Patch-ID# 107952-02
Keywords: mpi rsm robust descriptors mt thread
Synopsis: HPC 3.0: MPI library fixes
Date: Sep/15/99

Solaris Release: 2.6 7

SunOS Release: 5.6 5.7

Unbundled Product: HPC

Unbundled Release: 3.0

Relevant Architectures: sun4u

BugId's fixed with this patch: 4223734 4243364 4244462 4245555 4233062 4237970 4238508 4240198 4241267 4243344 4231997 4236310 4238484 4239092 4243060 4243904 4245135 4245148 4247029 4247703 4222546 4239495 4240478 4243764 4245133 4238806 4244101 4240932 4241032 4237925 4248856 4248999 4248857 4250047 4248208 4252527 4252893 4259229 4259523 4266878 

Changes incorporated in this version: 4248856 4248999 4248857 4250047 4248208 4252527 4252893 4259229 4259523 4266878 

Patches accumulated and obsoleted by this patch: 

Patches which conflict with this patch: 

Patches required with this patch: 

Obsoleted by: 

Files included with this patch: 

/opt/SUNWhpc/HPC3.0/lib/libmpi.so.1
/opt/SUNWhpc/HPC3.0/lib/libmpi_mt.so.1
/opt/SUNWhpc/HPC3.0/lib/tnf/libmpi.so.1
/opt/SUNWhpc/HPC3.0/lib/tnf/libmpi_mt.so.none 
/opt/SUNWhpc/HPC3.0/etc/sunhpc.hpc_rsmd1
/opt/SUNWhpc/HPC3.0/bin/hpc_rsmd

Problem Description: 

4248856 Client-Server program runs out of memory while doing MPI_Comm_accept 
4248999 Client-Server programs hang if you connect-disconnect-reconnect over RSM
4248857 Robust client-server programs incorrectly determine if they are fullconninit
4250047 Pipe size is computed incorrectly if MPI_RSM_PIPESIZE is set & num_stripes>=2  
4248208 MPI multi-threaded program fails making lazy connection over RSM 
4252527 broken fortran wrapper for MPI_Close_port
4252893 broken Fortran wrappers
4259229 MPI MT programs segfault with lazy connections over RSM on a 5.7 32-bit system 
4259523 Setting some shm env vars differently in client/server will cause errors
4266878 message-queue accessors for prism no longer work

(from 107952-01)

4223734 MPI programs hang running over RSM with lazy connections.
4243364 MPI_Comm_connect hangs when the establishment of a connection fails with RSM
4244462 RSM PM segv'ing in mpip_rsm_undoconns when a call to MPI_Comm_accept fails
4245555 Client/Server over RSM eats up File Descriptors.
4233062 lazy connection notification from hpc_rsmd to rsm-pm never happens
4237970 mutex is not released at error exit
4238508 call of acl() command with SETUP results in error
4240198 RSM data sent in first pbx is corrupted when robust operation is enabled
4241267 robust operation leaves unused RSM segments and defunct jobs in hpc_rsmd
4243344 lazy connection requests fail in hpc_rsmd built with SC5.0 compiler
4231997 RSM PM hangs in MPI_Send when MPI_POLLALL is set to 0
4236310 descheduling in libmpi_mt.so should call thr_yield() in addition to yield()      
4238484 Robust server hangs trying to rexmit to a client that has exited in RSM
4239092 MPI_Comm_errhandler_set is not threadsafe
4243060 32-bit client segfaults connecting to a 64-bit server under RSM
4243904 TNF probe MPI_Waitany_end reports "bytes" incorrectly
4245135 RSM PM does not correctly implement locking scheme for polling loops
4245148 SHM/RSM PM encounters a assertion failure as mpip_nactvreqs is dec to -1
4247029 Robust server receives corrupted msg from non-robust client over RSM
4247703 SPIND dies when running code over RSM with lazy_connecions and MPI_COSCHED=1
4222546 Truncated messages in threaded program can cause core dumps or hangs
4239495 RSM PM fails to free all segments while doing conn release under CRE
4240478 RSM PM calculates the high water mark for each connection incorrectly
4243764 libmpi does not handle RSMD error returns correctly for aborts
4245133 Code in mpip_shm_addconns unlocks a mutex that was never locked.
4238806 MPI_Info_set fails after 8 (key, value) pairs are stored
4244101 MPI_Test/MPI_Wait use wrong TNF probe argument
4240932 HPC 2.0 commutative user defined reduction will not work on HPC 3.0 cluster
4241032 A shared memory robust check has a very small chance of erroneously triggering
4237925 MPI_File_get_info is a collective call and should not be


Patch Installation Instructions: 
-------------------------------- 
Refer to the Install.info file for instructions on using the
generic 'installpatch' and 'backoutpatch' scripts provided with
each patch.  Any other special or non-generic installation
instructions should be described below as special instructions.


Special Install Instructions: 
----------------------------- 
None.