Fix posting long WQE lists for SRQ.
Signed-off-by: Michael S. Tsirkin [EMAIL PROTECTED]
Index: src/userspace/libmthca/src/srq.c
===
--- src/userspace/libmthca/src/srq.c(revision 4016)
+++ src/userspace/libmthca/src/srq.c
Thanks -- I had basically the same thing in my local working directory
but forgot to commit it.
- R.
___
openib-general mailing list
openib-general@openib.org
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit
At 02:09 PM 11/9/2005, Greg Lindahl wrote:
On Wed, Nov 09, 2005 at
01:57:06PM -0800, Michael Krause wrote:
What you indicate above is that RDS
will implement a resync of the two sides of the association to
determine
what has been successfully sent.
More accurate to say that it could
Move the computation of QP capabilities (max scatter/gather entries,
max inline data, etc) into the kernel, and have the uverbs module
return the values as part of the create QP response. This keeps
precise knowledge of device limits in the low-level kernel driver.
This requires an ABI bump, so
Now that ib_umad uses the new MAD sending interface, it no longer
needs its own L_Key. So just delete the array of MRs that it keeps.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/core/user_mad.c | 29 -
1 files changed, 4 insertions(+),
In Tavor mode, when posting a long list of receive work requests, a
doorbell must be rung every 256 requests. Add code to do this when
required.
Signed-off-by: Michael S. Tsirkin [EMAIL PROTECTED]
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/hw/mthca/mthca_qp.c |
Handle case where prod_index has wrapped around and become less than
cq-cons_index by checking that their difference as a signed int is
positive rather than comparing directly.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/hw/mthca/mthca_cq.c | 16 ++--
1
The previous umad deadlock fix left ib_umad_kill_port() still
vulnerable to deadlocking. This patch fixes that by downgrading our
lock to a read lock when we might end up trying to reacquire the lock
for reading.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
Hi,
I'm working with Cluster File Systems on lustre network drivers, including IB
drivers for the Voltaire, Infinicon and Topspin stacks. These are kernel
drivers which use RC QPs with VERBS for small message queueing and RDMA for
bulk transfers.
We're obviously looking at OpenIB Gen2, and I
Mike Krause wrote in response to Greg Lindahl:
If it is to be reasonably robust, then RDS should be required to
support
the resync between the two sides of the communication. This aligns
with the
stated objective of implementing reliability in one location in
Eric Barton wrote:
1. How stable is the CM API and is it supported by all OpenIB affiliated
vendors?
The IB CM API is stable. Changes might occur as a result of changes to the CM
protocol itself, but that effect is not limited to just the openib API.
The RDMA CMA API is fairly stable,
Yes, this is the case.
- Original Message -
From: Caitlin Bestler [EMAIL PROTECTED]
To: openib-general@openib.org
Sent: Thursday, November 10, 2005 1:48 PM
Subject: RE: [openib-general] [ANNOUNCE] Contribute RDS
(ReliableDatagramSockets) to OpenIB
Mike Krause wrote in response
Eric Barton wrote:
5. Should I pre-map all physical memory and do RDMA in page-sized fragments?
This avoids any mapping overhead at the expense of having much larger
numbers of queued RDMAs. Since I try to keep up to 8 (by default) 1MByte
RDMAs active concurrently to any individual
Yes, of course; I meant the QP.
Regarding the total number of outstanding RDMA work requests,
I can keep a separate cap on that, so if relatively few peers
are active, I push the maximum number of RDMAs at them, but if
many peers are active the number of active RDMAs per peer reduces.
However I
Eric However I guess this still means that CQ resources
Eric sufficient for the maximum number of RDMAs I _could_ queue
Eric have to be allocated...
In general there will be a relatively low limit on the maximum CQ
size. For example, the maximum CQ size on Mellanox HCAs is ~128K
James,
I fixed some problems with the free build openib_scm version. Also turned down
some debugging and
added some debug prints for uAT path records.
-arlin
Signed-off by: Arlin Davis [EMAIL PROTECTED]
Index: dapl/openib/dapl_ib_cm.c
Hi Eric... writing YAN (yet another NAL) I see :)
Eric 2. I'd like to scale to = 10,000 peer nodes; 1 RC QP per
Eric peer. Is this going to get me into trouble?
EricFor example, I currently create a single PD and CQ for
Eric everything, however the example I've seen
Linus, please pull from
master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband.git for-linus
This tree is also available from kernel.org mirrors at:
rsync://rsync.kernel.org/pub/scm/linux/kernel/git/roland/infiniband.git
for-linus
The pull will get the following changes:
Jack
On Tue, Sep 27, 2005 at 06:17:00PM -0700, Grant Grundler wrote:
Hi Michael,
I'm trying to collect a full set of netperf TCP_STREAM over SDP for
SVN r3547 on 2.6.13 kernel. But some netperf runs get no throughput.
Michael,
I was able to reproduce this problem with SVN r3984.
I've posted the
At 10:48 AM 11/10/2005, Caitlin Bestler wrote:
Mike Krause wrote in response to Greg Lindahl:
If it is to
be reasonably robust, then RDS should be required to
support
the resync between the two sides of the communication. This
aligns
with the
stated objective of implementing reliability in
My concern is the requirement that RDS resync the structures in the
face of failureand know whether to re-transmit or will deal with
duplicates. Having pre-posted bufferswill help enable the resync to
be accomplished but should not be equated to pre-post equalsone can deal
with
On Thu, 10 Nov 2005, Arlin Davis wrote:
James,
I fixed some problems with the free build openib_scm version. Also
turned down some debugging and added some debug prints for uAT path
records.
-arlin
Thanks Arlin. Committed in revision 4018.
Fixed the bit value for formating
indicator.
Arkady Kanevsky
email: [EMAIL PROTECTED]
Network
Appliance Inc.
phone: 781-768-5395
275 Totten
Pond Rd.
Fax:
781-895-1195
Waltham, MA
02451-2010
central phone: 781-768-5300
IP Address Support by InfiniBand CM_v3.pdf
Description: IP
From: Kanevsky, Arkady [mailto:[EMAIL PROTECTED]
Sent: Thursday, November 10, 2005 1:37 PM
It will be discussed at IBTA SWG meeting next week Tu.
Please, post your comments before that.
Looks fine to me overall. The only thing I would change is make the version
field 4 bits rather than
The libibverbs bits look mostly OK but:
. I don't see a file sa_kern-abi.h anywhere -- I think you forgot to
add it. Also, please name it sa-kern-abi.h (ie all '-'s) -- mixed
underscores and dashes are just too hard to type and look weird.
Or did you just put everything in kern-abi.h?
Michael Maybe its a naming thing? We can call the list
Michael iterator, does this make it less ugly?
I thought about this, but it feels like overkill for something pretty
simple. So how about just doing
/* put list of devices in list and return length of list */
extern int
. I don't see a file sa_kern-abi.h anywhere -- I think you forgot to
add it. Also, please name it sa-kern-abi.h (ie all '-'s) -- mixed
underscores and dashes are just too hard to type and look weird.
. Please add a ChangeLog entry covering the libibverbs changes.
Here's an updated patch
Roland Dreier wrote:
Or are the consts too confusing? Should we be a little less safe but
make it nice and simple and just do
The const confuses me somewhat.
extern int ibv_get_device_list(struct ibv_device ***list);
Is ***list really what we want here? Can we just get away with **list?
So how about just doing
/* put list of devices in list and return length of list */
extern int ibv_get_device_list(struct ibv_device * const **list);
/* free a list of devices from ibv_get_device_list */
extern void ibv_free_device_list(struct ibv_device * const *list);
I like it much
If you want to maximize consumer usable private data, then you can move the
version, IP version, protocol, source and destination ports into the service ID.
Separately, if there's any defined mapping to a service ID or set of service
IDs, then the service ID indicates the format of the private
Is ***list really what we want here? Can we just get away with **list?
Would something like:
struct ibv_device * ibv_get_device(index);
I would prefer one call to get the entire structure. Another option might
be:
struct ibv_device ** ibv_get_device()
where it returns a list which
Roland Dreier wrote:
Is ***list really what we want here? Can we just get away with **list?
Yes -- a single device is represented by a struct ibv_device *.
So an array of devices is represented by a struct ibv_device **.
And a pointer to such an array is struct ibv_device ***.
I
It seemed faintly preferable to tell the caller how big the array was
rather than forcing the caller to count for itself.
If you really wanted that, I would be more inclined towards:
struct ibv_device ** ibv_get_device(*length_ptr)
and if you do not want length, you could pass a null
Sean Hefty wrote:
Caitlin Bestler wrote:
Current CM software could generate the Serive ID. Therefore the fact
that the Private Data is in the new format cannot be part of the
Service ID. Otherwise I agree with your analysis that data can be
moved to the Serivce ID. Which is more valuable,
4
【受信メール1件】届きました。
『名前』:kirarin
『年齢』:27歳
『職業』:自営業
『年収』:1000万円
『写真』:あり
『一言』:正直に言うとエッチ希望なんです。10万円先払いしますのでここに連絡
くれませんか?連絡くるまで待ってます。090-8012-
☆こちらから無料返信☆
http://lov025.com/?senyoh
※現在、kirarinさんからの指名メールは貴方様への一通のみとなっております。
※番号の続きは本人掲示板にてご確認下さい。
☆yahooアドレスなどフリーメールアドレスからでも登録できます☆
拒否の方
On Wed, Nov 09, 2005 at 09:46:06AM +0200, Eitan Zahavi wrote:
Hi Hal,
I would like to bring this to MgtWG before we change anything.
IMO the situation when this happens is really not legal since if the
SM's are not coordinated at least in their SM_Key it will cause the two
masters on the
36 matches
Mail list logo