Measure the average stack size used by goroutines at every GC. When
starting a new goroutine, allocate an initial goroutine stack of that
average size. Intuition is that we'll waste at most 2x in stack space
because only half the goroutines can be below average. In turn, we
avoid some of the early stack growth / copying needed in the average
case.
More details in the design doc at: https://docs.google.com/document/d/1YDlGIdVTPnmUiTAavlZxBI1d9pwGQgZT7IKFKlIXohQ/edit?usp=sharing
name old time/op new time/op delta
Issue18138 95.3µs ± 0% 67.3µs ±13% -29.35% (p=0.000 n=9+10)
Fixes#18138
Change-Id: Iba34d22ed04279da7e718bbd569bbf2734922eaa
Reviewed-on: https://go-review.googlesource.com/c/go/+/345889
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
This gives explicit names to the possible states of throwing (-1, 0, 1).
m.throwing is now one of:
throwTypeOff: not throwing, previously == 0
throwTypeUser: user throw, previously == -1
throwTypeRuntime: runtime throw, previously == 1
For runtime throws, we now always include frame metadata and system
goroutines regardless of GOTRACEBACK to aid in debugging the runtime.
For user throws, we no longer include frame metadata or runtime frames,
unless GOTRACEBACK=system or higher.
For #51485.
Change-Id: If252e2377a0b6385ce7756b937929be4273a56c0
Reviewed-on: https://go-review.googlesource.com/c/go/+/390421
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Austin Clements <austin@google.com>
https://go-review.googlesource.com/c/go/+/383434 started using
atomic Load64 on this field, which breaks 32 bit platforms which
require 64-bit alignment of uint64s that are passed to atomic operations.
Not sure why this doesn't break everywhere, but I saw it break on
my laptop during all.bash.
Change-Id: Ida27b23068b3cc7208fce3c97b69a464ccf68209
Reviewed-on: https://go-review.googlesource.com/c/go/+/399754
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
A future change to gofmt will rewrite
// Doc comment.
//go:foo
to
// Doc comment.
//
//go:foo
Apply that change preemptively to all comments (not necessarily just doc comments).
For #51082.
Change-Id: Iffe0285418d1e79d34526af3520b415a12203ca9
Reviewed-on: https://go-review.googlesource.com/c/go/+/384260
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
A run of lines that are indented with any number of spaces or tabs
format as a <pre> block. This commit fixes various doc comments
that format badly according to that (standard) rule.
For example, consider:
// - List item.
// Second line.
// - Another item.
Because the - lines are unindented, this is actually two paragraphs
separated by a one-line <pre> block. This CL rewrites it to:
// - List item.
// Second line.
// - Another item.
Today, that will format as a single <pre> block.
In a future release, we hope to format it as a bulleted list.
Various other minor fixes as well, all in preparation for reformatting.
For #51082.
Change-Id: I95cf06040d4186830e571cd50148be3bf8daf189
Reviewed-on: https://go-review.googlesource.com/c/go/+/384257
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
In issue 50113, we see that a thread blocked in a system call can result
in a hang of AllThreadsSyscall. To resolve this, we must send a signal
to these threads to knock them out of the system call long enough to run
the per-thread syscall.
Stepping back, if we need to send signals anyway, it should be possible
to implement this entire mechanism on top of signals. This CL does so,
vastly simplifying the mechanism, both as a direct result of
newly-unnecessary code as well as some ancillary simplifications to make
things simpler to follow.
Major changes:
* The rest of the mechanism is moved to os_linux.go, with fields in mOS
instead of m itself.
* 'Fixup' fields and functions are renamed to 'perThreadSyscall' so they
are more precise about their purpose.
* Rather than getting passed a closure, doAllThreadsSyscall takes the
syscall number and arguments. This avoids a lot of hairy behavior:
* The closure may potentially only be live in fields in the M,
hidden from the GC. Not necessary with no closure.
* The need to loan out the race context. A direct RawSyscall6 call
does not require any race context.
* The closure previously conditionally panicked in strange
locations, like a signal handler. Now we simply throw.
* All manual fixup synchronization with mPark, sysmon, templateThread,
sigqueue, etc is gone. The core approach is much simpler:
doAllThreadsSyscall sends a signal to every thread in allm, which
executes the system call from the signal handler. We use (SIGRTMIN +
1), aka SIGSETXID, the same signal used by glibc for this purpose. As
such, we are careful to only handle this signal on non-cgo binaries.
Synchronization with thread creation is a key part of this CL. The
comment near the top of doAllThreadsSyscall describes the required
synchronization semantics and how they are achieved.
Note that current use of allocmLock protects the state mutations of allm
that are also protected by sched.lock. allocmLock is used instead of
sched.lock simply to avoid holding sched.lock for so long.
Fixes#50113
Change-Id: Ic7ea856dc66cf711731540a54996e08fc986ce84
Reviewed-on: https://go-review.googlesource.com/c/go/+/383434
Reviewed-by: Austin Clements <austin@google.com>
Trust: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
CL 339669 changed freedefer to not mention every field of _defer,
so no need to call it out in the _defer comment.
Change-Id: Id8b67ba2298685f609bf901b5948fd30666bd6e3
Reviewed-on: https://go-review.googlesource.com/c/go/+/382251
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
The netpoll code was written long ago, when the
only multiprocessors that Go ran on were x86.
It assumed that an atomic store would trigger a
full memory barrier and then used that barrier
to order otherwise racy access to a handful of fields,
including pollDesc.closing.
On ARM64, this code has finally failed, because
the atomic store is on a value completely unrelated
to any of the racily-accessed fields, and the ARMv8
hardware, unlike x86, is clever enough not to do a
full memory barrier for a simple atomic store.
We are seeing a constant background rate of trybot
failures where the net/http tests deadlock - a netpollblock
has clearly happened after the pollDesc has begun to close.
The code that does the racy reads is netpollcheckerr,
which needs to be able to run without acquiring a lock.
This CL fixes the race, without introducing unnecessary
inefficiency or deadlock, by arranging for every updater
of the relevant fields to publish a summary as a single
atomic uint32, and then having netpollcheckerr use a
single atomic load to fetch the relevant bits and then
proceed as before.
Fixes#45211 (until proven otherwise!).
Change-Id: Ib6788c8da4d00b7bda84d55ca3fdffb5a64c1a0a
Reviewed-on: https://go-review.googlesource.com/c/go/+/378234
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
Trust: Bryan Mills <bcmills@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
And then revert the bootstrap cmd directories and certain testdata.
And adjust tests as needed.
Not reverting the changes in std that are bootstrapped,
because some of those changes would appear in API docs,
and we want to use any consistently.
Instead, rewrite 'any' to 'interface{}' in cmd/dist for those directories
when preparing the bootstrap copy.
A few files changed as a result of running gofmt -w
not because of interface{} -> any but because they
hadn't been updated for the new //go:build lines.
Fixes#49884.
Change-Id: Ie8045cba995f65bd79c694ec77a1b3d1fe01bb09
Reviewed-on: https://go-review.googlesource.com/c/go/+/368254
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
This change adds two fields to gcControllerState: stackScan, used for
pacing decisions, and scannableStackSize, which directly tracks the
amount of space allocated for inuse stacks that will be scanned.
scannableStackSize is not updated directly, but is instead flushed from
each P when at an least 8 KiB delta has accumulated. This helps reduce
issues with atomics contention for newly created goroutines. Stack
growth paths are largely unaffected.
StackGrowth-48 51.4ns ± 0% 51.4ns ± 0% ~ (p=0.927 n=10+10)
StackGrowthDeep-48 6.14µs ± 3% 6.25µs ± 4% ~ (p=0.090 n=10+9)
CreateGoroutines-48 273ns ± 1% 273ns ± 1% ~ (p=0.676 n=9+10)
CreateGoroutinesParallel-48 65.5ns ± 5% 66.6ns ± 7% ~ (p=0.340 n=9+9)
CreateGoroutinesCapture-48 2.06µs ± 1% 2.07µs ± 4% ~ (p=0.217 n=10+10)
CreateGoroutinesSingle-48 550ns ± 3% 563ns ± 4% +2.41% (p=0.034 n=8+10)
For #44167.
Change-Id: Id1800d41d3a6c211b43aeb5681c57c0dc8880daf
Reviewed-on: https://go-review.googlesource.com/c/go/+/309589
Trust: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
The first field of the func data stored by the linker is the
entry PC for the function. Prior to this change, this was stored
as a relocation to the function. Change this to be an offset
relative to runtime.text.
This reduces the number of relocations on darwin/arm64 by about 10%.
It also slightly shrinks binaries:
file before after Δ %
addr2line 3803058 3791298 -11760 -0.309%
api 5140114 5104242 -35872 -0.698%
asm 4886850 4840626 -46224 -0.946%
buildid 2512466 2503042 -9424 -0.375%
cgo 4374770 4342274 -32496 -0.743%
compile 22920530 22769202 -151328 -0.660%
cover 4624626 4588242 -36384 -0.787%
dist 3217570 3205522 -12048 -0.374%
doc 3715026 3684498 -30528 -0.822%
fix 3148226 3119266 -28960 -0.920%
link 6350226 6313362 -36864 -0.581%
nm 3768850 3757106 -11744 -0.312%
objdump 4140594 4127618 -12976 -0.313%
pack 2227474 2218818 -8656 -0.389%
pprof 13598706 13506786 -91920 -0.676%
test2json 2497234 2487426 -9808 -0.393%
trace 10198066 10118498 -79568 -0.780%
vet 6930658 6889074 -41584 -0.600%
total 108055044 107366900 -688144 -0.637%
It should also incrementally speed up binary launching.
This is the first step towards removing enough relocations
that pages that were previously dirtied by the loader may remain clean,
which will offer memory savings useful in constrained environments.
Change-Id: Icfba55e696ba2f9c99c4f179125ba5a3ba4369c9
Reviewed-on: https://go-review.googlesource.com/c/go/+/351463
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
A subsequent change will alter the semantics of _func.entry.
To make that change obvious and clear, change _func.entry to a method,
and rename the field to _func.entryPC.
Change-Id: I05d66b54d06c5956d4537b0729ddf4290c3e2635
Reviewed-on: https://go-review.googlesource.com/c/go/+/351460
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
_func and funcinl are type-punned.
We distinguish them at runtime by inspecting the first word.
Prior to this change, we used 0 as the sentinel value
that means that a Func is a funcinl.
That worked because _func's first word is the functions' entry PC,
and 0 is not a valid PC. I plan to make *_func's entry PC relative
to the containing module. As a result, 0 will be a valid value,
for the first function in the module.
Switch to ^0 as the new sentinel value, which is neither a valid
entry PC nor a valid PC offset.
Change-Id: I4c718523a083ed6edd57767c3548640681993522
Reviewed-on: https://go-review.googlesource.com/c/go/+/351459
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
To measure all instructions having been completed before reading
the time stamp counter with RDTSC an instruction sequence that
has instruction stream serializing properties which guarantee
waiting until all previous instructions have been executed is
needed. This does not necessary mean to wait for all stores to
be globally visible.
This CL aims to remove vendor specific logic for determining the
instruction sequence with CPU feature flag checks that are
CPU vendor independent.
For intel LFENCE has the wanted properties at least
since it was introduced together with SSE2 support.
On AMD instruction stream serializing LFENCE is supported by setting
an MSR C001_1029[1]=1 on AMD family 10h/12h/14h/15h/16h/17h processors.
AMD family 0Fh/11h processors support LFENCE as serializing always.
AMD plans support for this MSR and access to this bit for all future processors.
Source: https://developer.amd.com/wp-content/resources/Managing-Speculation-on-AMD-Processors.pdf
Reading the MSR to determine LFENCE properties is not always possible
or reliable (hypervisors). The Linux kernel is relying on serializing
LFENCE on AMD CPUs since a commit in July 2019: https://lkml.org/lkml/2019/7/22/295
and the MSR C001_1029 to enable serialization has been set by default
with the Spectre v1 mitigations.
Using an MFENCE on AMD is waiting on previous instructions having been executed
but in addition also flushes store buffers.
To align the serialization properties without runtime detection
of CPU manufacturers we can use the newer RDTSCP instruction which
waits until all previous instructions have been executed.
RDTSCP is available on Intel since around 2008 and on AMD CPUs since
around 2006. Support for RDTSCP can be checked independently
of manufacturer by checking CPUID bits.
Using RDTSCP is the default in Linux to read TSC in program order
when the instruction is available.
e22ce8eb63/arch/x86/include/asm/msr.h (L231)
Change-Id: Ifa841843b9abb2816f8f0754a163ebf01385306d
Reviewed-on: https://go-review.googlesource.com/c/go/+/344429
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Martin Möhrmann <martin@golang.org>
Run-TryBot: Martin Möhrmann <martin@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Conflicts:
- src/go/types/check_test.go
CL 324730 on dev.typeparams changed the directory paths in TestCheck,
TestExamples, and TestFixedbugs and renamed checkFiles to testFiles;
whereas CL 337529 on master added a new test case just above them and
that used checkFiles.
Merge List:
+ 2021-08-12 46fd547d89 internal/goversion: update Version to 1.18
+ 2021-08-12 5805efc78e doc/go1.17: remove draft notice
+ 2021-08-12 39634e7dae CONTRIBUTORS: update for the Go 1.17 release
+ 2021-08-12 095bb790e1 os/exec: re-enable LookPathTest/16
+ 2021-08-11 dea23e9ca8 src/make.*: make --no-clean flag a no-op that prints a warning
+ 2021-08-11 d4c0ed26ac doc/go1.17: linker passes -I to extld as -Wl,--dynamic-linker
+ 2021-08-10 1f9c9d8530 doc: use "high address/low address" instead of "top/bottom"
+ 2021-08-09 f1dce319ff cmd/go: with -mod=vendor, don't panic if there are duplicate requirements
+ 2021-08-09 7aeaad5c86 runtime/cgo: when using msan explicitly unpoison cgoCallers
+ 2021-08-08 507cc341ec doc: add example for conversion from slice expressions to array ptr
+ 2021-08-07 891547e2d4 doc/go1.17: fix a typo introduced in CL 335135
+ 2021-08-06 8eaf4d16bc make.bash: do not overwrite GO_LDSO if already set
+ 2021-08-06 63b968f4f8 doc/go1.17: clarify Modules changes
+ 2021-08-06 70546f6404 runtime: allow arm64 SEH to be called if illegal instruction
+ 2021-08-05 fd45e267c2 runtime: warn that KeepAlive is not an unsafe.Pointer workaround
+ 2021-08-04 6e738868a7 net/http: speed up and deflake TestCancelRequestWhenSharingConnection
+ 2021-08-02 8a7ee4c51e io/fs: don't use absolute path in DirEntry.Name doc
+ 2021-07-31 b8ca6e59ed all: gofmt
+ 2021-07-30 b7a85e0003 net/http/httputil: close incoming ReverseProxy request body
+ 2021-07-29 70fd4e47d7 runtime: avoid possible preemption when returning from Go to C
+ 2021-07-28 9eee0ed439 cmd/go: fix go.mod file name printed in error messages for replacements
+ 2021-07-28 b39e0f461c runtime: don't crash on nil pointers in checkptrAlignment
+ 2021-07-27 7cd10c1149 cmd/go: use .mod instead of .zip to determine if version has go.mod file
+ 2021-07-27 c8cf0f74e4 cmd/go: add missing flag in UsageLine
+ 2021-07-27 7ba8e796c9 testing: clarify T.Name returns a distinct name of the running test
+ 2021-07-27 33ff155970 go/types: preserve untyped constants on the RHS of a shift expression
+ 2021-07-26 840e583ff3 runtime: correct variable name in comment
+ 2021-07-26 bfbb288574 runtime: remove adjustTimers counter
+ 2021-07-26 9c81fd53b3 cmd/vet: add missing copyright header
Change-Id: Ia80604d24c6f4205265683024e3100769cf32065
CL 339396 allowed stack copying on entry to and during freedefer, but
this introduced a subtle bug: if d is heap-allocated, and d.link
points to a stack-allocated defer, stack copying during freedefer can
briefly introduce a stale pointer, which the garbage collector can
discover and panic about. This happens because d has already been
unlinked from the defer chain when freedefer is called, so stack
copying won't update stack pointers in it.
Fix this by making freedefer nosplit again and immediately clearing
d.link.
This should fix the longtest builders, which currently fail on
GOMAXPROCS=2 runtime -cpu=1,2,4 -quick in the TestDeferHeapAndStack
test.
This seems like the simplest fix, but it just deals with the subtlety
rather than eliminating it. Really, every call site of freedefer (of
which there are surprisingly many) has hidden subtlety between
unlinking the defer and calling freedefer. We could consolidate the
subtlety into each call site by requiring that they unlink the defer
and set d.link to nil before calling freedefer. freedefer could check
this condition like it checks that various other fields have already
been zeroed. A more radical option is to replace freedefer with
"popDefer", which would both pop the defer off the link and take care
of freeing it. There would still be a brief moment of subtlety, but it
would be in one place, in popDefer. Annoyingly, *almost* every call to
freedefer just pops the defer from the head of the G's list, but
there's one place when handling open-coded defers where we have to
remove a defer from the middle of the list. I'm inclined to first fix
that subtlety by only expanding open-coded defer records when they're
at the head of the defer list, and then revisit the popDefer idea.
Change-Id: I3130d2542c01a421a5d60e8c31f5379263219627
Reviewed-on: https://go-review.googlesource.com/c/go/+/339730
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Prior to regabi, a deferred function could have any signature, so the
runtime always manipulated them as funcvals. Now, a deferred function
is always func(). Hence, this CL makes the runtime's manipulation of
deferred functions more type-safe by using func() directly instead of
*funcval.
Change-Id: Ib55f38ed49107f74149725c65044e4690761971d
Reviewed-on: https://go-review.googlesource.com/c/go/+/337650
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
In CL 336432 we changed adjusttimers so that it no longer cleared
timerModifiedEarliest if there were no timersModifiedEarlier timers.
This caused some Google internal tests to time out, presumably due
to the increased contention on timersLock. We can avoid that by
simply not skipping the loop in adjusttimers, which lets us safely
clear timerModifiedEarliest. And if we don't skip the loop, then there
isn't much reason to keep the count of timerModifiedEarlier timers at all.
So remove it.
The effect will be that for programs that create some timerModifiedEarlier
timers and then remove them all, the program will do an occasional
additional loop over all the timers. And, programs that have some
timerModifiedEarlier timers will always loop over all the timers,
without the quicker exit when they have all been seen. But the loops
should not occur all that often, due to timerModifiedEarliest.
For #47329
Change-Id: I7b244c1244d97b169a3c7fbc8f8a8b115731ddee
Reviewed-on: https://go-review.googlesource.com/c/go/+/337309
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
This avoids a race when a new timerModifiedEarlier timer is created by
a different goroutine.
Fixes#47329
Change-Id: I6f6c87b4a9b5491b201c725c10bc98e23e0ed9d1
Reviewed-on: https://go-review.googlesource.com/c/go/+/336432
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Conflicts:
- src/runtime/runtime2.go
On master, CL 317191 fixed the mentions of gc/reflect.go in comments
to reflectdata/reflect.go; but on dev.typeparams, CL 325921 fixed
that the same comment to reflect that deferstruct actually ended up
in ssagen/ssa.go.
Merge List:
+ 2021-07-08 296ddf2a93 net: filter bad names from Lookup functions instead of hard failing
+ 2021-07-08 ce76298ee7 Update oudated comment
+ 2021-07-08 2ca44fe221 doc/go1.17: linkify time.UnixMilli and time.UnixMicro
+ 2021-07-07 5c59e11f5e cmd/compile: remove special-casing of blank in types.sconv{,2}
+ 2021-07-07 b003a8b1ae cmd/compile: optimize types.sconv
+ 2021-07-07 11f5df2d67 cmd/compile: extract pkgqual from symfmt
+ 2021-07-07 991fd381d5 cmd/go: don't lock .mod and .sum files for read in overlay
+ 2021-07-07 186a3bb4b0 cmd/go/internal/modfetch/codehost: skip hg tests if no hg binary is present
+ 2021-07-07 00c00558e1 cmd/go/internal/modload: remove unused functions
+ 2021-07-07 f264879f74 cmd/go/internal/modload: fix an apparent typo in the AutoRoot comment
+ 2021-07-07 c96833e5ba doc: remove stale comment about arm64 port
Change-Id: I849046b6d8f7421f60323549f3f763ef418bf9e7
Update comment cause gc/select.go has been moved to walk/select.go and gc/reflect.go has been moved to reflectdata/reflect.go
Change-Id: I6894527e1e9dbca50ace92a51bf942f9495ce88c
GitHub-Last-Rev: 6d6a447144
GitHub-Pull-Request: golang/go#45976
Reviewed-on: https://go-review.googlesource.com/c/go/+/317191
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Michael Pratt <mpratt@google.com>
Temprary revert CL 325918.
Delve relies on the _defer.fn.fn field to get defer frames.
CL 325918 changes the type of _defer.fn to func(), which no
longer has an fn field.
Change-Id: If6c71b15a27bac579593f5273c9a49715e6e35b2
Reviewed-on: https://go-review.googlesource.com/c/go/+/327775
Trust: Cherry Mui <cherryyz@google.com>
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Now that deferred functions are always argumentless and defer
records are no longer with arguments, defer record can be fixed
size (just the _defer struct). This allows us to simplify the
allocation of defer records, specifically, remove the defer
classes and the pools of different sized defers.
Change-Id: Icc4b16afc23b38262ca9dd1f7369ad40874cf701
Reviewed-on: https://go-review.googlesource.com/c/go/+/326062
Trust: Cherry Mui <cherryyz@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
As deferred function now always has zero arguments, _defer.siz is
always 0 and can be removed.
Change-Id: Ibb89f65b2f9d2ba4aeabe50438cc3d4b6a88320b
Reviewed-on: https://go-review.googlesource.com/c/go/+/325921
Trust: Cherry Mui <cherryyz@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Previously it takes a *funcval, as it can be any function types.
Now it must be func(). Make it so.
Change-Id: I04273047b024386f55dbbd5fbda4767cbee7ac93
Reviewed-on: https://go-review.googlesource.com/c/go/+/325918
Trust: Cherry Mui <cherryyz@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Revive CL 310149
Change-Id: Ib4714ea5b2ade32c0f66edff841a79d8212bd79a
Reviewed-on: https://go-review.googlesource.com/c/go/+/313009
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
Trust: Michael Pratt <mpratt@google.com>
Trust: Michael Knyszek <mknyszek@google.com>
This change adds a metric to track scheduling latencies, defined as the
cumulative amount of time a goroutine spends being runnable before
running again. The metric is an approximations and samples instead of
trying to record every goroutine scheduling latency.
This change was primarily authored by mknyszek@google.com.
Change-Id: Ie0be7e6e7be421572eb2317d3dd8dd6f3d6aa152
Reviewed-on: https://go-review.googlesource.com/c/go/+/308933
Trust: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
The runtime support for syscall.AllThreadsSyscall() functions had
some corner case deadlock issues when signal handling was in use.
This was observed in at least 3 build test failures on ppc64 and
amd64 architecture CGO_ENABLED=0 builds over the last few months.
The fixes involve more controlled handling of signals while the
AllThreads mechanism is being executed. Further details are
discussed in bug #44193.
The all-threads syscall support is new in go1.16, so earlier
releases are not affected by this bug.
Fixes#44193
Change-Id: I01ba8508a6e1bb2d872751f50da86dd07911a41d
Reviewed-on: https://go-review.googlesource.com/c/go/+/305149
Reviewed-by: Michael Pratt <mpratt@google.com>
Trust: Michael Pratt <mpratt@google.com>
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Change-Id: Ida44a2e07f277bee8806538ecee4beee3474cf3d
Reviewed-on: https://go-review.googlesource.com/c/go/+/310149
Reviewed-by: Michael Pratt <mpratt@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Trust: Michael Pratt <mpratt@google.com>
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
The debug call tests currently assume that the target Go function is
ABI0; this is clearly no longer true when we switch to the new ABI, so
make the tests set up argument register state in the debug call handler
and copy back results returned in registers.
A small snag in calling a Go function that follows the new ABI is that
the debug call protocol depends on the AX register being set to a
specific value as it bounces in and out of the handler, but this
register is part of the new register ABI, so results end up being
clobbered. Use R12 instead.
Next, the new desugaring behavior for "go" statements means that
newosproc1 must always call a function with no frame; if it takes any
arguments, it closes over them and they're passed in the context
register. Currently when debugCallWrap creates a new goroutine, it uses
newosproc1 directly and passes a non-zero-sized frame, so that needs to
be updated. To fix this, briefly use the g's param field which is
otherwise only used for channels to pass an explicitly allocated object
containing the "closed over" variables. While we could manually do the
desugaring ourselves (we cannot do so automatically because the Go
compiler prevents heap-allocated closures in the runtime), that bakes in
more ABI details in a place that really doesn't need to care about them.
Finally, there's an old bug here where the context register was set up
in CX, so technically closure calls never worked. Oops. It was otherwise
harmless for other types of calls before, but now CX is an argument
register, so now that interferes with regular calls, too.
For #40724.
Change-Id: I652c25ed56a25741bb04c24cfb603063c099edde
Reviewed-on: https://go-review.googlesource.com/c/go/+/309169
Trust: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Alessandro Arzilli <alessandro.arzilli@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
On Windows, TLS is uninitialized for C threads calling into Go code.
In this path, before calling into user Go code, we call into needm which
runs without an m, but whose purpose is to pick one up. While in Go
code, we may occasionally restore the G register from TLS for a number
of reasons. Rather than try to flag all these cases, given that needm
(and its callees) are already somewhat special, just set up a dummy TLS
space for it that's read-only. If it ever actually tries to write to
this space (it shouldn't), it will fail loudly. Otherwise, code that
restores the G register will simply load a zero value, but that's OK
since needm is careful never to require the G at any point, because it
doesn't yet have a valid G. Furthermore, by the time needm returns, it
will have set up TLS properly for a Windows C thread, so there's no need
to do anything extra afterwards.
For #40724.
Change-Id: I34e8095059817e4ee663505e89cda8785b634b98
Reviewed-on: https://go-review.googlesource.com/c/go/+/307872
Trust: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This moves all remaining GOEXPERIMENT flags into the objabi.Experiment
struct, drops the "_enabled" from their name, and makes them all bool
typed.
We also drop DebugFlags.Fieldtrack because the previous CL shifted the
one test that used it to use GOEXPERIMENT instead.
Change-Id: I3406fe62b1c300bb4caeaffa6ca5ce56a70497fe
Reviewed-on: https://go-review.googlesource.com/c/go/+/302389
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Since allocation for p struct will be rounded up to the next size class,
the two relevant adjacent classes for this case are 9728 bytes and 10240 bytes.
A p is currently 10072 bytes, so it gets rounded up to 10240 bytes when we allocate one,
So the pad in p struct is unnecessary, eliminate it and add comments for
warning the false sharing.
Change-Id: Iae8b32931d1beddbfff1f58044d8401703da6407
Reviewed-on: https://go-review.googlesource.com/c/go/+/268759
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Ian Lance Taylor <iant@golang.org>
Frame pointers were already enabled on linux, darwin, ios,
but not freebsd, android, openbsd, netbsd.
But the space was reserved on all platforms, leading to
two different arm64 framepointer conditions in different
parts of the code, one of which had no name
(framepointer_enabled || GOARCH == "arm64",
which might have been "framepointer_space_reserved").
So on the disabled systems, the stack layouts were still
set up for frame pointers and the only difference was not
actually maintaining the FP register in the generated code.
Reduce complexity by just enabling the frame pointer
completely on all the arm64 systems.
This commit passes on freebsd, android, netbsd.
I have not been able to try it on openbsd.
This CL is part of a stack adding windows/arm64
support (#36439), intended to land in the Go 1.17 cycle.
This CL is, however, not windows/arm64-specific.
It is cleanup meant to make the port (and future ports) easier.
Change-Id: I83bd23369d24b76db4c6a648fa74f6917819a093
Reviewed-on: https://go-review.googlesource.com/c/go/+/288814
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
The runtime traceback code has its own definition of which functions
mark the top frame of a stack, separate from the TOPFRAME bits that
exist in the assembly and are passed along in DWARF information.
It's error-prone and redundant to have two different sources of truth.
This CL provides the actual TOPFRAME bits to the runtime, so that
the runtime can use those bits instead of reinventing its own category.
This CL also adds a new bit, SPWRITE, which marks functions that
write directly to SP (anything but adding and subtracting constants).
Such functions must stop a traceback, because the traceback has no
way to rederive the SP on entry. Again, the runtime has its own definition
which is mostly correct, but also missing some functions. During ordinary
goroutine context switches, such functions do not appear on the stack,
so the incompleteness in the runtime usually doesn't matter.
But profiling signals can arrive at any moment, and the runtime may
crash during traceback if it attempts to unwind an SP-writing frame
and gets out-of-sync with the actual stack. The runtime contains code
to try to detect likely candidates but again it is incomplete.
Deriving the SPWRITE bit automatically from the actual assembly code
provides the complete truth, and passing it to the runtime lets the
runtime use it.
This CL is part of a stack adding windows/arm64
support (#36439), intended to land in the Go 1.17 cycle.
This CL is, however, not windows/arm64-specific.
It is cleanup meant to make the port (and future ports) easier.
Change-Id: I227f53b23ac5b3dabfcc5e8ee3f00df4e113cf58
Reviewed-on: https://go-review.googlesource.com/c/go/+/288800
Trust: Russ Cox <rsc@golang.org>
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Jason A. Donenfeld <Jason@zx2c4.com>
Document what the values in internal/sys mean.
Remove various special cases for arm64 in the code using StackAlign.
Delete Uintreg - it was for GOARCH=amd64p32,
which was specific to GOOS=nacl and has been retired.
This CL is part of a stack adding windows/arm64
support (#36439), intended to land in the Go 1.17 cycle.
This CL is, however, not windows/arm64-specific.
It is cleanup meant to make the port (and future ports) easier.
Change-Id: I40e8fa07b4e192298b6536b98a72a751951a4383
Reviewed-on: https://go-review.googlesource.com/c/go/+/288795
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
This merge had two conflicts to resolve:
1. The embed code on master had somewhat substantially diverged, so
this CL tediously backported the changes to dev.regabi. In particular,
I went through all of the embed changes to gc/{embed,noder,syntax}.go
and made sure the analogous code on dev.regabi in noder/noder.go and
staticdata/embed.go mirrors it.
2. The init-cycle reporting code on master was extended slightly to
track already visited declarations to avoid exponential behavior. The
same fix is applied on dev.regabi, just using ir.NameSet instead of
map[ir.Node]bool.
Conflicts:
- src/cmd/compile/internal/gc/embed.go
- src/cmd/compile/internal/gc/noder.go
- src/cmd/compile/internal/gc/syntax.go
- src/cmd/compile/internal/pkginit/initorder.go
- src/embed/internal/embedtest/embed_test.go
- src/go/types/stdlib_test.go
Merge List:
+ 2021-01-22 dab3e5affe runtime: switch runtime to libc for openbsd/amd64
+ 2021-01-22 a1b53d85da cmd/go: add documentation for test and xtest fields output by go list
+ 2021-01-22 b268b60774 runtime: remove pthread_kill/pthread_self for openbsd
+ 2021-01-22 ec4051763d runtime: fix typo in mgcscavenge.go
+ 2021-01-22 7ece3a7b17 net/http: fix flaky TestDisableKeepAliveUpgrade
+ 2021-01-22 50cba0506f time: clarify Timer.Reset behavior on AfterFunc Timers
+ 2021-01-22 cf10e69f17 doc/go1.16: mention net/http.Transport.GetProxyConnectHeader
+ 2021-01-22 ec1b945265 doc/go1.16: mention path/filepath.WalkDir
+ 2021-01-22 11def3d40b doc/go1.16: mention syscall.AllThreadsSyscall
+ 2021-01-21 07b0235609 doc/go1.16: add notes about package-specific fs.FS changes
+ 2021-01-21 e2b4f1fea5 doc/go1.16: minor formatting fix
+ 2021-01-21 9f43a9e07b doc/go1.16: mention new debug/elf constants
+ 2021-01-21 3c2f11ba5b cmd/go: overwrite program name with full path
+ 2021-01-21 953d1feca9 all: introduce and use internal/execabs
+ 2021-01-21 b186e4d70d cmd/go: add test case for cgo CC setting
+ 2021-01-21 5a8a2265fb cmd/cgo: report exec errors a bit more clearly
+ 2021-01-21 46e2e2e9d9 cmd/go: pass resolved CC, GCCGO to cgo
+ 2021-01-21 3d40895e36 runtime: switch openbsd/arm64 to pthreads
+ 2021-01-21 d95ca91380 crypto/elliptic: fix P-224 field reduction
+ 2021-01-20 ecf4ebf100 cmd/internal/moddeps: check content of all modules in GOROOT
+ 2021-01-20 d2d155d1ae runtime: don't adjust timer pp field in timerWaiting status
+ 2021-01-20 803d18fc6c cmd/go: set Incomplete field on go list output if no files match embed
+ 2021-01-20 6e243ce71d cmd/go: have go mod vendor copy embedded files in subdirs
+ 2021-01-20 be28e5abc5 cmd/go: fix mod_get_fallback test
+ 2021-01-20 928bda4f4a runtime: convert openbsd/amd64 locking to libc
+ 2021-01-19 824f2d635c cmd/go: allow go fmt to complete when embedded file is missing
+ 2021-01-19 0575e35e50 cmd/compile: require 'go 1.16' go.mod line for //go:embed
+ 2021-01-19 ccb2e90688 cmd/link: exit before Asmb2 if error
+ 2021-01-19 ca5774a5a5 embed: treat uninitialized FS as empty
+ 2021-01-19 d047c91a6c cmd/link,runtime: switch openbsd/amd64 to pthreads
+ 2021-01-19 61debffd97 runtime: factor out usesLibcall
+ 2021-01-19 9fed39d281 runtime: factor out mStackIsSystemAllocated
+ 2021-01-18 dbab079835 runtime: free Windows event handles after last lock is dropped
+ 2021-01-18 5a8fbb0d2d os: do not close syscall.Stdin in TestReadStdin
+ 2021-01-15 682a1d2176 runtime: detect errors in DuplicateHandle
+ 2021-01-15 9f83418b83 cmd/link: remove GOROOT write in TestBuildForTvOS
+ 2021-01-15 ec9470162f cmd/compile: allow embed into any string or byte slice type
+ 2021-01-15 54198b04db cmd/compile: disallow embed of var inside func
+ 2021-01-15 b386c735e7 cmd/go: fix go generate docs
+ 2021-01-15 bb5075a525 syscall: remove RtlGenRandom and move it into internal/syscall
+ 2021-01-15 1deae0b597 os: invoke processKiller synchronously in testKillProcess
+ 2021-01-15 ff196c3e84 crypto/x509: update iOS bundled roots to version 55188.40.9
+ 2021-01-14 e125ccd10e cmd/go: in 'go mod edit', validate versions given to -retract and -exclude
+ 2021-01-14 eb330020dc cmd/dist, cmd/go: pass -arch for C compilation on Darwin
+ 2021-01-14 84e8a06f62 cmd/cgo: remove unnecessary space in cgo export header
+ 2021-01-14 0c86b999c3 cmd/test2json: document passing -test.paniconexit0
+ 2021-01-14 9135795891 cmd/go/internal/load: report positions for embed errors
+ 2021-01-14 d9b79e53bb cmd/compile: fix wrong complement for arm64 floating-point comparisons
+ 2021-01-14 c73232d08f cmd/go/internal/load: refactor setErrorPos to PackageError.setPos
+ 2021-01-14 6aa28d3e06 go/build: report positions for go:embed directives
+ 2021-01-13 7eb31d999c cmd/go: add hints to more missing sum error messages
+ 2021-01-12 ba76567bc2 cmd/go/internal/modload: delete unused *mvsReqs.next method
+ 2021-01-12 665def2c11 encoding/asn1: document unmarshaling behavior for IMPLICIT string fields
+ 2021-01-11 81ea89adf3 cmd/go: fix non-script staleness checks interacting badly with GOFLAGS
+ 2021-01-11 759309029f doc: update editors.html for Go 1.16
+ 2021-01-11 c3b4c7093a cmd/internal/objfile: don't require runtime.symtab symbol for XCOFF
+ 2021-01-08 59bfc18e34 cmd/go: add hint to read 'go help vcs' to GOVCS errors
+ 2021-01-08 cd6f3a54e4 cmd/go: revise 'go help' documentation for modules
+ 2021-01-08 6192b98751 cmd/go: make hints in error messages more consistent
+ 2021-01-08 25886cf4bd cmd/go: preserve sums for indirect deps fetched by 'go mod download'
+ 2021-01-08 6250833911 runtime/metrics: mark histogram metrics as cumulative
+ 2021-01-08 8f6a9acbb3 runtime/metrics: remove unused StopTheWorld Description field
+ 2021-01-08 6598c65646 cmd/compile: fix exponential-time init-cycle reporting
+ 2021-01-08 fefad1dc85 test: fix timeout code for invoking compiler
+ 2021-01-08 6728118e0a cmd/go: pass signals forward during "go tool"
+ 2021-01-08 e65c543f3c go/build/constraint: add parser for build tag constraint expressions
+ 2021-01-08 0c5afc4fb7 testing/fstest,os: clarify racy behavior of TestFS
+ 2021-01-08 32afcc9436 runtime/metrics: change unit on *-by-size metrics to match bucket unit
+ 2021-01-08 c6513bca5a io/fs: minor corrections to Glob doc
+ 2021-01-08 304f769ffc cmd/compile: don't short-circuit copies whose source is volatile
+ 2021-01-08 ae97717133 runtime,runtime/metrics: use explicit histogram boundaries
+ 2021-01-08 a9ccd2d795 go/build: skip string literal while findEmbed
+ 2021-01-08 d92f8add32 archive/tar: fix typo in comment
+ 2021-01-08 cab1202183 cmd/link: accept extra blocks in TestFallocate
+ 2021-01-08 ee4d32249b io/fs: minor corrections to Glob release date
+ 2021-01-08 54bd1ccce2 cmd: update to latest golang.org/x/tools
+ 2021-01-07 9ec21a8f34 Revert "reflect: support multiple keys in struct tags"
+ 2021-01-07 091414b5b7 io/fs: correct WalkDirFunc documentation
+ 2021-01-07 9b55088d6b doc/go1.16: add release note for disallowing non-ASCII import paths
+ 2021-01-07 fa90aaca7d cmd/compile: fix late expand_calls leaf type for OpStructSelect/OpArraySelect
+ 2021-01-07 7cee66d4cb cmd/go: add documentation for Embed fields in go list output
+ 2021-01-07 e60cffa4ca html/template: attach functions to namespace
+ 2021-01-07 6da2d3b7d7 cmd/link: fix typo in asm.go
+ 2021-01-07 df81a15819 runtime: check mips64 VDSO clock_gettime return code
+ 2021-01-06 4787e906cf crypto/x509: rollback new CertificateRequest fields
+ 2021-01-06 c9658bee93 cmd/go: make module suggestion more friendly
+ 2021-01-06 4c668b25c6 runtime/metrics: fix panic message for Float64Histogram
+ 2021-01-06 d2131704a6 net/http/httputil: fix deadlock in DumpRequestOut
+ 2021-01-05 3e1e13ce6d cmd/go: set cfg.BuildMod to "readonly" by default with no module root
+ 2021-01-05 0b0d004983 cmd/go: pass embedcfg to gccgo if supported
+ 2021-01-05 1b85e7c057 cmd/go: don't scan gccgo standard library packages for imports
+ 2021-01-05 6b37b15d95 runtime: don't take allglock in tracebackothers
+ 2021-01-04 9eef49cfa6 math/rand: fix typo in comment
+ 2021-01-04 b01fb2af9e testing/fstest: fix typo in error message
+ 2021-01-01 3dd5867605 doc: 2021 is the Year of the Gopher
+ 2020-12-31 95ce805d14 io/fs: remove darwin/arm64 special condition
+ 2020-12-30 20d0991b86 lib/time, time/tzdata: update tzdata to 2020f
+ 2020-12-30 ed301733bb misc/cgo/testcarchive: remove special flags for Darwin/ARM
+ 2020-12-30 0ae2e032f2 misc/cgo/test: enable TestCrossPackageTests on darwin/arm64
+ 2020-12-29 780b4de16b misc/ios: fix wording for command line instructions
+ 2020-12-29 b4a71c95d2 doc/go1.16: reference misc/ios/README for how to build iOS programs
+ 2020-12-29 f83e0f6616 misc/ios: add to README how to build ios executables
+ 2020-12-28 4fd9455882 io/fs: fix typo in comment
Change-Id: I2f257bbc5fbb05f15c2d959f8cfe0ce13b083538
Went in a semi-automated way through the clearest renames of functions,
and updated comments and error messages where it made sense.
Change-Id: Ied8e152b562b705da7f52f715991a77dab60da35
Reviewed-on: https://go-review.googlesource.com/c/go/+/284216
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
tracebackothers is called from fatal throw/panic.
A fatal throw may be taken with allglock held (notably in the allocator
when allglock is held), which would cause a deadlock in tracebackothers
when we try to take allglock again. Locking allglock here is also often
a lock order violation w.r.t. the locks held when throw was called.
Avoid the deadlock and ordering issues by skipping locking altogether.
It is OK to miss concurrently created Gs (which are generally avoided by
freezetheworld(), and which were possible previously anyways if created
after the loop).
Fatal throw/panic freezetheworld(), which should freeze other threads
that may be racing to modify allgs. However, freezetheworld() does _not_
guarantee that it stops all other threads, so we can't simply drop the
lock.
Fixes#42669
Updates #43175
Change-Id: I657aec46ed35fd5d1b3f1ba25b500128ab26b088
Reviewed-on: https://go-review.googlesource.com/c/go/+/270861
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Trust: Michael Pratt <mpratt@google.com>
This change modifies the consistent stats implementation to keep the
per-P sequence counter on each P instead of each mcache. A valid mcache
is not available everywhere that we want to call e.g. allocSpan, as per
issue #42339. By decoupling these two, we can add a mechanism to allow
contexts without a P to update stats consistently.
In this CL, we achieve that with a mutex. In practice, it will be very
rare for an M to update these stats without a P. Furthermore, the stats
reader also only needs to hold the mutex across the update to "gen"
since once that changes, writers are free to continue updating the new
stats generation. Contention could thus only arise between writers
without a P, and as mentioned earlier, those should be rare.
A nice side-effect of this change is that the consistent stats acquire
and release API becomes simpler.
Fixes#42339.
Change-Id: Ied74ab256f69abd54b550394c8ad7c4c40a5fe34
Reviewed-on: https://go-review.googlesource.com/c/go/+/267158
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Trust: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
Background mark workers perform per-P marking work. Currently each
worker is assigned a P at creation time. The worker "attaches" to the P
via p.gcBgMarkWorker, making itself (usually) available to
findRunnableGCWorker for scheduling GC work.
While running gcMarkDone, the worker "detaches" from the P (by clearing
p.gcBgMarkWorker), since it may park for other reasons and should not be
scheduled by findRunnableGCWorker.
Unfortunately, this design is complex and difficult to reason about. We
simplify things by changing the design to eliminate the hard P
attachment. Rather than workers always performing work from the same P,
workers perform work for whichever P they find themselves on. On park,
the workers are placed in a pool of free workers, which each P's
findRunnableGCWorker can use to run a worker for its P.
Now if a worker parks in gcMarkDone, a P may simply use another worker
from the pool to complete its own work.
The P's GC worker mode is used to communicate the mode to run to the
selected worker. It is also used to emit the appropriate worker
EvGoStart tracepoint. This is a slight change, as this G may be
preempted (e.g., in gcMarkDone). When it is rescheduled, the trace
viewer will show it as a normal goroutine again. It is currently a bit
difficult to connect to the original worker tracepoint, as the viewer
does not display the goid for the original worker (though the data is in
the trace file).
Change-Id: Id7bd3a364dc18a4d2b1c99c4dc4810fae1293c1b
Reviewed-on: https://go-review.googlesource.com/c/go/+/262348
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Trust: Michael Pratt <mpratt@google.com>
Following golang.org/cl/259578, findrunnable still must touch every
other P in checkTimers in order to look for timers to steal. This scales
poorly with GOMAXPROCS and potentially performs poorly by pulling remote
Ps into cache.
Add timerpMask, a bitmask that tracks whether each P may have any timers
on its timer heap.
Ideally we would update this field on any timer add / remove to always
keep it up to date. Unfortunately, updating a shared global structure is
antithetical to sharding timers by P, and doing so approximately doubles
the cost of addtimer / deltimer in microbenchmarks.
Instead we only (potentially) clear the mask when the P goes idle. This
covers the best case of avoiding looking at a P _at all_ when it is idle
and has no timers. See the comment on updateTimerPMask for more details
on the trade-off. Future CLs may be able to expand cases we can avoid
looking at the timers.
Note that the addition of idlepMask to p.init is a no-op. The zero value
of the mask is the correct init value so it is not necessary, but it is
included for clarity.
Benchmark results from WakeupParallel/syscall/pair/race/1ms (see
golang.org/cl/228577). Note that these are on top of golang.org/cl/259578:
name old msec new msec delta
Perf-task-clock-8 244 ± 4% 246 ± 4% ~ (p=0.841 n=5+5)
Perf-task-clock-16 247 ±11% 252 ± 4% ~ (p=1.000 n=5+5)
Perf-task-clock-32 270 ± 1% 268 ± 2% ~ (p=0.548 n=5+5)
Perf-task-clock-64 302 ± 3% 296 ± 1% ~ (p=0.222 n=5+5)
Perf-task-clock-128 358 ± 3% 352 ± 2% ~ (p=0.310 n=5+5)
Perf-task-clock-256 483 ± 3% 458 ± 1% -5.16% (p=0.008 n=5+5)
Perf-task-clock-512 663 ± 1% 612 ± 4% -7.61% (p=0.008 n=5+5)
Perf-task-clock-1024 1.06k ± 1% 0.95k ± 2% -10.24% (p=0.008 n=5+5)
Updates #28808
Updates #18237
Change-Id: I4239cd89f21ad16dfbbef58d81981da48acd0605
Reviewed-on: https://go-review.googlesource.com/c/go/+/264477
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Trust: Michael Pratt <mpratt@google.com>
