that's consistent with GDB failing to access *bd. But GDB behavior is also consistent with a mapping existing without a backing page. To tell the difference, on would probably need to look at /proc/XXX/maps file at failure time, to see if there is a mapping covering
#12690: Segmentation fault in GHC runtime system under low memory with USE_LARGE_ADDRESS_SPACE -----------------------------------+-------------------------------------- Reporter: pggiarrusso | Owner: (none) Type: bug | Status: new Priority: high | Milestone: Component: Runtime System | Version: 8.0.1 Resolution: | Keywords: Operating System: Linux | Architecture: x86_64 (amd64) Type of failure: Runtime crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -----------------------------------+-------------------------------------- Comment (by pggiarrusso): the address in bd. Regardless, I'd patch osCommitMemory no matter what, unless somebody can explain why it's correct to ignore a failure from mmap. If that needs saying: yes, mmap can fail, even with default settings of https://www.kernel.org/doc/Documentation/vm/overcommit-accounting according to those docs. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/12690#comment:3 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler