Octavian Purdila wrote:
On Wednesday 03 February 2010 06:30:07 you wrote:
This patch introduces /proc/sys/net/ipv4/ip_local_reserved_ports,
it can be used like ip_local_port_range, but this is used to
reserve ports for third-party applications which use fixed
port numbers within ip_local_port_r
Completion channels are created with ia_open for CNO events and
with ep_create in cases where DAT allows EP(qp) to be created with
no EVD(cq) and IB doesn't. These completion channels need to be
destroyed at close along with a CQ for the "EP without EVD" case.
Signed-off-by: Arlin Davis
---
dap
On 02/03/2010 09:41 AM, Or Gerlitz wrote:
@@ -526,17 +523,8 @@ void iser_snd_completion(struct iser_tx_
kmem_cache_free(ig.desc_cache, tx_desc);
}
- if (atomic_read(&iser_conn->ib_conn->post_send_buf_count) ==
- ISER_QP_MAX_REQ_DTOS)
- resume
> +int mlx4_bitmap_init_no_mask(struct mlx4_bitmap *bitmap, u32 num,
> + u32 reserved_bot, u32 reserved_top)
> +{
> +u32 num_rounded = roundup_pow_of_two(num);
> +return mlx4_bitmap_init(bitmap, num_rounded, num_rounded - 1,
> +reser
Update Copyright file and include the 3 license files in distribution
Signed-off-by: Arlin Davis
---
COPYING | 16
Makefile.am |3 +++
dapl.spec.in |2 +-
3 files changed, 8 insertions(+), 13 deletions(-)
diff --git a/COPYING b/COPYING
index 2012c04..35ba319 100
Update Copyright file and include the 3 license files in distribution
Signed-off-by: Arlin Davis
---
COPYING | 16
Makefile.am |3 +++
dapl.spec.in |2 +-
3 files changed, 8 insertions(+), 13 deletions(-)
diff --git a/COPYING b/COPYING
index 2012c04..35ba319 100
> I think I have uncovered a couple of bugs in my review (not surprising,
> since you
> most likely could not try out the code).
Thanks. Yes, this was all compile-tested only.
> 1. the xrcd refcnt is incremented when creating an xrc qp and xrc srq, but
> is unconditionally
>decrement
On Wed, Feb 3, 2010 at 4:30 PM, Or Gerlitz wrote:
>
> The following patch set removes some in efficiencies in the iser data path
> through simplification and reducing the amount of code, using less atomic
> operations, avoiding TX interrupts, moving to iscsi passthrough mode,
> etc. I did my best
On Wed, Feb 3, 2010 at 11:59 AM, Yevgeny Kliteynik
wrote:
> On 03/Feb/10 11:43, Eli Dorfman (Voltaire) wrote:
>>
>> Currently there is a duplication of port group definition in OpenSM
>> configuration files.
>> Using a single configuration file for port groups will make it easier to
>> manage and
currently iser recv completion flow takes the session lock twice.
optimize it to avoid the first one by letting iser_task_rdma_finalize()
be called only from the cleanup_task callback invoked by iscsi_free_task,
thus reducing the contention on the session lock between the scsi
command submission to
currently iser recv completion flow takes the session lock twice.
optimize it to avoid the first one by letting iser_task_rdma_finalize()
be called only from the cleanup_task callback invoked by iscsi_free_task,
thus reducing the contention on the session lock between the scsi
command submission to
The libiscsi passthrough mode invokes the transport xmit calls directly
without passing through an internal queue, now when the "cant_sleep"
prerequisite of iscsi_alloc_session is met, move to use it. Since the
libibscsi queue isn't used in passthrough mode, the code that schedules
the xmitworker i
remove unnecessary checks for the IB connection state and for QP
overflow, as conn state changes are reported by iser to libiscsi
and handled there. QP overflow is theoretically possible only when
unsolicited data-outs are used, its being checked and handled
by HW drivers.
Signed-off-by: Or Gerlit
Two minor flows in iser's data path still use allocations,
move them to be atomic as a preperation step towards moving
to use libiscsi passthrough flow.
Signed-off-by: Or Gerlitz
---
drivers/infiniband/ulp/iser/iser_initiator.c |2 +-
drivers/infiniband/ulp/iser/iser_memory.c|4 ++--
Simplify and shrink the logic/code used for the send descriptors.
Changes include removal of struct iser_dto which is unnecessary
abstraction, use struct iser_regd_buf only for handling SCSI commands,
use dma_sync instead of dma_map/unmap, etc.
Signed-off-by: Or Gerlitz
---
drivers/infiniband/ul
Use a different CQ for send completions, where send completions are
being polled by the interrupt driven recv completion handler.
As such, interrupts aren't used for the send CQ.
Signed-off-by: Or Gerlitz
---
drivers/infiniband/ulp/iser/iscsi_iser.h |3
drivers/infiniband/ulp/iser/iser_verb
With both the posting and reaping of recv buffers being in the
completion path, their outstanding number counter need not be atomic.
Signed-off-by: Or Gerlitz
---
drivers/infiniband/ulp/iser/iscsi_iser.h |2 +-
drivers/infiniband/ulp/iser/iser_initiator.c |6 +++---
drivers/infiniba
Currently, the recv buffer posting logic is based on the transactional
nature of iser which allows for posting a buffer before sending a PDU.
Change this to post only when the number of outstanding recv buffers is
below a water mark and in a batched manner, thus simplifying and
optimizing the data
towards a major change in the recv buffer posting logic, with which
the problem commit bba7ebb "avoid recv buffer exhaustion caused by
unexpected PDUs" comes to solve doesn't exist any more, revert it.
Signed-off-by: Or Gerlitz
CC: David Disseldorp
CC: Ken Sandars
---
drivers/infiniband/ulp/is
The following patch set removes some in efficiencies in the iser data path
through simplification and reducing the amount of code, using less atomic
operations, avoiding TX interrupts, moving to iscsi passthrough mode,
etc. I did my best to build it as a sequence of patches and not as one
big re-wr
On Wed, Feb 03, 2010 at 03:17:17PM +0100, Rui Machado wrote:
>
> Great, thanks. I increased the log_mtts_per_seg to 5 and was able to
> register more memory. But one question remaining: what is then the
> maximum memory that one is able to register? 5 is the maximum value to
> this parameter, so w
Hi there,
>>
>> The objective is to be able to memory-pin large portion of a machine
>> with lots of RAM, otherwise we are limited to about 16GB.
>> Does anyone has an hint or thoughts on what might be the problem?
>
> For this you can use the parameter log_mtts_per_seg which allows to
> define th
On Wed, Feb 03, 2010 at 11:02:07AM +0100, Rui Machado wrote:
>
> on trying to change the mlx4_core parameters log_num_mtt and
> log_num_mpt to 21 and 18 respectively, I see the following message:
The MPTs and MTTs are managed by bitmap allocators within the mlx4
core drivers. These allocators req
On 15:29 Tue 02 Feb , Yevgeny Kliteynik wrote:
> Hi Sasha,
>
> Fixing a bug in cl_ptr_vector_find_from_end(), discovered
> by compiler's warning:
>
> cl_ptr_vector.c: In function 'cl_ptr_vector_find_from_end':
> cl_ptr_vector.c:305: warning: operation on 'i' may be undefined
>
> Signed-off-b
Hi Sasha,
On 03/Feb/10 12:51, Sasha Khapyorsky wrote:
Hi Yevgeny,
On 15:24 Tue 02 Feb , Yevgeny Kliteynik wrote:
There are couple of problematic castings in complib/cl_passivelock.h:
(cl_status_t) pthread_rwlock_*()
Turns out that compiler doesn't like casting directly from
int to enum,
On Wednesday 03 February 2010 06:30:07 you wrote:
> This patch introduces /proc/sys/net/ipv4/ip_local_reserved_ports,
> it can be used like ip_local_port_range, but this is used to
> reserve ports for third-party applications which use fixed
> port numbers within ip_local_port_range.
>
> This onl
Optimize port selection logic when port LID is explicitly requested
or wildcarded. Remove unneeded variable.
Signed-off-by: Sasha Khapyorsky
---
opensm/opensm/osm_sa_pkey_record.c | 19 +--
1 files changed, 5 insertions(+), 14 deletions(-)
diff --git a/opensm/opensm/osm_sa_pk
Memory leak is possible in case of port list allocation failure. Fix
this by proper cleanup.
Signed-off-by: Sasha Khapyorsky
---
opensm/opensm/osm_mcast_mgr.c |2 ++
1 files changed, 2 insertions(+), 0 deletions(-)
diff --git a/opensm/opensm/osm_mcast_mgr.c b/opensm/opensm/osm_mcast_mgr.c
On 14:46 Mon 01 Feb , Yevgeny Kliteynik wrote:
> osm_get_physp_by_mad_addr() function is defined twice in
> the same header file. Removing one of the definitions
> and fixing compilation warning.
>
> Signed-off-by: Yevgeny Kliteynik
Applied. Thanks.
Sasha
--
To unsubscribe from this list: s
Hi Yevgeny,
On 15:24 Tue 02 Feb , Yevgeny Kliteynik wrote:
>
> There are couple of problematic castings in complib/cl_passivelock.h:
>
> (cl_status_t) pthread_rwlock_*()
>
> Turns out that compiler doesn't like casting directly from
> int to enum, probably because int can be negative:
>
>
From: Slava Strebkov
Proposed new algorithm for calculation of root switch for multicast
spanning tree. Only edge switches(those connected to hosts or routers)
and switches - multicast members themselves are involved in root
calculation. This gives improvement, especially on large fabrics, since
On 10:39 Thu 28 Jan , Hal Rosenstock wrote:
> On Wed, Jan 27, 2010 at 5:45 AM, Sasha Khapyorsky wrote:
> > On 13:59 Wed 20 Jan , Sasha Khapyorsky wrote:
> >> On 13:32 Wed 20 Jan , Slava Strebkov wrote:
> >> > "average hops" was chosen instead of "max hops" because in root weight
> >> >
Hi list,
on trying to change the mlx4_core parameters log_num_mtt and
log_num_mpt to 21 and 18 respectively, I see the following message:
mlx4_core :07:00.0: Failed to initialize memory region table, aborting.
The objective is to be able to memory-pin large portion of a machine
with lots of
On 03/Feb/10 11:43, Eli Dorfman (Voltaire) wrote:
Currently there is a duplication of port group definition in OpenSM
configuration files.
Using a single configuration file for port groups will make it easier to manage
and will allow using group name as keyword in other configuration files (i.e
Hi Eli,
On 16:13 Tue 02 Feb , Eli Dorfman (Voltaire) wrote:
>
> In this case the GID in the data details is the MGID and
> not the source gid. So the SM gid should be taken as the source gid.
There the source port is detected in order P_Key matching check between
source and trap destination
Currently there is a duplication of port group definition in OpenSM
configuration files.
Using a single configuration file for port groups will make it easier to manage
and will allow using group name as keyword in other configuration files (i.e.
partitions.conf, qos-policy.conf and opensm.conf)
Check whether the loading is needed in the state mgr.
Signed-off-by: Yevgeny Kliteynik
---
Changes since v1:
stupid copy-paste artifact - failed compilation
opensm/opensm/osm_sa.c|6 --
opensm/opensm/osm_state_mgr.c | 11 +++
2 files changed, 7 insertions(+),
Check whether the loading is needed in the state mgr.
Signed-off-by: Yevgeny Kliteynik
---
opensm/opensm/osm_sa.c|6 --
opensm/opensm/osm_state_mgr.c | 11 +++
2 files changed, 7 insertions(+), 10 deletions(-)
diff --git a/opensm/opensm/osm_sa.c b/opensm/opensm/osm_sa.
Sean Hefty wrote:
Will Mellanox be adding this option to the IB spec rather than keep it
as vendor proprietary ?
Along with that question, what is the use case for this feature? The only
benefit mentioned was saving a few bytes of memory, at the cost of carrying
extra network headers. Atomics
39 matches
Mail list logo