[Git][ghc/ghc][wip/tsan/fix-races] 74 commits: EPA: Simplify GHC/Parser.y comb4/comb5

Ben Gamari (@bgamari) gitlab at gitlab.haskell.org
Wed Aug 2 20:24:47 UTC 2023



Ben Gamari pushed to branch wip/tsan/fix-races at Glasgow Haskell Compiler / GHC


Commits:
355e1792 by Alan Zimmerman at 2023-07-26T10:17:32-04:00
EPA: Simplify GHC/Parser.y comb4/comb5

Use the HasLoc instance from Ast.hs to allow comb4/comb5 to work with
anything with a SrcSpan

Also get rid of some more now unnecessary reLoc calls.

- - - - -
9393df83 by Gavin Zhao at 2023-07-26T10:18:16-04:00
compiler: make -ddump-asm work with wasm backend NCG

Fixes #23503.

Now the `-ddump-asm` flag is respected in the wasm backend
NCG, so developers can directly view the generated ASM instead of
needing to pass `-S` or `-keep-tmp-files` and manually find & open
the assembly file.

Ideally, we should be able to output the assembly files in smaller
chunks like in other NCG backends. This would also make dumping assembly
stats easier. However, this would require a large refactoring, so for
short-term debugging purposes I think the current approach works fine.

Signed-off-by: Gavin Zhao <git at gzgz.dev>

- - - - -
79463036 by Krzysztof Gogolewski at 2023-07-26T10:18:54-04:00
llvm: Restore accidentally deleted code in 0fc5cb97

Fixes #23711

- - - - -
20db7e26 by Rodrigo Mesquita at 2023-07-26T10:19:33-04:00
configure: Default missing options to False when preparing ghc-toolchain Targets

This commit fixes building ghc with 9.2 as the boostrap compiler.

The ghc-toolchain patch assumed all _STAGE0 options were available, and
forgot to account for this missing information in 9.2.

Ghc 9.2 does not have in settings whether ar supports -l, hence can't
report it with --info (unliked 9.4 upwards).

The fix is to default the missing information (we default "ar supports
-l" and other missing options to False)

- - - - -
fac9e84e by Naïm Favier at 2023-07-26T10:20:16-04:00
docs: Fix typo
- - - - -
503fd647 by Bartłomiej Cieślar at 2023-07-26T17:23:10-04:00
This MR is an implementation of the proposal #516.

It adds a warning -Wincomplete-record-selectors for usages of a record
field access function (either a record selector or getField @"rec"),
while trying to silence the warning whenever it can be sure that a constructor
without the record field would not be invoked (which would otherwise cause
the program to fail). For example:

    data T = T1 | T2 {x :: Bool}

    f a = x a -- this would throw an error

    g T1 = True
    g a = x a -- this would not throw an error

    h :: HasField "x" r Bool => r -> Bool
    h = getField @"x"

    j :: T -> Bool
    j = h -- this would throw an error because of the `HasField`
          -- constraint being solved

See the tests DsIncompleteRecSel* and TcIncompleteRecSel for more examples of the warning.
See Note [Detecting incomplete record selectors] in GHC.HsToCore.Expr for implementation details

- - - - -
af6fdf42 by Arnaud Spiwack at 2023-07-26T17:23:52-04:00
Fix user-facing label in MR template

- - - - -
5d45b92a by Matthew Pickering at 2023-07-27T05:46:46-04:00
ci: Test bootstrapping configurations with full-ci and on marge batches

There have been two incidents recently where bootstrapping has been
broken by removing support for building with 9.2.*.

The process for bumping the minimum required version starts with bumping
the configure version and then other CI jobs such as the bootstrap jobs
have to be updated. We must not silently bump the minimum required
version.

Now we are running a slimmed down validate pipeline it seems worthwile
to test these bootstrap configurations in the full-ci pipeline.

- - - - -
25d4fee7 by Matthew Pickering at 2023-07-27T05:46:46-04:00
bootstrap: Remove ghc-9_2_* plans

We are anticipating shortly making it necessary to use ghc-9.4 to boot
the compiler.

- - - - -
2f66da16 by Matthew Pickering at 2023-07-27T05:46:46-04:00
Update bootstrap plans for ghc-platform and ghc-toolchain dependencies

Fixes #23735

- - - - -
c8c6eab1 by Matthew Pickering at 2023-07-27T05:46:46-04:00
bootstrap: Disable -selftest flag from bootstrap plans

This saves on building one dependency (QuickCheck) which is unecessary
for bootstrapping.

- - - - -
a80ca086 by Bodigrim at 2023-07-27T05:47:26-04:00
Link reference paper and package from System.Mem.{StableName,Weak}

- - - - -
a5319358 by David Knothe at 2023-07-28T13:13:10-04:00
Update Match Datatype

EquationInfo currently contains a list of the equation's patterns together with a CoreExpr that is to be evaluated after a successful match on this equation.
All the match-functions only operate on the first pattern of an equation - after successfully matching it, match is called recursively on the tail of the pattern list.
We can express this more clearly and make the code a little more elegant by updating the datatype of EquationInfo as follows:

data EquationInfo
    = EqnMatch { eqn_pat = Pat GhcTc, eqn_rest = EquationInfo }
    | EqnDone { eqn_rhs = MatchResult CoreExpr }

An EquationInfo now explicitly exposes its first pattern which most functions operate on, and exposes the equation that remains after processing the first pattern. An EqnDone signifies an empty equation where the CoreExpr can now be evaluated.

- - - - -
86ad1af9 by David Binder at 2023-07-28T13:13:53-04:00
Improve documentation for Data.Fixed

- - - - -
f8fa1d08 by Ben Gamari at 2023-07-28T13:14:31-04:00
ghc-prim: Use C11 atomics

Previously `ghc-prim`'s atomic wrappers used the legacy `__sync_*`
family of C builtins. Here we refactor these to rather use the
appropriate C11 atomic equivalents, allowing us to be more explicit
about the expected ordering semantics.

- - - - -
0bfc8908 by Finley McIlwaine at 2023-07-28T18:46:26-04:00
Include -haddock in DynFlags fingerprint

The -haddock flag determines whether or not the resulting .hi files
contain haddock documentation strings. If the existing .hi files do
not contain haddock documentation strings and the user requests them,
we should recompile.

- - - - -
40425c50 by Andreas Klebinger at 2023-07-28T18:47:02-04:00
Aarch64 NCG: Use encoded immediates for literals.

Try to generate

    instr x2, <imm>

instead of

    mov x1, lit
    instr x2, x1

When possible. This get's rid if quite a few redundant
mov instructions.

I believe this causes a metric decrease for LargeRecords as
we reduce register pressure.

-------------------------
Metric Decrease:
    LargeRecord
-------------------------

- - - - -
e9a0fa3f by Bodigrim at 2023-07-28T18:47:42-04:00
Bump filepath submodule to 1.4.100.4

Resolves #23741

Metric Decrease:
    MultiComponentModules
    MultiComponentModulesRecomp
    MultiLayerModules
    MultiLayerModulesRecomp
    T10421
    T12234
    T12425
    T13035
    T13701
    T13719
    T16875
    T18304
    T18698a
    T18698b
    T21839c
    T9198
    TcPlugin_RewritePerf
    hard_hole_fits

Metric decrease on Windows can be probably attributed to https://github.com/haskell/filepath/pull/183

- - - - -
ee93edfd by Bodigrim at 2023-07-28T18:48:21-04:00
Add since pragmas to GHC.IO.Handle.FD

- - - - -
d0369802 by Simon Peyton Jones at 2023-07-30T09:24:48+01:00
Make the occurrence analyser smarter about join points

This MR addresses #22404.  There is a big Note

   Note [Occurrence analysis for join points]

that explains it all.  Significant changes

* New field occ_join_points in OccEnv

* The NonRec case of occAnalBind splits into two cases:
  one for existing join points (which does the special magic for
  Note [Occurrence analysis for join points], and one for other
  bindings.

* mkOneOcc adds in info from occ_join_points.

* All "bring into scope" activity is centralised in the
  new function `addInScope`.

* I made a local data type LocalOcc for use inside the occurrence analyser
  It is like OccInfo, but lacks IAmDead and IAmALoopBreaker, which in turn
  makes computationns over it simpler and more efficient.

* I found quite a bit of allocation in GHC.Core.Rules.getRules
  so I optimised it a bit.

More minor changes

* I found I was using (Maybe Arity) a lot, so I defined a new data
  type JoinPointHood and used it everwhere.  This touches a lot of
  non-occ-anal files, but it makes everything more perspicuous.

* Renamed data constructor WithUsageDetails to WUD, and
  WithTailUsageDetails to WTUD

This also fixes #21128, on the way.

--------- Compiler perf -----------
I spent quite a time on performance tuning, so even though it
does more than before, the occurrence analyser runs slightly faster
on average.  Here are the compile-time allocation changes over 0.5%

      CoOpt_Read(normal) ghc/alloc    766,025,520    754,561,992  -1.5%
CoOpt_Singletons(normal) ghc/alloc    759,436,840    762,925,512  +0.5%
     LargeRecord(normal) ghc/alloc  1,814,482,440  1,799,530,456  -0.8%
       PmSeriesT(normal) ghc/alloc     68,159,272     67,519,720  -0.9%
          T10858(normal) ghc/alloc    120,805,224    118,746,968  -1.7%
          T11374(normal) ghc/alloc    164,901,104    164,070,624  -0.5%
          T11545(normal) ghc/alloc     79,851,808     78,964,704  -1.1%
          T12150(optasm) ghc/alloc     73,903,664     71,237,544  -3.6% GOOD
          T12227(normal) ghc/alloc    333,663,200    331,625,864  -0.6%
          T12234(optasm) ghc/alloc     52,583,224     52,340,344  -0.5%
          T12425(optasm) ghc/alloc     81,943,216     81,566,720  -0.5%
          T13056(optasm) ghc/alloc    294,517,928    289,642,512  -1.7%
      T13253-spj(normal) ghc/alloc    118,271,264     59,859,040 -49.4% GOOD
          T15164(normal) ghc/alloc  1,102,630,352  1,091,841,296  -1.0%
          T15304(normal) ghc/alloc  1,196,084,000  1,166,733,000  -2.5%
          T15630(normal) ghc/alloc    148,729,632    147,261,064  -1.0%
          T15703(normal) ghc/alloc    379,366,664    377,600,008  -0.5%
          T16875(normal) ghc/alloc     32,907,120     32,670,976  -0.7%
          T17516(normal) ghc/alloc  1,658,001,888  1,627,863,848  -1.8%
          T17836(normal) ghc/alloc    395,329,400    393,080,248  -0.6%
          T18140(normal) ghc/alloc     71,968,824     73,243,040  +1.8%
          T18223(normal) ghc/alloc    456,852,568    453,059,088  -0.8%
          T18282(normal) ghc/alloc    129,105,576    131,397,064  +1.8%
          T18304(normal) ghc/alloc     71,311,712     70,722,720  -0.8%
         T18698a(normal) ghc/alloc    208,795,112    210,102,904  +0.6%
         T18698b(normal) ghc/alloc    230,320,736    232,697,976  +1.0%  BAD
          T19695(normal) ghc/alloc  1,483,648,128  1,504,702,976  +1.4%
          T20049(normal) ghc/alloc     85,612,024     85,114,376  -0.6%
         T21839c(normal) ghc/alloc    415,080,992    410,906,216  -1.0% GOOD
           T4801(normal) ghc/alloc    247,590,920    250,726,272  +1.3%
           T6048(optasm) ghc/alloc     95,699,416     95,080,680  -0.6%
            T783(normal) ghc/alloc    335,323,384    332,988,120  -0.7%
           T9233(normal) ghc/alloc    709,641,224    685,947,008  -3.3% GOOD
           T9630(normal) ghc/alloc    965,635,712    948,356,120  -1.8%
           T9675(optasm) ghc/alloc    444,604,152    428,987,216  -3.5% GOOD
           T9961(normal) ghc/alloc    303,064,592    308,798,800  +1.9%  BAD
           WWRec(normal) ghc/alloc    503,728,832    498,102,272  -1.1%

               geo. mean                                          -1.0%
               minimum                                           -49.4%
               maximum                                            +1.9%

In fact these figures seem to vary between platforms; generally worse
on i386 for some reason.  The Windows numbers vary by 1% espec in
benchmarks where the total allocation is low. But the geom mean stays
solidly negative, which is good.  The "increase/decrease" list below
covers all platforms.

The big win on T13253-spj comes because it has a big nest of join
points, each occurring twice in the next one.  The new occ-anal takes
only one iteration of the simplifier to do the inlining; the old one
took four.  Moreover, we get much smaller code with the new one:

  New: Result size of Tidy Core
    = {terms: 429, types: 84, coercions: 0, joins: 14/14}

  Old: Result size of Tidy Core
    = {terms: 2,437, types: 304, coercions: 0, joins: 10/10}

--------- Runtime perf -----------
No significant changes in nofib results, except a 1% reduction in
compiler allocation.

Metric Decrease:
    CoOpt_Read
    T13253-spj
    T9233
    T9630
    T9675
    T12150
    T21839c
    LargeRecord
    MultiComponentModulesRecomp
    T10421
    T13701
    T10421
    T13701
    T12425

Metric Increase:
    T18140
    T9961
    T18282
    T18698a
    T18698b
    T19695

- - - - -
42aa7fbd by Julian Ospald at 2023-07-30T17:22:01-04:00
Improve documentation around IOException and ioe_filename

See:

* https://github.com/haskell/core-libraries-committee/issues/189
* https://github.com/haskell/unix/pull/279
* https://github.com/haskell/unix/pull/289

- - - - -
33598ecb by Sylvain Henry at 2023-08-01T14:45:54-04:00
JS: implement getMonotonicTime (fix #23687)

- - - - -
d2bedffd by Bartłomiej Cieślar at 2023-08-01T14:46:40-04:00
Implementation of the Deprecated Instances proposal #575

This commit implements the ability to deprecate certain instances,
which causes the compiler to emit the desired deprecation message
whenever they are instantiated. For example:

  module A where
  class C t where
  instance {-# DEPRECATED "dont use" #-} C Int where

  module B where
  import A
  f :: C t => t
  f = undefined
  g :: Int
  g = f -- "dont use" emitted here

The implementation is as follows:
  - In the parser, we parse deprecations/warnings attached to instances:

      instance {-# DEPRECATED "msg" #-} Show X
      deriving instance {-# WARNING "msg2" #-} Eq Y

    (Note that non-standalone deriving instance declarations do not support
    this mechanism.)

  - We store the resulting warning message in `ClsInstDecl` (respectively, `DerivDecl`).
    In `GHC.Tc.TyCl.Instance.tcClsInstDecl` (respectively, `GHC.Tc.Deriv.Utils.newDerivClsInst`),
    we pass on that information to `ClsInst` (and eventually store it in `IfaceClsInst` too).

  - Finally, when we solve a constraint using such an instance, in
    `GHC.Tc.Instance.Class.matchInstEnv`, we emit the appropriate warning
    that was stored in `ClsInst`.
    Note that we only emit a warning when the instance is used in a different module
    than it is defined, which keeps the behaviour in line with the deprecation of
    top-level identifiers.

Signed-off-by: Bartłomiej Cieślar <bcieslar2001 at gmail.com>

- - - - -
d5a65af6 by Ben Gamari at 2023-08-01T14:47:18-04:00
compiler: Style fixes

- - - - -
7218c80a by Ben Gamari at 2023-08-01T14:47:19-04:00
rts: Fix implicit cast

This ensures that Task.h can be built with a C++ compiler.

- - - - -
d6d5aafc by Ben Gamari at 2023-08-01T14:47:19-04:00
testsuite: Fix warning in hs_try_putmvar001

- - - - -
d9eddf7a by Ben Gamari at 2023-08-01T14:47:19-04:00
testsuite: Add AtomicModifyIORef test

- - - - -
f9eea4ba by Ben Gamari at 2023-08-01T14:47:19-04:00
rts: Introduce NO_WARN macro

This allows fine-grained ignoring of warnings.

- - - - -
497b24ec by Ben Gamari at 2023-08-01T14:47:19-04:00
rts: Simplify atomicModifyMutVar2# implementation

Previously we would perform a redundant load in the non-threaded RTS in
atomicModifyMutVar2# implementation for the benefit of the non-moving
GC's write barrier. Eliminate this.

- - - - -
52ee082b by Ben Gamari at 2023-08-01T14:47:19-04:00
rts: Introduce more principled fence operations

- - - - -
cd3c0377 by Ben Gamari at 2023-08-01T14:47:19-04:00
rts: Introduce SET_INFO_RELAXED

- - - - -
6df2352a by Ben Gamari at 2023-08-01T14:47:19-04:00
rts: Style fixes

- - - - -
4ef6f319 by Ben Gamari at 2023-08-01T14:47:19-04:00
codeGen/tsan: Rework handling of spilling

- - - - -
f9ca7e27 by Ben Gamari at 2023-08-01T14:47:19-04:00
hadrian: More debug information

- - - - -
df4153ac by Ben Gamari at 2023-08-01T14:47:19-04:00
Improve TSAN documentation

- - - - -
fecae988 by Ben Gamari at 2023-08-01T14:47:19-04:00
hadrian: More selective TSAN instrumentation

- - - - -
465a9a0b by Alan Zimmerman at 2023-08-01T14:47:56-04:00
EPA: Provide correct annotation span for ImportDecl

Use the whole declaration, rather than just the span of the 'import'
keyword.

Metric Decrease:
   T9961
   T5205
Metric Increase:
  T13035

- - - - -
ae63d0fa by Bartłomiej Cieślar at 2023-08-01T14:48:40-04:00
Add cases to T23279: HasField for deprecated record fields

This commit adds additional tests from ticket #23279 to ensure that we don't
regress on reporting deprecated record fields in conjunction with HasField,
either when using overloaded record dot syntax or directly through `getField`.

Fixes #23279

- - - - -
00fb6e6b by Andreas Klebinger at 2023-08-01T14:49:17-04:00
AArch NCG: Pure refactor

Combine some alternatives. Add some line breaks for overly long lines

- - - - -
8f3b3b78 by Andreas Klebinger at 2023-08-01T14:49:54-04:00
Aarch ncg: Optimize immediate use for address calculations

When the offset doesn't fit into the immediate we now just reuse the
general getRegister' code path which is well optimized to compute the
offset into a register instead of a special case for CmmRegOff.

This means we generate a lot less code under certain conditions which is
why performance metrics for these improve.

-------------------------
Metric Decrease:
    T4801
    T5321FD
    T5321Fun
-------------------------

- - - - -
74a882dc by MorrowM at 2023-08-02T06:00:03-04:00
Add a RULE to make lookup fuse

See https://github.com/haskell/core-libraries-committee/issues/175

Metric Increase:
    T18282

- - - - -
cca74dab by Ben Gamari at 2023-08-02T06:00:39-04:00
hadrian: Ensure that way-flags are passed to CC

Previously the way-specific compilation flags (e.g. `-DDEBUG`,
`-DTHREADED_RTS`) would not be passed to the CC invocations. This meant
that C dependency files would not correctly reflect
dependencies predicated on the way, resulting in the rather
painful #23554.

Closes #23554.

- - - - -
622b483c by Jaro Reinders at 2023-08-02T06:01:20-04:00
Native 32-bit Enum Int64/Word64 instances

This commits adds more performant Enum Int64 and Enum Word64 instances
for 32-bit platforms, replacing the Integer-based implementation.

These instances are a copy of the Enum Int and Enum Word instances with
minimal changes to manipulate Int64 and Word64 instead.

On i386 this yields a 1.5x performance increase and for the JavaScript
back end it even yields a 5.6x speedup.

Metric Decrease:
    T18964

- - - - -
c8bd7fa4 by Sylvain Henry at 2023-08-02T06:02:03-04:00
JS: fix typos in constants (#23650)

- - - - -
b9d5bfe9 by Josh Meredith at 2023-08-02T06:02:40-04:00
JavaScript: update MK_TUP macros to use current tuple constructors (#23659)

- - - - -
28211215 by Matthew Pickering at 2023-08-02T06:03:19-04:00
ci: Pass -Werror when building hadrian in hadrian-ghc-in-ghci job

Warnings when building Hadrian can end up cluttering the output of HLS,
and we've had bug reports in the past about these warnings when building
Hadrian. It would be nice to turn on -Werror on at least one build of
Hadrian in CI to avoid a patch introducing warnings when building
Hadrian.

Fixes #23638

- - - - -
aca20a5d by Ben Gamari at 2023-08-02T06:03:55-04:00
codeGen: Ensure that TSAN is aware of writeArray# write barriers

By using a proper release store instead of a fence.

- - - - -
453c0531 by Ben Gamari at 2023-08-02T06:03:55-04:00
codeGen: Ensure that array reads have necessary barriers

This was the cause of #23541.

- - - - -
93a0d089 by Arnaud Spiwack at 2023-08-02T06:04:37-04:00
Add test for #23550

- - - - -
6a2f4a20 by Arnaud Spiwack at 2023-08-02T06:04:37-04:00
Desugar non-recursive lets to non-recursive lets (take 2)

This reverts commit 522bd584f71ddeda21efdf0917606ce3d81ec6cc. And
takes care of the case that I missed in my previous attempt. Namely
the case of an AbsBinds with no type variables and no dictionary
variable.

Ironically, the comment explaining why non-recursive lets were
desugared to recursive lets were pointing specifically at this case
as the reason. I just failed to understand that it was until Simon PJ
pointed it out to me.

See #23550 for more discussion.

- - - - -
ff81d53f by jade at 2023-08-02T06:05:20-04:00
Expand documentation of List & Data.List

This commit aims to improve the documentation and examples
of symbols exported from Data.List

- - - - -
fa4e5913 by Jade at 2023-08-02T06:06:03-04:00
Improve documentation of Semigroup & Monoid

This commit aims to improve the documentation of various symbols
exported from Data.Semigroup and Data.Monoid

- - - - -
63ec05f1 by Ben Gamari at 2023-08-02T15:10:09-04:00
rts: Fix data race in threadPaused

This only affects an assertion in the debug RTS and only needs relaxed
ordering.

- - - - -
fe81ea2e by Ben Gamari at 2023-08-02T15:10:09-04:00
cmm: Introduce MO_RelaxedRead

In hand-written Cmm it can sometimes be necessary to atomically load
from memory deep within an expression (e.g. see the `CHECK_GC` macro).
This MachOp provides a convenient way to do so without breaking the
expression into multiple statements.

- - - - -
02708e02 by Ben Gamari at 2023-08-02T15:10:09-04:00
rts: Silence spurious data races in ticky counters

Previously we would use non-atomic accesses when bumping ticky counters,
which would result in spurious data race reports from ThreadSanitizer
when the threaded RTS was in use.

- - - - -
afeda4aa by Ben Gamari at 2023-08-02T15:10:09-04:00
codeGen: Use relaxed accesses in ticky bumping

- - - - -
4d9c9d61 by Ben Gamari at 2023-08-02T15:10:09-04:00
rts: Fix data race in Interpreter's preemption check

- - - - -
da9fb4e3 by Ben Gamari at 2023-08-02T15:10:09-04:00
rts: Fix data race in threadStatus#

- - - - -
e59cc61a by Ben Gamari at 2023-08-02T15:10:10-04:00
rts: Fix data race in CHECK_GC

- - - - -
a740278e by Ben Gamari at 2023-08-02T15:10:10-04:00
base: use atomic write when updating timer manager

- - - - -
5d989bbe by Ben Gamari at 2023-08-02T15:10:10-04:00
Use relaxed atomics to manipulate TSO status fields

- - - - -
e82f189f by Ben Gamari at 2023-08-02T15:10:10-04:00
rts: Add necessary barriers when manipulating TSO owner

- - - - -
cf8406f1 by Ben Gamari at 2023-08-02T15:10:10-04:00
rts: Fix synchronization on thread blocking state

- - - - -
b4eef55a by Ben Gamari at 2023-08-02T15:10:10-04:00
rts: Use relaxed ordering on dirty/clean info tables updates

When changing the dirty/clean state of a mutable object we needn't have
any particular ordering.

- - - - -
2b751375 by Ben Gamari at 2023-08-02T15:10:10-04:00
Wordsmith TSAN Note

- - - - -
344d1b43 by Ben Gamari at 2023-08-02T15:10:10-04:00
codeGen: Use relaxed-read in closureInfoPtr

- - - - -
67009487 by Ben Gamari at 2023-08-02T15:10:10-04:00
Fix thunk update ordering

Previously we attempted to ensure soundness of concurrent thunk update
by synchronizing on the access of the thunk's info table pointer field.
This was believed to be sufficient since the indirectee (which may
expose a closure allocated by another core) would not be examined
until the info table pointer update is complete.

However, it turns out that this can result in data races in the presence
of multiple threads racing a update a single thunk. For instance,
consider this interleaving under the old scheme:

            Thread A                             Thread B
            ---------                            ---------
    t=0     Enter t
      1     Push update frame
      2     Begin evaluation

      4     Pause thread
      5     t.indirectee=tso
      6     Release t.info=BLACKHOLE

      7     ... (e.g. GC)

      8     Resume thread
      9     Finish evaluation
      10    Relaxed t.indirectee=x

      11                                         Load t.info
      12                                         Acquire fence
      13                                         Inspect t.indirectee

      14    Release t.info=BLACKHOLE

Here Thread A enters thunk `t` but is soon paused, resulting in `t`
being lazily blackholed at t=6. Then, at t=10 Thread A finishes
evaluation and updates `t.indirectee` with a relaxed store.

Meanwhile, Thread B enters the blackhole. Under the old scheme this
would introduce an acquire-fence but this would only synchronize with
Thread A at t=6. Consequently, the result of the evaluation, `x`, is not
visible to Thread B, introducing a data race.

We fix this by treating the `indirectee` field as we do all other
mutable fields. This means we must always access this field with
acquire-loads and release-stores.

See #23185.

- - - - -
191f3c52 by Ben Gamari at 2023-08-02T15:10:10-04:00
STM: Use acquire loads when possible

Full sequential consistency is not needed here.

- - - - -
4c46c71f by Ben Gamari at 2023-08-02T15:10:10-04:00
rts/Interpreter: Fix data race

- - - - -
28b74190 by Ben Gamari at 2023-08-02T15:10:10-04:00
rts/Messages: Fix data race

- - - - -
80197851 by Ben Gamari at 2023-08-02T15:10:10-04:00
rts/Prof: Fix data race

- - - - -
e99add7c by Ben Gamari at 2023-08-02T15:10:10-04:00
rts: Use fence rather than redundant load

Previously we would use an atomic load to ensure acquire ordering.
However, we now have `ACQUIRE_FENCE_ON`, which allows us to express this
more directly.

- - - - -
3dccbfe2 by Ben Gamari at 2023-08-02T15:10:10-04:00
rts: Fix data races in profiling timer

- - - - -
a7761236 by Ben Gamari at 2023-08-02T15:10:10-04:00
rts/RaiseAsync: Drop redundant release fence

- - - - -


27 changed files:

- .gitlab-ci.yml
- .gitlab/merge_request_templates/Default.md
- compiler/GHC/Cmm/Expr.hs
- compiler/GHC/Cmm/Info.hs
- compiler/GHC/Cmm/MachOp.hs
- compiler/GHC/Cmm/Parser.y
- compiler/GHC/Cmm/ThreadSanitizer.hs
- compiler/GHC/CmmToAsm.hs
- compiler/GHC/CmmToAsm/AArch64/CodeGen.hs
- compiler/GHC/CmmToAsm/AArch64/Instr.hs
- compiler/GHC/CmmToAsm/AArch64/Ppr.hs
- compiler/GHC/CmmToAsm/AArch64/Regs.hs
- compiler/GHC/CmmToAsm/PPC/CodeGen.hs
- compiler/GHC/CmmToAsm/Wasm.hs
- compiler/GHC/CmmToAsm/Wasm/FromCmm.hs
- compiler/GHC/CmmToAsm/X86/CodeGen.hs
- compiler/GHC/CmmToC.hs
- compiler/GHC/CmmToLlvm/CodeGen.hs
- compiler/GHC/Core/ConLike.hs
- compiler/GHC/Core/InstEnv.hs
- compiler/GHC/Core/Lint.hs
- compiler/GHC/Core/Opt/CSE.hs
- compiler/GHC/Core/Opt/DmdAnal.hs
- compiler/GHC/Core/Opt/Exitify.hs
- compiler/GHC/Core/Opt/FloatIn.hs
- compiler/GHC/Core/Opt/FloatOut.hs
- compiler/GHC/Core/Opt/OccurAnal.hs


The diff was not included because it is too large.


View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/d001bd8446b1e539016064db84c1adc7776d83e1...a77612362422963349cd18d1f2dc36c2ba1f38e9

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/d001bd8446b1e539016064db84c1adc7776d83e1...a77612362422963349cd18d1f2dc36c2ba1f38e9
You're receiving this email because of your account on gitlab.haskell.org.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20230802/7701ff1b/attachment-0001.html>


More information about the ghc-commits mailing list