GetQueuedCompletionStatusEx has a ~16ms timeout resolution. Use a
WaitCompletionPacket associated with the I/O Completion Port (IOCP)
and a high resolution timer so the IOCP is signaled on timer expiry,
therefore improving the GetQueuedCompletionStatusEx timeout resolution.
BenchmarkSleep from the time package shows an important improvement:
goos: windows
goarch: amd64
pkg: time
cpu: Intel(R) Core(TM) i7-10850H CPU @ 2.70GHz
│ old.txt │ new.txt │
│ sec/op │ sec/op vs base │
Sleep-12 1258.5µ ± 5% 250.7µ ± 1% -80.08% (p=0.000 n=20)
Fixes#44343.
Change-Id: I79fc09e34dddfc49e0e23c3d1d0603926c22a11d
Reviewed-on: https://go-review.googlesource.com/c/go/+/488675
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Quim Muntal <quimmuntal@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Bryan Mills <bcmills@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
This is another attempt at CL 558895, but without adding stale pollDescs
protection, which deviates from the original goal of the CL and adds
complexity without proper testing.
It is currently not possible to distinguish between a netpollBreak,
an internal/poll WSA operation, and an external WSA operation (as
in #58870). This can cause spurious wakeups when external WSA operations
are retrieved from the queue, as they are treated as netpollBreak
events.
This CL makes use of completion keys to identify the source of the
event.
While here, fix TestWSASocketConflict, which was not properly
exercising the "external WSA operation" case.
Change-Id: I91f746d300d32eb7fed3c8f27266fef379360d98
Reviewed-on: https://go-review.googlesource.com/c/go/+/561895
Reviewed-by: Than McIntosh <thanm@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
This reverts commit 29746b4814bb76a3d5a36fc86967b358f11bbb93.
Reason for revert: Windows builders are flaky since this CL. Needs investigation. See https://build.golang.org/log/70d5d039b57b505870c9cc4e61de320df06a6f3a.
Change-Id: I8a5874bb057785497d03b9450819578de7faeb47
Reviewed-on: https://go-review.googlesource.com/c/go/+/561276
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
It is currently not possible to distinguish between a netpollBreak
entry and an entry initiated by external WSA operations (as in #58870).
This CL sets a unique completion key when posting the
netpollBreak entry so that it can be identified as such.
Change-Id: I8e74a7ddc607dc215d6ed8c59d5c3cf47ec8dc62
Reviewed-on: https://go-review.googlesource.com/c/go/+/558895
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
On Windows, the netpoll is currently coupled with the websocket usage
in the internal/poll package.
This CL moves the websocket handling out of the runtime and puts it into
the internal/poll package, which already contains most of the async I/O
logic for websockets.
This is a good refactor per se, as the Go runtime shouldn't know about
websockets. In addition, it will make it easier (in a future CL) to only
load ws2_32.dll when the Go program actually uses websockets.
Change-Id: Ic820872cf9bdbbf092505ed7f7504edb6687735e
Reviewed-on: https://go-review.googlesource.com/c/go/+/556936
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
The runtime was adjusting netpollWaiters before the waiting
goroutines were marked as ready. This could cause the scheduler
to report a deadlock because there were no goroutines ready to run.
Keeping netpollWaiters non-zero ensures that at least one goroutine
will call netpoll(-1) from findRunnable.
This does mean that if a program has network activity for a while
and then never has it again, and also has no timers, then we can leave
an M stranded in a call to netpoll from which it will never return.
At least this won't be a common case. And it's not new; this has been
a potential problem for some time.
Fixes#61454
Change-Id: I17c7f891c2bb1262fda12c6929664e64686463c8
Reviewed-on: https://go-review.googlesource.com/c/go/+/511455
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Auto-Submit: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
It is possible for a netpoll file to be closed and for the pollDesc
to be reused while a netpoll is running. This normally only causes
spurious wakeups, but if there is an error on the old file then the
new file can be incorrectly marked as having an error.
Fix this problem on most systems by introducing an fd sequence field
and using that as a tag in a taggedPointer. The taggedPointer is
stored in epoll or kqueue or whatever is being used. If the taggedPointer
returned by the kernel has a tag that does not match the fd
sequence field, the notification is for a closed file, and we
can ignore it. We check the tag stored in the pollDesc, and we also
check the tag stored in the pollDesc.atomicInfo.
This approach does not work on 32-bit systems where the kernel
only provides a 32-bit field to hold a user value. On those systems
we continue to use the older method without the sequence protection.
This is not ideal, but it is not an issue on Linux because the kernel
provides a 64-bit field, and it is not an issue on Windows because
there are no poller errors on Windows. It is potentially an issue
on *BSD systems, but on those systems we already call fstat in newFile
in os/file_unix.go to avoid adding non-pollable files to kqueue.
So we currently don't know of any cases that will fail.
Fixes#59545
Change-Id: I9a61e20dc39b4266a7a2978fc16446567fe683ac
Reviewed-on: https://go-review.googlesource.com/c/go/+/484837
Auto-Submit: Ian Lance Taylor <iant@golang.org>
Auto-Submit: Ian Lance Taylor <iant@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Orlando Labao <orlando.labao43@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
TryBot-Bypass: Ian Lance Taylor <iant@google.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
TryBot-Bypass: Ian Lance Taylor <iant@golang.org>
In the case where a user program requests overlapped I/O directly on a
handlethat is managed by the runtime, it is possible that
runtime.netpoll will attempt to dereference a pointer with an invalid
value. This CL prevents the runtime from accessing the invalid pointer
value by adding a special key to each overlapped I/O operation that it
creates.
Fixes#58870
Co-authored-by: quimmuntal@gmail.com
Change-Id: Ib58ee757bb5555efba24c29101fc6d1a0dedd61a
Reviewed-on: https://go-review.googlesource.com/c/go/+/482495
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: David Chase <drchase@google.com>
Auto-Submit: Ian Lance Taylor <iant@google.com>
Reviewed-by: Quim Muntal <quimmuntal@gmail.com>
Run-TryBot: Quim Muntal <quimmuntal@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
There's no need for netpollWakeSig to use a uintptr type, a uint32 is enough.
Relevant CL: CL 212737
Change-Id: Ide24478b217a02bad62f7e000a9680c26a8c5366
Reviewed-on: https://go-review.googlesource.com/c/go/+/227798
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
There might be some concurrent (maybe not concurrent, just sequential but in a short time window) and duplicate calls to `netpollBreak`, trying to wake up a net-poller. If one has called `netpollBreak` and that waking event hasn't been received by epollwait/kevent/..., then the subsequent calls of `netpollBreak` ought to be ignored or in other words, these calls should be converged into one.
Benchmarks go1.13.5 darwin/amd64:
benchmark-func time/op (old) time/op (new) delta
BenchmarkNetpollBreak-4 29668ns ±1% 3131ns ±2% -89.45%
mem/B (old) mem/B (new) delta
154B ±13% 0B ±0% -100%
Change-Id: I3cf757a5d6edc5a99adad7aea3baee4b7f2a8f5c
GitHub-Last-Rev: 15bcfbab8a5db51f65da01315a5880a5dbf9e028
GitHub-Pull-Request: golang/go#36294
Reviewed-on: https://go-review.googlesource.com/c/go/+/212737
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
We used to fall back to GetQueuedCompletionStatus if
GetQueuedCompletionStatus was not available, but as of Go 1.11 we
require Windows 7 or later, so GetQueuedCompletionStatusEx is always
available.
Fixes#37957
Change-Id: I7d8d49a92ab7b1f5afdc54a442f696aaf4a5168e
Reviewed-on: https://go-review.googlesource.com/c/go/+/225059
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
The new netpollBreak function can be used to interrupt a blocking netpoll.
This function is not currently used; it will be used by later CLs.
Updates #27707
Change-Id: I5cb936609ba13c3c127ea1368a49194fc58c9f4d
Reviewed-on: https://go-review.googlesource.com/c/go/+/171824
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
This new facility will be used by future CLs in this series.
Change the only blocking call to netpoll to do the right thing when
netpoll returns an empty list.
Updates #6239
Updates #27707
Change-Id: I58b3c2903eda61a3698b1a4729ed0e81382bb1ed
Reviewed-on: https://go-review.googlesource.com/c/go/+/171821
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
netpoll is perhaps one of the most confusing uses of G lists currently
since it passes around many lists as bare *g values right now.
Switching to gList makes it much clearer what's an individual g and
what's a list.
Change-Id: I8d8993c4967c5bae049c7a094aad3a657928ba6c
Reviewed-on: https://go-review.googlesource.com/129397
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
Make netpollopen return what Windows GetLastError API returns.
It is probably copy / paste error from long time ago.
Change-Id: I28f78718c15fef3e8b5f5d11a259533d7e9c6185
Reviewed-on: https://go-review.googlesource.com/72592
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
This changes the os package to use the runtime poller for file I/O
where possible. When a system call blocks on a pollable descriptor,
the goroutine will be blocked on the poller but the thread will be
released to run other goroutines. When using a non-pollable
descriptor, the os package will continue to use thread-blocking system
calls as before.
For example, on GNU/Linux, the runtime poller uses epoll. epoll does
not support ordinary disk files, so they will continue to use blocking
I/O as before. The poller will be used for pipes.
Since this means that the poller is used for many more programs, this
modifies the runtime to only block waiting for the poller if there is
some goroutine that is waiting on the poller. Otherwise, there is no
point, as the poller will never make any goroutine ready. This
preserves the runtime's current simple deadlock detection.
This seems to crash FreeBSD systems, so it is disabled on FreeBSD.
This is issue 19093.
Using the poller on Windows requires opening the file with
FILE_FLAG_OVERLAPPED. We should only do that if we can remove that
flag if the program calls the Fd method. This is issue 19098.
Update #6817.
Update #7903.
Update #15021.
Update #18507.
Update #19093.
Update #19098.
Change-Id: Ia5197dcefa7c6fbcca97d19a6f8621b2abcbb1fe
Reviewed-on: https://go-review.googlesource.com/36800
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
This will make it possible to use the poller with the os package.
This is a lot of code movement but the behavior is intended to be
unchanged.
Update #6817.
Update #7903.
Update #15021.
Update #18507.
Change-Id: I1413685928017c32df5654ded73a2643820977ae
Reviewed-on: https://go-review.googlesource.com/36799
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
This CL revises CL 7504 to use explicitly uintptr types for the
struct fields that are going to be updated sometimes without
write barriers. The result is that the fields are now updated *always*
without write barriers.
This approach has two important properties:
1) Now the GC never looks at the field, so if the missing reference
could cause a problem, it will do so all the time, not just when the
write barrier is missed at just the right moment.
2) Now a write barrier never happens for the field, avoiding the
(correct) detection of inconsistent write barriers when GODEBUG=wbshadow=1.
Change-Id: Iebd3962c727c0046495cc08914a8dc0808460e0e
Reviewed-on: https://go-review.googlesource.com/9019
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Rename "gothrow" to "throw" now that the C version of "throw"
is no longer needed.
This change is purely mechanical except in panic.go where the
old version of "throw" has been deleted.
sed -i "" 's/[[:<:]]gothrow[[:>:]]/throw/g' runtime/*.go
Change-Id: Icf0752299c35958b92870a97111c67bcd9159dc3
Reviewed-on: https://go-review.googlesource.com/2150
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
