On Mon, Sep 21, 2009 at 05:39:11PM +0300, Moni Shoua wrote:
So there is only one question left. Do we also need the other patch?
Yes, we need something like the other patch to handle the case where
the SM is unwilling to create a group for some reason. Right now that
is exactly the same failure
On Mon, Sep 28, 2009 at 10:49:23PM +0200, Pavel Machek wrote:
I don't remember seeing discussion of this on lkml. Yes it is in
-next...
eg http://lkml.org/lkml/2009/7/31/197 and followups, or search for v2
and earlier patches.
Well... it seems little overspecialized. Just
On Thu, Oct 01, 2009 at 03:50:07PM -0700, Ira Weiny wrote:
On Wed, 30 Sep 2009 20:57:52 -0600
Jason Gunthorpe jguntho...@obsidianresearch.com wrote:
On Wed, Sep 30, 2009 at 06:31:26PM -0700, Ira Weiny wrote:
Now I likely would agree with Ira that moving ib_types.h to libibumad
On Mon, Oct 05, 2009 at 10:43:44AM -0700, Sean Hefty wrote:
Export rdma_set_ib_paths to user space to allow applications to
manually set the IB path used for connections. This allows
alternative ways for a user space application or library to obtain
path record information, including
On Tue, Oct 06, 2009 at 12:05:02PM -0700, Sean Hefty wrote:
From user space, the call sequence does not change. The user calls
rdma_resolve_addr, rdma_resolve_route, rdma_connect, etc. It is up to the
librdmacm to perform the resolution. Today, the resolution request is simply
passed down
On Tue, Oct 06, 2009 at 03:53:21PM -0700, Sean Hefty wrote:
Actually, thinking about it some more, that would be very helpful. As
I said before, I have worked on apps using IB CM. The only reason is
to have complete control over the addressing. If I could use RDMA CM
API in some kind of AF_GID
On Tue, Oct 06, 2009 at 06:20:05PM -0700, Sean Hefty wrote:
There are 3 interfaces of interest here. The librdmacm API, the rdma_ucm user
to kernel interface, and the rdma_cm interface. These patches are looking to
change the rdma_ucm interface. I want to avoid changing the API or behavior
On Tue, Oct 06, 2009 at 12:05:02PM -0700, Sean Hefty wrote:
Ideally the best approach would be to have a mux at the ib_mad level. We
could
allow a user space application to intercept all outbound MADs for a given
class
and/or attribute. Unlike the present snooping of mads, this would
On Mon, Oct 12, 2009 at 08:19:44PM +0200, Ingo Molnar wrote:
After that point the scheme is perfectly lossless.
Well if it can OOM it's not lossless, obviously. You just define event
loss to be equivalent to Destruction of the universe. ;-)
It can't OOM once the ummunotify registration is
On Mon, Oct 12, 2009 at 10:52:59AM -0700, David J. Wilder wrote:
It is not, IPv6 link local addresses must be scoped. rping is
parsing the address with getaddrinfo, that does correctly set the
sin6_scope_id value in the sockaddr.
ping6 is scoping the address (setting sin6_scope_id) by
On Mon, Oct 12, 2009 at 10:20:46PM +0200, Ingo Molnar wrote:
It might be more acceptable because the flag-hint mechanism can at most
cause over-flushing - while with perf events we might miss to invalidate
a range altogether.
Right. Overflushing is not important, but missing an event
On Mon, Oct 12, 2009 at 03:29:38PM -0700, Smith, Stan wrote:
If __linux__ doesn't work for you, then please create a Linux Platform define
I can use.
Pretty much all the patches I've seen you make should be guarded by
__WIN__, you shouldn't be using __linux__.
opensm and the other OFA stuff
On Tue, Oct 13, 2009 at 08:40:06AM +0200, Ingo Molnar wrote:
* Jason Gunthorpe jguntho...@obsidianresearch.com wrote:
On Mon, Oct 12, 2009 at 10:20:46PM +0200, Ingo Molnar wrote:
It might be more acceptable because the flag-hint mechanism can at most
cause over-flushing - while
On Tue, Oct 13, 2009 at 03:09:40PM -0700, David J. Wilder wrote:
Here is a patch to addr6_resolve_remote() to correctly handle link-local
address.
It should cover all the conditions Jason described.
Looks pretty good to me, definitely on the right track.
Hmm..
Actually, upon comparing to
On Wed, Oct 14, 2009 at 10:30:05AM -0700, David J. Wilder wrote:
This looks good. Once concern, it may be obtuse, but if both the src and
dst are link-local addresses should only one need to be scoped? This
patch will required the src to always be scoped when using link local.
The TCPv6
with that.
Codes should have been be positive for alignment with POSIX, but
it is much too late for that..
Signed-off-by: Jason Gunthorpe jguntho...@obsidianresearch.com
---
src/cm.c | 30 +++---
1 files changed, 15 insertions(+), 15 deletions(-)
librdmacm has the same
to return errno than to fixup the cases that don't,
so lets stick with that.
Codes should have been be positive for alignment with POSIX, but
it is much too late for that..
Signed-off-by: Jason Gunthorpe jguntho...@obsidianresearch.com
src/cm.c | 30 +++---
1 files
On Fri, Oct 09, 2009 at 02:48:03PM -0700, Sean Hefty wrote:
Before spending any more time on this patch series, is there any disagreement
to
accepting this patch (as is or slightly modified) upstream?
Can you please have some way for this to pass APM data and the
reversible GMP path as well?
On Tue, Oct 20, 2009 at 11:08:58AM -0700, Sean Hefty wrote:
Looking on kernel cma.c format_hdr code it first branches on the address
family and next of the port space.
Going with your proposed flow, I understand that an app call to
rdma_resolve_addr will be broken down to rdma_bind_addr, ACM
return 0.
Signed-off-by: Jason Gunthorpe jguntho...@obsidianresearch.com
---
examples/rping.c | 22 ++---
man/rdma_cm.7| 16
src/cma.c| 94 -
3 files changed, 68 insertions(+), 64 deletions(-)
I think
On Tue, Oct 20, 2009 at 01:05:15PM -0700, Sean Hefty wrote:
I agree. But we can still support AF_IB/PS_TCP by simply assigning
the service ID correctly. rdma_bind_addr only needs to fill in a
service id if one is not given. This should enable
'one is not given' == 0 service ID?
yes
On Tue, Oct 20, 2009 at 01:48:34PM -0700, Sean Hefty wrote:
Private data:
- AF_IB/PS_TCP - the kernel munges the private data to be compatible
with AF_INET/PS_TCP, but otherwise is the same.
- AF_IB/PS_IB - the kernel doesn't touch the private data.
I was thinking AF_IB/* - kernel doesn't
On Wed, Oct 21, 2009 at 05:40:30PM -0700, Sean Hefty wrote:
Even so, it still seems OK to me:
Path:
addr4_resolve_remote
$ ip route get 10.0.0.11 from 192.168.122.1
local 10.0.0.11 from 192.168.122.1 dev lo
srcIP = 192.168.122.1
rdma_translate_ip(dst_ip = 10.0.0.11)
On Wed, Oct 21, 2009 at 06:07:54PM -0700, Sean Hefty wrote:
I'm reluctant to override fields like this to save 4 bytes. The
clarity and extensibility of using an additional flags field seems
worth it to me, and the processing code is not complex. I cannot think
of a motivation to save the 4
On Wed, Oct 21, 2009 at 11:49:52PM -0700, Sean Hefty wrote:
This is actually something of a mandatory notion to implement the
full generality of the IB CM protocol which allows the CM REP to
contain a port GUID of another port on the same node (multi-port
APM is an IB feature). So
On Thu, Oct 22, 2009 at 01:10:09AM -0700, Sean Hefty wrote:
+static int ucma_set_ib_path(struct ucma_context *ctx,
+ struct ib_path_rec_data *path_data, size_t optlen)
+{
+ struct ib_sa_path_rec sa_path;
+ struct rdma_cm_event event;
+ int ret;
+
+
On Thu, Oct 22, 2009 at 10:52:13AM -0700, Sean Hefty wrote:
+static int ucma_set_ib_path(struct ucma_context *ctx,
+ struct ib_path_rec_data *path_data, size_t optlen)
+{
+ struct ib_sa_path_rec sa_path;
+ struct rdma_cm_event event;
+ int ret;
+
+ if
On Thu, Oct 22, 2009 at 11:28:11AM -0700, Sean Hefty wrote:
I don't like the idea of the kernel silently ignoring the alternate path.
Returning an error seems like a better idea.
Then provide a way for userspace to know WTF to do. Without a
negotiation process this is now an 'impossible to use
On Sun, Oct 25, 2009 at 01:25:21PM +0200, Or Gerlitz wrote:
well, you didn't address some of my comments (not the ice-cream
ones...), which come to say that this wouldn't be inter-operable if for
one side you convert INET/TCP to IB/IB and for the other side you don't
(e.g userA/userB
On Sun, Oct 25, 2009 at 01:52:13PM +0200, Or Gerlitz wrote:
Jason, Have you even looked into or tested any of the bonding
load-balancing modes with ipoib? some/most of them are not applicable to
IPoIB and I don't think that the ones which may be such were ever
tested.
I was saying that
- Fold addr_send_arp into addr_resolve so that it uses the correct
dst structure.
Based on work from David J. Wilder dwil...@us.ibm.com
Signed-off-by: Jason Gunthorpe jguntho...@obsidianresearch.com
Reported-by: David J. Wilder dwil...@us.ibm.com
Reported-by: leo.tomi...@oracle.com
---
drivers
On Wed, Oct 28, 2009 at 10:05:19AM -0700, Sean Hefty wrote:
Can you explain how rdma_resolve_addr is used in conjunction with
multicast? I do not understand what the dest would be. Is it just a man
page typo?
A UD endpoint can communicate using multicast and to other UD
endpoints. A user
on the state of the rdma cm id. The librdmacm
already invokes this after rdma_resolve_addr completes.
Great, I didn't realize that was there. No further comments from me then
Reviewed-By: Jason Gunthorpe jguntho...@obsidianresearch.com
Jason
--
To unsubscribe from this list: send the line unsubscribe
(probably ESPIPE).
Signed-off-by: Jason Gunthorpe jguntho...@obsidianresearch.com
---
examples/asyncwatch.c |2 +-
examples/device_list.c|2 +-
examples/devinfo.c|4 ++--
examples/rc_pingpong.c|2 +-
examples/srq_pingpong.c |2 +-
examples/uc_pingpong.c
9f3a76deb5bfda0f8243eadfa024eb547c03f583 Mon Sep 17 00:00:00 2001
From: Jason Gunthorpe jguntho...@obsidianresearch.com
Date: Mon, 2 Nov 2009 11:23:38 -0700
Subject: [PATCH] RDMA CM: Fix AF_INET6 support in multicast joining
If joining to an AF_INET6 address we need to map the address to a MGID
On Tue, Nov 03, 2009 at 08:43:01AM -0800, Sean Hefty wrote:
What's missing is Jason's patch to fix the IPv6 mapping, and a way to extend
the
rdma_cm to support the full range of unmapped addresses. I just haven't been
able to get to either of these yet.
My feeling is when AF_IB is
RDMA CM treats AF_INET6 addresses that are either 0 or prefixed with
FF1x:A01B::/32 as MGIDs, but the detection for the prefix was buggy,
fix it up.
Signed-off-by: Jason Gunthorpe jguntho...@obsidianresearch.com
---
drivers/infiniband/core/cma.c |2 +-
1 files changed, 1 insertions(+), 1
On Sun, Nov 08, 2009 at 08:25:55AM +0200, Or Gerlitz wrote:
ACM is intended to be a service that's used by the librdmacm to resolve
address mappings and routes. Trying to have ACM use the librdmacm ends up
with a circular dependency. That's the part I'm trying to avoid.
fail-enough, I
On Mon, Nov 09, 2009 at 09:44:31AM +0200, Or Gerlitz wrote:
No rdma_resolve_addr2 is needed the one that exists now has
source addresses specified, I don't see that extra info is needed for
AF_INET that was resolved with rdma_getaddrinfo is this AF_IB specific?
The extra info in
On Tue, Nov 10, 2009 at 01:12:17PM -0800, Sean Hefty wrote:
How can we make progress? What are your thoughts?
I'm actually working on this today. I've taken Jason's patch as a
starting point, and I'm breaking it into separate patches for
merging and testing. I hope to post the patches
On Wed, Nov 11, 2009 at 05:44:59PM -0500, Richard Frank wrote:
Would anyone like to through out the list of HCAs that do this... I
can guess at a few... and can ask the vendors directly.. if not.. .
It would be much nicer to not hardcode names of adapters.. but that won't
stop us.. :)
On Wed, Nov 11, 2009 at 03:22:50PM -0800, Ralph Campbell wrote:
While this is true for SLtoVL, we create other files which are
device specific under the port directory too.
It seems like we might need to introduce a callback into the driver to
create the port specific sysfs files.
Maybe give
On Sat, Nov 21, 2009 at 12:17:32PM +0100, Bart Van Assche wrote:
ib_post_send() has to request a completion notification for each WR,
which has a negative performance impact. My opinion is that the
current behavior makes ib_post_send() easier to implement, while the
behavior specified in the
On Sat, Nov 21, 2009 at 09:37:29PM +0100, Bart Van Assche wrote:
It would be more useful IMHO if ib_post_send() would not
post any WR instead of posting part of the WR's passed to
ib_post_send() when the 'queue full' condition is hit. Note: SRPT uses
a single SRQ per HCA, even when multiple
On Mon, Nov 16, 2009 at 04:12:07PM -0800, Sean Hefty wrote:
+static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
+struct sockaddr *addr)
+{
+ struct sockaddr_in6 *sin6;
+
+ if (addr-sa_family != AF_INET6)
+ return 0;
+
+ sin6
On Mon, Nov 23, 2009 at 02:07:09PM -0800, Sean Hefty wrote:
The main missing test from the new version is:
if (dev_addr-bound_dev_if
dev_addr-bound_dev_if != addr6-sin6_scope_id)
return -EINVAL;
Done on the 2nd src and dest
On Mon, Nov 23, 2009 at 07:43:34PM -0800, Sean Hefty wrote:
The sin6_scope_id must be ignored in all cases except LL
addresses. Fixing this will break the path that sets the bound_dev_if
starting from rdma_resolve_addr. This is why my version was setting
bound_dev_if directly in
On Tue, Nov 24, 2009 at 06:23:15PM -0500, Jeff Squyres wrote:
2. I am somewhat confused by the overloading of the term transport.
It appears that a device will have
ibv_device.transport_type==IBV_TRANSPORT_IB for both IB and RDMAOE
devices. The only way to tell the difference is to
On Tue, Nov 24, 2009 at 09:12:53PM -0500, Jeff Squyres wrote:
On Nov 24, 2009, at 7:11 PM, Jason Gunthorpe wrote:
Is the same true for openmpi? If you try to run it as is on a RDMAOE
interface will it work? If not I think that alone should kill this
idea..
OMPI uses RDMACM (among
On Wed, Nov 25, 2009 at 04:41:08PM +0200, Eli Cohen wrote:
On Wed, Nov 25, 2009 at 09:30:40AM -0500, Jeff Squyres wrote:
In practice, we have seen that applications *do* need to query the
transport type -- at least (real) IB vs. iWARP. It is your
expectation that IB and IBoE will
On Mon, Nov 30, 2009 at 03:34:06PM +0200, Eli Cohen wrote:
If we change struct ibv_port_attr transport field from enum to uint8,
we eliminate binary compatibility problems. That's because the previous
filed is aligned to 16 bits address so that leaves us 16 bits more.
Dealing with ABI
On Mon, Nov 30, 2009 at 10:50:02AM -0800, Roland Dreier wrote:
If we change struct ibv_port_attr transport field from enum to uint8,
we eliminate binary compatibility problems. That's because the previous
filed is aligned to 16 bits address so that leaves us 16 bits more.
diff
On Tue, Dec 01, 2009 at 06:22:06PM +0200, Liran Liss wrote:
Dealing with ABI compatability is a different issue, this new scheme
is API incompatible due to the change in semantics for existing values.
For rdmacm applications, there are no semantic changes between IB and
RDMAoE.
So? There
On Fri, Dec 04, 2009 at 08:03:31PM -0800, Roland Dreier wrote:
then I think legacy apps should be OK (port_attr size doesn't change,
binary compat is still there), and new apps that do check link_layer
should also be OK ... if they use an old library and/or old driver,
they'll see
On Wed, Dec 09, 2009 at 11:06:41AM -0800, Roland Dreier wrote:
It looks good to me. Thanks, I will take it for RDMAoE.
Great... as Jason suggested, please also add in the appropriate reserved
fields to pad the struct to a 32 bit boundary and zero them in the
wrapper. So if there is a
On Thu, Dec 10, 2009 at 07:05:36PM +0200, Eli Cohen wrote:
here is the patch I prepared based on the discussions we had. The patch is
based on the last
rdmaoe/libibverbs patch I sent. libmlx4 was modified too, a trivial
change that changes name. Both fixes were push to OFED. I will send a
On Thu, Dec 10, 2009 at 11:14:55PM +0200, Eli Cohen wrote:
On Thu, Dec 10, 2009 at 10:33:53AM -0700, Jason Gunthorpe wrote:
Could you prepare this based on Roland's tree? This patch won't apply.
I quote two patches, one for libibverbs based on 74638ac, and the
other for libmlx4 based
On Thu, Dec 10, 2009 at 01:57:11PM -0800, Roland Dreier wrote:
Maybe I'm wrong but I don't like setting don't know magically to IB
behind the scenes.
Well, it isn't just don't know it also means the kernel doesn't
support the link_layer query. The kernels that don't support
link_layer also
On Tue, Dec 15, 2009 at 09:18:19AM -0800, Ira Weiny wrote:
On Tue, 15 Dec 2009 10:15:32 -0700
Jason Gunthorpe jguntho...@obsidianresearch.com wrote:
However, I don't understand the comment Only set HopLimit if going
through a
router?
This is from '#ifdef ROUTER_EXP' days
On Wed, Dec 16, 2009 at 03:06:37PM -0600, frank zago wrote:
Hello,
The development tree for a soft RDMA transport over Ethernet driver
(rxe) is available in the OFA git repository. This is a work in progress
but has enough functionality for people interested in looking at it to
be
On Thu, Dec 17, 2009 at 03:43:56PM -0600, Robert Pearson wrote:
I agree about the polynomial. That would have been nice. As currently
proposed the embedding of the IB transport in Ethernet frames preserves the
IB ICRC as part of the transport and trades the VCRC for Ethernet's CRC32
over the
On Thu, Dec 17, 2009 at 05:00:26PM -0600, Robert Pearson wrote:
Of course we need to interoperate with ConnectX-en and it, as far as I know,
only knows how to compute the VCRC as though the packet was going to get
sent to IB including a phantom 12 byte LRH that is filled with 1's without
any
On Tue, Jan 05, 2010 at 07:27:57PM -0500, Aaron Knister wrote:
As an aside I would love to know how to pull an infiniband interface
into a bridge.
bridging only works between networks with the same L2 address scheme
and support in the bridging driver for that address scheme.
The vnic schemes
On Thu, Jan 07, 2010 at 08:50:47AM -0800, Sean Hefty wrote:
+route-path_rec-hop_limit = 2;
The reason is that ib_init_ah_from_path() will not set IB_AH_GRH for
hop_limit smaller then 2, and since that GRH is required in RoCEE, and
since this is specific to RoCEE, I put 2 to make
On Tue, Jan 12, 2010 at 02:57:35PM -0800, Roland Dreier wrote:
I doubt you could benchmark the overhead of calling ib_post_recv() in
the full SRP protocol. Really, I bet it's less than 100 nanoseconds to
form the work request and call ib_post_recv(). Maybe I'm wrong but I
really expect the
On Wed, Jan 13, 2010 at 08:23:27AM +0100, Bart Van Assche wrote:
On Wed, Jan 13, 2010 at 12:24 AM, Jason Gunthorpe
jguntho...@obsidianresearch.com wrote:
[ ... ]
Also, I couldn't tell for sure from a cursory examination of the
patch, but the initiator must be designed to not stall
On Wed, Jan 13, 2010 at 07:57:26PM +0100, Bart Van Assche wrote:
On Wed, Jan 13, 2010 at 7:16 PM, Jason Gunthorpe
jguntho...@obsidianresearch.com wrote:
On Wed, Jan 13, 2010 at 08:23:27AM +0100, Bart Van Assche wrote:
On Wed, Jan 13, 2010 at 12:24 AM, Jason Gunthorpe
jguntho
On Tue, Jan 19, 2010 at 05:09:49PM -0800, Roland Dreier wrote:
The last time I tried to use it the kernel began reporting lots of
OOM events (2.6.30 stock). I thought this was well known because CM
mode uses high order allocations??
That's not well-known to me. What's the backtrace
On Mon, Feb 01, 2010 at 03:05:13PM +0200, Yevgeny Kliteynik wrote:
In general, here's the problem:
We have cl_file.h and cl_file_osd.h.
cl_file.h has include directive for cl_file_osd.h
cl_file.h has the following definition of function:
int foo();
cl_file_osd.h has another
On Tue, Feb 02, 2010 at 03:11:26PM +0200, Yevgeny Kliteynik wrote:
If no inline version is defined then the compiler just emits a normal
function call, if an inline version is defined then the compiler might
use it.
Thanks for the idea.
I read some documentation about it, and it does look
On Fri, Feb 05, 2010 at 12:32:51PM -0600, Steve Wise wrote:
I think we should remove the feature of allowing binds to 127.0.0.1
altogether based on Jeff's arguments and my assertion that 127.0.0.1 is
a sw-loopback mechanism anyway...
I don't agree, the kernel should be free to provide a
On Fri, Feb 05, 2010 at 03:08:10PM -0500, Jeff Squyres wrote:
On Feb 5, 2010, at 1:56 PM, Jason Gunthorpe wrote:
I think we should remove the feature of allowing binds to 127.0.0.1
altogether based on Jeff's arguments and my assertion that 127.0.0.1 is
a sw-loopback mechanism anyway
On Tue, Feb 09, 2010 at 05:01:21PM -0500, Jeff Squyres wrote:
1. Is this now the recommended way to find all the IP interfaces that support
RDMA:
- loop over all local IP addresses
- if 127.0.0.1/8, skip
- try to rdma_bind_addr()
- if it succeeds and verbs ptr is != NULL, it's an RDMA
On Mon, Feb 15, 2010 at 02:56:56PM +0200, Jack Morgenstein wrote:
If I put the XRC srq fields after the pthread_cond_t, the RHEL4/5
incompatibility kicks in, in a big way. If I put them before the
pthread_cond_t, we still have a problem with events_completed, as
indicated by Gleb Nabokov of
On Tue, Feb 16, 2010 at 03:29:48PM -0800, Andy Grover wrote:
Right now, RDS follows each RDMA write op with a Send op, which 1)
causes an interrupt and 2) includes the info we need to call
ib_dma_sync_sg_for_cpu() for the target of the rdma write.
We want to omit the Send. If we don't do the
On Tue, Feb 16, 2010 at 10:40:45PM -0800, Paul Grun wrote:
Two advantages come to mind vs an RDMA Write followed by a SEND:
Using a SEND will consume a second WQE on the send side, and the
synchronizing SEND will cause an entire new transaction, which will
consume a(n infinitesimally) small
On Sat, Feb 27, 2010 at 11:38:37AM +0100, Jiri Pirko wrote:
The problem this statement is trying to solve had to do with bonding
creating multicast addresess for ethernet rather than infiniband in
some cases. This happens because bonding makes a device that switches
from ethernet to
On Wed, Mar 17, 2010 at 10:48:30AM -0700, Sean Hefty wrote:
Would you accept a patch to libibverbs sa.h to add the wire format for a path
record, shown below? I'm not ready to submit a patch yet, just need to find a
place for this.
Could we use the bitfield versions of this I posted awhile
On Thu, Mar 18, 2010 at 10:49:01AM -0700, Roland Dreier wrote:
Including this structure is no problem in principle.
Could we use the bitfield versions of this I posted awhile back? :)
I don't have that patch handy, but bitfields often end up ugly with
endian stuff, and also some
On Mon, Mar 22, 2010 at 02:21:39PM +0100, Jiri Pirko wrote:
Finally this bit can be removed. Currently, after the bonding driver is
changed/fixed (32a806c194ea112cfab00f558482dd97bee5e44e net-next-2.6),
that's not possible for an addr with different length than dev-addr_len
to be present in
On Mon, Mar 22, 2010 at 06:26:14PM +0100, Jiri Pirko wrote:
Mon, Mar 22, 2010 at 05:59:16PM CET, jguntho...@obsidianresearch.com wrote:
On Mon, Mar 22, 2010 at 02:21:39PM +0100, Jiri Pirko wrote:
Finally this bit can be removed. Currently, after the bonding driver is
changed/fixed
On Tue, Mar 23, 2010 at 12:06:50PM -0400, Jeff Squyres wrote:
IBM has found a resource that they think will be able to progress Roland's
ummunotify work.
After a few discussions in Sonoma last week and some off-list emails, here's
what we decided:
1. Take Roland's last code drop
On Tue, Mar 23, 2010 at 01:17:40PM -0400, Jeff Squyres wrote:
On Mar 23, 2010, at 12:59 PM, Jason Gunthorpe wrote:
The main reason for the new FD is so it can be polled on..
What do you poll on the fd for?
With ummunotify, you only read() from the fd when (counter !=
last_counter
On Tue, Mar 23, 2010 at 03:17:12PM -0400, Jeff Squyres wrote:
If you don't think that is worth doing it
does simplify things alot, just add two new verbs calls:
ibv_set_mmu_counter(verbs, my_counter);
ibv_get_mmu_notifications(verbs, my_list, sizeof(my_list));
I have no real opinion
On Tue, Mar 23, 2010 at 04:01:21PM -0400, Jeff Squyres wrote:
On Mar 23, 2010, at 3:52 PM, Jason Gunthorpe wrote:
ibv_set_mmu_counter(verbs, my_counter);
ibv_get_mmu_notifications(verbs, my_list, sizeof(my_list));
These are not hiding mmap/read, they are new uverbs 'syscalls
On Tue, Mar 23, 2010 at 10:55:08PM -0700, Roland Dreier wrote:
I would prefer to do this by adding a new verbs call that returns a fd
directly. Ie use ib_uverbs_alloc_event_file and act like
ibv_create_comp_channel.
The main reason for the new FD is so it can be polled on..
On Tue, Mar 23, 2010 at 10:59:42PM -0700, Roland Dreier wrote:
That is all definitely doable. I wonder if it's better to get rid of
the dedicated fd though. After all, having the fd means a fancy app can
do poll() or sigio or whatever internally. Being able to integrate into
an fd-driven
On Thu, Mar 25, 2010 at 12:05:33PM -0700, Sean Hefty wrote:
+ case AF_IB:
+ ((struct sockaddr_ib *) addr)-sib_sid =
+ cpu_to_be64(((u64) id_priv-id.ps 16) + ntohs(port));
+ break;
+ }
Could you elaborate a bit on how you are mixing the
On Thu, Mar 25, 2010 at 08:56:23PM -0700, Sean Hefty wrote:
Is there any reason the port space has to be known when the cm_id is
created but before bind?
No - but it is still required for transport neutrality. rdma_create_id()
simply stores the value.
To correct this slightly, a user can
On Fri, Mar 26, 2010 at 10:17:13AM -0700, Sean Hefty wrote:
Consider UDP, an application can create an id and join multicast groups using
AF_INET, AF_INET6, and/or AF_IB - whatever addressing scheme is most
convenient.
Actually, the port space doesn't matter when specifying a multicast
On Fri, Mar 26, 2010 at 05:11:04PM -0700, Sean Hefty wrote:
Actually, the port space doesn't matter when specifying a multicast group.
So,
I can't think of any real advantage to supporting AF_IB/PS_TCP.
Okay - I found one advantage - fewer changes. :b
Lol!
I've started adding
On Mon, Mar 29, 2010 at 12:01:07PM -0700, Roland Dreier wrote:
The rdma_cm might be able to support this if the port space were
separated based
on the address family, depending on how PS IB ends up.
I think separate port spaces is the correct solution.
This gets a bit tricky --
On Mon, Mar 29, 2010 at 02:51:46PM -0500, Steve Wise wrote:
Yeah, exactly, it is very complex and there is a real need for
things pretending to be IP to capture all this subtlety. The details
can't just be skipped over, people will notice :(
Though, I'm also not entirely certain that
On Thu, Apr 01, 2010 at 10:08:41AM -0700, Sean Hefty wrote:
Include RDMA_PS_IB checks when setting the qkey for UD QPs.
I know we talked about this, but seeing this patch makes me ask again,
should the QKEY be part of sockaddr_ib? Or at least be settable somehow?
Jason
--
To unsubscribe from
On Thu, Apr 01, 2010 at 11:00:44AM -0700, Sean Hefty wrote:
I know we talked about this, but seeing this patch makes me ask again,
should the QKEY be part of sockaddr_ib? Or at least be settable somehow?
I think so. I'm not sure of the best approach. With these patches,
the qkey seems to
On Thu, Apr 01, 2010 at 10:46:06AM -0700, Sean Hefty wrote:
} else if ((addr-sa_family == AF_INET6)) {
ipv6_ib_mc_map(sin6-sin6_addr, dev_addr-broadcast, mc_map);
- if (id_priv-id.ps == RDMA_PS_UDP)
+ if (id_priv-id.ps == RDMA_PS_UDP || id_priv-id.ps ==
On Wed, Apr 07, 2010 at 10:12:43AM -0700, Sean Hefty wrote:
RDMA requires the user to allocate hardware resources before
establishing a connection. To support this, the user must know
the source address that the connection will use to reach the
remote endpoint. Modify rdma_getaddrinfo to
On Wed, Apr 07, 2010 at 10:12:44AM -0700, Sean Hefty wrote:
+ * The rdma_cm_id will be set to use synchronous operations (connect,
+ * listen, and get_request). To convert to synchronous operation, the
^
asynchronous?
Jason
--
To
On Wed, Apr 07, 2010 at 12:37:03PM -0700, Roland Dreier wrote:
No, there is no mmap. Like this:
u64 my_counter = 0;
ibv_set_mmu_counter(verbs, my_counter);
[..]
while (my_counter != last_my_counter) {
last_my_counter = my_counter;
On Wed, Apr 07, 2010 at 12:54:56PM -0700, Sean Hefty wrote:
I haven't looked through everything you posted to make a suggestion
here, but this bothers me..
The resources should be allocated after the rdma_bind syscall, prior to
listen/accept or connect, IMHO.
How does tha rai-ai_src_addr
1 - 100 of 1197 matches
Mail list logo