Re: [PATCH libibverbs] init.c: increase sysfs read buffer size to 16

[Jeff Squyres (jsquyres)]

Any further comments on this?

Doug -- does it look ok to you?


> On Dec 7, 2015, at 5:27 AM, Haggai Eran  wrote:
> 
> On 12/04/2015 01:09 AM, Jeff Squyres wrote:
>> The default value of 8 is too small to read
>> /sys/class/infiniband/usnic_x/node_type, which contains "6: usNIC
>> UDP".  Per a7a73a8c1b39362f1701256bc772d82847832f9c, the too-small
>> buffer causes a stderr warning to be emitted from ibv_devinfo when
>> reading usNIC devices.
>> 
>> This commit therefore increases the buffer size to 16, which is long
>> enough to read the usNIC node_type value.
> 
> Reviewed-by: Haggai Eran 


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH v3] libibverbs init.c: conditionally emit warning if no userspace driver found

[Jeff Squyres (jsquyres)]

On Jun 17, 2015, at 10:25 AM, Doug Ledford dledf...@redhat.com wrote:
 
 The patch is accepted, I just haven’t pushed it out yet.

Is there a timeline for when this patch will be available in the upstream git 
repo and released in a new version of libibverbs?

I ask because we'd like to see this patch get into upstream distro libibverbs 
releases.  Once that happens, we can start planning the end of the horrible 
hackarounds we had to put into place (e.g., in Open MPI) to suppress the 
misleading libibverbs output.

Thanks!

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

N�r��yb�X��ǧv�^�)޺{.n�+{��ٚ�{ay�ʇڙ�,j��f���h���z��w���
���j:+v���w�j�mzZ+�ݢj��!�i

Re: [PATCH v3] libibverbs init.c: conditionally emit warning if no userspace driver found

[Jeff Squyres (jsquyres)]

Ping.

This is just a periodic query to see if there has been any progress on 
accepting this patch into libibverbs.



 On Jun 3, 2015, at 12:50 PM, Doug Ledford dledf...@redhat.com wrote:
 
 On Mon, 2015-06-01 at 22:02 +, Jeff Squyres (jsquyres) wrote:
 On May 22, 2015, at 9:44 AM, Doug Ledford dledf...@redhat.com wrote:
 
 Did that happen yet?
 
 I don't think so.  I didn't file a specific ticket for it at k.o yet
 (the k.o tickets take a while to process, so I didn't want to file it
 until after the comment period here on list).
 
 Ping.
 
 This is just a periodic query to see if there has been any progress on 
 accepting this patch into libibverbs.
 
 
 I have a ticket with kernel.org helpdesk to change the permissions on
 the libibverbs.git repo, and they are waiting on Roland to ACK the
 change.  Until then, I can't do much.
 
 -- 
 Doug Ledford dledf...@redhat.com
  GPG KeyID: 0E572FDD
 


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH v3] libibverbs init.c: conditionally emit warning if no userspace driver found

[Jeff Squyres (jsquyres)]

On May 22, 2015, at 9:44 AM, Doug Ledford dledf...@redhat.com wrote:
 
 Did that happen yet?
 
 I don't think so.  I didn't file a specific ticket for it at k.o yet
 (the k.o tickets take a while to process, so I didn't want to file it
 until after the comment period here on list).

Ping.

This is just a periodic query to see if there has been any progress on 
accepting this patch into libibverbs.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH v3] libibverbs init.c: conditionally emit warning if no userspace driver found

[Jeff Squyres (jsquyres)]

On May 20, 2015, at 1:11 PM, Doug Ledford dledf...@redhat.com wrote:
 
 The location of the upstream sources and tarballs would not change.
 Neither the git repo nor the tarball repo were like the kernel.  The
 upstream kernel.org git repo Roland had, had his name in the repo.  So
 it had to change.  But the libibverbs repo is in a generic location.
 There is no need to change it, only to change the permissions on the git
 repo to allow the new maintainer to push directly into it.  

Did that happen yet?

 Ditto with
 the upload/download space on openfabrics.org/downloads/verbs.

It looks like someone did part of this on flatbed -- you own the download 
directory but none of the files, and they are all 644.  So I took the liberty 
of chown'ing them all to you:


$ hostname; pwd; ls -la
flatbed.openfabrics.org
/var/www/html/downloads/verbs
total 6096
drwxr-xr-x.  2 dledford ofed   4096 May  7  2014 .
drwxrwxr-x. 55 apache   ofed   4096 Feb 13 07:31 ..
-rw-r--r--.  1 dledford ofed 347508 Mar 14  2006 libibverbs-1.0.2.tar.gz
-rw-r--r--.  1 dledford ofed 349439 May  2  2006 libibverbs-1.0.3.tar.gz
-rw-r--r--.  1 dledford ofed 360410 Oct 31  2006 libibverbs-1.0.4.tar.gz
-rw-r--r--.  1 dledford ofed 359902 Jun 18  2007 libibverbs-1.0.5.tar.gz
-rw-r--r--.  1 dledford ofed 321835 Aug 29  2005 libibverbs-1.0-rc1.tar.gz
-rw-r--r--.  1 dledford ofed 338537 Oct  2  2005 libibverbs-1.0-rc3.tar.gz
-rw-r--r--.  1 dledford ofed 341792 Oct 28  2005 libibverbs-1.0-rc4.tar.gz
-rw-r--r--.  1 dledford ofed 347699 Feb 17  2006 libibverbs-1.0-rc7.tar.gz
-rw-r--r--.  1 dledford ofed 384743 Jun 18  2007 libibverbs-1.1.1.tar.gz
-rw-r--r--.  1 dledford ofed 394618 Apr 18  2008 libibverbs-1.1.2.tar.gz
-rw-r--r--.  1 dledford ofed 359331 Oct 29  2009 libibverbs-1.1.3.tar.gz
-rw-r--r--.  1 dledford ofed 362475 Jun  3  2010 libibverbs-1.1.4.tar.gz
-rw-r--r--.  1 dledford ofed 364219 Jun 28  2011 libibverbs-1.1.5.tar.gz
-rw-r--r--.  1 dledford ofed 387794 Dec 21  2011 libibverbs-1.1.6.tar.gz
-rw-r--r--.  1 dledford ofed 391812 May 28  2013 libibverbs-1.1.7.tar.gz
-rw-r--r--   1 dledford ofed 406548 May  5  2014 libibverbs-1.1.8.tar.gz
-rw-r--r--.  1 dledford ofed 384656 Apr 24  2007 libibverbs-1.1.tar.gz
-rw-r--r--   1 dledford ofed   3957 May  7  2014 README.html
-rw-r--r--.  1 dledford ofed 60 Mar 12  2008 WEB_README
-

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs init.c: remove stderr warnings if no userspace driver found

[Jeff Squyres (jsquyres)]

On May 9, 2015, at 8:04 AM, Yann Droneaud ydrone...@opteya.com wrote:
 
 Le vendredi 08 mai 2015 à 11:21 -0700, Jeff Squyres a écrit :
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 
 This is a little short for an explanation: what was the issue with the
 error messages ?

Cisco has stopped shipping its libibverbs usnic driver, although we are still 
using the kernel driver in the /sys/class/infiniband space (since it's the only 
way to be upstream).  Specifically: instead of using libibverbs for userspace 
access, we are now using libfabric.

That is: it's not a warning or an error if libibverbs cannot find a userspace 
driver for kernel devices.  Indeed, returning a num_devices of 0 is sufficient 
-- the middleware shouldn't be unconditionally printing out stderr message; let 
the upper layer application do that (if it wants to).

FWIW, Sean just removed a similar set of stderr warnings from librdmacm:

   
http://git.openfabrics.org/?p=~shefty/librdmacm.git;a=commitdiff;h=2b2aad809afc56fa3157f5cf99036f92b9c90f16

 -free(sysfs_dev);
 
 I believe this free() was necessary to not leak some memory.

Ah -- I mis-read the loop.  I'll re-submit with the loop still there, but just 
removing the fprintf block.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

N�r��yb�X��ǧv�^�)޺{.n�+{��ٚ�{ay�ʇڙ�,j��f���h���z��w���
���j:+v���w�j�mzZ+�ݢj��!�i

Re: [PATCH libibverbs V2] Add new verb: uv_query_port_max_datagram()

[Jeff Squyres (jsquyres)]

Bump.  This is V2 of the patch, which removes the ABI issue: libibverbs 
directly calls the command in the kernel (without going through the provider 
plugin).

On Aug 21, 2013, at 5:22 PM, Jeff Squyres jsquy...@cisco.com wrote:

 Per lengthy discussion on the linux-rdma list, add a new verb to get
 max datagram size (in bytes) since the methods for retrieving MTU
 values are limited to a finite enum set, and are difficult to change
 for backwards compatibility reasons.  
 
 Also add corresponding command: uv_cmd_query_port_max_datagram().
 Since this is a new verb, there was no need to add a _V2 enum for the
 command macro, which required adding a UB_INIT_CMD_RESP() macro.
 
 If the kernel does not support the new QUERY_PORT_MAX_DATAGRAM
 command, fall back to returning the int-ized MTU enum from
 ibv_cmd_query_port().
 
 Note that the name for this verb was chosen with the following
 rationale:
 
 * After discussion with Roland, use the prefix uv instead of ibv,
  since this verb is generic to both Ethernet, InfiniBand, and
  whatever other transports are underneath.
 * query was used (vs. get) because it invokes a command (vs. a
  struct lookup)
 
 If the community likes this approach, I'll send the corresponding
 kernel patch.
 
 Difference from V1 
 ==
 Do not add this verb to the devops struct (because that would break ABI).
 Instead, just have uv_query_port_max_datagram() directly invoke 
 uv_cmd_query_port_max_datagram().
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 Makefile.am  |  3 +-
 examples/devinfo.c   |  7 +
 include/infiniband/driver.h  |  4 +++
 include/infiniband/kern-abi.h| 17 +++-
 include/infiniband/verbs.h   |  6 
 man/uv_query_port_max_datagram.3 | 59 
 src/cmd.c| 54 
 src/ibverbs.h|  8 ++
 src/libibverbs.map   |  2 ++
 src/verbs.c  | 13 +
 10 files changed, 171 insertions(+), 2 deletions(-)
 create mode 100644 man/uv_query_port_max_datagram.3
 
 diff --git a/Makefile.am b/Makefile.am
 index 40e83be..51fe5d5 100644
 --- a/Makefile.am
 +++ b/Makefile.am
 @@ -54,7 +54,8 @@ man_MANS = man/ibv_asyncwatch.1 man/ibv_devices.1 
 man/ibv_devinfo.1 \
 man/ibv_post_srq_recv.3 man/ibv_query_device.3 man/ibv_query_gid.3
 \
 man/ibv_query_pkey.3 man/ibv_query_port.3 man/ibv_query_qp.3  \
 man/ibv_query_srq.3 man/ibv_rate_to_mult.3 man/ibv_reg_mr.3   
 \
 -man/ibv_req_notify_cq.3 man/ibv_resize_cq.3 man/ibv_rate_to_mbps.3
 +man/ibv_req_notify_cq.3 man/ibv_resize_cq.3 man/ibv_rate_to_mbps.3   
 \
 +man/uv_query_port_max_datagram.3
 
 DEBIAN = debian/changelog debian/compat debian/control debian/copyright \
 debian/ibverbs-utils.install debian/libibverbs1.install \
 diff --git a/examples/devinfo.c b/examples/devinfo.c
 index ff078e4..f51620b 100644
 --- a/examples/devinfo.c
 +++ b/examples/devinfo.c
 @@ -209,6 +209,7 @@ static int print_hca_cap(struct ibv_device *ib_dev, 
 uint8_t ib_port)
   struct ibv_port_attr port_attr;
   int rc = 0;
   uint8_t port;
 + uint32_t max_datagram;
   char buf[256];
 
   ctx = ibv_open_device(ib_dev);
 @@ -298,6 +299,11 @@ static int print_hca_cap(struct ibv_device *ib_dev, 
 uint8_t ib_port)
   fprintf(stderr, Failed to query port %u props\n, 
 port);
   goto cleanup;
   }
 + rc = uv_query_port_max_datagram(ctx, port, max_datagram);
 + if (rc) {
 + fprintf(stderr, Failed to query port %u max datagram 
 size\n, port);
 + goto cleanup;
 + }
   printf(\t\tport:\t%d\n, port);
   printf(\t\t\tstate:\t\t\t%s (%d)\n,
  port_state_str(port_attr.state), port_attr.state);
 @@ -305,6 +311,7 @@ static int print_hca_cap(struct ibv_device *ib_dev, 
 uint8_t ib_port)
  mtu_str(port_attr.max_mtu), port_attr.max_mtu);
   printf(\t\t\tactive_mtu:\t\t%s (%d)\n,
  mtu_str(port_attr.active_mtu), port_attr.active_mtu);
 + printf(\t\t\tmax_datagram_size:\t%u\n, max_datagram);
   printf(\t\t\tsm_lid:\t\t\t%d\n, port_attr.sm_lid);
   printf(\t\t\tport_lid:\t\t%d\n, port_attr.lid);
   printf(\t\t\tport_lmc:\t\t0x%02x\n, port_attr.lmc);
 diff --git a/include/infiniband/driver.h b/include/infiniband/driver.h
 index 9a81416..6e1236c 100644
 --- a/include/infiniband/driver.h
 +++ b/include/infiniband/driver.h
 @@ -67,6 +67,10 @@ int ibv_cmd_query_device(struct ibv_context *context,
 int ibv_cmd_query_port(struct ibv_context *context, uint8_t port_num,
  struct ibv_port_attr *port_attr,
  struct ibv_query_port *cmd, size_t cmd_size);
 +int 

Re: [PATCH libibverbs] Add new verb: uv_query_port_max_datagram()

[Jeff Squyres (jsquyres)]

On Aug 19, 2013, at 4:19 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 What about doing query port in this case and returning that value,
 decoded to an enum? Otherwise apps have to include that logic anyhow.
 
 I'm assuming the kernel will do basically the same?
 
 Bascially, the only failure for this call should be due to a bad port
 number..


Sure, can do.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] Add new verb: uv_query_port_max_datagram()

[Jeff Squyres (jsquyres)]

On Aug 19, 2013, at 5:18 PM, Hefty, Sean sean.he...@intel.com wrote:

 Bumped the ABI version to 7 (the new verb will return -ENOSYS if
 abi_verb is  7).
 
 How does this break the ABI?


It doesn't *break* the ABI, but it does add a new downcall into the kernel.  
That requires bumping the ABI version to 7, no?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] Add new verb: uv_query_port_max_datagram()

[Jeff Squyres (jsquyres)]

On Aug 19, 2013, at 6:07 PM, Hefty, Sean sean.he...@intel.com wrote:

 It doesn't *break* the ABI, but it does add a new downcall into the kernel.
 That requires bumping the ABI version to 7, no?
 
 No - adding a new command is fine.  Older kernels will return ENOSYS if that 
 command is not supported.  In that case, you can handle things like Jason 
 suggested.


Gotcha.  I'll adjust the patch.

Any other feedback?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: Add the use of IBV_SEND_INLINE to example pingpong programs

[Jeff Squyres (jsquyres)]

4th bump...

On Jul 10, 2013, at 4:32 PM, Jeff Squyres jsquy...@cisco.com wrote:

 If the send size is less than the cap.max_inline_data reported by the
 qp, use the IBV_SEND_INLINE flag.  This now only shows the example of
 using ibv_query_qp(), it also reduces the latency time shown by the
 pingpong programs when the sends can be inlined.
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 examples/rc_pingpong.c  | 18 +-
 examples/srq_pingpong.c | 19 +--
 examples/uc_pingpong.c  | 17 -
 examples/ud_pingpong.c  | 18 +-
 4 files changed, 51 insertions(+), 21 deletions(-)
 
 diff --git a/examples/rc_pingpong.c b/examples/rc_pingpong.c
 index 15494a1..a8637a5 100644
 --- a/examples/rc_pingpong.c
 +++ b/examples/rc_pingpong.c
 @@ -65,6 +65,7 @@ struct pingpong_context {
   struct ibv_qp   *qp;
   void*buf;
   int  size;
 + int  send_flags;
   int  rx_depth;
   int  pending;
   struct ibv_port_attr portinfo;
 @@ -319,8 +320,9 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   if (!ctx)
   return NULL;
 
 - ctx-size = size;
 - ctx-rx_depth = rx_depth;
 + ctx-size   = size;
 + ctx-send_flags = IBV_SEND_SIGNALED;
 + ctx-rx_depth   = rx_depth;
 
   ctx-buf = memalign(page_size, size);
   if (!ctx-buf) {
 @@ -367,7 +369,8 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   }
 
   {
 - struct ibv_qp_init_attr attr = {
 + struct ibv_qp_attr attr;
 + struct ibv_qp_init_attr init_attr = {
   .send_cq = ctx-cq,
   .recv_cq = ctx-cq,
   .cap = {
 @@ -379,11 +382,16 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   .qp_type = IBV_QPT_RC
   };
 
 - ctx-qp = ibv_create_qp(ctx-pd, attr);
 + ctx-qp = ibv_create_qp(ctx-pd, init_attr);
   if (!ctx-qp)  {
   fprintf(stderr, Couldn't create QP\n);
   goto clean_cq;
   }
 +
 + ibv_query_qp(ctx-qp, attr, IBV_QP_CAP, init_attr);
 + if (init_attr.cap.max_inline_data = size) {
 + ctx-send_flags |= IBV_SEND_INLINE;
 + }
   }
 
   {
 @@ -508,7 +516,7 @@ static int pp_post_send(struct pingpong_context *ctx)
   .sg_list= list,
   .num_sge= 1,
   .opcode = IBV_WR_SEND,
 - .send_flags = IBV_SEND_SIGNALED,
 + .send_flags = ctx-send_flags,
   };
   struct ibv_send_wr *bad_wr;
 
 diff --git a/examples/srq_pingpong.c b/examples/srq_pingpong.c
 index 6e00f8c..552a144 100644
 --- a/examples/srq_pingpong.c
 +++ b/examples/srq_pingpong.c
 @@ -68,6 +68,7 @@ struct pingpong_context {
   struct ibv_qp   *qp[MAX_QP];
   void*buf;
   int  size;
 + int  send_flags;
   int  num_qp;
   int  rx_depth;
   int  pending[MAX_QP];
 @@ -350,9 +351,10 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   if (!ctx)
   return NULL;
 
 - ctx-size = size;
 - ctx-num_qp   = num_qp;
 - ctx-rx_depth = rx_depth;
 + ctx-size   = size;
 + ctx-send_flags = IBV_SEND_SIGNALED;
 + ctx-num_qp = num_qp;
 + ctx-rx_depth   = rx_depth;
 
   ctx-buf = memalign(page_size, size);
   if (!ctx-buf) {
 @@ -413,7 +415,8 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   }
 
   for (i = 0; i  num_qp; ++i) {
 - struct ibv_qp_init_attr attr = {
 + struct ibv_qp_attr attr;
 + struct ibv_qp_init_attr init_attr = {
   .send_cq = ctx-cq,
   .recv_cq = ctx-cq,
   .srq = ctx-srq,
 @@ -424,11 +427,15 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   .qp_type = IBV_QPT_RC
   };
 
 - ctx-qp[i] = ibv_create_qp(ctx-pd, attr);
 + ctx-qp[i] = ibv_create_qp(ctx-pd, init_attr);
   if (!ctx-qp[i])  {
   fprintf(stderr, Couldn't create QP[%d]\n, i);
   goto clean_qps;
   }
 + ibv_query_qp(ctx-qp[i], attr, IBV_QP_CAP, init_attr);
 + if (init_attr.cap.max_inline_data = size) {
 + ctx-send_flags |= IBV_SEND_INLINE;
 + }
   }
 
   for (i = 0; i  num_qp; ++i) {
 @@ -568,7 +575,7 @@ static int pp_post_send(struct 

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jul 23, 2013, at 9:26 AM, Jeff Squyres (jsquyres) jsquy...@cisco.com wrote:

 .. and UD is the least abstracted transport, so existing apps won't
 support Jeff's new NIC anyhow, MTU is the least of their problems.
 
 Existing apps with existing transports see the same old values.


Bump.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jul 30, 2013, at 12:44 PM, Christoph Lameter c...@gentwo.org wrote:

 What in the world does that mean? I am an oldtimer I guess. Seems that
 this is something that can be done in the newfangled forum? How does this
 affect mailing lists?


I'm not sure what you're asking me; please see the prior posts on this thread 
that describes the MTU issue and why we still need a solution.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: Add the use of IBV_SEND_INLINE to example pingpong programs

[Jeff Squyres (jsquyres)]

Bump bump bump.

I know this isn't a huge / important patch, but it is a small thing that does 
decrease the latency reported by these example programs.


On Jul 10, 2013, at 4:32 PM, Jeff Squyres jsquy...@cisco.com wrote:

 If the send size is less than the cap.max_inline_data reported by the
 qp, use the IBV_SEND_INLINE flag.  This not only shows the example of
 using ibv_query_qp(), it also reduces the latency time shown by the
 pingpong programs when the sends can be inlined.
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 examples/rc_pingpong.c  | 18 +-
 examples/srq_pingpong.c | 19 +--
 examples/uc_pingpong.c  | 17 -
 examples/ud_pingpong.c  | 18 +-
 4 files changed, 51 insertions(+), 21 deletions(-)
 
 diff --git a/examples/rc_pingpong.c b/examples/rc_pingpong.c
 index 15494a1..a8637a5 100644
 --- a/examples/rc_pingpong.c
 +++ b/examples/rc_pingpong.c
 @@ -65,6 +65,7 @@ struct pingpong_context {
   struct ibv_qp   *qp;
   void*buf;
   int  size;
 + int  send_flags;
   int  rx_depth;
   int  pending;
   struct ibv_port_attr portinfo;
 @@ -319,8 +320,9 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   if (!ctx)
   return NULL;
 
 - ctx-size = size;
 - ctx-rx_depth = rx_depth;
 + ctx-size   = size;
 + ctx-send_flags = IBV_SEND_SIGNALED;
 + ctx-rx_depth   = rx_depth;
 
   ctx-buf = memalign(page_size, size);
   if (!ctx-buf) {
 @@ -367,7 +369,8 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   }
 
   {
 - struct ibv_qp_init_attr attr = {
 + struct ibv_qp_attr attr;
 + struct ibv_qp_init_attr init_attr = {
   .send_cq = ctx-cq,
   .recv_cq = ctx-cq,
   .cap = {
 @@ -379,11 +382,16 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   .qp_type = IBV_QPT_RC
   };
 
 - ctx-qp = ibv_create_qp(ctx-pd, attr);
 + ctx-qp = ibv_create_qp(ctx-pd, init_attr);
   if (!ctx-qp)  {
   fprintf(stderr, Couldn't create QP\n);
   goto clean_cq;
   }
 +
 + ibv_query_qp(ctx-qp, attr, IBV_QP_CAP, init_attr);
 + if (init_attr.cap.max_inline_data = size) {
 + ctx-send_flags |= IBV_SEND_INLINE;
 + }
   }
 
   {
 @@ -508,7 +516,7 @@ static int pp_post_send(struct pingpong_context *ctx)
   .sg_list= list,
   .num_sge= 1,
   .opcode = IBV_WR_SEND,
 - .send_flags = IBV_SEND_SIGNALED,
 + .send_flags = ctx-send_flags,
   };
   struct ibv_send_wr *bad_wr;
 
 diff --git a/examples/srq_pingpong.c b/examples/srq_pingpong.c
 index 6e00f8c..552a144 100644
 --- a/examples/srq_pingpong.c
 +++ b/examples/srq_pingpong.c
 @@ -68,6 +68,7 @@ struct pingpong_context {
   struct ibv_qp   *qp[MAX_QP];
   void*buf;
   int  size;
 + int  send_flags;
   int  num_qp;
   int  rx_depth;
   int  pending[MAX_QP];
 @@ -350,9 +351,10 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   if (!ctx)
   return NULL;
 
 - ctx-size = size;
 - ctx-num_qp   = num_qp;
 - ctx-rx_depth = rx_depth;
 + ctx-size   = size;
 + ctx-send_flags = IBV_SEND_SIGNALED;
 + ctx-num_qp = num_qp;
 + ctx-rx_depth   = rx_depth;
 
   ctx-buf = memalign(page_size, size);
   if (!ctx-buf) {
 @@ -413,7 +415,8 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   }
 
   for (i = 0; i  num_qp; ++i) {
 - struct ibv_qp_init_attr attr = {
 + struct ibv_qp_attr attr;
 + struct ibv_qp_init_attr init_attr = {
   .send_cq = ctx-cq,
   .recv_cq = ctx-cq,
   .srq = ctx-srq,
 @@ -424,11 +427,15 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   .qp_type = IBV_QPT_RC
   };
 
 - ctx-qp[i] = ibv_create_qp(ctx-pd, attr);
 + ctx-qp[i] = ibv_create_qp(ctx-pd, init_attr);
   if (!ctx-qp[i])  {
   fprintf(stderr, Couldn't create QP[%d]\n, i);
   goto clean_qps;
   }
 + ibv_query_qp(ctx-qp[i], attr, IBV_QP_CAP, init_attr);
 + if (init_attr.cap.max_inline_data = size) {
 + ctx-send_flags |= 

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jul 18, 2013, at 12:50 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 We need it for UD for our upcoming device, however, because the MTU
 is the only way to get the max message size.
 
 .. and UD is the least abstracted transport, so existing apps won't
 support Jeff's new NIC anyhow, MTU is the least of their problems.
 
 Existing apps with existing transports see the same old values.


...so how do we move forward?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: Add the use of IBV_SEND_INLINE to example pingpong programs

[Jeff Squyres (jsquyres)]

Bump bump.

On Jul 10, 2013, at 4:32 PM, Jeff Squyres jsquy...@cisco.com wrote:

 If the send size is less than the cap.max_inline_data reported by the
 qp, use the IBV_SEND_INLINE flag.  This now only shows the example of
 using ibv_query_qp(), it also reduces the latency time shown by the
 pingpong programs when the sends can be inlined.
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 examples/rc_pingpong.c  | 18 +-
 examples/srq_pingpong.c | 19 +--
 examples/uc_pingpong.c  | 17 -
 examples/ud_pingpong.c  | 18 +-
 4 files changed, 51 insertions(+), 21 deletions(-)
 
 diff --git a/examples/rc_pingpong.c b/examples/rc_pingpong.c
 index 15494a1..a8637a5 100644
 --- a/examples/rc_pingpong.c
 +++ b/examples/rc_pingpong.c
 @@ -65,6 +65,7 @@ struct pingpong_context {
   struct ibv_qp   *qp;
   void*buf;
   int  size;
 + int  send_flags;
   int  rx_depth;
   int  pending;
   struct ibv_port_attr portinfo;
 @@ -319,8 +320,9 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   if (!ctx)
   return NULL;
 
 - ctx-size = size;
 - ctx-rx_depth = rx_depth;
 + ctx-size   = size;
 + ctx-send_flags = IBV_SEND_SIGNALED;
 + ctx-rx_depth   = rx_depth;
 
   ctx-buf = memalign(page_size, size);
   if (!ctx-buf) {
 @@ -367,7 +369,8 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   }
 
   {
 - struct ibv_qp_init_attr attr = {
 + struct ibv_qp_attr attr;
 + struct ibv_qp_init_attr init_attr = {
   .send_cq = ctx-cq,
   .recv_cq = ctx-cq,
   .cap = {
 @@ -379,11 +382,16 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   .qp_type = IBV_QPT_RC
   };
 
 - ctx-qp = ibv_create_qp(ctx-pd, attr);
 + ctx-qp = ibv_create_qp(ctx-pd, init_attr);
   if (!ctx-qp)  {
   fprintf(stderr, Couldn't create QP\n);
   goto clean_cq;
   }
 +
 + ibv_query_qp(ctx-qp, attr, IBV_QP_CAP, init_attr);
 + if (init_attr.cap.max_inline_data = size) {
 + ctx-send_flags |= IBV_SEND_INLINE;
 + }
   }
 
   {
 @@ -508,7 +516,7 @@ static int pp_post_send(struct pingpong_context *ctx)
   .sg_list= list,
   .num_sge= 1,
   .opcode = IBV_WR_SEND,
 - .send_flags = IBV_SEND_SIGNALED,
 + .send_flags = ctx-send_flags,
   };
   struct ibv_send_wr *bad_wr;
 
 diff --git a/examples/srq_pingpong.c b/examples/srq_pingpong.c
 index 6e00f8c..552a144 100644
 --- a/examples/srq_pingpong.c
 +++ b/examples/srq_pingpong.c
 @@ -68,6 +68,7 @@ struct pingpong_context {
   struct ibv_qp   *qp[MAX_QP];
   void*buf;
   int  size;
 + int  send_flags;
   int  num_qp;
   int  rx_depth;
   int  pending[MAX_QP];
 @@ -350,9 +351,10 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   if (!ctx)
   return NULL;
 
 - ctx-size = size;
 - ctx-num_qp   = num_qp;
 - ctx-rx_depth = rx_depth;
 + ctx-size   = size;
 + ctx-send_flags = IBV_SEND_SIGNALED;
 + ctx-num_qp = num_qp;
 + ctx-rx_depth   = rx_depth;
 
   ctx-buf = memalign(page_size, size);
   if (!ctx-buf) {
 @@ -413,7 +415,8 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   }
 
   for (i = 0; i  num_qp; ++i) {
 - struct ibv_qp_init_attr attr = {
 + struct ibv_qp_attr attr;
 + struct ibv_qp_init_attr init_attr = {
   .send_cq = ctx-cq,
   .recv_cq = ctx-cq,
   .srq = ctx-srq,
 @@ -424,11 +427,15 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   .qp_type = IBV_QPT_RC
   };
 
 - ctx-qp[i] = ibv_create_qp(ctx-pd, attr);
 + ctx-qp[i] = ibv_create_qp(ctx-pd, init_attr);
   if (!ctx-qp[i])  {
   fprintf(stderr, Couldn't create QP[%d]\n, i);
   goto clean_qps;
   }
 + ibv_query_qp(ctx-qp[i], attr, IBV_QP_CAP, init_attr);
 + if (init_attr.cap.max_inline_data = size) {
 + ctx-send_flags |= IBV_SEND_INLINE;
 + }
   }
 
   for (i = 0; i  num_qp; ++i) {
 @@ -568,7 +575,7 @@ static int pp_post_send(struct 

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jul 17, 2013, at 5:44 PM, Steve Wise sw...@opengridcomputing.com wrote:

 The iwarp drivers just report the nearest mtu enum.  Apps don't need it for 
 iwarp like they do for ib.


For RC, it doesn't matter much.  So the fact that RoCE and iWARP lie about 
their MTU isn't a huge deal.  It's wrong, but it doesn't matter much.

We need it for UD for our upcoming device, however, because the MTU is the only 
way to get the max message size.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jul 16, 2013, at 10:47 AM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 A source change is completely unvaoidable. Supporting the new MTU
 values requires updated source.


I don't really care one way or the other; I'll submit whatever patch people 
want.  :-)

But FWIW, I tend to believe the Doug/Jason position:

- MTU really needs to be a plain integer (not an enum)
- forcing application source change/adaptation is the safest way to move forward
- doing it this way preserves ABI, so existing binaries are safe

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

Bump.

On Jul 10, 2013, at 8:14 AM, Jeff Squyres (jsquyres) jsquy...@cisco.com wrote:

 On Jul 8, 2013, at 1:26 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
 wrote:
 
 Jeff's patch doesn't break old binaries, old binaries, running with
 normal IB MTUs work fine. The structure layouts all stay the same,
 etc.
 
 
 FWIW, I did a simple test to confirm this.  I installed a stock git HEAD 
 libibverbs into $HOME/libibverbs-HEAD and a libibverbs with the MTU patch in 
 $HOME/libibverbs-mtu-patch.  The mlx4 driver was installed into both trees (I 
 used some fairly old Mellanox HCAs+Dell servers for this test).
 
 This is the base case:
 
 -
 [5:06] dell012:~ ❯❯❯ cd libibverbs-HEAD
 [5:07] dell012:~/libibverbs-HEAD ❯❯❯ ldd bin/ibv_rc_pingpong
   linux-vdso.so.1 =  (0x2aacb000)
   libibverbs.so.1 = /home/jsquyres/libibverbs-HEAD/lib/libibverbs.so.1 
 (0x2accd000)
   libpthread.so.0 = /lib64/libpthread.so.0 (0x2aeec000)
   libdl.so.2 = /lib64/libdl.so.2 (0x2b109000)
   libc.so.6 = /lib64/libc.so.6 (0x2b30e000)
   /lib64/ld-linux-x86-64.so.2 (0x2aaab000)
 [5:07] dell012:~/libibverbs-HEAD ❯❯❯ ./bin/ibv_rc_pingpong dell011
  local address:  LID 0x0004, QPN 0x04004a, PSN 0xc08742, GID ::
  remote address: LID 0x0019, QPN 0x20004a, PSN 0x44c48e, GID ::
 8192000 bytes in 0.02 seconds = 4170.28 Mbit/sec
 1000 iters in 0.02 seconds = 15.72 usec/iter
 -
 
 Works fine.  Now let's use the same libibverbs-HEAD rc pingpong binary, but 
 with the MTU-patched libibverbs.so:
 
 -
 [5:07] dell012:~/libibverbs-HEAD ❯❯❯ mv lib/libibverbs.so.1 
 lib/libibverbs.so.1-bogus
 [5:07] dell012:~/libibverbs-HEAD ❯❯❯ export 
 LD_LIBRARY_PATH=$HOME/libibverbs-mtu-patch/lib
 [5:08] dell012:~/libibverbs-HEAD ❯❯❯ ldd bin/ibv_rc_pingpong
   linux-vdso.so.1 =  (0x2aacb000)
   libibverbs.so.1 = 
 /home/jsquyres/libibverbs-mtu-patch/lib/libibverbs.so.1 (0x2accd000)
   libpthread.so.0 = /lib64/libpthread.so.0 (0x2aeed000)
   libdl.so.2 = /lib64/libdl.so.2 (0x2b10a000)
   libc.so.6 = /lib64/libc.so.6 (0x2b30e000)
   /lib64/ld-linux-x86-64.so.2 (0x2aaab000)
 [5:08] dell012:~/libibverbs-HEAD ❯❯❯ ./bin/ibv_rc_pingpong dell011
  local address:  LID 0x0004, QPN 0x08004a, PSN 0x65391c, GID ::
  remote address: LID 0x0019, QPN 0x24004a, PSN 0x7d137e, GID ::
 8192000 bytes in 0.02 seconds = 4163.39 Mbit/sec
 1000 iters in 0.02 seconds = 15.74 usec/iter
 -
 
 Still works fine.


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

Re: [PATCH] libibverbs: Add the use of IBV_SEND_INLINE to example pingpong programs

[Jeff Squyres (jsquyres)]

Bump.

On Jul 10, 2013, at 4:32 PM, Jeff Squyres jsquy...@cisco.com wrote:

 If the send size is less than the cap.max_inline_data reported by the
 qp, use the IBV_SEND_INLINE flag.  This now only shows the example of
 using ibv_query_qp(), it also reduces the latency time shown by the
 pingpong programs when the sends can be inlined.
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 examples/rc_pingpong.c  | 18 +-
 examples/srq_pingpong.c | 19 +--
 examples/uc_pingpong.c  | 17 -
 examples/ud_pingpong.c  | 18 +-
 4 files changed, 51 insertions(+), 21 deletions(-)
 
 diff --git a/examples/rc_pingpong.c b/examples/rc_pingpong.c
 index 15494a1..a8637a5 100644
 --- a/examples/rc_pingpong.c
 +++ b/examples/rc_pingpong.c
 @@ -65,6 +65,7 @@ struct pingpong_context {
   struct ibv_qp   *qp;
   void*buf;
   int  size;
 + int  send_flags;
   int  rx_depth;
   int  pending;
   struct ibv_port_attr portinfo;
 @@ -319,8 +320,9 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   if (!ctx)
   return NULL;
 
 - ctx-size = size;
 - ctx-rx_depth = rx_depth;
 + ctx-size   = size;
 + ctx-send_flags = IBV_SEND_SIGNALED;
 + ctx-rx_depth   = rx_depth;
 
   ctx-buf = memalign(page_size, size);
   if (!ctx-buf) {
 @@ -367,7 +369,8 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   }
 
   {
 - struct ibv_qp_init_attr attr = {
 + struct ibv_qp_attr attr;
 + struct ibv_qp_init_attr init_attr = {
   .send_cq = ctx-cq,
   .recv_cq = ctx-cq,
   .cap = {
 @@ -379,11 +382,16 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   .qp_type = IBV_QPT_RC
   };
 
 - ctx-qp = ibv_create_qp(ctx-pd, attr);
 + ctx-qp = ibv_create_qp(ctx-pd, init_attr);
   if (!ctx-qp)  {
   fprintf(stderr, Couldn't create QP\n);
   goto clean_cq;
   }
 +
 + ibv_query_qp(ctx-qp, attr, IBV_QP_CAP, init_attr);
 + if (init_attr.cap.max_inline_data = size) {
 + ctx-send_flags |= IBV_SEND_INLINE;
 + }
   }
 
   {
 @@ -508,7 +516,7 @@ static int pp_post_send(struct pingpong_context *ctx)
   .sg_list= list,
   .num_sge= 1,
   .opcode = IBV_WR_SEND,
 - .send_flags = IBV_SEND_SIGNALED,
 + .send_flags = ctx-send_flags,
   };
   struct ibv_send_wr *bad_wr;
 
 diff --git a/examples/srq_pingpong.c b/examples/srq_pingpong.c
 index 6e00f8c..552a144 100644
 --- a/examples/srq_pingpong.c
 +++ b/examples/srq_pingpong.c
 @@ -68,6 +68,7 @@ struct pingpong_context {
   struct ibv_qp   *qp[MAX_QP];
   void*buf;
   int  size;
 + int  send_flags;
   int  num_qp;
   int  rx_depth;
   int  pending[MAX_QP];
 @@ -350,9 +351,10 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   if (!ctx)
   return NULL;
 
 - ctx-size = size;
 - ctx-num_qp   = num_qp;
 - ctx-rx_depth = rx_depth;
 + ctx-size   = size;
 + ctx-send_flags = IBV_SEND_SIGNALED;
 + ctx-num_qp = num_qp;
 + ctx-rx_depth   = rx_depth;
 
   ctx-buf = memalign(page_size, size);
   if (!ctx-buf) {
 @@ -413,7 +415,8 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   }
 
   for (i = 0; i  num_qp; ++i) {
 - struct ibv_qp_init_attr attr = {
 + struct ibv_qp_attr attr;
 + struct ibv_qp_init_attr init_attr = {
   .send_cq = ctx-cq,
   .recv_cq = ctx-cq,
   .srq = ctx-srq,
 @@ -424,11 +427,15 @@ static struct pingpong_context *pp_init_ctx(struct 
 ibv_device *ib_dev, int size,
   .qp_type = IBV_QPT_RC
   };
 
 - ctx-qp[i] = ibv_create_qp(ctx-pd, attr);
 + ctx-qp[i] = ibv_create_qp(ctx-pd, init_attr);
   if (!ctx-qp[i])  {
   fprintf(stderr, Couldn't create QP[%d]\n, i);
   goto clean_qps;
   }
 + ibv_query_qp(ctx-qp[i], attr, IBV_QP_CAP, init_attr);
 + if (init_attr.cap.max_inline_data = size) {
 + ctx-send_flags |= IBV_SEND_INLINE;
 + }
   }
 
   for (i = 0; i  num_qp; ++i) {
 @@ -568,7 +575,7 @@ static int pp_post_send(struct 

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jul 8, 2013, at 1:26 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 Jeff's patch doesn't break old binaries, old binaries, running with
 normal IB MTUs work fine. The structure layouts all stay the same,
 etc.


FWIW, I did a simple test to confirm this.  I installed a stock git HEAD 
libibverbs into $HOME/libibverbs-HEAD and a libibverbs with the MTU patch in 
$HOME/libibverbs-mtu-patch.  The mlx4 driver was installed into both trees (I 
used some fairly old Mellanox HCAs+Dell servers for this test).

This is the base case:

-
[5:06] dell012:~ ❯❯❯ cd libibverbs-HEAD
[5:07] dell012:~/libibverbs-HEAD ❯❯❯ ldd bin/ibv_rc_pingpong
linux-vdso.so.1 =  (0x2aacb000)
libibverbs.so.1 = /home/jsquyres/libibverbs-HEAD/lib/libibverbs.so.1 
(0x2accd000)
libpthread.so.0 = /lib64/libpthread.so.0 (0x2aeec000)
libdl.so.2 = /lib64/libdl.so.2 (0x2b109000)
libc.so.6 = /lib64/libc.so.6 (0x2b30e000)
/lib64/ld-linux-x86-64.so.2 (0x2aaab000)
[5:07] dell012:~/libibverbs-HEAD ❯❯❯ ./bin/ibv_rc_pingpong dell011
  local address:  LID 0x0004, QPN 0x04004a, PSN 0xc08742, GID ::
  remote address: LID 0x0019, QPN 0x20004a, PSN 0x44c48e, GID ::
8192000 bytes in 0.02 seconds = 4170.28 Mbit/sec
1000 iters in 0.02 seconds = 15.72 usec/iter
-

Works fine.  Now let's use the same libibverbs-HEAD rc pingpong binary, but 
with the MTU-patched libibverbs.so:

-
[5:07] dell012:~/libibverbs-HEAD ❯❯❯ mv lib/libibverbs.so.1 
lib/libibverbs.so.1-bogus
[5:07] dell012:~/libibverbs-HEAD ❯❯❯ export 
LD_LIBRARY_PATH=$HOME/libibverbs-mtu-patch/lib
[5:08] dell012:~/libibverbs-HEAD ❯❯❯ ldd bin/ibv_rc_pingpong
linux-vdso.so.1 =  (0x2aacb000)
libibverbs.so.1 = 
/home/jsquyres/libibverbs-mtu-patch/lib/libibverbs.so.1 (0x2accd000)
libpthread.so.0 = /lib64/libpthread.so.0 (0x2aeed000)
libdl.so.2 = /lib64/libdl.so.2 (0x2b10a000)
libc.so.6 = /lib64/libc.so.6 (0x2b30e000)
/lib64/ld-linux-x86-64.so.2 (0x2aaab000)
[5:08] dell012:~/libibverbs-HEAD ❯❯❯ ./bin/ibv_rc_pingpong dell011
  local address:  LID 0x0004, QPN 0x08004a, PSN 0x65391c, GID ::
  remote address: LID 0x0019, QPN 0x24004a, PSN 0x7d137e, GID ::
8192000 bytes in 0.02 seconds = 4163.39 Mbit/sec
1000 iters in 0.02 seconds = 15.74 usec/iter
-

Still works fine.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

N�r��yb�X��ǧv�^�)޺{.n�+{��ٚ�{ay�ʇڙ�,j��f���h���z��w���
���j:+v���w�j�mzZ+�ݢj��!�i

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jul 5, 2013, at 3:11 PM, Roland Dreier rol...@purestorage.com wrote:

 So what happens if I have an old application binary, and I run against
 a new libibverbs without recompiling?
 
 Also it seems that I'm forced to change my source code to be able to
 compile against new libibverbs?


I previously sent an ABI-preserving version of this patch, but it was hated by 
Doug Ledford and (eventually) Jason Gunthorpe.  

After long discussion (see thread starting here: 
http://www.spinics.net/lists/linux-rdma/msg15951.html), they decided that they 
wanted a clean break that forces both source code and ABI changes, which 
resulted in this patch.

I personally don't care which way this goes; I just want the ability to have 
non-enum MTU values.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

Bump.

On Jul 2, 2013, at 8:31 AM, Jeff Squyres jsquy...@cisco.com wrote:

 (Previous patch did not include updates for the man pages)
 
 Keep IBV_MTU_* enums values as they are, but pass MTU values around as
 a struct containing a single int.  
 
 Per lengthy discusson on the linux-rdma list, this patch introdces a
 source code incompatibility.  Although legacy applications can
 continue to use the enum values, they will need to be updated to use
 the struct.  Newer applications are encouraged to use arbitrary int
 values, not the MTU enums (e.g., 1024, 1500, 9000).
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 Makefile.am|  3 +-
 examples/devinfo.c | 20 +++--
 examples/pingpong.c| 12 
 examples/pingpong.h|  1 -
 examples/rc_pingpong.c | 10 +++
 examples/srq_pingpong.c| 10 +++
 examples/uc_pingpong.c | 10 +++
 examples/ud_pingpong.c |  2 +-
 include/infiniband/verbs.h | 61 +--
 man/ibv_modify_qp.3|  2 +-
 man/ibv_mtu_to_num.3   | 71 ++
 man/ibv_query_port.3   |  4 +--
 man/ibv_query_qp.3 |  2 +-
 src/cmd.c  |  8 +++---
 src/marshall.c |  2 +-
 15 files changed, 160 insertions(+), 58 deletions(-)
 create mode 100644 man/ibv_mtu_to_num.3
 
 diff --git a/Makefile.am b/Makefile.am
 index 40e83be..1159e55 100644
 --- a/Makefile.am
 +++ b/Makefile.am
 @@ -54,7 +54,8 @@ man_MANS = man/ibv_asyncwatch.1 man/ibv_devices.1 
 man/ibv_devinfo.1 \
 man/ibv_post_srq_recv.3 man/ibv_query_device.3 man/ibv_query_gid.3
 \
 man/ibv_query_pkey.3 man/ibv_query_port.3 man/ibv_query_qp.3  \
 man/ibv_query_srq.3 man/ibv_rate_to_mult.3 man/ibv_reg_mr.3   
 \
 -man/ibv_req_notify_cq.3 man/ibv_resize_cq.3 man/ibv_rate_to_mbps.3
 +man/ibv_req_notify_cq.3 man/ibv_resize_cq.3 man/ibv_rate_to_mbps.3  \
 +man/ibv_mtu_to_num.3
 
 DEBIAN = debian/changelog debian/compat debian/control debian/copyright \
 debian/ibverbs-utils.install debian/libibverbs1.install \
 diff --git a/examples/devinfo.c b/examples/devinfo.c
 index ff078e4..e8fb27e 100644
 --- a/examples/devinfo.c
 +++ b/examples/devinfo.c
 @@ -111,18 +111,6 @@ static const char *atomic_cap_str(enum ibv_atomic_cap 
 atom_cap)
   }
 }
 
 -static const char *mtu_str(enum ibv_mtu max_mtu)
 -{
 - switch (max_mtu) {
 - case IBV_MTU_256:  return 256;
 - case IBV_MTU_512:  return 512;
 - case IBV_MTU_1024: return 1024;
 - case IBV_MTU_2048: return 2048;
 - case IBV_MTU_4096: return 4096;
 - default:   return invalid MTU;
 - }
 -}
 -
 static const char *width_str(uint8_t width)
 {
   switch (width) {
 @@ -301,10 +289,10 @@ static int print_hca_cap(struct ibv_device *ib_dev, 
 uint8_t ib_port)
   printf(\t\tport:\t%d\n, port);
   printf(\t\t\tstate:\t\t\t%s (%d)\n,
  port_state_str(port_attr.state), port_attr.state);
 - printf(\t\t\tmax_mtu:\t\t%s (%d)\n,
 -mtu_str(port_attr.max_mtu), port_attr.max_mtu);
 - printf(\t\t\tactive_mtu:\t\t%s (%d)\n,
 -mtu_str(port_attr.active_mtu), port_attr.active_mtu);
 + printf(\t\t\tmax_mtu:\t\t%d (%d)\n,
 +ibv_mtu_to_num(port_attr.max_mtu), 
 port_attr.max_mtu.mtu);
 + printf(\t\t\tactive_mtu:\t\t%d (%d)\n,
 + ibv_mtu_to_num(port_attr.active_mtu), 
 port_attr.active_mtu.mtu);
   printf(\t\t\tsm_lid:\t\t\t%d\n, port_attr.sm_lid);
   printf(\t\t\tport_lid:\t\t%d\n, port_attr.lid);
   printf(\t\t\tport_lmc:\t\t0x%02x\n, port_attr.lmc);
 diff --git a/examples/pingpong.c b/examples/pingpong.c
 index 90732ef..d1c22c9 100644
 --- a/examples/pingpong.c
 +++ b/examples/pingpong.c
 @@ -36,18 +36,6 @@
 #include stdio.h
 #include string.h
 
 -enum ibv_mtu pp_mtu_to_enum(int mtu)
 -{
 - switch (mtu) {
 - case 256:  return IBV_MTU_256;
 - case 512:  return IBV_MTU_512;
 - case 1024: return IBV_MTU_1024;
 - case 2048: return IBV_MTU_2048;
 - case 4096: return IBV_MTU_4096;
 - default:   return -1;
 - }
 -}
 -
 uint16_t pp_get_local_lid(struct ibv_context *context, int port)
 {
   struct ibv_port_attr attr;
 diff --git a/examples/pingpong.h b/examples/pingpong.h
 index 9cdc03e..91d217b 100644
 --- a/examples/pingpong.h
 +++ b/examples/pingpong.h
 @@ -35,7 +35,6 @@
 
 #include infiniband/verbs.h
 
 -enum ibv_mtu pp_mtu_to_enum(int mtu);
 uint16_t pp_get_local_lid(struct ibv_context *context, int port);
 int pp_get_port_info(struct ibv_context *context, int port,
struct ibv_port_attr *attr);
 diff --git a/examples/rc_pingpong.c b/examples/rc_pingpong.c
 index 15494a1..a7e1836 100644
 --- a/examples/rc_pingpong.c
 +++ b/examples/rc_pingpong.c
 @@ -78,7 +78,7 @@ struct pingpong_dest {
 };
 
 static 

Re: [PATCH V2] libibverbs: Allow arbitrary int values for MTU.

[Jeff Squyres (jsquyres)]

On Jun 21, 2013, at 5:20 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 Jeff: If you are still reading -

I am still reading, just didn't have much to contribute until now.  :-)

 one concrete suggestion, I think, is
 to ensure compile-time failure when the new-format MTU variable is
 touched.  This is trivially done by wrapping it in a struct:
 
 struct ibv_mtu_t {int __mtu;};

Sure, I can work up a patch that does this.

Do others agree?  Roland?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jun 18, 2013, at 2:49 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 +int num_to_ibv_mtu(int num);
 
 Probably should be ibv_num_to_mtu() to keep with the naming pattern..


New patch coming momentarily, but I wanted to comment on this one: 

I used the name num_to_ibv_mtu because it is in the spirit of the other 
enum-to-int/int-to-enum function pair naming conventions:

int ibv_rate_to_mult(enum ibv_rate rate);
enum ibv_rate mult_to_ibv_rate(int mult);

int ibv_rate_to_mbps(enum ibv_rate rate);
enum ibv_rate mbps_to_ibv_rate(int mbps);

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jun 20, 2013, at 1:09 PM, Hefty, Sean sean.he...@intel.com wrote:

 int ibv_rate_to_mult(enum ibv_rate rate);
 enum ibv_rate mult_to_ibv_rate(int mult);
 
 int ibv_rate_to_mbps(enum ibv_rate rate);
 enum ibv_rate mbps_to_ibv_rate(int mbps);
 
 libibverbs uses the ibv_ prefix for pretty much everything.

...except for those 2 functions above (mbps_to_ibv_rate and mult_to_ibv_rate).  
See:

https://git.kernel.org/cgit/libs/infiniband/libibverbs.git/tree/include/infiniband/verbs.h#n392

and

https://git.kernel.org/cgit/libs/infiniband/libibverbs.git/tree/include/infiniband/verbs.h#n379

respectively.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: Allow arbitrary int values for MTU

[Jeff Squyres (jsquyres)]

On Jun 20, 2013, at 4:40 PM, Doug Ledford dledf...@redhat.com wrote:

 {
  static char str[16];
  snprintf(str, sizeof(str), %d, ibv_mtu_to_num(max_mtu));
return str;
 }
 
 That is not, however, multi-thread safe nor advisable unless you clearly
 indicate in the man page to the function that subsequent calls to the
 function wipe out the result of previous calls.  It's not even single
 thread safe if you have more than one interface and don't know that
 later calls wipe this buffer out.  Best to avoid library routines such
 as this.

This is in the devinfo.c program (which is single-threaded), not in the library 
itself.

But regardless, this whole function went away in V2 of the patch.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 12, 2013, at 5:17 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 Yes, it can, via MAP_FIXED. There are lots of fun tricks you can play
 using that.


You're missing the point.

Normal users (i.e., MPI users) don't do that.  They call malloc() and they get 
what they get.

The whole point of upper-layer APIs is that they hide all the network stuff 
from the application programmer.  Verbs is *hard* for the mere mortal to 
program.  MPI can do a great deal to hide the complexities of verbs from app 
developers, but one major concession that MPI (intentionally) made is that the 
*application provides the buffer*, not MPI.

Hence, we're stuck with what buffers the user passes in.

This is the root of the whole MPI has a registration cache issue.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 10, 2013, at 11:56 AM, Liran Liss lir...@mellanox.com wrote:

 Register all address space is the moral equivalent of not having userspace
 registration, so let's talk about it in those terms.  Specifically, there's 
 a subtle
 difference between:
 
 a) telling verbs to register (0...2^64)
   -- Which is weird because it tells verbs to register memory that isn't in 
 my
 address space
 
 Another way to look at it is specify IO access permissions for address 
 space ranges.
 This could be useful to implement a buffer pool to be used for a specific MR 
 only, yet still map/unmap memory within this pool on the fly to optimize 
 physical memory utilization.
 In this case, you would provide smaller ranges than 2^64...


Hmm; I'm not sure I understand.

Userspace doesn't control what virtual addresses it gets back from mmap/etc.  
So how is what you're talking about different than regular/reactive memory 
registration? (vs. pre-emptively registering a whole pile of memory that 
doesn't exist yet)

Specifically: I'm confused because you said you could (preemptively) register 
some small regions (that assumedly don't yet exist in your virtual memory 
address space) and use them as memory pools.  But given that userspace doesn't 
control its virtual address ranges, I'm not sure how that's useful.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 10, 2013, at 1:26 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 I agree that pushing all registration issues out of the application
 and (somewhere) into the verbs stack would be a nice solution.
 
 Well, it creates a mess in another sense, because now you've lost
 context. When your MPI goes to do a 1byte send the kernel may well
 prefetch a few megabytes of page tables, whereas an implementation in
 userspace still has the context and can say, no I don't need that..

It seems like there are Big Problems on either side of this problem (userspace 
and kernel).

I thought that ummunotify was a good balance between the two -- MPI kept its 
registration caches (which are annoying, but we have long-since understood that 
*someone* has to maintain them), but it gets a bulletproof way to keep them 
coherent.  That is what is missing in today's solutions: bulletproofness (plus 
we have to use the horrid glibc malloc hooks, which are deprecated and are 
going away).

 That being said, everyone I've talked to about ODP finds it very,
 very strange that the kernel would keep memory registrations around
 for memory that is no longer part of a process.  Not only does it
 
 MRs are badly named. They are not 'memory registrations'. They are
 'address registrations'. Don't conflat address === memory in your
 head, then it seems weird :)
 
 The memory the address space points to is flexible.
 
 The address space is tied to the lifetime of the process.
 
 It doesn't matter if there is no memory mapped to the address space,
 the address space is still there.
 
 Liran had a good example. You can register address space and then use
 mmap/munmap/MAP_FIXED to mess around with where it points to

...but this is not how people write applications.  Real apps use malloc (and 
some direct mmap, and perhaps even some shared memory).  They don't pay 
attention to the contiguiousness (is that a word?) of memory/addresses in the 
large scale.  To be clear: the most tightly bound codes *do* actually care 
about cache hits and locality, but that's in the small scale -- not in the 
large scale.  I would find it hard to believe that a real code would pay 
attention to where in its address range a given malloc() returns, for example.

*That's* what makes this whole concept weird.

It seems like this is a perfect kernel space concept, but is quite foreign to 
userspace developers.

 A practical example of using this would be to avoid the need to send
 scatter buffer pointers to the remote. The remote writes into a memory
 ring and the ring is made 'endless' by clever use of remapping.

I don't understand -- please explain your example a bit more...?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 7, 2013, at 4:57 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 We talked about this at the MPI Forum this week; it doesn't seem
 like ODP fixes any MPI problems.
 
 ODP without 'register all address space' changes the nature of the
 problem, and fixes only one problem.

I agree that pushing all registration issues out of the application and 
(somewhere) into the verbs stack would be a nice solution.

 You do need to cache registrations, and all the tuning parameters (how
 much do I cache, how long do I hold it for, etc, etc) all still apply.
 
 What goes away (is fixed) is the need for intercepts and the need to
 purge address space from the cache because the backing registration
 has become non-coherent/invalid. Registrations are always
 coherent/valid with ODP.

 This cache, and the associated optimization problem, can never go
 away. With a 'register all of memory' semantic the cache can move into
 the kernel, but the performance implication and overheads are all
 still present, just migrated.

Good summary; and you corrected some of my mistakes -- thanks.

That being said, everyone I've talked to about ODP finds it very, very strange 
that the kernel would keep memory registrations around for memory that is no 
longer part of a process.  Not only does it lead to the new memory is 
magically already registered semantic that I find weird, it's just plain *odd* 
for the kernel to maintain state for something that doesn't exist any more.  It 
feels dirty.

Sidenote: I was just informed today that the current way MPI implementations 
implement registration cache coherence (glibc malloc hooks) has been deprecated 
and will be removed from glibc 
(http://sourceware.org/ml/libc-alpha/2011-05/msg00103.html).  This really puts 
on the pressure to find a new / proper solution.

 What MPI wants is:
 
 1. verbs for ummunotify-like functionality
 2. non-blocking memory registration verbs; poll the cq to know when it has 
 completed
 
 To me, ODP with an additional 'register all address space' semantic, plus
 an asynchronous prefetch does both of these for you.
 
 1. ummunotify functionality and caching is now in the kernel, under
   ODP. RDMA access to an 'all of memory' registration always does the
   right thing.

Register all address space is the moral equivalent of not having userspace 
registration, so let's talk about it in those terms.  Specifically, there's a 
subtle difference between:

a) telling verbs to register (0...2^64) 
   -- Which is weird because it tells verbs to register memory that isn't in 
my address space
b) telling verbs that the app doesn't want to handle registration
   -- How that gets implemented is not important (from userspace's point of 
view) -- if the kernel chooses to implement that by registering non-existent 
memory, that's the kernel's problem

I guess I'm arguing that registering non-existent memory is not the Right Thing.

Regardless of what solution is devised for registered memory management 
(ummunotify, ODP, or something else), a non-blocking verb for registering 
memory would still be a Very Useful Thing.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 6, 2013, at 4:33 PM, Jeff Squyres (jsquyres) jsquy...@cisco.com wrote:

 I don't think this covers other memory regions, like those added via mmap, 
 right?


We talked about this at the MPI Forum this week; it doesn't seem like ODP fixes 
any MPI problems.

1. MPI still has to have a memory registration cache, because 
ibv_reg_mr(0...sbrk()) doesn't cover the stack or mmap'ed memory, etc.

2. MPI still has to intercept (at least) munmap().

3. Having mmap/malloc/etc. return new memory that may already be registered 
because of a prior memory registration and subsequent munmap/free/etc. is just 
plain weird.  Worse, if we re-register it, ref counts could go such that the 
actual registration will never actually expire until the process dies (which 
could lead to processes with abnormally large memory footprints, because they 
never actually let go of memory because it's still registered).

4. Even if MPI checks the value of sbrk() and re-registers (0...sbrk()) when 
sbrk() increases, this would seem to create a lot of work for the kernel -- 
which is both slow and synchronous.  Example:

a = malloc(5GB);
MPI_Send(a, 1, MPI_CHAR, ...); // MPI sends 1 byte

Then the MPI_Send of 1 byte will have to pay the cost of registering 5GB of new 
memory.

-

Unless we understand this wrong (and there's definitely a chance that we do!), 
it doesn't sound like ODP solves anything for MPI.  Especially since HPC 
applications almost never swap (in fact, swap is usually disabled in HPC 
environments).

What MPI wants is:

1. verbs for ummunotify-like functionality
2. non-blocking memory registration verbs; poll the cq to know when it has 
completed

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 10:52 PM, Haggai Eran hagg...@mellanox.com wrote:

 Haggai: A verb to resize a registration would probably be a helpful
 step. MPI could maintain one registration that covers the sbrk
 region and one registration that covers the heap, much easier than
 searching tables and things.
 
 That's a nice idea. Even without this verb, I think it is possible to
 develop a registration cache that covers those regions though. When you
 find out you have some part of your region not registered, you can
 register a new, larger region that covers everything you need. For new
 operations you only use the newer region. Once the previous, smaller
 region is not used, you de-register it.


I'm not sure what you mean.  Are you saying I should do something like this:

MPI_Init() {
// the first MPI function invoked
  mpi_sbrk_save = sbrk();
  ibv_reg_mr(..., 0, mpi_sbrk_save, ...);
  ...
}

MPI_Send(buffer, ...) {
  if (mpi_sbrk_save != sbrk())
  mpi_sbrk_save = sbrk();
  ibv_rereg_mr(..., 0, mpi_sbrk_save, ...);
  ...
}

I don't think this covers other memory regions, like those added via mmap, 
right?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 12:14 AM, Haggai Eran hagg...@mellanox.com wrote:

 Hmm; I'm confused.  How does this fix the 
 MPI-needs-to-intercept-freed-memory problem?
 Well, there is no problem if an application frees registered memory (in
 an on-demand paging memory region) and that memory is returned to the
 OS. The OS will invalidate these pages, and the HCA will no longer be
 able to use them. This means that the registration cache doesn't have to
 de-register memory immediately when it is freed.


(must... resist... urge... to... throw... furniture...)

This is why features should not be introduced to solve MPI problems without an 
understanding of what the MPI problems are.  :-)  Please go talk to the 
Mellanox MPI team.

Forgive me for being frustrated; memory registration and all the pain that it 
entails was highlighted as ***the #1 problem*** by *5 major MPI 
implementations* at the Sonoma 2009 workshop (see 
https://www.openfabrics.org/resources/document-downloads/presentations/doc_download/301-mpi-update-and-requirements-panel-all-presentations.html,
 starting at slide 7 in the openmpi slide deck).  

Why don't we have something like ummunotify yet?
Why don't we have non-blocking memory registration yet?
...etc.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 6:39 AM, Haggai Eran hagg...@mellanox.com wrote:

 Perhaps I'm missing something, but I believe ODP deals with the first
 two problems in the list (slide 8), even if it doesn't solve them
 completely.

Unfortunately, it does not.  If we could register(0 ... 2^64) and never have to 
worry about registered memory, that might be cool (depending on how that 
actually works) -- more below.

See this blog post that describes the freed registered memory issue:


http://blogs.cisco.com/performance/registered-memory-rma-rdma-and-mpi-implementations/

and consider the following valid user code:

a = malloc(x);// a gets (va=0x100, pa=0x12345) back from malloc
MPI_Send(a, ...); // MPI registers 0x100 for len=x, and saves (0x100,x) in reg 
cache
free(a);
a = malloc(x);// a gets (va=0x100, pa=0x98765) back from malloc
MPI_Send(a, ...); // MPI sees a=0x100 and things that it is already registered
// ...kaboom

In short, MPI has to intercept free/sbrk/whatever so that it can update its 
registration cache.

 In the future we want to implement an implicit memory region covering
 the entire process address space, thus eliminating the need for memory
 registration almost completely (you might still want memory
 registration, or memory windows, in order to control permissions of
 remote operations).

This would be great, as long as it's fast, transparent, and has no subtle 
implementation effects (like causing additional RNR NAKs for pages that are 
still in memory, which, according to your descriptions, it sounds like it 
won't).

 We can also allow fork to work with our implementation. Copy-on-write
 will work with ODP regions by invalidating the HCA's page tables before
 modifying the pages to be read-only. A page fault from the HCA can then
 refill the pages, or even break COW in case of a write.

That would be cool, too.  fork() has been a continuing problem -- solving that 
problem would be wonderful.

If this ODP stuff becomes a new verb, it would be good:

- if these fork-fixing / register-infinite capabilities can be queried at run 
time (maybe on ibv_device_cap_flags?) so that ULPs can know to use this 
functionality
- if driver owners can get a heads up so that they can know to implement it

 Why don't we have something like ummunotify yet?
 I think that the problem we are trying to solve is better handled inside
 the kernel. If you are going to change the HCA's memory mappings, you'd
 have to go through the kernel anyway.

If/when you allow registering all memory, then I think you're right -- the 
MPI-must-intercept-free/sbrk-whatever issue may go away (that's why I started 
this thread asking about register(0 .. 2^64)).  But without that, unless I'm 
missing something, I don't think it solves the MPI-must-catch-free-sbrk-etc. 
issues...?  And therefore, having some kind of ummunotify-like functionality as 
a verb would be a Very Good Thing.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: A possible solution for allowing arbitrary MTU values.

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 9:46 AM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 No, this too big of an ABI break, and silent at that..
 
 The IBA values have to continue to be accepted and exported in all
 cases so the ABI stays the same, which is what I thought was agreed
 on??


Can this go to a libibverbs 2.0, where it would be palatable to have an ABI 
break?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: A possible solution for allowing arbitrary MTU values.

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 10:19 AM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 The concept of a libibverbs 2.0 has been NAK's by pretty much everyone
 involved. This is why we are suffering with the complex extension
 mechanism.

Are you saying that libibverbs must always always always be backwards 
compatible, and there will never be an ABI break at any version in the future?

 The mixed approach that was brought up, where values like 1500 were
 passed as 1500, and values like 1024 were passed as 3 seemed doable to
 me. Did you see a problem with it for your use?

It just seems overly complex in terms of implementation.

 Thoughts:
 - 1024 and 3 both mean 1024, the library must accept both values,
   it should only ever return 3 though.

Why?  If the caller can pass in 1024, it seems like 1024 should be able to be 
passed out, too.

 - 1500/etc means 1500, the libray can return that.
 - Make a ibv_from/to_mtu inline function to translate from bytes to
   the encoded MTU value.
 - Switch ibv_mtu from a enum to a typedef int ibv_mtu

That also breaks ABI, doesn't it?

 Jason


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 10:14 AM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 a = malloc(x);// a gets (va=0x100, pa=0x12345) back from malloc
 MPI_Send(a, ...); // MPI registers 0x100 for len=x, and saves (0x100,x) in 
 reg cache
 free(a);
 a = malloc(x);// a gets (va=0x100, pa=0x98765) back from malloc
 MPI_Send(a, ...); // MPI sees a=0x100 and things that it is already 
 registered
 // ...kaboom
 
 ODP is supposed to completely solve this problem. The HCA's view and
 Kernels view of virtual to physical mapping becomes 100% synchronized,
 and there is no 'kaboom'. The kernel updates the HCA after the free,
 and after the 2nd malloc to 100% match the current virtual memory map
 in the process.

Are you saying that the 2nd malloc will magically be registered (with the new 
physical address)?

 AFAIK the ummunotify user space API was nak'd by the core kernel
 guys.

It was NAK'ed by Linus, saying fix your own network stack; this is not needed 
in the general purpose part of the kernel (remember that Roland initially 
developed this as a standalone, non-IB-related kernel module).  

 I got the impression people thought it would be acceptable as a
 rdma API, not a general API. So it is waiting on someone to recast the
 function within verbs to make progress...

'zactly.  Roland has this ummunot branch in his git tree, where he is in the 
middle of incorporating this functionality from the original ummunotify 
standalone kernel module into libibverbs and ibcore.

I started this thread asking the status of that branch.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH] libibverbs: A possible solution for allowing arbitrary MTU values.

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 11:11 AM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 I won't say never, but this is what people want. Bumping the soname is
 seen as too difficult now.

Gotcha.  

Ok, so my patch is a non-starter.

 Thoughts:
 - 1024 and 3 both mean 1024, the library must accept both values,
  it should only ever return 3 though.
 
 Why?  If the caller can pass in 1024, it seems like 1024 should be
 able to be passed out, too.
 
 If the caller passes in 1024 then it is probably OK to return 1024,
 but you have to keep track of that specially. That seems more complex
 than just always returning 3. 3 is guarenteed compatible with all
 users.
 
 Old users will test directly against 3.
 New users will call ibv_from_mtu which tests against 3 as well.


Ok.

I'll take a to-do to work up a new patch -- probably not until next week.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 11:18 AM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 Are you saying that the 2nd malloc will magically be registered
 (with the new physical address)?
 
 Yes, that is the whole point.

Interesting.

 ODP fundamentally fixes the *bug* where the HCA's view of process
 memory can become inconsistent with the kernel's view.

Hum.  I was under the impression that with today's code (i.e., not ODP), if you

a = malloc(N);
ibv_reg_mr(..., a, N, ...);
free(a);

(assuming that the memory actually left the process at free)

Then the relevant kernel verbs driver was notified, and would unregister that 
device.  ...but I'm an MPI guy, not a kernel guy -- it seems like you're saying 
that my impression was wrong (which doesn't currently matter because we 
intercept free/sbrk and unregister such memory, anyway).

 'magically be registered' is the wrong way to think about it - the
 registration of VA=0x100 is simply kept, and any change to the
 underlying physical mapping of the VA is synchronized with the HCA.

What happens if you:

a = malloc(N * page_size);
ibv_reg_mr(..., a, N * page_size, ...);
free(a);
// incoming RDMA arrives targeted at buffer a

Or if you:

a = malloc(N * page_size);
ibv_reg_mr(..., a, N * page_size, ...);
free(a);
a = malloc(N / 2 * page_size);
// incoming RDMA arrives targeted at buffer a that is of length (N*page_size)

It does seem quite odd, abstractly speaking, that a registration would survive 
a free/re-malloc (which is arguably a different buffer).

That being said, it still seems like MPI needs a registration cache.  It is 
several good steps forward if we don't need to intercept free/sbrk/whatever, 
but when MPI_Send(buf, ...) is invoked, we still have to check that the entire 
buf is registered.  If ibv_reg_mr(..., 0, 2^64, ...) was supported, that would 
obviate the entire need for registration caches.  That would be wonderful.

 Right, this was discussed at the Enterprise Summit a few weeks
 ago. I'm sure Roland would welcome patches...


That's why I asked at the beginning of this thread.  He didn't provide any 
details about what still needs to be done, though.  :-)

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 5, 2013, at 12:05 PM, Jason Gunthorpe jguntho...@obsidianresearch.com 
wrote:

 It does seem quite odd, abstractly speaking, that a registration
 would survive a free/re-malloc (which is arguably a different
 buffer).
 
 Not at all: the purpose of the registration is to allow access via
 RDMA to a portion of the process's address space. The address space
 doesn't change, but what it is mapped to can vary.

I still think it's really weird.  When I do this:

a = malloc(N);
ibv_reg_mr(..., a, N, ...);
free(a);
b = malloc(M);

If b just happens to be partially or wholly registered by some quirk of the 
malloc() system (i.e., some/all of the virtual address space in b happens to 
have been covered by a prior malloc/ibv_reg_mr)... that's just weird.

 If ibv_reg_mr(...,
 0, 2^64, ...) was supported, that would obviate the entire need for
 registration caches.  That would be wonderful.
 
 Yes, except that this shifts around where the registration overhead
 ends up. Basically the HCA driver now has the registration cache you
 had in MPI, and all the same overheads still exist.

There's fewer verbs drivers than applications, right?

 Haggai: A verb to resize a registration would probably be a helpful
 step. MPI could maintain one registration that covers the sbrk
 region and one registration that covers the heap, much easier than
 searching tables and things.

If we still have to register buffers piecemeal, a non-blocking registration 
verb would be quite helpful.

 Also bear in mind that all RDMA access protections will be disabled if
 you register the entire process VM, the remote(s) can scribble/read
 everything..


No problem for MPI/HPC...  :-)

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 4, 2013, at 2:54 AM, Haggai Eran hagg...@mellanox.com wrote:

 We wish to get there eventually. In our current implementation you still
 have to register an on-demand memory region explicitly. The difference
 between a regular memory region is that the pages in the region aren't
 pinned.

Does this mean that an MPI implementation still has to register memory upon 
usage, and maintain its own registered memory cache?

 We chose to support only 2 concurrent page faults per QP since this
 allows us to maintain order between the QP's operations and the
 user-space code using it.


I talked to someone who was at the OpenFabrics workshop and saw the ODP 
presentation in person; he tells me that a fault will be incurred when a page 
is not in the HCA's TLB cache (vs. when a registered page is not in memory and 
must be swapped back in), and that this will trigger an RNR NAK.

Is this correct?

He was very concerned about what the size of the TLB on the HCA, and therefore 
what the actual run-time behavior would be for sending around large messages 
via MPI -- i.e., would RDMA'ing 1GB messages now incur this 
HCA-must-reload-its-TLB-and-therefore-incur-RNR-NAKs behavior?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On Jun 4, 2013, at 4:50 AM, Haggai Eran hagg...@mellanox.com wrote:

 Does this mean that an MPI implementation still has to register memory upon 
 usage, and maintain its own registered memory cache?
 Yes. However, since registration doesn't pin memory, you can leave
 registered memory regions in the cache for longer periods, and you can
 register larger memory regions without needing to back them with
 physical memory.

Hmm; I'm confused.  How does this fix the MPI-needs-to-intercept-freed-memory 
problem?

 
 We chose to support only 2 concurrent page faults per QP since this
 allows us to maintain order between the QP's operations and the
 user-space code using it.
 
 
 I talked to someone who was at the OpenFabrics workshop and saw the ODP 
 presentation in person; he tells me that a fault will be incurred when a 
 page is not in the HCA's TLB cache (vs. when a registered page is not in 
 memory and must be swapped back in), and that this will trigger an RNR NAK.
 
 Is this correct?
 
 Our HCAs use their own page tables, in addition to a TLB cache. A miss
 in the TLB cache that can be filled from the HCA's page tables will not
 cause an RNR NAK, since the HCA can fill it relatively fast without the
 help of the operating system. If the page is missing from the HCA's page
 table though it will trigger a page fault and ask the OS to bring that
 page. Since this might take longer, in these cases we send an RNR NAK.

Ok.

But the primary use case I care about is fixing the 
MPI-needs-to-intercept-freed-memory problem, and it doesn't sounds like ODP 
fixes this.

 He was very concerned about what the size of the TLB on the HCA, and 
 therefore what the actual run-time behavior would be for sending around 
 large messages via MPI -- i.e., would RDMA'ing 1GB messages now incur this 
 HCA-must-reload-its-TLB-and-therefore-incur-RNR-NAKs behavior?
 
 We have a mechanism to prefetch the pages needed for a large message
 upon the first page fault, which can also help amortizing the cost of
 the page fault for larger messages.


Ok, thanks.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On May 29, 2013, at 1:53 AM, Or Gerlitz or.gerl...@gmail.com wrote:

 Have you looked on ODP? see
 https://www.openfabrics.org/resources/document-downloads/presentations/doc_download/568-on-demand-paging-for-user-space-networking.html


Is the idea behind ODP that, at the beginning of time, you register the entire 
memory space (i.e., NULL to 2^64) and then never worry about registered memory?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On May 30, 2013, at 1:09 AM, Or Gerlitz ogerl...@mellanox.com wrote:

 Has this been run by the MPI implementor community?
 
 The team that works on this here isn't ready for submission, so community 
 runs were not made yet

If this is a solution to an MPI problem, it would seem like a good idea to run 
the specifics of this proposal to the MPI *implementor* community first (not 
*users*).

I say this because Mellanox also proposed the concept of a shared send queue 
as a solution to MPI RC scalability problems a while ago (around about the time 
XRC first debuted, IIRC?), and the MPI community universally hated it.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On May 29, 2013, at 4:53 AM, Or Gerlitz or.gerl...@gmail.com wrote:

 Have you looked on ODP? see
 https://www.openfabrics.org/resources/document-downloads/presentations/doc_download/568-on-demand-paging-for-user-space-networking.html


Is this upstream?

Has this been run by the MPI implementor community?

The limitation of a max of 2 concurrent page faults seems fairly significant.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: Status of ummunot branch?

[Jeff Squyres (jsquyres)]

On May 28, 2013, at 1:52 PM, Roland Dreier rol...@purestorage.com wrote:

 Haven't touched it in quite a while except to keep it building.  Needs
 work to finish up.

What kinds of things still need to be done?  (I don't know if we could work on 
this or not; just asking to scope out what would need to be done at this point)

Has anything been done on the userspace side?

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 1/2] libibverbs: Use autoreconf in autogen.sh

[Jeff Squyres (jsquyres)]

On May 1, 2013, at 11:30 AM, Doug Ledford dledf...@redhat.com wrote:

 This is fine with me, however, I think you also need to bump the
 autotools version to the latest upstream.  The automated checkers in our
 build environment is spitting out errors about a number of upstream
 packages where the autotools used to configure the package does not
 include proper arm support.  The latest autotools bring in all of the
 forthcoming arm variants.  So I would like to see both of these things done.

Are you referring to the version of Autotools that Roland uses to create his 
tarballs?

Because I have no control over that.  :-)


 On Apr 25, 2013, at 11:38 AM, Jeff Squyres (jsquyres) jsquy...@cisco.com 
 wrote:
 
 Bump.
 
 On Apr 22, 2013, at 1:41 PM, Jeff Squyres jsquy...@cisco.com wrote:
 
 The old sequence of Autotools commands listed in autogen.sh is no
 longer correct.  Instead, just use the single autoreconf command,
 which will invoke all the Right Autotools commands in the correct
 order.
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 autogen.sh | 6 +-
 1 file changed, 1 insertion(+), 5 deletions(-)
 
 diff --git a/autogen.sh b/autogen.sh
 index fd47839..6c9233e 100755
 --- a/autogen.sh
 +++ b/autogen.sh
 @@ -1,8 +1,4 @@
 #! /bin/sh
 
 set -x
 -aclocal -I config
 -libtoolize --force --copy
 -autoheader
 -automake --foreign --add-missing --copy
 -autoconf
 +autoreconf -ifv -I config
 -- 
 1.8.1.1
 
 
 
 -- 
 Jeff Squyres
 jsquy...@cisco.com
 For corporate legal information go to: 
 http://www.cisco.com/web/about/doing_business/legal/cri/
 
 --
 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
 
 
 


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 2/2] Ad IB_MTU_1500|9000 enums.

[Jeff Squyres (jsquyres)]

On Apr 22, 2013, at 4:00 PM, Doug Ledford dledf...@redhat.com wrote:

 2. Change all instances of ib_mtu/ibv_mtu to an int.  Code such as 
 switch(mtu) case IBV_MTU_1024: ... will need to be updated to 
 switch(mtu) case 1024: 
 
 I was actually thinking that an ibverbs API version 2.0 might be an
 interesting way to go.  The proliferation of non-IB link layers
 providing the verbs API make some of the original assumptions of IB link
 layer in the original API obsolete.  But, if we were to do that, I'd
 take some time to really think the issue over and try to catch all of
 the needed updates in one go.


In addition to the MTU, another obvious issue is the active_speed attribute on 
the ibv_port_attr.  On the kernel side, it's an enum (IB_SPEED_SDR through 
IB_SPEED_EDR), but there's no corresponding enum names in libibverbs.  

It would be good to make this value a non-enum-int, too.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 1/2] libibverbs: Use autoreconf in autogen.sh

[Jeff Squyres (jsquyres)]

Bump bump.  :-)

On Apr 25, 2013, at 11:38 AM, Jeff Squyres (jsquyres) jsquy...@cisco.com 
wrote:

 Bump.
 
 On Apr 22, 2013, at 1:41 PM, Jeff Squyres jsquy...@cisco.com wrote:
 
 The old sequence of Autotools commands listed in autogen.sh is no
 longer correct.  Instead, just use the single autoreconf command,
 which will invoke all the Right Autotools commands in the correct
 order.
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 autogen.sh | 6 +-
 1 file changed, 1 insertion(+), 5 deletions(-)
 
 diff --git a/autogen.sh b/autogen.sh
 index fd47839..6c9233e 100755
 --- a/autogen.sh
 +++ b/autogen.sh
 @@ -1,8 +1,4 @@
 #! /bin/sh
 
 set -x
 -aclocal -I config
 -libtoolize --force --copy
 -autoheader
 -automake --foreign --add-missing --copy
 -autoconf
 +autoreconf -ifv -I config
 -- 
 1.8.1.1
 
 
 
 -- 
 Jeff Squyres
 jsquy...@cisco.com
 For corporate legal information go to: 
 http://www.cisco.com/web/about/doing_business/legal/cri/
 
 --
 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


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 1/2] libibverbs: Use autoreconf in autogen.sh

[Jeff Squyres (jsquyres)]

Bump.

On Apr 22, 2013, at 1:41 PM, Jeff Squyres jsquy...@cisco.com wrote:

 The old sequence of Autotools commands listed in autogen.sh is no
 longer correct.  Instead, just use the single autoreconf command,
 which will invoke all the Right Autotools commands in the correct
 order.
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 autogen.sh | 6 +-
 1 file changed, 1 insertion(+), 5 deletions(-)
 
 diff --git a/autogen.sh b/autogen.sh
 index fd47839..6c9233e 100755
 --- a/autogen.sh
 +++ b/autogen.sh
 @@ -1,8 +1,4 @@
 #! /bin/sh
 
 set -x
 -aclocal -I config
 -libtoolize --force --copy
 -autoheader
 -automake --foreign --add-missing --copy
 -autoconf
 +autoreconf -ifv -I config
 -- 
 1.8.1.1
 


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 1/2] Use autoreconf in autogen.sh

[Jeff Squyres (jsquyres)]

On Apr 19, 2013, at 8:19 PM, Hefty, Sean sean.he...@intel.com wrote:

 It may help if you identify the library this patch is against.  :)

3rd time sending will be the charm...  :-)

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 2/2] Ad IB_MTU_1500|9000 enums.

[Jeff Squyres (jsquyres)]

On Apr 22, 2013, at 1:30 PM, Doug Ledford dledf...@redhat.com wrote:

 However, for some reason I had it
 in my mind when I was reading the patch that it was against libibverbs.
 That's what I get for staying up late and reviewing when I'm tired :-/


There were other patches against libibverbs that were submitted at the same 
time.

That being said, I see two obvious ways to go forward, both of which have 
pros/cons:

1. Extend the enum ib_mtu to include new enum values for 1500 and 9000 -- 
probably with a different prefix to indicate that they're not IBTA-sanctioned 
values (note that this will also require corresponding changes in libibverbs, 
since MTU values get passed up from kernel to userspace).

PRO: fixes the immediate problem
PRO: probably the lowest impact solution; just adding some more enum values
CON: weird naming (IB_ and RDMA_ prefixes in the same ib_mtu enum; probably 
something similar in userspace)
CON: doesn't do anything to address other MTU values (e.g., what if someone has 
an MTU of 1498?)

2. Change all instances of ib_mtu/ibv_mtu to an int.  Code such as switch(mtu) 
case IBV_MTU_1024: ... will need to be updated to switch(mtu) case 1024: 

PRO: solves the problem for all MTU values
PRO: eliminates the enum-to-int translation functions
CON: much driver code will need to be updated per above, and also update logic 
checking for out-of-bounds MTU calues
CON: similarly, userspace apps will need to be updated; it might be worthwhile 
to bump libibverbs to 2.x, and then intentionally change the MTU field names in 
ibv_port_attr and ibv_qp_attr so that apps using those fields will fail to 
compile with libibverbs 2.x (and therefore forcibly realize they need to adapt 
to the new int MTU values)

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 2/2] Ad IB_MTU_1500|9000 enums.

[Jeff Squyres (jsquyres)]

On Apr 12, 2013, at 11:40 AM, Jeff Squyres (jsquyres) jsquy...@cisco.com 
wrote:

 As an aside I like the use of RDMA_MTU_* for these values.  Again to 
 distinguish them from the IBTA values.  But I know that is poor form.
 
 So what's the right way to move forward on this?  Is it this:
 
 enum ib_mtu {
   IB_MTU_256  = 1,
   IB_MTU_512  = 2,
   IB_MTU_1024 = 3,
   IB_MTU_2048 = 4,
   IB_MTU_4096 = 5,
   RDMA_MTU_1500 = 1500,
   RDMA_MTU_9000 = 9000
 };


Bump.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 1/2] Use autoreconf in autogen.sh

[Jeff Squyres (jsquyres)]

Bump.

Any thoughts on these two patches?  They're pretty trivial, enable use with 
modern versions of Autotools, and now feature the proper Signed-off-by line.


On Apr 13, 2013, at 8:15 AM, Jeff Squyres jsquy...@cisco.com wrote:

 The old sequence of Autotools commands listed in autogen.sh is no
 longer correct.  Instead, just use the single autoreconf command,
 which will invoke all the Right Autotools commands in the correct
 order.
 
 Signed-off-by: Jeff Squyres jsquy...@cisco.com
 ---
 autogen.sh | 6 +-
 1 file changed, 1 insertion(+), 5 deletions(-)
 
 diff --git a/autogen.sh b/autogen.sh
 index fd47839..6c9233e 100755
 --- a/autogen.sh
 +++ b/autogen.sh
 @@ -1,8 +1,4 @@
 #! /bin/sh
 
 set -x
 -aclocal -I config
 -libtoolize --force --copy
 -autoheader
 -automake --foreign --add-missing --copy
 -autoconf
 +autoreconf -ifv -I config
 -- 
 1.8.1.1
 


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 2/2] Ad IB_MTU_1500|9000 enums.

[Jeff Squyres (jsquyres)]

On Apr 9, 2013, at 10:44 PM, Weiny, Ira ira.we...@intel.com wrote:

 As an aside I like the use of RDMA_MTU_* for these values.  Again to 
 distinguish them from the IBTA values.  But I know that is poor form.


So what's the right way to move forward on this?  Is it this:

enum ib_mtu {
   IB_MTU_256  = 1,
   IB_MTU_512  = 2,
   IB_MTU_1024 = 3,
   IB_MTU_2048 = 4,
   IB_MTU_4096 = 5,
   RDMA_MTU_1500 = 1500,
   RDMA_MTU_9000 =  9000
};

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 2/2] Ad IB_MTU_1500|9000 enums.

[Jeff Squyres (jsquyres)]

On Apr 8, 2013, at 6:16 PM, Hefty, Sean sean.he...@intel.com wrote:

 Why can't IB_MTU_1500 = 1500?


It certainly could.  Additionally, since Roland was a little concerned about 
the IB prefix (since 1500 and 9000 are not IBTA-sanctioned MTUs), they could 
have a different prefix -- perhaps RDMA_MTU_1500.  

Although I admit that it would be weird to have an enum that contains values 
with different prefixes:

enum ib_mtu {
IB_MTU_256  = 1,
IB_MTU_512  = 2,
IB_MTU_1024 = 3,
IB_MTU_2048 = 4,
IB_MTU_4096 = 5,
RDMA_MTU_1500 = 1500,
RDMA_MTU_9000 = 9000
};

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 1/4] Add IBV_*_USNIC enums for the Cisco Ethernet Virtual NIC.

[Jeff Squyres (jsquyres)]

On Apr 5, 2013, at 4:40 PM, Roland Dreier rol...@purestorage.com wrote:

 I think the idea is that without context, it's hard to know if adding
 these enums makes sense or not.  And I'm sorry but I'm not that
 sympathetic to my code isn't ready but you have to take this
 out-of-context patch so I can meet Red Hat's arbitrary schedule.


Ok, fair enough.  It'll be a few weeks before we can submit usnic.ko, so I'll 
re-bring up the IBV_NODE_VENDOR/related patches then.

I think the MTU discussion is still relevant, however -- there seems to be a 
larger design issue there.  I'll go reply separately on that thread.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 2/2] Ad IB_MTU_1500|9000 enums.

[Jeff Squyres (jsquyres)]

On Apr 4, 2013, at 1:57 PM, Weiny, Ira ira.we...@intel.com wrote:

 In hindsight, the user space API never should have exposed the mtu as an
 enum...
 
 Since an enum is an int, and we're never going to have anything with an mtu
 = 5 bytes, couldn't we just store all new mtu values directly as their byte
 value?
 
 That seems like a pretty good idea.


Agreed, but changing to an int would seem to have some fairly serious backwards 
compatibility issues.

What is the right way to move forward here?

Just to re-state: our issue is that there does not seem to be any other way to 
get the max UD message size without knowing the actual MTU (are we incorrect 
about that?).  Hence, using the IB-defined values is not really sufficient.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 3/4] Use autoreconf in autogen.sh

[Jeff Squyres (jsquyres)]

Roland --

If there are no objections, can this patch (and patch 4 of this set: 
https://patchwork.kernel.org/patch/2387321/) be committed?  Neither should not 
have any real impact other than the modernization of the libibverbs build 
system.


On Apr 3, 2013, at 9:06 AM, Jeff Squyres jsquy...@cisco.com wrote:

 The old sequence of Autotools commands listed in autogen.sh is no
 longer correct.  Instead, just use the single autoreconf command,
 which will invoke all the Right Autotools commands in the correct
 order.
 
 ---
 autogen.sh | 6 +-
 1 file changed, 1 insertion(+), 5 deletions(-)
 
 diff --git a/autogen.sh b/autogen.sh
 index fd47839..6c9233e 100755
 --- a/autogen.sh
 +++ b/autogen.sh
 @@ -1,8 +1,4 @@
 #! /bin/sh
 
 set -x
 -aclocal -I config
 -libtoolize --force --copy
 -autoheader
 -automake --foreign --add-missing --copy
 -autoconf
 +autoreconf -ifv -I config
 -- 
 1.8.1.1
 


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

Re: [PATCH 1/4] Add IBV_*_USNIC enums for the Cisco Ethernet Virtual NIC.

[Jeff Squyres (jsquyres)]

Forgive the top reply; I'm actually on vacation this week and currently only 
have email access on my phone...

I'm not sure what you're asking me to do. Are you asking us to submit our 
known-buggy-and-not-yet-complete driver just to get two enums approved?

Sent from my phone. No type good. 

On Apr 4, 2013, at 5:27 PM, Or Gerlitz or.gerl...@gmail.com wrote:

 Jeff Squyres (jsquyres) jsquy...@cisco.com wrote:
 
 Sure.  For a little background, the 2nd-generation Cisco VIC has been 
 available
 since last year (IIRC): http://www.cisco.com/en/US/products/ps10277
 /prod_module_series_home.html.  It's a converged 10G Ethernet adapter 
 available  in a variety of form factors (e.g., 2x10G on PCIe and Mezz).
 
 After some off-list discussion with Roland, we chose to create new 
 IBV_*_USNIC
 enums because none of the current enums were accurate for our device.  It's 
 an
 Ethernet NIC, but it's not an RNIC.  It's an Ethernet-based transport, but 
 it's not
 iWARP.
 
 
 The reason we're asking for these IBV_*_USNIC enums now -- before we've 
 submitted the driver -- is because we're targeting getting our driver 
 included in RHEL 6.5.  There's a bit of a chicken-and-egg issue here: 
 they'll accept our patches for a new hardware driver while that driver is 
 being worked upstream.  But they (rightfully) won't accept patches to IB 
 core and libibverbs until they've been vetted by the community.  Hence, even 
 though our driver is slowly working its way through QA and not available 
 yet, we wanted to submit these new enums upstream for community approval so 
 that they can be included in RHEL 6.5.
 
 Does that help?
 
 yes it does, but I still think we need to see the driver code in order
 to conduct proper /better review and maybe even accept the proposed
 changes to the IB core. You can submit it as RFC which means you can
 look on it, and give me comments, but don't pick it up yet
 
 Or.
 --
 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
--
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

Re: [PATCH 1/4] Add IBV_*_USNIC enums for the Cisco Ethernet Virtual NIC.

[Jeff Squyres (jsquyres)]

Per my previous email, forgive my top reply...

RDMA_NODE_VENDOR would be great, actually. Should I work up a patch for that?

Sent from my phone. No type good. 

On Apr 4, 2013, at 10:32 AM, Hefty, Sean sean.he...@intel.com wrote:

 The reason we're asking for these IBV_*_USNIC enums now -- before we've
 submitted the driver -- is because we're targeting getting our driver 
 included
 in RHEL 6.5.  There's a bit of a chicken-and-egg issue here: they'll accept 
 our
 patches for a new hardware driver while that driver is being worked upstream.
 But they (rightfully) won't accept patches to IB core and libibverbs until
 they've been vetted by the community.  Hence, even though our driver is 
 slowly
 working its way through QA and not available yet, we wanted to submit these 
 new
 enums upstream for community approval so that they can be included in RHEL 
 6.5.
 
 I understand the issue.
 
 In the end, these are kernel changes with no actual users of those changes... 
  But then they are also just small changes to a framework...
 
 Just thinking aloud here, but what if we added 'RDMA_NODE_VENDOR' instead?  
 Then other fields, such as transport, become vendor specific.
 
 - 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

Re: [PATCH 1/4] Add IBV_*_USNIC enums for the Cisco Ethernet Virtual NIC.

[Jeff Squyres (jsquyres)]

On Apr 3, 2013, at 2:45 PM, Or Gerlitz or.gerl...@gmail.com wrote:

 Jeff, I agree with Sean, there's not much point to review/discuss
 these general/pre-step patches without seeing some actual device
 specific kernel (if there are such or user space code if there aren't
 any kernel ones) code. e.g you can submit the two kernel pre-step
 patches as the two first pieces in a series that has the driver code.


Unfortunately, not yet.

I just sent another mail that explained our rationale: our kernel driver and 
libibverbs plugin code are working their way through QA.  It'll take a little 
time before we can submit good patches for these.  The main driving factor for 
submitting these new enums is so that they can be included in RHEL 6.5.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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

New patches

[Jeff Squyres (jsquyres)]

I'm about to send some patches for libibverbs and Roland's infiniband kernel 
git tree.  The patches fit into two general categories:

1. Add enums for Cisco's Ethernet Virtual NIC (it's not an RNIC and therefore 
doesn't fit the RNIC/IWARP enums).  Also add enums for 1500 and 9000 MTUs.

2. Minor modernization of the GNU Autotools usage in libibverbs.

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

--
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