909206 - jemalloc.c:6769: Failed assertion: "stats->mapped >= stats->allocated + stats->waste + stats->page_cache + stats->bookkeeping

Status: RESOLVED WORKSFORME

(Core :: Memory Allocator, defect)

Description

[Nicholas Nethercote [inactive]]

jemalloc.c:6769: Failed assertion: "stats->mapped >= stats->allocated + stats->waste + stats->page_cache + stats->bookkeeping"
Hit MOZ_CRASH() at /home/njn/moz/mi1/memory/build/jemalloc_config.c:33

Comment 1

[Nicholas Nethercote [inactive]]

I'm hitting this in a local build every time I hit "Measure" in about:memory.

Here are the numbers from one run.

(gdb) p *stats
$1 = {opt_abort = 1 '\001', opt_junk = 1 '\001', opt_utrace = 0 '\000', 
  opt_sysv = 0 '\000', opt_xmalloc = 0 '\000', opt_zero = 0 '\000', 
  narenas = 1, balance_threshold = 18446744073709551615, quantum = 16, 
  small_max = 512, large_max = 1040384, chunksize = 1048576, dirty_max = 1024, 
  mapped = 56623104, allocated = 52504864, waste = 2909920, 
  page_cache = 1478656, bookkeeping = 4096}
(gdb) p stats->mapped
$2 = 56623104
(gdb) p stats->allocated + stats->waste + stats->page_cache + stats->bookkeeping
$3 = 56897536

Comment 2

[Nicholas Nethercote [inactive]]

> I'm hitting this in a local build every time I hit "Measure" in about:memory.

Just to clarify... I'm not getting this on a trunk build.  But I'm working on a patch that removes nsIMemoryReporter and I'm getting it consistently there.  I haven't changed the innards of HeapAllocatedReporter.  However, it's pretty likely that the memory reporters are running in a substantially different order than they were previously.  In particular, the HeapAllocatedReporter may well be getting called much later than it previously was.

Comment 3

[Mike Hommey [:glandium]]

This could be because of bug 908971.

Comment 4

[Justin Lebar (not reading bugmail)]

That's in jemalloc_print_stats, which isn't the same as jemalloc_stats_impl.

Comment 5

[Nicholas Nethercote [inactive]]

Haven't seen this for a long time.