Gareth Rees
687f5e0294
Fix issues identified in review by apt.
...
Copied from Perforce
Change: 193982
2018-06-19 14:51:56 +01:00
Gareth Rees
0662e96899
Refactor atfork code to improve separation of concerns.
...
Copied from Perforce
Change: 193806
2018-06-15 10:58:33 +01:00
Gareth Rees
a865019b13
Update design history. restore accidentally removed check.
...
Copied from Perforce
Change: 193787
2018-06-14 17:55:03 +01:00
Gareth Rees
61e24c399a
Take locks before fork(), release them in the parent and reinitialize them in the child.
...
Add forktest target to the Xcode project.
Write design for fork safety and cross-reference from the code.
Update release notes.
Copied from Perforce
Change: 193760
2018-06-14 16:25:08 +01:00
Gareth Rees
2ccdd80953
Extend forktest so that it does allocation before the fork and collection afterwards; this ensures that we hit the read barrier and thus exercise the exception handling.
...
After forking in the child, re-create exception port and exception thread, and re-register the (only) thread with that port.
Copied from Perforce
Change: 193755
2018-06-14 12:54:35 +01:00
Gareth Rees
47fc093662
Add fork test case (fails on os x if pthread_atfork is not called).
...
pthread_atfork handlers on OS X: in the child, update the mach port for the forking thread and move all other threads to the dead ring.
Copied from Perforce
Change: 193746
2018-06-13 17:09:07 +01:00
Gareth Rees
3389ec396f
Add "cold end" and "hot end" to the glossary index.
...
Copied from Perforce
Change: 192660
ServerID: perforce.ravenbrook.com
2016-10-20 13:52:30 +01:00
Gareth Rees
d83934de69
Remove return statements that have no effect.
...
Copied from Perforce
Change: 192529
ServerID: perforce.ravenbrook.com
2016-10-13 15:17:50 +01:00
Gareth Rees
437bf9c6a4
Rename mutatorfaultcontext to mutatorcontext because this data structure is not only used to store the context of a fault, but also to store the context of a thread that has been suspended.
...
Copied from Perforce
Change: 192523
ServerID: perforce.ravenbrook.com
2016-10-13 14:23:39 +01:00
Gareth Rees
34be22b383
Remove undocumented global mps_exception_info. now that the mps is open source, there is no difficulty in debugging the mutator context.
...
Copied from Perforce
Change: 192518
ServerID: perforce.ravenbrook.com
2016-10-13 13:45:31 +01:00
Gareth Rees
62be09b5e9
Revert incorrect change to globalspreparetodestroy (see change 192365) that made the mps assert when destroying an arena with uncollected finalization messages.
...
Copied from Perforce
Change: 192473
ServerID: perforce.ravenbrook.com
2016-09-27 11:57:19 +01:00
Gareth Rees
cb0c6eabef
Catch-up merge from custom/cet/branch/2016-09-13/job004006 to branch/2016-09-06/job004006.
...
Copied from Perforce
Change: 192351
ServerID: perforce.ravenbrook.com
2016-09-13 15:33:31 +01:00
Gareth Rees
d56741e1ea
In order to be able to test mps_arena_busy in all build configurations, we must always take and release the arena lock, even in the single-threaded case. we gain some additional checking by doing this, at low cost (since in this build configuration the locks are generic and so just increment/decrement a count of claims).
...
In the CONFIG_POLL_NONE we configuration, we would still like to check that no traces are busy when leaving the arena, but since we now call ArenaLeave in this configurations, move the assertion to ShieldLeave.
In ArenaDestroy, call ArenaLeave, don't just release the lock.
Copied from Perforce
Change: 192267
ServerID: perforce.ravenbrook.com
2016-09-09 13:01:25 +01:00
Gareth Rees
01c13bf1b5
New functions mps_arena_busy and mps_arena_postmortem, plus test coverage.
...
Copied from Perforce
Change: 192247
ServerID: perforce.ravenbrook.com
2016-09-08 13:55:43 +01:00
Gareth Rees
c737a08bc2
Fix assertion text in the manual.
...
Place a cross-reference by each assertion the appears in the manual, to assist in keeping the manual up to date.
Copied from Perforce
Change: 192097
ServerID: perforce.ravenbrook.com
2016-09-02 14:11:08 +01:00
Gareth Rees
0297b380b9
Clear the emergency before removing the trace from busytraces, to avoid violating <code/global.c#emergency.invariant>.
...
Copied from Perforce
Change: 191774
ServerID: perforce.ravenbrook.com
2016-05-02 13:06:07 +01:00
Richard Brooksby
546e082125
Removing bogus arenatrivdescribe and turning arenadescribe the right way in.
...
Copied from Perforce
Change: 191480
ServerID: perforce.ravenbrook.com
2016-04-21 15:27:41 +01:00
Richard Brooksby
02a8064a07
Moving location dependency history into its own structure to reduce arenastruct bloat.
...
Copied from Perforce
Change: 191417
ServerID: perforce.ravenbrook.com
2016-04-21 09:05:34 +01:00
Gareth Rees
0d0eb93f01
Merge branch/2016-04-19/job004011.
...
Copied from Perforce
Change: 191374
ServerID: perforce.ravenbrook.com
2016-04-20 16:00:43 +01:00
Gareth Rees
c1aeb1f67a
Ensure that at most one collection of the world can be started in a call to arenapoll. this avoids a loop if the live set is large enough to provoke the "dynamic criterion".
...
Copied from Perforce
Change: 191274
ServerID: perforce.ravenbrook.com
2016-04-19 18:17:09 +01:00
Gareth Rees
aabe4992eb
Compile out statistic declarations in non-statistical varieties (at the cost of some syntax checking).
...
Add some missing STATISTIC guards where statistics are used.
Add design documentation.
Avoid unnecessary distinction between STATISTIC and STATISTIC_STAT (both result in statements).
Copied from Perforce
Change: 191178
ServerID: perforce.ravenbrook.com
2016-04-16 12:07:10 +01:00
Richard Brooksby
247396690d
Catch-up merge from master sources to branch/2016-03-11/shield-coalesce.
...
Copied from Perforce
Change: 190373
ServerID: perforce.ravenbrook.com
2016-03-24 11:08:24 +00:00
Richard Brooksby
428e8bae92
Fixing build errors on windows (w3i6mv).
...
tagtest was missing from makefiles.
Implicit conversion warning from Clock to double.
Copied from Perforce
Change: 190372
ServerID: perforce.ravenbrook.com
2016-03-24 11:04:58 +00:00
Richard Brooksby
a03174304d
Tearing down shield correctly.
...
Copied from Perforce
Change: 190225
ServerID: perforce.ravenbrook.com
2016-03-20 01:01:54 +00:00
Richard Brooksby
e09dbaab0f
Promoting shield to first class structure.
...
Copied from Perforce
Change: 190219
ServerID: perforce.ravenbrook.com
2016-03-20 00:32:43 +00:00
Richard Brooksby
ff69c9af04
Separating flag for segment in shield cache from depth, clarifying code and optimising segfree somewhat.
...
Copied from Perforce
Change: 190194
ServerID: perforce.ravenbrook.com
2016-03-19 13:42:09 +00:00
Richard Brooksby
9172773f9c
Updating shield design and actually checking invariants in code.
...
Copied from Perforce
Change: 190192
ServerID: perforce.ravenbrook.com
2016-03-19 10:15:10 +00:00
Richard Brooksby
641cdc164c
Catch-up merge from master sources to branch/2016-03-11/shield-coalesce.
...
Copied from Perforce
Change: 190150
ServerID: perforce.ravenbrook.com
2016-03-17 13:53:40 +00:00
Richard Brooksby
662a1b1b9b
Allocate the shield cache. also improve commentary quite a bit.
...
Copied from Perforce
Change: 190147
ServerID: perforce.ravenbrook.com
2016-03-17 13:37:40 +00:00
Richard Brooksby
9583c0f68a
Merging branch mps/branch/2016-03-13/without-reservation into master sources.
...
Copied from Perforce
Change: 190106
ServerID: perforce.ravenbrook.com
2016-03-15 13:19:27 +00:00
Richard Brooksby
d4a055f70b
Merging branch/2016-03-12/pause into the master sources.
...
Copied from Perforce
Change: 190053
ServerID: perforce.ravenbrook.com
2016-03-15 06:31:08 +00:00
Gareth Rees
e034e7aeea
Address points made by rb in review < https://info.ravenbrook.com/mail/2016/03/14/18-48-19/0/ >
...
Copied from Perforce
Change: 190031
ServerID: perforce.ravenbrook.com
2016-03-14 20:10:14 +00:00
Richard Brooksby
07987ab579
Eliminating withreservoirpermit and all its variants.
...
Copied from Perforce
Change: 190005
ServerID: perforce.ravenbrook.com
2016-03-13 21:04:01 +00:00
Richard Brooksby
ca7dcd913e
Basic removal of the reservoir pool from code and design.
...
Copied from Perforce
Change: 190000
ServerID: perforce.ravenbrook.com
2016-03-13 20:32:24 +00:00
Gareth Rees
167408b7e0
Implement pause time control.
...
Copied from Perforce
Change: 189921
ServerID: perforce.ravenbrook.com
2016-03-12 22:08:26 +00:00
Richard Brooksby
a3efd2d694
Catch-up merge from branch/2015-08-06/config to remove arenaconfigure.
...
Copied from Perforce
Change: 189335
ServerID: perforce.ravenbrook.com
2016-02-25 22:59:06 +00:00
Richard Brooksby
c015827c36
Catch-up merge from master sources.
...
Copied from Perforce
Change: 189265
ServerID: perforce.ravenbrook.com
2016-02-17 11:04:51 +00:00
Richard Brooksby
5dccd2e006
Catch-up merge from masters.
...
Copied from Perforce
Change: 188928
ServerID: perforce.ravenbrook.com
2016-01-20 11:04:09 +00:00
Richard Brooksby
eac348d664
Catch-up merge from masters.
...
Copied from Perforce
Change: 188921
ServerID: perforce.ravenbrook.com
2016-01-19 16:22:39 +00:00
Gareth Rees
aa6a027c68
Catch-up merge from branch/2015-08-11/policy to branch/2015-08-25/tradeoff.
...
Copied from Perforce
Change: 188250
ServerID: perforce.ravenbrook.com
2015-09-04 19:13:28 +01:00
Gareth Rees
2514663693
Move policy for mps_arena_step to policy module.
...
Copied from Perforce
Change: 188238
ServerID: perforce.ravenbrook.com
2015-09-04 15:25:37 +01:00
Gareth Rees
30a7fa542d
Fix bug introduced by change 188219: the end of the interval is not the same as the available collection time.
...
Copied from Perforce
Change: 188223
ServerID: perforce.ravenbrook.com
2015-09-04 08:53:23 +01:00
Gareth Rees
b9cd26b30a
There can only be an emergency when a trace is busy, so check this.
...
Copied from Perforce
Change: 188220
ServerID: perforce.ravenbrook.com
2015-09-03 20:14:06 +01:00
Gareth Rees
158c7758d7
Arenastep considers collecting the world after finishing a trace (not just on entry if no traces are busy).
...
ArenaStep does not care about quanta, so call TraceAdvance instead of TracePoll.
Copied from Perforce
Change: 188219
ServerID: perforce.ravenbrook.com
2015-09-03 19:11:49 +01:00
Gareth Rees
7bf1664f5e
Separate the values "more work to do?" and "amount of work done" in tracepoll. previously, the code used "amount of work done > 0" when it needed "more work to do?" but that's not right, because on the last two calls to traceadvance, no "work" is done (because reclaim work is not measured), but there may still be more work to do.
...
Copied from Perforce
Change: 188218
ServerID: perforce.ravenbrook.com
2015-09-03 15:51:37 +01:00
Gareth Rees
b4db3bd09f
Introduce new type work representing a measure of work done by the collector. use this systematically to make the code clearer.
...
Copied from Perforce
Change: 188210
ServerID: perforce.ravenbrook.com
2015-09-03 13:01:55 +01:00
Gareth Rees
bea6e91d73
Accumulate trace metrics in arenapark.
...
Copied from Perforce
Change: 188208
ServerID: perforce.ravenbrook.com
2015-09-03 11:50:28 +01:00
Gareth Rees
9dd49aa48e
New functions policypoll and policypollagain.
...
Update design.strategy.
Copied from Perforce
Change: 188165
ServerID: perforce.ravenbrook.com
2015-08-25 13:27:17 +01:00
Gareth Rees
d506458eea
New function policycollectiontime estimates the time needed to collect the arena. mysterious constants are given names and moved to config.h.
...
Copied from Perforce
Change: 188160
ServerID: perforce.ravenbrook.com
2015-08-21 15:27:14 +01:00
Richard Brooksby
1a730f9e83
Clarifying situations in which arenaaccess might find the barrier already down. see < https://info.ravenbrook.com/mail/2015/07/14/15-08-55/0/ >.
...
Copied from Perforce
Change: 188033
ServerID: perforce.ravenbrook.com
2015-07-14 17:18:00 +01:00