Project

General

Profile

Bug #1749

CMA fails at Runtime when OS doesn't provide permissions

Added by Nitin Bhat 14 days ago. Updated 10 days ago.

Status:
Merged
Priority:
High
Assignee:
Category:
Machine Layers
Target version:
Start date:
12/04/2017
Due date:
% Done:

0%

Tags:

Description

Ronak saw this issue on his laptop with a netlrts-linux-x86_64 build.

Although the issue is being identified and handled, there's nothing much the user can do to fix it. A check for permissions during configure can help disable CMA on machines that support it but do not provide permissions for it.

------------- Processor 0 Exiting: Called CmiAbort ------------
Reason: Can't use CMA for SHM! Permissions Denied

History

#1 Updated by Sam White 13 days ago

  • Priority changed from Normal to High
  • Category set to Machine Layers
  • Tags changed from #lrts to #lrts, #cma

#2 Updated by Sam White 12 days ago

Related: the print statement that says if CMA is enabled should only be printed once on PE 0, but currently it gets printed multiple times:

$ ./charmrun +p4 ./pgm +vp4 ++local
Charmrun> scalable start enabled. 
Charmrun> started all node programs in 0.006 seconds.
Charm++> Running in non-SMP mode: numPes 4
Converse/Charm++ Commit ID: v6.8.2-138-g529f8574d
Charm++> cma enabled for within node transfers using the zerocopy API
Charm++> cma enabled for within node transfers using the zerocopy API
Charm++> cma enabled for within node transfers using the zerocopy API
Charm++> cma enabled for within node transfers using the zerocopy API
Charm++> scheduler running in netpoll mode.
CharmLB> Load balancer assumes all CPUs are same.
Charm++> Running on 1 unique compute nodes (8-way SMP).
Charm++> cpu topology info is gathered in 0.001 seconds.
[Partition 0][Node 0] End of program

Also we should capitalize "CMA"

#3 Updated by Sam White 12 days ago

From Core meeting: we'd like to add a runtime check for CMA support. This should be fixed with high priority, since it breaks all Charm++ programs on some machines.

#4 Updated by Nitin Bhat 11 days ago

  • Status changed from New to Implemented

#5 Updated by Nitin Bhat 10 days ago

  • Status changed from Implemented to Merged

Also available in: Atom PDF