Hi, Moni
On 06/10/2015 11:13 AM, Moni Shoua wrote:
Registering an event handler is done for a device. This device may have
one RoCE port (no SA cap) and one InfiniBand port (has SA cap).
Therefore, warning from the event handler about a specific port that
doesn't have SA cap is correct but
On 6/9/2015 10:27 AM, Matan Barak wrote:
On 6/9/2015 12:37 AM, Hefty, Sean wrote:
Previously, every vendor implemented its net device notifiers in its
own
driver. This introduces a huge code duplication as figuring
28 files changed, 2253 insertions(+), 860 deletions(-)
How does
Registering an event handler is done for a device. This device may have
one RoCE port (no SA cap) and one InfiniBand port (has SA cap).
Therefore, warning from the event handler about a specific port that
doesn't have SA cap is correct but pollutes the kernel log without a
need.
Signed-off-by:
On Mon, May 11, 2015 at 1:04 PM, Yuval Shaia yuval.sh...@oracle.com wrote:
By default, IPoIB-CM driver uses 64k MTU. Larger MTU gives better performance.
This MTU plus overhead puts the memory allocation for IP based packets at 32
4k pages (order 5), which have to be contiguous.
When the
The Subnet Administrator (SA) is not a component of the RoCE spec.
Therefore, it should not be a capability of a RoCE port.
Signed-off-by: Moni Shoua mo...@mellanox.com
---
include/rdma/ib_verbs.h | 1 -
1 file changed, 1 deletion(-)
diff --git a/include/rdma/ib_verbs.h
On 6/10/2015 4:26 AM, Christoph Lameter wrote:
I have no problem with a bare metal interface exposing this. But pretendin=
g that it's generic and that this is the one and only way that this could b=
e implemented doesn't make it so.
This is a way it was implemented and its usable. Shooting
On 6/10/2015 8:35 AM, Moni Shoua wrote:
The Subnet Administrator (SA) is not a component of the RoCE spec.
Therefore, it should not be a capability of a RoCE port.
Change-Id: Iadfaa56bdc9f6e28f46d009064c2d15969293cf7
Please remove the internal Gerrit IDs we use prior to sending patches out
On Mon, May 11, 2015 at 1:04 PM, Yuval Shaia yuval.sh...@oracle.com wrote:
By default, IPoIB-CM driver uses 64k MTU. Larger MTU gives better performance.
This MTU plus overhead puts the memory allocation for IP based packets at 32
4k pages (order 5), which have to be contiguous.
When the
On 6/9/2015 10:57 AM, kaike@intel.com wrote:
From: Kaike Wan kaike@intel.com
A SA cache is undeniably critical for fabric scalability and performance.
In user space, the ibacm application provides a good example of pathrecord
cache for address and route resolution. With the recent
-Original Message-
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 9:37 AM
A SA cache is undeniably critical for fabric scalability and performance.
In user space, the ibacm application provides a good example of
pathrecord cache for
On 6/9/2015 9:52 PM, Bob Ciotti wrote:
We have an issue where lustre servers and clients cannot talk to each
other.
There are about 11,000 clients all trying to connect to a server that
just been rebooted
(nbp6-oss3 in this example)
pfe21 is a lustre client thats trying to remount the
On 6/10/2015 10:22 AM, Wan, Kaike wrote:
-Original Message-
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 9:37 AM
A SA cache is undeniably critical for fabric scalability and performance.
In user space, the ibacm application provides a good
On Wed, Jun 10, 2015 at 11:53:15AM +0300, Or Gerlitz wrote:
Jason, can you ack that this post addressed your comments?
Well, I asked for a cleanup series, multiple times, and this is the
closest things have got.
It isn't really a cleanup because the whole gid table is new code and
has latent
From: Ira Weiny ira.we...@intel.com
Hi Ira,
OPA cannot impersonate IB; OPA node and link types have to be designated as
such.
In terms of MAD processing flows, both explicit (as in the handle_opa_smi()
call below) and implicit code paths (which share IB flows - there are several
cases) must
On Tue, Jun 09, 2015 at 08:33:12PM -0700, Mark Hairgrove wrote:
On Tue, 9 Jun 2015, Jerome Glisse wrote:
On Mon, Jun 08, 2015 at 06:54:29PM -0700, Mark Hairgrove wrote:
Can you clarify how that's different from mmu_notifiers? Those are also
embedded into a driver-owned struct.
A SA cache is undeniably critical for fabric scalability and
performance.
In user space, the ibacm application provides a good example of
pathrecord cache for address and route resolution. With the recent
implementation of the provider architecture, ibacm offers more
extensibility as
On 6/10/2015 11:21 AM, Hefty, Sean wrote:
While this appears to address the current upstream use model for ACM
with it's multicast overlay backend where PRs are static, it does not
appear to address PR changes.
Although this ties into ibacm, from the viewpoint of the kernel, there's no
Registering an event handler is done for a device. This device may
have one RoCE port (no SA cap) and one InfiniBand port (has SA cap).
Therefore, warning from the event handler about a specific port that
doesn't have SA cap is correct but pollutes the kernel log without a
need.
Registering an event handler is done for a device. This device may have
one RoCE port (no SA cap) and one InfiniBand port (has SA cap).
Therefore, warning from the event handler about a specific port that
doesn't have SA cap is correct but pollutes the kernel log without a
need.
Maybe
On Wed, 2015-06-10 at 11:43 +0300, Moni Shoua wrote:
The Subnet Administrator (SA) is not a component of the RoCE spec.
Therefore, it should not be a capability of a RoCE port.
Signed-off-by: Moni Shoua mo...@mellanox.com
---
include/rdma/ib_verbs.h | 1 -
1 file changed, 1 deletion(-)
There are multiple problems with libfrabric related to the use cases in my
area. Most of all the lack of multicast support. Then there is the build
up of software bloat on top. The interest here is in low latency
operations. Redenzvous and other new features are really not wanted if
they
RDMA_CM_EVENT_UNREACHABLE is indicated when there are timeouts in
underlying CM protocol exchange. I suspect that the server is really
busy and doesn't respond to the low level CM MADs in a timely manner.
RDMA CM (and other kernel ULPs like IPoIB and SRP use hard coded local
and remote
Use kernel.h macro definition.
Thanks to Julia Lawall for Coccinelle scripting support.
Signed-off-by: Fabian Frederick f...@skynet.be
---
drivers/infiniband/hw/mthca/mthca_profile.c | 8 ++--
1 file changed, 2 insertions(+), 6 deletions(-)
diff --git
On 6/10/2015 11:49 AM, Wan, Kaike wrote:
A SA cache is undeniably critical for fabric scalability and
performance.
In user space, the ibacm application provides a good example of
pathrecord cache for address and route resolution. With the recent
implementation of the provider architecture,
On Wed, Jun 10, 2015 at 05:04:55PM +, Hefty, Sean wrote:
Not in the patches themselves but in the general issue when a PR changes.
Do you think this needs addressing or are things fine as they are now ?
IMO, I think it needs addressing in terms of can the proposed
netlink
On 6/10/2015 3:10 PM, Jason Gunthorpe wrote:
On Wed, Jun 10, 2015 at 01:47:36PM -0400, Hal Rosenstock wrote:
On 6/9/2015 10:57 AM, kaike@intel.com wrote:
From: Kaike Wan kaike@intel.com
This patch routes a SA pathrecord query to netlink first
Should only unicast PRs be done in this
On 6/10/2015 1:04 PM, Hefty, Sean wrote:
Not in the patches themselves but in the general issue when a PR changes.
Do you think this needs addressing or are things fine as they are now ?
IMO, I think it needs addressing in terms of can the proposed netlink
architecture and design
+/* Local Service Reversible attribute */ struct
+rdma_nla_ls_reversible {
+ __u32 reversible;
+};
Isn't __u8 sufficient for reversible ?
Certainly enough. However, reversible is __u32 in struct
ib_user_path_rec and int in struct ib_sa_path_rec.
OK; I hadn't double
This series does not attempt to optimize the kernel needing to know that a
PR
has been updated. There are existing mechanisms for that.
Does this exist in the kernel?
--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to majord...@vger.kernel.org
On Wed, Jun 10, 2015 at 9:49 PM, Jason Gunthorpe
jguntho...@obsidianresearch.com wrote:
On Wed, Jun 10, 2015 at 06:08:30PM +0300, Matan Barak wrote:
It isn't really a cleanup because the whole gid table is new code and
has latent elements for rocev2 - this is why it is so much bigger than
it
On Wed, 2015-06-10 at 12:56 -0600, Jason Gunthorpe wrote:
On Wed, Jun 10, 2015 at 02:37:26PM -0400, Doug Ledford wrote:
On Wed, 2015-06-10 at 06:30 +, Liran Liss wrote:
From: Ira Weiny ira.we...@intel.com
Hi Ira,
OPA cannot impersonate IB; OPA node and link types have to
From: Ira Weiny ira.we...@intel.com
For devices which support OPA MADs
1) Use previously defined SMP support functions.
2) Pass correct base version to ib_create_send_mad when processing OPA MADs.
3) Process out_mad_key_index returned by agents for a response. This is
necessary
On Wed, Jun 10, 2015 at 01:47:36PM -0400, Hal Rosenstock wrote:
On 6/9/2015 10:57 AM, kaike@intel.com wrote:
From: Kaike Wan kaike@intel.com
This patch routes a SA pathrecord query to netlink first
Should only unicast PRs be done in this manner or should API support
enabling
On 6/10/2015 2:31 PM, Wan, Kaike wrote:
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 1:47 PM
On 6/9/2015 10:57 AM, kaike@intel.com wrote:
From: Kaike Wan kaike@intel.com
This patch adds netlink defines for SA client, local service group,
On Wed, 2015-06-10 at 06:30 +, Liran Liss wrote:
From: Ira Weiny ira.we...@intel.com
Hi Ira,
OPA cannot impersonate IB; OPA node and link types have to be designated as
such.
In terms of MAD processing flows, both explicit (as in the handle_opa_smi()
call below) and implicit code
Not in the patches themselves but in the general issue when a PR changes.
Do you think this needs addressing or are things fine as they are now ?
IMO, I think it needs addressing in terms of can the proposed netlink
architecture and design accommodate this sort of request in the future? We
On Wed, Jun 10, 2015 at 06:30:58AM +, Liran Liss wrote:
From: Ira Weiny ira.we...@intel.com
Hi Ira,
OPA cannot impersonate IB; OPA node and link types have to be designated as
such.
This was discussed at length and we agreed that the kernel would have explicit
capabilities
On Wed, Jun 10, 2015 at 06:08:30PM +0300, Matan Barak wrote:
It isn't really a cleanup because the whole gid table is new code and
has latent elements for rocev2 - this is why it is so much bigger than
it should be.
I disagree. Could you please point on anything that is RoCE V2 specific?
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 1:47 PM
On 6/9/2015 10:57 AM, kaike@intel.com wrote:
From: Kaike Wan kaike@intel.com
This patch adds netlink defines for SA client, local service group,
local service operations, and related
On 06/10/2015 06:35 AM, Hal Rosenstock wrote:
On 6/9/2015 9:52 PM, Bob Ciotti wrote:
We have an issue where lustre servers and clients cannot talk to each
other.
There are about 11,000 clients all trying to connect to a server that
just been rebooted
(nbp6-oss3 in this example)
pfe21 is a
On Wed, Jun 10, 2015 at 02:37:26PM -0400, Doug Ledford wrote:
On Wed, 2015-06-10 at 06:30 +, Liran Liss wrote:
From: Ira Weiny ira.we...@intel.com
Hi Ira,
OPA cannot impersonate IB; OPA node and link types have to be
designated as such. In terms of MAD processing flows, both
On 6/9/2015 10:57 AM, kaike@intel.com wrote:
From: Kaike Wan kaike@intel.com
This patch routes a SA pathrecord query to netlink first
Should only unicast PRs be done in this manner or should API support
enabling for unicast and/or multicast ?
AFAIK kernel doesn't query multicast PRs
On Wed, Jun 10, 2015 at 10:39:49AM -0400, Hal Rosenstock wrote:
On 6/10/2015 10:22 AM, Wan, Kaike wrote:
-Original Message-
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 9:37 AM
A SA cache is undeniably critical for fabric
On Wed, Jun 10, 2015 at 09:34:58PM +, Hefty, Sean wrote:
I agree. I just wanted to make sure that there wasn't some feature
regarding PRs, such as unpath, that a kernel client would lose
(i.e. it is currently implemented) by changing how the PRs are
retrieved. Basically nothing breaks
Not directly. IPoIB treats it that way. I guess to be safe.
Officially one should register for UnPath/RePath traps. But no one has
ever implemented that.
To be clear I am agreeing with Hal that having some sort of update
signal would be nice. But I don't think that must be done before
On Wed, Jun 10, 2015 at 11:19:03PM +0300, Matan Barak wrote:
Sure gid_type is gone, but I didn't say roceve2 specific, I said
latent elements. ie I'm assuming reasons for the scary locking are
because the ripped out rocev2 code needed it? And some of the
complexity that looks pointless
On 6/10/2015 3:10 PM, Jason Gunthorpe wrote:
On Wed, Jun 10, 2015 at 01:47:36PM -0400, Hal Rosenstock wrote:
On 6/9/2015 10:57 AM, kaike@intel.com wrote:
From: Kaike Wan kaike@intel.com
This patch routes a SA pathrecord query to netlink first
Should only unicast PRs be done
This series does not attempt to optimize the kernel needing to know
that a PR has been updated. There are existing mechanisms for that.
Does this exist in the kernel?
At least some support, yes. For example client reregister marks all IPoIB
paths as invalid.
Reregister indicates
This series does not attempt to optimize the kernel needing to
know that a PR has been updated. There are existing mechanisms for
that.
Does this exist in the kernel?
At least some support, yes. For example client reregister marks all
IPoIB paths as invalid.
Reregister
This series does not attempt to optimize the kernel needing to know
that a PR has been updated. There are existing mechanisms for that.
Does this exist in the kernel?
At least some support, yes. For example client reregister marks all IPoIB
paths as invalid.
Ira
--
To unsubscribe
On Wed, 2015-06-10 at 09:00 -0600, Jason Gunthorpe wrote:
On Wed, Jun 10, 2015 at 11:53:15AM +0300, Or Gerlitz wrote:
Jason, can you ack that this post addressed your comments?
Well, I asked for a cleanup series, multiple times, and this is the
closest things have got.
It isn't really
On Wed, 2015-06-10 at 12:49 -0600, Jason Gunthorpe wrote:
On Wed, Jun 10, 2015 at 06:08:30PM +0300, Matan Barak wrote:
It isn't really a cleanup because the whole gid table is new code and
has latent elements for rocev2 - this is why it is so much bigger than
it should be.
I disagree.
Here Sean reacts to our RoCEv2 patches (same thread as the one I sent before
with Jason's feedback).
Sean is not even making technical statements to dismiss our patches. He has
been rejecting all our previous revs of this set with loose FUD comments. In
this case he chose to count the lines of
On Wed, 10 Jun 2015, Jerome Glisse wrote:
[...]
Like said, just ignore current code it is utterly broken in so many way
when it comes to lifetime. I screw that part badly when reworking the
patchset, i was focusing on other part.
I fixed that in my tree, i am waiting for more review on
On Tue, 2015-06-09 at 18:23 +0530, Hariprasad Shenai wrote:
Hi,
This patch series adds support for user mode bar2 mappings for T4 adapter
and also adds support for bar2 qid densities exceeding page size.
This patch series has been created against Doug's github tree 'for-4.1'
branch and
On Thu, Jun 11, 2015 at 12:49:59AM -0400, Doug Ledford wrote:
fact that the mlx4 driver and the ocrdma driver had their own gid
management code, there were some distinct differences between the two.
The gid at index 0 never matched up in my testing for example. One
supported bonding, the
On Wed, 2015-06-10 at 21:57 -0600, Jason Gunthorpe wrote:
On Wed, Jun 10, 2015 at 09:06:28PM -0400, Doug Ledford wrote:
People tend to push the patches should be small, self contained,
incremental ideal. In some cases, that gets carried to an extreme. In
this case, patch 1 introduces
On Wed, 2015-06-10 at 12:13 +0300, Moni Shoua wrote:
Registering an event handler is done for a device. This device may have
one RoCE port (no SA cap) and one InfiniBand port (has SA cap).
Therefore, warning from the event handler about a specific port that
doesn't have SA cap is correct but
On Fri, 2015-06-05 at 15:47 +0100, Colin King wrote:
From: Colin Ian King colin.k...@canonical.com
A reorganisation of the PD allocation and deallocation in commit
9ba1377daa (RDMA/ocrdma: Move PD resource management to driver.)
introduced a double free on pd, as detected by static analysis
On Wed, Jun 10, 2015 at 08:15:56PM -0400, Doug Ledford wrote:
I'm not sure the complexity here is latent RoCEv2 stuff versus simple
over-design.
Well, for instance, the wrong RCU locking around
table-data_vec[ix].attr.ndev appears to exist to support find_gid
when called with
On Mon, Jun 08, 2015 at 05:12:15PM +0300, Matan Barak wrote:
From: Somnath Kotur somnath.ko...@emulex.com
1.Check and set port capability flags to indicate RoCEV2 support.
??? This series has nothing to with rocev2 now, what is this about?
mutex_init(dev-dev_lock);
- dev-sgid_tbl
On Mon, Jun 08, 2015 at 05:12:06PM +0300, Matan Barak wrote:
drivers/infiniband/core/core_priv.h | 26 ++
drivers/infiniband/core/device.c | 77 +
I wouldn't mind seeing the core portion which consists of adding
the get_netdev be it's own little mini-series of three, adding
On Wed, 2015-06-10 at 23:17 +, Diego Crupnicoff wrote:
Here Sean reacts to our RoCEv2 patches (same thread as the one I sent before
with Jason's feedback).
Sean is not even making technical statements to dismiss our patches. He has
been rejecting all our previous revs of this set with
On Wed, Jun 10, 2015 at 09:06:28PM -0400, Doug Ledford wrote:
People tend to push the patches should be small, self contained,
incremental ideal. In some cases, that gets carried to an extreme. In
this case, patch 1 introduces one side of the locking and patch 3 and 5
introduce the other
-Original Message-
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 10:40 AM
-Original Message-
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 9:37 AM
A SA cache is undeniably critical
Sean, this change is needed b/c two drivers have (mlx4 and ocrda) and
more two to come soon (mlx5 and soft-Roce) would have the very same
logic of constructing the port GID table according to netdev events and
such, no point in repeating this logic/code over and over.
Matan explained why we
On 6/10/2015 11:07 AM, Wan, Kaike wrote:
-Original Message-
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 10:40 AM
-Original Message-
From: Hal Rosenstock [mailto:h...@dev.mellanox.co.il]
Sent: Wednesday, June 10, 2015 9:37 AM
A
On 6/10/2015 6:09 PM, Hefty, Sean wrote:
Sean, this change is needed b/c two drivers have (mlx4 and ocrda) and
more two to come soon (mlx5 and soft-Roce) would have the very same
logic of constructing the port GID table according to netdev events and
such, no point in repeating this logic/code
While this appears to address the current upstream use model for ACM
with it's multicast overlay backend where PRs are static, it does not
appear to address PR changes.
Although this ties into ibacm, from the viewpoint of the kernel, there's no
requirement on the user space implementation.
On 6/10/2015 6:00 PM, Jason Gunthorpe wrote:
On Wed, Jun 10, 2015 at 11:53:15AM +0300, Or Gerlitz wrote:
Jason, can you ack that this post addressed your comments?
Well, I asked for a cleanup series, multiple times, and this is the
closest things have got.
It isn't really a cleanup
70 matches
Mail list logo