| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
| |
And add or adjust comments at the use sites of awaitEvent.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The blocked_queue_{hd,tl} and the sleeping_queue are currently
cooperatively managed between the scheduler and (some but not all of)
the non-threaded I/O manager implementations.
They lived as global vars with the scheduler, but are poked by I/O
primops and the I/O manager backends.
This patch is a step on the path towards making the management of I/O or
timer blocking belong to the I/O managers and not the scheduler.
Specifically, this patch moves the {blocked,sleeping}_queue from being
global vars in the scheduler to being members of the CapIOManager struct
within each Capability. They are not yet exclusively used by the I/O
managers: they are still poked from a couple other places, notably in
the scheduler before calling awaitEvent.
|
| |
|
|
|
|
|
|
|
|
| |
Since we don't intend to ever change the incoming string, declare this
to be true.
Also, in the POSIX implementation, the argument is no longer `STG_UNUSED`
(since ee0deb8054da2a597fc5624469b4c44fd769ada2) in any code path.
See: https://gitlab.haskell.org/ghc/ghc/-/commit/ee0deb8054da2a597fc5624469b4c44fd769ada2#note_460080
|
| |
|
|
|
|
|
|
|
|
|
|
| |
As advertized by `autoreconf`:
> All current systems provide time.h; it need not be checked for.
Hence, remove the check for it in `configure.ac` and remove conditional
inclusion of the header in `HAVE_TIME_H` blocks where applicable.
The `time.h` header was being included in various source files without a
`HAVE_TIME_H` guard already anyway.
|
| |
|
|
|
|
|
| |
This eliminates the thread label HashTable and instead tracks this
information in the TSO, allowing us to use proper StgArrBytes arrays for
backing the label and greatly simplifying management of object lifetimes
when we expose them to the user with the coming `threadLabel#` primop.
|
| | |
|
| |
|
|
|
| |
This introduces a global hook which is called when an exception is
thrown during finalization.
|
| |
|
|
|
| |
Since f6e366c058b136f0789a42222b8189510a3693d1 setExecutable has been
dead code. Drop it.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Here we refactor WinIO's IO completion scheme, squashing a memory leak
and fixing #18382.
To fix #18382 we drop the special thread status introduced for IoPort
blocking, BlockedOnIoCompletion, as well as drop the non-threaded RTS's
special dead-lock detection logic (which is redundant to the GC's
deadlock detection logic), as proposed in #20947.
Previously WinIO relied on foreign import ccall "wrapper" to create an
adjustor thunk which can be attached to the OVERLAPPED structure passed
to the operating system. It would then use foreign import ccall
"dynamic" to back out the original continuation from the adjustor. This
roundtrip is significantly more expensive than the alternative, using a
StablePtr. Furthermore, the implementation let the adjustor leak,
meaning that every IO request would leak a page of memory.
Fixes T18382.
|
| | |
|
| |
|
|
|
|
|
| |
Just a tiny cleanup inspired by the following comment:
https://gitlab.haskell.org/ghc/ghc/-/issues/19437#note_334271
I was just getting familiar with rts code base so I
thought might as well do this.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Many small things to fix:
* Hadrian: platform triple is "x86_64-w64-mingw32" and this wasn't recognized by
Hadrian (note "w64" instead of "unknown")
* Hadrian was using the build platform ("isWindowsHost") to detect
the use of the Windows toolchain, which was wrong. We now use the
"targetOs" setting.
* Hadrian was doing the same thing for Darwin so we fixed both at once,
even if cross-compilation to Darwin is unlikely to happen afaik (cf
"osxHost" vs "osxTarget" changes)
* Hadrian: libffi name was computed in two different places and one of
them wasn't taking the different naming on Windows into account.
* Hadrian was passing "-Irts/include" when building the stage1 compiler
leading to the same error as in #18143 (which is using make).
stage1's RTS is stage0's one so mustn't do this.
* Hadrian: Windows linker doesn't seem to support "-zorigin" so we
don't pass it (similarly to Darwin)
* Hadrian: hsc2hs in cross-compilation mode uses a trick (taken from
autoconf): it defines "static int test_array[SOME_EXPR]" where
SOME_EXPR is a constant expression. However GCC reports an error
because SOME_EXPR is supposedly not constant. This is fixed by using
another method enabled with the `--via-asm` flag of hsc2hs. It has been
fixed in `make` build system (5f6fcf7808b16d066ad0fb2068225b3f2e8363f7)
but not in Hadrian.
* Hadrian: some packages are specifically built only on Windows but they
shouldn't be when building a cross-compiler (`touchy` and
`ghci-wrapper`). We now correctly detect this case and disable these
packages.
* Base: we use `iNVALID_HANDLE_VALUE` in a few places. It fixed some
hsc2hs issues before we switched to `--via-asm` (see above). I've kept
these changes are they make the code nicer.
* Base: `base`'s configure tries to detect if it is building for Windows
but for some reason the `$host_alias` value is `x86_64-windows` in my
case and it wasn't properly detected.
* Base: libraries/base/include/winio_structs.h imported "Windows.h" with
a leading uppercase. It doesn't work on case-sensitive systems when
cross-compiling so we have to use "windows.h".
* RTS: rts/win32/ThrIOManager.c was importin "rts\OSThreads.h" but this
path isn't valid when cross-compiling. We replaced "\" with "/".
* DeriveConstants: this tool derives the constants from the target
RTS header files. However these header files define `StgAsyncIOResult`
only when `mingw32_HOST_OS` is set hence it seems we have to set it
explicitly.
Note that deriveConstants is called more than once (why? there is
only one target for now so it shouldn't) and in the second case this
value is correctly defined (probably coming indirectly from the import
of "rts/PosixSource.h"). A better fix would probably be to disable the
unneeded first run of deriveconstants.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Previously we relied on the caller to check the return value from
broadcastCondition and friends, most of whom neglected to do so. Given
that these functions should not fail anyways, I've opted to drop the
return value entirely and rather move the result check into the
OSThreads functions.
This slightly changes the semantics of timedWaitCondition which now
returns false only in the case of timeout, rather than any error as
previously done.
|
| |
|
|
|
| |
Previously we were treating the thread ID as a HANDLE, but it is not. We
must first OpenThread.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
| |
They are not part of the IOManager interface used within the rest of the
RTS. They are the part of the interface of specific I/O manager
implementations.
They are no longer called directly elsewhere in the RTS, and are now
only called by the dispatch functions in IOManager.c
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Naming is hard. Where we want to get to is to have a clear internal and
external API for the IO manager within the RTS. What we have right now
is just the external API (used in base for the Haskell side of the
threaded IO manager impls) living in includes/rts/IOManager.h.
We want to add a clear RTS internal API, which really ought to live in
rts/IOManager.h. Several people think it's too confusing to have both:
* includes/rts/IOManager.h for the external API
* rts/IOManager.h for the internal API
So the plan is to add rts/IOManager.{h,c} as the internal parts, and
rename the external part to be includes/rts/IOInterface.h.
It is admittidly not great to have .h files in includes/rts/ called
"interface" since by definition, every .h fle under includes/ is an
interface!
Alternative naming scheme suggestions welcome!
|
| |
|
|
|
| |
It is only for MIO, and we want to use the generic name IOManager for
the name of the common parts of the interface and dispatch.
|
| | |
|
| |
|
|
| |
Use the GHC wrappers instead of <assert.h>.
|
| |
|
|
|
|
|
| |
As noted in #18991, we would previously allocate heap in low memory.
Due to this the linker, which typically *needs* low memory, would end up
competing with the heap. In longer builds we end up running out of
low memory entirely, leading to linking failures.
|
| |\ |
|
| | |
| |
| |
| |
| | |
Previously we would take all capabilities but fail to join on the thread
itself, potentially resulting in a leaked thread.
|
| |/ |
|
| | |
|
| |
|
|
| |
These slipped through CI.
|
| |
|
|
|
| |
This can help in debugging RTS memory leaks since all allocations go
through the same interface.
|
| |
|
|
|
|
|
|
| |
Since switching to the two-step allocator, the `outofmem` test fails via
`osCommitMemory` failing to commit. However, this was previously exiting
with `EXIT_FAILURE`, rather than `EXIT_HEAPOVERFLOW`. I think the latter
is a more reasonable exit code for this case and matches the behavior on
POSIX platforms.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We used a variable to keep track of situations where we got
entries from the IO port, but all of them had already been
canceled. While we can avoid some work that way this case
seems quite rare.
So we give up on tracking this and instead always assume at
least one of the returned entries is valid.
If that's not the case no harm is done, we just perform some
additional work. But it makes the runner easier to reason about.
In particular we don't need to care if another thread modifies
oustanding_requests after we return from waiting on the IO Port.
|
| |
|
|
|
|
|
|
| |
Only use a single communication point (registerAlertableWait) to inform
the C side aobut both timeouts to use as well as outstanding requests.
Also queue a haskell processing thread after each return from alertable
waits. This way there is no risk of us missing a timer event.
|
| | |
|
| | |
|
| |
|
|
|
| |
This will unblock the IO thread sooner hopefully leading to higher
throughput in some situations.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We now set a flag in the IO thread. The scheduler when looking for work
will check the flag and create/queue threads accordingly.
We used to create these in the IO thread. This improved performance
but caused frequent segfaults. Thread creation/allocation is only safe to
do if nothing currently accesses the storeagemanager. However without
locks in the non-threaded runtime this can't be guaranteed.
This shouldn't change performance all too much.
In the past we had:
* IO: Create/Queue thread.
* Scheduler: Runs a few times. Eventually picks up IO processing thread.
Now it's:
* IO: Set flag to queue thread.
* Scheduler: Pick up flag, if set create/queue thread. Eventually picks up IO processing thread.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As far as I know C(99) gives no guarantees for code like
bool condition;
...
while(condition)
sleep();
that condition will be updated if it's changed by another thread.
So we are explicit here and mark it as volatile, this will force
a reload from memory on each iteration.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
According to phyx these should only be read and written once per
object. Not neccesarily in that order.
To strengthen that guarantee the primitives will now throw an
exception if we violate this invariant.
As a consequence we can eliminate some code from their primops.
In particular code dealing with multiple queued readers/writers
now simply checks the invariant and throws an exception if it
was violated. That is in contrast to mvars which will do things
like wake up all readers, queue multi writers etc.
|
| | |
|
| | |
|
| | |
|
| | |
|
