Add a new call to libibumad that allows a user to register
for snooping MADs. Snooped MADs are returned to the user
through the umad_recv routine.
A user can specify a filter for snooped MADs through the
new API. To support possible future extensions to the filter,
the size of the structure is a
In short, this series ports the kernel madeye debug utility to
user space. It builds on the updated kernel MAD snooping functionality
patches recently submitted to this list. In length, the series
adds the following to the ib-mgmt tree:
* Adds the ability to snoop MADs to libibumad.
* Adds new h
> Sean has been working on some stuff to allow for GID addressing with
> RDMA CM's API but I don't think it is finished yet (?)
In short, this is complete, just waiting for review and merging upstream.
- Sean
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of
> If you want to do non-blocking accept .. I'm not sure that is possible
> with RDMA CM.
non-blocking 'accept' functionality can be achieved by using get_event and
select/poll on the associated fd.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message
Thanks Dave, I took your branch into my for-next.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
thanks, applied.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
thanks, applied.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
I applied this, but can you check if there are any alignment/size issues
with 32-bit userspace on 64-bit kernel? Specifically:
> struct iwch_create_cq_resp {
> __u64 key;
> __u32 cqid;
> __u32 size_log2;
> +__u32 memsize;
> };
this structure now has size 20 on i386 (32
The immediate word for RDMA_WRITE_ONLY_WITH_IMMEDIATE was being
extracted from the wrong location in the header.
Thanks to Jason Gunthorpe for
finding this.
Signed-off-by: Ralph Campbell
---
drivers/infiniband/hw/qib/qib_rc.c |5 -
drivers/infiniband/hw/qib/qib_uc.c |6 --
2 fi
From: Jason Gunthorpe
Clean up properly if pci_set_consistent_dma_mask() fails.
Signed-off-by: Jason Gunthorpe
---
drivers/infiniband/hw/qib/qib_pcie.c |4 +++-
1 files changed, 3 insertions(+), 1 deletions(-)
diff --git a/drivers/infiniband/hw/qib/qib_pcie.c
b/drivers/infiniband/hw/qib
On Fri, Oct 22, 2010 at 03:29:54PM -0700, Ralph Campbell wrote:
> Some PCIe root complex chip sets don't support advanced error reporting.
> Allow the driver to load OK if pci_enable_pcie_error_reporting() fails.
Thanks
If you use this solution you'll want this hunk as well to avoid a leak:
diff
Trying to use a 64 bit kernel with a 32 bit userspace results in
local protection violations reported to the CQ. This is caused by
a difference in padding for ipath_rwe, so make the padding 64 bit
uses explicit.
Signed-off-by: Jason Gunthorpe
---
src/ipathverbs.h |1 +
1 files changed, 1 ins
Some PCIe root complex chip sets don't support advanced error reporting.
Allow the driver to load OK if pci_enable_pcie_error_reporting() fails.
Signed-off-by: Ralph Campbell
---
drivers/infiniband/hw/qib/qib_pcie.c |4 +++-
1 files changed, 3 insertions(+), 1 deletions(-)
diff --git a/dri
If CONFIG_PCI_MSI is not set, and a QLE7140 is present, the
pointer "dd" was uninitialized.
Signed-off-by: Ralph Campbell
---
drivers/infiniband/hw/qib/qib_init.c |1 +
1 files changed, 1 insertions(+), 0 deletions(-)
diff --git a/drivers/infiniband/hw/qib/qib_init.c
b/drivers/infiniband/
I don't think it needs the extra blank lines but I agree the code is correct.
From: Jason Gunthorpe [jguntho...@obsidianresearch.com]
Sent: Friday, October 22, 2010 3:00 PM
To: Ralph Campbell; RDMA list
Subject: [PATCH] [IB/QIB] Process RDMA WRITE ONLY with
On Fri, Oct 22, 2010 at 05:07:23PM -0400, Adit Ranadive wrote:
> I have been building a server client type of application based on top
> of RDMA/Verbs user libraries.
You can use select/poll/etc on CQ's after the connection has been
accepted, see ibv_create_comp_channel.
If you want to do non-bl
See table 35 in IBA - the header order for RDMA_WRITE_ONLY_WITH_IMMEDIATE
and SEND_LAST_WITH_IMMEDIATE is different, the RDMA_WRITE_ONLY has
a RETH header before the immediate data, so we need a different code path
to extract the immediate data.
I tested this with a userspace app that does RDMA_WR
On Fri, Oct 22, 2010 at 05:25:25PM -0400, Adit Ranadive wrote:
> Is there a way to use the rdma connection manager library without the
> IPoIB up and running?
Sean has been working on some stuff to allow for GID addressing with
RDMA CM's API but I don't think it is finished yet (?)
The only other
pci_enable_pcie_error_reporting() is optional.
From: linux-rdma-ow...@vger.kernel.org [linux-rdma-ow...@vger.kernel.org] On
Behalf Of Jason Gunthorpe [jguntho...@obsidianresearch.com]
Sent: Friday, October 22, 2010 2:36 PM
To: Ralph Campbell
Cc: RDMA list
S
Acked-by: Ralph Campbell
From: Jason Gunthorpe [jguntho...@obsidianresearch.com]
Sent: Friday, October 22, 2010 1:41 PM
To: Ralph Campbell; RDMA list
Subject: [PATCH] [IB/QIB] Fix QIB early error reporting prints
Noticed this odd looking thing in dmesg:
i
On Fri, Oct 22, 2010 at 02:28:54PM -0700, Ralph Campbell wrote:
> I'm not sure this is the right change. I agree there are a lot of
> clean up on error bugs. pci_set_consistent_dma_mask() isn't optional
> because the chip doesn't support 32-bit PCIe addressing.
Do you think pci_enable_pcie_error_
> Is there a way to use the rdma connection manager library without the
> IPoIB up and running?
Use IBoE or an iWARP adapter? ;)
More seriously, on IB, no, you must configure IPoIB to use the RDMA CM.
The whole point of the RDMA CM is to use IP addressing and ARP to
resolve addresses, which is
> Roland, this patch was intended to be sent with the whole series but
> somehow did not find its way out. I hope it you can intergrate it too.
Sure, looks fine. I think I have all the IBoE stuff in my for-next
branch now.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma"
I'm not sure this is the right change. I agree there are a lot of clean up on
error
bugs. pci_set_consistent_dma_mask() isn't optional because the chip doesn't
support 32-bit PCIe addressing.
I'll look it over and try to come up with something better.
I'll submit a patch for the CONFIG_PCI_MSI no
Is there a way to use the rdma connection manager library without the
IPoIB up and running?
Thanks,
Adit
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
> @@ -1801,6 +1831,11 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct
> ib_send_wr *wr,
> ctrl->owner_opcode = mlx4_ib_opcode[wr->opcode] |
> (ind & qp->sq.wqe_cnt ? cpu_to_be32(1 << 31) : 0) | blh;
>
> +if (vlan < 0x1000) {
> +
Hello,
I have been building a server client type of application based on top
of RDMA/Verbs user libraries.
The issue that Im running into is how to multiplex between multiple
clients connecting to the server using rdma_listen & rdma_connect
functions.
Looking at he man pages the rdma_listen call s
The idea was to have QPs allocated to queues with specific CPUs
processing the queue "local" to the process that created it.
Since IRQs get distributed to all CPUs by default (at least with irqbalance)
and the fact that the CPU that created the QP may not be the CPU doing
the posts, it didn't help.
Simultaneous DREQ processing from user and CM thread caused some improper
state change on UCM. State change can incorrectly change from FREE back
to DISC in certain corner cases. Add checking on internal disconnect call
to prevent double callback events and improper state change.
For SCM, a remote
This seems to be the intention of the code, since the jump to bail
is missing. PCI-E advanced error reporting seems optional, but
I wonder if pci_set_consistent_dma_mask is also optional?
This also fixes one case where the PCI region is leaked during
device startup. qib_init_one assumes that qib_p
Noticed this odd looking thing in dmesg:
ib_qib :02:00.0: <3>ib_qib: Unable to enable pcie error reporting: -5
Which is due to a bad use of dev_info.
Signed-off-by: Jason Gunthorpe
---
drivers/infiniband/hw/qib/qib.h |2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a
Hi Ralph,
It doesn't look like smp_processor_id is callable the way QIB does, I
get kernel BUGS:
This is a stock upstream with CONFIG_PREEMPT/DEBUG_PREEMPT
$ uname -a
Linux ib6 2.6.35.7 #17 SMP PREEMPT Thu Oct 7 15:38:53 MDT 2010 x86_64 GNU/Linux
[ 36.769466] BUG: using smp_processor_id() in p
Signed-off-by: Arlin Davis
---
dapl/openib_cma/cm.c |9 -
1 files changed, 4 insertions(+), 5 deletions(-)
diff --git a/dapl/openib_cma/cm.c b/dapl/openib_cma/cm.c
index 9338d55..1eb7aed 100644
--- a/dapl/openib_cma/cm.c
+++ b/dapl/openib_cma/cm.c
@@ -105,7 +105,6 @@ static void dapl
Signed-off-by: Arlin Davis
---
dapl/common/dapl_debug.c | 13 +++--
1 files changed, 11 insertions(+), 2 deletions(-)
diff --git a/dapl/common/dapl_debug.c b/dapl/common/dapl_debug.c
index 904d075..e1a940c 100644
--- a/dapl/common/dapl_debug.c
+++ b/dapl/common/dapl_debug.c
@@ -36,6 +
Signed-off-by: Arlin Davis
---
dapl/common/dapl_evd_dequeue.c | 11 ++-
1 files changed, 6 insertions(+), 5 deletions(-)
diff --git a/dapl/common/dapl_evd_dequeue.c b/dapl/common/dapl_evd_dequeue.c
index 7632fe1..f31e497 100644
--- a/dapl/common/dapl_evd_dequeue.c
+++ b/dapl/common/da
The QP's need to be flushed and processed via EVD's during
the EP (QP) destroy to avoid an error on poll_cq. IBAL
provider was not moving to ERR state during QP destroy.
Better flush CQ processing was added and pushed down to the provider
level via dapls_ib_qp_free() where it can move QP to ERR, f
Add new DAPL_UCM_TX_BURST environment variable, default=50.
Every 50 posted send messages will signal event which
is 10 percent bursting of default 500 message limit.
Signed-off-by: Arlin Davis
---
dapl/openib_common/dapl_ib_common.h |1 +
dapl/openib_ucm/cm.c| 27 +
Can you please resubmit both patches with signed-off-by lines?
Thanks,
Sean
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Eli Cohen wrote:
> [...] Address resolution is done atomically in the
> case of a link local address or a multicast GID and otherwise -EINVAL is
> returned. mlx4 transport packets were changed too to accommodate for IBoE.
> Multicast groups attach/detach calls dev_mc_add/remove to update the NIC'
Since an IB port may use either IB or Ethernet as its link layer, add
the file /sys/class/infiniband//ports//link_layer to
provide the configured link layer for the port.
Signed-off-by: Eli Cohen
---
Roland, this patch was intended to be sent with the whole series but
somehow did not find its way
On Thu, Oct 21, 2010 at 09:15:35PM -0700, Roland Dreier wrote:
> We don't need ICRC in the header_buf, do we? We're talking about
> MLX4_IB_UD_HEADER_SIZE, which is used to hold the packet header we get
> from ib_ud_header_pack(). The ICRC is accounted for in a different
> segment of the WQE.
>
make install fails if the include files in the install prefix
include/rdma,infiniband already exist. install claims that the
and file are the same and exits with an error.
This patch modifies Makefile.am so that the rdma and infiniband include
files explicitly reference the source directory rat
On Oct 22, 2010 12:00 PM, Jonathan Rosser
wrote:
> Arithmetic on void* pointers generates a compiler warning, and
> projects
> that include rdma/rdma_verbs.h and compile with -Werror -Wall will
> fail
> to build.
> ---
> include/rdma/rdma_verbs.h |2 +-
> 1 files changed, 1 insertions(+),
Arithmetic on void* pointers generates a compiler warning, and projects
that include rdma/rdma_verbs.h and compile with -Werror -Wall will fail
to build.
---
include/rdma/rdma_verbs.h |2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/include/rdma/rdma_verbs.h b/include/rd
44 matches
Mail list logo