quoting Pradeep Satyanarayana [EMAIL PROTECTED]:
Subject: Re: IPOIB CM (NOSRQ)[PATCH V4] patch for review
Michael S. Tsirkin wrote:
Quoting Pradeep Satyanarayana [EMAIL PROTECTED]:
Subject: Re: IPOIB CM (NOSRQ)[PATCH V4] patch for review
Michael S. Tsirkin wrote:
Quoting Pradeep
Hi Vu,
Thanks for your replay. But I have got something wrong when using two
initiators.
Two initiators and one target are at three separated nodes. The first
initiator connected to the target correctly. However, the second one
was aborted 1 minute after its login, and then required to
This email was generated automatically, please do not reply
Common build parameters: --with-ipoib-mod --with-sdp-mod --with-srp-mod
--with-user_mad-mod --with-user_access-mod --with-mthca-mod --with-core-mod
--with-addr_trans-mod --with-rds-mod --with-cxgb3-mod
Passed:
Passed on i686 with
Hi Hal,
Hal Rosenstock wrote:
Hi Yevgeny,
On Mon, 2007-05-14 at 10:07, Yevgeny Kliteynik wrote:
Hi Hal.
[snip]
Date: 03/30/2007 12:24:12 AM
OpenSM: Handle conf file open failures better
diff --git a/osm/opensm/osm_subnet.c b/osm/opensm/osm_subnet.c
index 46315a5..746fbd1 100644
---
Sasha Khapyorsky wrote:
On 14:20 Tue 15 May , Yevgeny Kliteynik wrote:
I think that the message should appear when OpenSM *does* find cached
option file, and no message should appear when such file wasn't found
(which is the most common use case).
AFAIK OpenSM which used in the labs'
Hi,
Will apps built with OFED 1.1 verbs.h run on an OFED 1.2 install ?
-Bill___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit
SERAH ALI ASSOCIATES
NOTARY PUBLIC CORPERATE ATTORNEY
II6 WINCHESEA STREET , LONDON.
UK.
ATTN
RE-ESTATE OF LATE ABDULAZEE AHMED HAMZA HABIB OF IRAQ
We are attorneys and executors of the estate of late Ahmed Hamza Habib of
Iraq. Who is the
Quoting Roland Dreier [EMAIL PROTECTED]:
Subject: Re: [PATCH] libmlx4: WQE shift calculation
the code that calculates WQ size is quite different between kernel and
user. I think that writing the code in a way that will allow to copy it
as is between kernel and user is in place. Would
On Tue, 2007-05-15 at 07:26 -0700, Roland Dreier wrote:
the code that calculates WQ size is quite different between kernel and
user. I think that writing the code in a way that will allow to copy it
as is between kernel and user is in place. Would like me to send such a
patch?
First I
On Tue, 2007-05-15 at 10:05, Hal Rosenstock wrote:
As discussed last month, the following changes have now been made for IB
management (master branch of my management git tree):
In order to better match package names, the following directory names have
been changed from-to:
osm-opensm
Here are the patches as you asked.
I changed the code to use strncmp as suggested by Sasha.
Philippe
Hal Rosenstock a écrit :
Hi Philippe,
On Mon, 2007-05-14 at 11:26, Philippe Gregoire wrote:
This time , with the definitive patch (sorry)
Can you resubmit this with a S-O-B line ?
First I should add the case that triggered his patch: the userspace code
calculated a smaller buffer size than the kernel code, which caused
get_user_pages() to fail since part of the buffer did not belong to the
process's address space.
OK, in this case it seems the bug is in the kernel
Thanks for the reminder. I put them in the wrong folder but I think I
found them now.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit
Ofed 1.2 rc3 is running IPoIB slower that rc2 or earlier
I use to get ~800MB/sec no tuning and now I get ~650MB/sec ??
Ideas??
thanks
At 09:44 AM 5/14/2007, Moni Levy wrote:
On 5/14/07, Tziporet Koren
mailto:[EMAIL PROTECTED][EMAIL PROTECTED] wrote:
Major limitations and known issues:
...and this for libmlx4?
diff --git a/src/mlx4.h b/src/mlx4.h
index c4d389f..1e92b88 100644
--- a/src/mlx4.h
+++ b/src/mlx4.h
@@ -65,6 +65,20 @@
# define wmb() mb()
#endif
+#ifndef wc_wmb
+
+#if defined(__i386__)
+#define wc_wmb() asm volatile(lock; addl $0,0(%%esp) ::: memory)
+#elif
memcpy(ctx-bf_page + ctx-bf_offset, ctrl, align(size * 16,
64));
By the way, why are we aligning the size of the WQE we copy to 64
bytes? I copied this from Jack's code but I don't see anything that
requires it. We already have:
if (nreq == 1 inl size 1 size
Quoting Roland Dreier [EMAIL PROTECTED]:
Subject: Re: libmlx4 wc flash
By the way, I just re-checked and it seems that WC support first
appeared in Pentium II systems. So I think we should be able to
use sfence if WC is enabled.
That's actually doubly wrong: WC support was added in
Quoting Roland Dreier [EMAIL PROTECTED]:
Subject: Re: libmlx4 wc flash
memcpy(ctx-bf_page + ctx-bf_offset, ctrl, align(size * 16,
64));
By the way, why are we aligning the size of the WQE we copy to 64
bytes? I copied this from Jack's code but I don't see anything that
so what it comes down to, is that if we assume that WC
will *only be enabled through PAT* then it's safe to use sfence
in this case. Right?
I don't think we can really make any assumptions about what
instructions a 32-bit x86 processor has available. Who knows what
wacky stuff VIA or
Quoting Roland Dreier [EMAIL PROTECTED]:
Subject: Re: libmlx4 wc flash
...and this for libmlx4?
diff --git a/src/mlx4.h b/src/mlx4.h
index c4d389f..1e92b88 100644
--- a/src/mlx4.h
+++ b/src/mlx4.h
@@ -65,6 +65,20 @@
# define wmb() mb()
#endif
+#ifndef wc_wmb
+
+#if
Once the processor has started to evict data from the WC buffer into system
memory, it will make a bus-transaction style decision based on how much of
the
buffer contains valid data. If the buffer is full (for example, all bytes are
valid) the processor will execute a burst-write
memcpy(ctx-bf_page + ctx-bf_offset, ctrl, align(size * 16,
64));
By the way, if we have an SQ with 32-byte WQEs, and we do blueflame
from a WQE at the end of the buffer, we might end up reading off the
end of the buffer. Not very likely, I guess.
I wonder if memset(,0,) for the
I don't think it works this way: if PAT is programmed to UC,
I think you get UC access with movntq. No?
You're right -- I misremembered what the non-temporal stuff does, but
I just checked and the manual says:
The memory type of the region being written to can override the
Quoting Roland Dreier [EMAIL PROTECTED]:
Subject: Re: libmlx4 wc flash
memcpy(ctx-bf_page + ctx-bf_offset, ctrl, align(size
* 16, 64));
By the way, if we have an SQ with 32-byte WQEs, and we do blueflame
from a WQE at the end of the buffer, we might end up reading
Do you have any idea how WC works on ppc? Is the lwsync instruction
necessary/sufficient to flush WC buffers?
Donnu yet. I hear Jack here plans to start looking at ppc RSN.
--
MST
___
general mailing list
general@lists.openfabrics.org
So it looks like we can start by:
1. Change the user code to pass the size to kernel
2. Fix calculations in kernel.
Would like me to send patches or do you prefer to add your code? If you
prefer to code this can you tell when that would be?
It would be great if you implement it.
I would like to propose a better interface. What if there were a generic
DM agent in the kernel that provided an API for target devices (kernel and
user) to register IOC's with it? It might look something like this:
A generic DM makes sense.
There are existing interfaces / implementations
27 matches
Mail list logo
