Arlin Davis [Tue, 14 Jul 2015 21:58:32 +0000 (14:58 -0700)]
mcm,mpxyd: fix dreq processing to defer QP flush when proxy WRs still pending
The proxy will now defer DREQ flushing of proxy QPs if PI and PO
data engines have outstanding requests. Add mcm_qp_busy routine
for checking PI and PO data engines. When MIC calls disconnect
always send DREQ up to proxy in order to handle deferred flush
of proxy side posted rcv messages.
Change QP free to modify both local and proxy QPs and check for
outstanding rcv message before qp_destroy to avoid infinite wait
in dapls_ep_flush_cqs.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Amir Hanania [Wed, 17 Jun 2015 17:12:24 +0000 (10:12 -0700)]
mcm: bug fixes for non-inline devices
mcm proxy mi_send_pi setup registered WR structure properly for no
inline data support but incorrectly overwrote sg.addr with WR
WR structure on stack.
qp create didn't check for no inline and setup create accordingly
Signed-off-by: Amir Hanania <amir.hanania@intel.com> Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Fri, 5 Jun 2015 19:14:37 +0000 (12:14 -0700)]
mpxyd,mcm: RDMA write with immed data not signaled on request side
With eager completions set, the wc_flags is not set properly on event.
With eager completions no set, the proxy CQ reference is incorrect
and event is forwarded to MCM receive EVD instead of transmit EVD.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Wed, 20 May 2015 18:43:03 +0000 (11:43 -0700)]
mcm: add HST side provider support for device without inline data capability
Add registered WR buffers for HST->MXS (proxy in) mode
when inline data is not supported by device. Use registered
memory for source WR buffer instead of stack when sending
RDMA write request to peer proxy-in service.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Mon, 18 May 2015 21:51:08 +0000 (14:51 -0700)]
ucm: CM changes for UD extended port space and indexer
Tested on 1200n 28ppn cluster, AlltoAll Intel MPI, UD mode.
Both static and dynamic modes, over 500m connections.
Change port manager to indexer and service ID manager
to bitarray indexer. Reduces footprint for service IDs
and allow direct lookup on CM messages.
New insert, remove, lookup functions for processing ID
based CM objects. Inbound requests, with the exception
of new CM requests, will no longer parse list but
use hash table lookups.
AH caching is now used to prevent unnecessarily
creating multiple AH's for same QP destination.
Add 24-bit port space support to CM processing code and
to wire protocol via DCM message reserve space.
Add version check to limit to 16-bit for backward compatibility.
Bump CM protocol version to 8 for xport and rtns fields.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Mon, 18 May 2015 21:21:07 +0000 (14:21 -0700)]
ucm: optimizations for large scale UD communication management
AH caching per QP, AH space set to 48K for LID unicast
Bump port space up to 24 bits
Reduce CM object and reduce private data to 68 bytes
Add xport space and rtns to DCM reserve fields.
New indexer macros for port space hash table management
Add hash table storage to ibtrans device objects
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Thu, 12 Feb 2015 20:21:37 +0000 (15:21 -0500)]
mpxyd: add support for devices without inline data support
Add function to check for inline support during device open.
If inline data is not supported, the CM service and Proxy
data mover will not use inline data option on small IO.
The PO->PI service will now allocate and register necessary
memory to send mcm_wr_rx and mcm_wc_rx operations from
registered memory locations if inline data not supported.
If inline is supported, no extra memory will be allocated
and src buffer will be built on stack as before.
Cleanup some build warnings.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Thu, 22 Jan 2015 23:49:25 +0000 (15:49 -0800)]
openib: add inline data support check during device open
Not all rdma devices support inline data, however without
a verbs device attribute the only way to determine
support is with a QP create with max_inline_send set.
Add a common function to verify inline data support
before setting default to 64 bytes.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Mon, 15 Dec 2014 20:15:54 +0000 (12:15 -0800)]
dapl: mpxyd service changes to support multi-thread single-core option
The proxy service has been changed to reduce the number of cores required
on the host side. Provides new option, via mpxyd.conf, to use single-core
and allow system adminitrator to bind to specific core id for all Intel
Xeon Phi adapters in the platform.
mcm_affinity = 2 will set to single core (per Intel Xeon Phi).
mcm_affinity_base_mic will set to specific core for all adapters.
Best performance can be acheived with mcm_affinity = 2 and
mcm_affinity_base_mic == 0. This option will cause single core
to remain busy, polling operations from clients, as long
as long as device is open and being used by clients for data
transfers.
Arlin Davis [Mon, 15 Dec 2014 20:05:33 +0000 (12:05 -0800)]
dapl: add rdma_write_imm and write only option to dtest
New write_only (-w) option with rdma_write_imm can
be used with providers that support IB extensions.
Allows more options for write bandwith profiling
with immediate data and signaling rate options
to increase write data rates, especially on MIC
clients that use proxy services.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Fri, 21 Nov 2014 22:26:40 +0000 (14:26 -0800)]
mpxyd: DTO completion ERR: status 12, op RDMA_WRITE running MPI alltoall test
Running MIC scale-up configuration with mcm provider on a MXS node
instead of shm causes DTO error due to heavy use of proxy-in buffer pools.
Hit corner case where proxy buffer management hd ptr crossed tl
ptr due to 64 byte alignment on start when hd < 64 bytes behind tl.
Add additional checking on PO and PI buffer management to handle
the case of HD passing TL on start locations. Also changed PO
processing to hold lock until hd ptr is registered with buf_wc slot
management to preserve order of memory usage across threads.
Reduced the size of WC queue for PO and PI buffer management.
Profiling, via MCM_PROFILE, was added to monitor and trigger buffer
management errors.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Tue, 30 Sep 2014 21:07:52 +0000 (14:07 -0700)]
dtestx: update IB extension example test with new v2.0.9 features
Add support for new IB extensions for CM and AH resource cleanup.
Check for v2.0.9 and call dat_ib_ud_cm_free after connection
establishment and dat_ib_ud_ah_free after all data has been
transfered on UD endpoints.
Also add socket based address exchange to eliminate the need
to include lid and qpn parameters on the client side.
Change the multiple EP mode to send from EP 0 to EP[0-3] on
server side and EP[0-3] to EP[0-3] on client side.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Amir Hanania [Thu, 25 Sep 2014 23:32:06 +0000 (16:32 -0700)]
common: add srq support for openib verbs providers
Add necessary components and hooks to support ib_verbs shared
receive queues for both RC and UD QP's. External interfaces
were already provided per DAT 2.0 specification but internal
support was missing.
A new dtestsrq will be provided with package for testing and
example code.
Arlin Davis [Thu, 25 Sep 2014 23:06:33 +0000 (16:06 -0700)]
openib: add IB UD cm_free/ah_free extension support in UCM provider
Make changes to UCM provider for new CM and AH destroy extensions.
Allow consumer to schedule CM object destroy after CM connection
event has been processed. Active side will put CM object in
TIMEWAIT in case RTU is dropped, passive side can schedule
CM object destroy immediatly when called. In the case where
consumer requests CM object destroy, the provider will remove
all internal references to AH since consumer will call AH
destroy directly when finished with UD sends.
All other providers, MCM, CMA, SCM will return UNSUPPORTED
if new extensions are called.
See dtestx source for code examples of new extensions.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Thu, 25 Sep 2014 22:42:38 +0000 (15:42 -0700)]
extension: add IB UD extensions to reduce provider CM and AH memory footprint
dat_ib_ud_cm_free, dat_ib_ud_ah_free added to allow consumers
the option to free provider CM and AH objects, related to AH resolution,
immediately after consuming CONN events instead of waiting for
EP destroy. With existing UD service providers the CM and AH objects
are linked to EP and not destroyed until consumer calls dat_ep_free.
dat_ib_ud_cm_free() frees CM object after AH and private data are copied
and stored by consumer. Provider will destroy internal object
and memory associated with CM and AH resolution.
MAY be called after CM establishment and before EP destroyed
dat_ib_ud_ah_free() destroys UD Address Handle (AH).
MUST be called after all UD sends are complete and
before UD EP is destroyed.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
This patch adds the dapl_os_atopmic_inc, dapl_os_atomic_dec,
and dapl_os_atomic_assign function implementatios to the dapl
userspace package to provide the DAPL API support on the s390x
platform by adding Assembler language implemenation of those
platform specific functions.
Signed-off-by: Alexey Ishchuk <aishchuk@linux.vnet.ibm.com> Acked-by: Arlin Davis <arlin.r.davis@intel.com>
Amir Hanania [Tue, 26 Aug 2014 22:41:10 +0000 (15:41 -0700)]
dtest server exchange connection info with client
The server and client create connection for the server to send the setup info to the client.
When using dtest, the client only needs to use -h <hostname/IP address> option and it will get the rest of the info from the server.
Signed-off-by: Amir Hanania <amir.hanania@intel.com>
Arlin Davis [Mon, 25 Aug 2014 15:59:50 +0000 (08:59 -0700)]
mcm: implement proxy mix_prov_attr function, add fields CPU model and family
Provide MIC consumers with a provider specific query for proxy CPU model and family
to identify platform type from MIC side. Supported in MCM provider only.
The following provider specific name attributes were added to MCM:
DAT_IB_PROXY_CPU_FAMILY
DAT_IB_PROXY_CPU_MODEL
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Mon, 11 Aug 2014 16:49:08 +0000 (09:49 -0700)]
mpxyd: change MIC cpu_mask to per numa node instead of adapter
The proxy processing threads for multiple cards in same socket will overlap
same cpu cores with existing cpumask per adapter. Change thread affinity
and cpumask to a per socket method.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Fri, 1 Aug 2014 17:54:14 +0000 (10:54 -0700)]
mpxyd: MXS based alltoall benchmark hangs or returns post_send timeout
Clean-up shared proxy buffer slot management during IO completions.
Current code adjusts proxy buffer tail, using m_idx, incorrectly
if freeing multiple in order buffer slots. Also, when processing
immediate in-order slot, m_po_buf_tl() failed to continue parsing
list to free other in-order !busy slots.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Mon, 21 Jul 2014 18:18:07 +0000 (11:18 -0700)]
Add new MIC RDMA proxy service daemon (MPXYD)
New service created to support MIC based proxy RDMA. Includes
services to manage connectivity of multi-path heterogeneous
endpoints and use data paths based on platform constraints.
It will create and manage multiple QP's per endpoint if needed. This
allows optimal performance per direction based on various platform
constraints. For example, if the MIC is on same socket as HCA, only
proxy out is needed and not proxy in. In this case, data can go direct
from MPXYD->MIC. However, if the MIC is on a different CPU socket
from HCA, the provider will use both proxy out and proxy in services
to avoid additional constraints of the server platform.
The MCM provider and MPXYD will support connections between
MIC and non MIC endpoints.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Mon, 21 Jul 2014 17:58:37 +0000 (10:58 -0700)]
add new dapl MIC provider (MCM) to support MIC RDMA proxy services
Provider supports all modes of connectivity and will setup data paths
based on endpoint locality and platform constraints. Provides
transparent DAT API support for RDMA writes, RDMA write with
immediate data, Sends, and Recvs. No RDMA read or atomic support.
To use MCM provider an application can use the new ofa-v2-mcm
device definations in dat.conf. Intel MPSS is required for
for MCM provider build and usage.
The following shows connectivity modes and data paths:
HST -> HST to HCA
MSS -> MIC to HCA same socket
MXS -> MIC to HCA cross socket
Arlin Davis [Mon, 21 Jul 2014 15:03:46 +0000 (08:03 -0700)]
MCM: new MIC provider and proxy service definitions
Definitions for MIC Proxy RDMA services
MCM <-> MPXYD over SCI (Symmetric Communications InterFace) - ops, cm, events
MCM <-> MCM over IB - CM, WR/WC proxy-in and proxy-out wire protocol
This service enables MIC based DAPL provider (MCM) to use
proxy data service (host CPU) for SND/RCV and RDMA write operations.
RDMA reads and atomics are not supported. This service communicates within
within a server platform over PCI-E bus using SCIF and a MCM specific
MIX (MIC exchange) messaging protocol. The MCM provider uses a new MCM
CM protocol on the wire along with a Proxy WR/WC protocol.
This service is designed to improved bandwidth on larger IO
when direct MIC based IO is contrained.
This new MCM provider maintains the DAT level API semantics, including
strict ordering requirements of data flow. RDMA write with immediate
data is the only IB extension supported.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Fri, 18 Jul 2014 15:51:00 +0000 (08:51 -0700)]
common: add CQ,QP,MR abstractions for new MIC provider and data proxy service
The new MIC (many integrated core) based provider (MCM) has the capability to
shadow QPs,CQs,MRs on the host side of the platform for optimial performance
based on locality of endpoints and platform contraints. Each endpoint (DAPL_EP),
transparent to consumer, may have multiple connections via MCM provider.
openib_common ib_cq/ib_qp code base has been expanded, MCM only, to support
separate send and receive channels per endpoint.
openib_common dapl_mr code base has been expanded, MCM only, to support
MIC base DMA interfaces for MIC to HOST communications.
openib_common post_send,post_recv inline code base, MCM only, has been
modified to proxy data services via the new MCM provider.
dapl_ib_async_str added for better logging across openib providers.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>
Arlin Davis [Fri, 11 Jul 2014 15:39:01 +0000 (08:39 -0700)]
new lightweight open_query/close_query IB extension for fast attribute query
Consumers that need provider attributes must do a full device open
in order to get any provider/device information. With so many static device
entries in /etc/dat.conf consumers are building classification
mechanisms to identify provider type, locality, name, device
mode, and decide which device is appropriate. The existing DAT interface
doesn't provide a lightweight mechanism for queries.
The following fast query functions have been added to dat_ib_extensions.h:
In addition, DAT extension interface, dat_extension_op, has been
expanded to include new internal calls to handle quick provider load
and function linkage via udat_extension_open, and udat_extension_close
functions. Extended operations needing DAT open/close services need
to be defined from a DAT_OPEN_EXTENSION_BASE or DAT_CLOSE_EXTENSION_BASE
respectively.
NOTE: The ia_handle returned with open query must be closed with subsequent
close_query and not used with any other dat_ia_ operations. Attribute
storage from query_open is not valid after close_query call.
The IB extensions have been rolled to version 2.0.8 with this new API.
The changes are backward compatible.
Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>