This change makes it so that the GC pacer's trigger ratio can never fall
below 0.6. Upcoming changes to the allocator make it significantly more
scalable and thus much faster in certain cases, creating a large gap
between the performance of allocation and scanning. The consequence of
this is that the trigger ratio can drop very low (0.07 was observed) in
order to drop GC utilization. A low trigger ratio like this results in a
high amount of black allocations, which causes the live heap to appear
larger, and thus the heap, and RSS, grows to a much higher stable point.
This change alleviates the problem by placing a lower bound on the
trigger ratio. The expected (and confirmed) effect of this is that
utilization in certain scenarios will no longer converge to the expected
25%, and may go higher. As a result of this artificially high trigger
ratio, more time will also be spent doing GC assists compared to
dedicated mark workers, since the GC will be on for an artifically short
fraction of time (artificial with respect to the pacer). The biggest
concern of this change is that allocation latency will suffer as a
result, since there will now be more assists. But, upcoming changes to
the allocator reduce the latency enough to outweigh the expected
increase in latency from this change, without the blowup in RSS observed
from the changes to the allocator.
Updates #35112.
Change-Id: Idd7c94fa974d0de673304c4397e716e89bfbf09b
Reviewed-on: https://go-review.googlesource.com/c/go/+/200439
Reviewed-by: Austin Clements <austin@google.com>
The js.Value struct now contains a pointer, so a finalizer can
determine if the value is not referenced by Go any more.
Unfortunately this breaks Go's == operator with js.Value. This change
adds a new Equal method to check for the equality of two Values.
This is a breaking change. The == operator is now disallowed to
not silently break code.
Additionally the helper methods IsUndefined, IsNull and IsNaN got added.
Fixes#35111
Change-Id: I58a50ca18f477bf51a259c668a8ba15bfa76c955
Reviewed-on: https://go-review.googlesource.com/c/go/+/203600
Run-TryBot: Richard Musiol <neelance@gmail.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This is a rough attempt at restoring -m=2 escape analysis diagnostics
on par with those that were available with esc.go. It's meant to be
simple and non-invasive.
For example, given this random example from bytes/reader.go:
138 func (r *Reader) WriteTo(w io.Writer) (n int64, err error) {
...
143 b := r.s[r.i:]
144 m, err := w.Write(b)
esc.go used to report:
bytes/reader.go:138:7: leaking param content: r
bytes/reader.go:138:7: from r.s (dot of pointer) at bytes/reader.go:143:8
bytes/reader.go:138:7: from b (assigned) at bytes/reader.go:143:4
bytes/reader.go:138:7: from w.Write(b) (parameter to indirect call) at bytes/reader.go:144:19
With this CL, escape.go now reports:
bytes/reader.go:138:7: parameter r leaks to {heap} with derefs=1:
bytes/reader.go:138:7: flow: b = *r:
bytes/reader.go:138:7: from r.s (dot of pointer) at bytes/reader.go:143:8
bytes/reader.go:138:7: from r.s[r.i:] (slice) at bytes/reader.go:143:10
bytes/reader.go:138:7: from b := r.s[r.i:] (assign) at bytes/reader.go:143:4
bytes/reader.go:138:7: flow: {heap} = b:
bytes/reader.go:138:7: from w.Write(b) (call parameter) at bytes/reader.go:144:19
Updates #31489.
Change-Id: I0c2b943a0f9ce6345bfff61e1c635172a9290cbb
Reviewed-on: https://go-review.googlesource.com/c/go/+/196959
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: David Chase <drchase@google.com>
Add a 'single lane' SIMD implemementation of the single byte count
function for use on machines that support the vector facility. This
allows up to 16 bytes to be counted per loop iteration.
We can probably improve performance further by adding more 'lanes'
(i.e. counting more bytes in parallel) however this will increase
the complexity of the function so I'm not sure it is worth doing
yet.
name old speed new speed delta
pkg:strings goos:linux goarch:s390x
CountByte/10 789MB/s ± 0% 1131MB/s ± 0% +43.44% (p=0.000 n=9+9)
CountByte/32 936MB/s ± 0% 3236MB/s ± 0% +245.87% (p=0.000 n=8+9)
CountByte/4096 1.06GB/s ± 0% 21.26GB/s ± 0% +1907.07% (p=0.000 n=10+10)
CountByte/4194304 1.06GB/s ± 0% 20.54GB/s ± 0% +1838.50% (p=0.000 n=10+10)
CountByte/67108864 1.06GB/s ± 0% 18.31GB/s ± 0% +1629.51% (p=0.000 n=10+10)
pkg:bytes goos:linux goarch:s390x
CountSingle/10 800MB/s ± 0% 986MB/s ± 0% +23.21% (p=0.000 n=9+10)
CountSingle/32 925MB/s ± 0% 2744MB/s ± 0% +196.55% (p=0.000 n=9+10)
CountSingle/4K 1.26GB/s ± 0% 19.44GB/s ± 0% +1445.59% (p=0.000 n=10+10)
CountSingle/4M 1.26GB/s ± 0% 20.28GB/s ± 0% +1510.26% (p=0.000 n=8+10)
CountSingle/64M 1.23GB/s ± 0% 17.78GB/s ± 0% +1350.67% (p=0.000 n=9+10)
Change-Id: I230d57905db92a8fdfc50b1d5be338941ae3a7a1
Reviewed-on: https://go-review.googlesource.com/c/go/+/199979
Run-TryBot: Michael Munday <mike.munday@ibm.com>
Reviewed-by: Keith Randall <khr@golang.org>
TestArchiveBuildInvokeWithExec is failing on darwin due to
duplicated symbols, because the C definition (int fortytwo;) is
copied to two generated cgo sources. In fact, this test is about
building c-archive, but doesn't need to import "C". Removed the
"C" import.
Change-Id: I3a17546e01272a7ae37e6417791ab949fb44597e
Reviewed-on: https://go-review.googlesource.com/c/go/+/205278
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
When dropping a P, if it has any timers, and if some thread is
sleeping in the netpoller, wake the netpoller to run the P's timers.
This mitigates races between the netpoller deciding how long to sleep
and a new timer being added.
In sysmon, if all P's are idle, check the timers to decide how long to sleep.
This avoids oversleeping if no thread is using the netpoller.
This can happen in particular if some threads use runtime.LockOSThread,
as those threads do not block in the netpoller.
Also, print the number of timers per P for GODEBUG=scheddetail=1.
Before this CL, TestLockedDeadlock2 would fail about 1% of the time.
With this CL, I ran it 150,000 times with no failures.
Updates #6239
Updates #27707Fixes#35274Fixes#35288
Change-Id: I7e5193e6c885e567f0b1ee023664aa3e2902fcd1
Reviewed-on: https://go-review.googlesource.com/c/go/+/204800
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
This CL makes these changes to the hash/maphash API to make it fit a bit
more into the standard library:
- Move some of the package doc onto type Hash, so that `go doc maphash.Hash` shows it.
- Instead of having identical AddBytes and Write methods,
standardize on Write, the usual name for this function.
Similarly, AddString -> WriteString, AddByte -> WriteByte.
- Instead of having identical Hash and Sum64 methods,
standardize on Sum64 (for hash.Hash64). Dropping the "Hash" method
also helps because Hash is usually reserved to mean the state of a
hash function (hash.Hash etc), not the hash value itself.
- Make an uninitialized hash.Hash auto-seed with a random seed.
It is critical that users not use the same seed for all hash functions
in their program, at least not accidentally. So the Hash implementation
must either panic if uninitialized or initialize itself.
Initializing itself is less work for users and can be done lazily.
- Now that the zero hash.Hash is useful, drop maphash.New in favor of
new(maphash.Hash) or simply declaring a maphash.Hash.
- Add a [0]func()-typed field to the Hash so that Hashes cannot be compared.
(I considered doing the same for Seed but comparing seeds seems OK.)
- Drop the integer argument from MakeSeed, to match the original design
in golang.org/issue/28322. There is no point to giving users control
over the specific seed bits, since we want the interpretation of those
bits to be different in every different process. The only thing users
need is to be able to create a new random seed at each call.
(Fixes a TODO in MakeSeed's public doc comment.)
This API is new in Go 1.14, so these changes do not violate the compatibility promise.
Fixes#35060.
Fixes#35348.
Change-Id: Ie6fecc441f3f5ef66388c6ead92e875c0871f805
Reviewed-on: https://go-review.googlesource.com/c/go/+/205069
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
We absorbed Not into most integer comparisons but not into pointer
and floating point equality checks.
The new cases trigger more than 300 times during make.bash.
Change-Id: I77c6b31fcacde10da5470b73fc001a19521ce78d
Reviewed-on: https://go-review.googlesource.com/c/go/+/200618
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
CL 200958 adds skipping empty init function feature without any tests
for it. A codegen test sounds ideal, but it's unlikely that we can make
one for now, so use a program to manipulate runtime/proc.go:initTask
directly.
Updates #34869
Change-Id: I2683b9a1ace36af6861af02a3a9fb18b3110b282
Reviewed-on: https://go-review.googlesource.com/c/go/+/204217
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Speed up nanotime1 and walltime1 on MIPS64 with vDSO, just like the
other vDSO-enabled targets.
Benchmark numbers on Loongson 3A3000 (GOARCH=mips64le, 1.4GHz) against
current master:
benchmark old ns/op new ns/op delta
BenchmarkNow 868 293 -66.24%
BenchmarkNowUnixNano 851 296 -65.22%
Performance hit on fallback case, tested by using a wrong vDSO symbol name:
benchmark old ns/op new ns/op delta
BenchmarkNow 868 889 +2.42%
BenchmarkNowUnixNano 851 893 +4.94%
Change-Id: Ibfb48893cd060536359863ffee7624c00def646b
GitHub-Last-Rev: 03a58ac2e4
GitHub-Pull-Request: golang/go#35181
Reviewed-on: https://go-review.googlesource.com/c/go/+/203578
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
When linking a Go archive, if the archiver invocation is the very last
thing that needs to happen in the link (no "atexit" cleanups required
remove the locally created tmpdir) then call syscall.Exec to invoke
the archiver command instead of the usual exec.Command. This has the
effect of reducing peak memory use for the linker overall, since we
don't be holding onto all of the linker's live memory while the
archiver is running.
Change-Id: Ibbe22d8d67a70cc2a4f91c68aab56d19fb77c393
Reviewed-on: https://go-review.googlesource.com/c/go/+/203821
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
When we do a successful recover of a panic, we resume normal execution by
returning from the frame that had the deferred call that did the recover (after
executing any remaining deferred calls in that frame).
However, suppose we have called runtime.Goexit and there is a panic during one of the
deferred calls run by the Goexit. Further assume that there is a deferred call in
the frame of the Goexit or a parent frame that does a recover. Then the recovery
process will actually resume normal execution above the Goexit frame and hence
abort the Goexit. We will not terminate the thread as expected, but continue
running in the frame above the Goexit.
To fix this, we explicitly create a _panic object for a Goexit call. We then
change the "abort" behavior for Goexits, but not panics. After a recovery, if the
top-level panic is actually a Goexit that is marked to be aborted, then we return
to the Goexit defer-processing loop, so that the Goexit is not actually aborted.
Actual code changes are just panic.go, runtime2.go, and funcid.go. Adjusted the
test related to the new Goexit behavior (TestRecoverBeforePanicAfterGoexit) and
added several new tests of aborted panics (whose behavior has not changed).
Fixes#29226
Change-Id: Ib13cb0074f5acc2567a28db7ca6912cfc47eecb5
Reviewed-on: https://go-review.googlesource.com/c/go/+/200081
Run-TryBot: Dan Scales <danscales@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
From CL 199979, I noticed that there were some
instructions not covered by the test cases. Added those in this CL.
Additional tests for assembly instructions are also added
based on suggestions made during the review of this CL.
Previously, VSB and VSH are not included in asmz.go, they were also
added in this patch.
Change-Id: I6060a9813b483a161d61ad2240c30eec6de61536
Reviewed-on: https://go-review.googlesource.com/c/go/+/203721
Reviewed-by: Michael Munday <mike.munday@ibm.com>
We used to pass -no_pie to external linker on darwin/arm, which
is incompatible with -fembed-bitcode. CL 201358 attempted to
remove the -no_pie flag, but it resulted the darwin linker to
complain about absolute addressing in TEXT segment.
On darwin/arm, we already get away from absolute addressing in
the TEXT section. The complained absolute addressing is in
RODATA, which was embedded in the TEXT segment. This CL moves
RODATA to the DATA segment, like what we already did on ARM64
and on AMD64 in c-archive/c-shared buildmodes for the same reason.
So there is no absolute addressing in the TEXT segment, which
allows us to remove -no_pie flag.
Fixes#35252.
Updates #32963.
Change-Id: Id6e3a594cb066d257d4f58fadb4a3ee4672529f7
Reviewed-on: https://go-review.googlesource.com/c/go/+/205060
Reviewed-by: Elias Naur <mail@eliasnaur.com>
POWER9 (ISA 3.0) introduced a new format of load/store instructions to
implement indexed load/store quadword, using an immediate value instead
of a register index.
This change adds support for this new instruction encoding and adds the
new load/store quadword instructions (lxv/stxv) to the assembler.
This change also adds the missing XX1-form loads/stores (halfword and byte)
included in ISA 3.0.
Change-Id: Ibcdf53c342d7a352d64a9403c2fe7b25be9c3b24
Reviewed-on: https://go-review.googlesource.com/c/go/+/200399
Run-TryBot: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
R_ADDRCUOFF refers to the offset from the start of the
compilation unit. Now that we could have multiple compilation
units per package, we should use the start of the compilation
unit, instead of the start of the package.
Change-Id: I3d6a3c5a5c47c9ccf88091ed05413ced826f9934
Reviewed-on: https://go-review.googlesource.com/c/go/+/204828
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Currently, at compile time we emit a function DIE with '"".' in
the function name, and we expand it at link time, with a really
ugly function. We can just expand it at compile time instead.
This way, we don't need to modify the symbol content at link time,
and also no need to allocate memory for that.
Keep the linker expansion, in case the compiler is invoked
without the import path.
Change-Id: Id53cd2e2d3eb61efceb8d44479c4b6ef890baa43
Reviewed-on: https://go-review.googlesource.com/c/go/+/204826
Reviewed-by: Than McIntosh <thanm@google.com>
Replace
buf := [HUGE_CONST]*T)(unsafe.Pointer(p))[:]
with
buf := [HUGE_CONST]*T)(unsafe.Pointer(p))[:n:n]
Pointer p points to n of T elements. New unsafe pointer conversion
logic verifies that both first and last elements point into the
same Go variable. And this change adjusts all code to comply with
this rule.
Verified by running
go test -a -short -gcflags=all=-d=checkptr crypto/x509
The test does not fail even with original version of this code. I
suspect it is because all variables I changed live outside of Go
memory. But I am just guessing, I don't really know how pointer
checker works.
Updates golang/go#34972
Change-Id: Ibc33fdc9e2023d9b14905c9badf2f0b683999ab8
Reviewed-on: https://go-review.googlesource.com/c/go/+/204621
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
We don't async preempt assembly functions. We do that by checking
whether the function has a local pointer map, and assume it is
an assembly (or, non-Go) function if there isn't one. However,
assembly functions marked with NO_LOCAL_POINTERS still have local
pointer maps, and we wouldn't identify them. For them, check for
the special pointer map runtime.no_pointers_stackmap as well, and
treat them as not async preemptible.
Change-Id: I1301e3b4d35893c31c4c5a5147a0d775987bd6f4
Reviewed-on: https://go-review.googlesource.com/c/go/+/202337
Reviewed-by: Austin Clements <austin@google.com>
This adds a test of preempting a loop containing no synchronous safe
points for STW and stack scanning.
We couldn't add this test earlier because it requires scheduler, STW,
and stack scanning preemption to all be working.
For #10958, #24543.
Change-Id: I73292db78ca3d14aab11bdafd26d03986920ef0a
Reviewed-on: https://go-review.googlesource.com/c/go/+/201777
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This adds signal-based preemption to preemptone.
Since STW and forEachP ultimately use preemptone, this also makes
these work with async preemption.
This also makes freezetheworld more robust so tracebacks from fatal
panics should be far less likely to report "goroutine running on other
thread; stack unavailable".
For #10958, #24543. (This doesn't fix it yet because asynchronous
preemption only works on POSIX platforms on 386 and amd64 right now.)
Change-Id: If776181dd5a9b3026a7b89a1b5266521b95a5f61
Reviewed-on: https://go-review.googlesource.com/c/go/+/201762
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This adds support for pausing a running G by sending a signal to its
M.
The main complication is that we want to target a G, but can only send
a signal to an M. Hence, the protocol we use is to simply mark the G
for preemption (which we already do) and send the M a "wake up and
look around" signal. The signal checks if it's running a G with a
preemption request and stops it if so in the same way that stack check
preemptions stop Gs. Since the preemption may fail (the G could be
moved or the signal could arrive at an unsafe point), we keep a count
of the number of received preemption signals. This lets stopG detect
if its request failed and should be retried without an explicit
channel back to suspendG.
For #10958, #24543.
Change-Id: I3e1538d5ea5200aeb434374abb5d5fdc56107e53
Reviewed-on: https://go-review.googlesource.com/c/go/+/201760
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This adds support for scanning the stack when a goroutine is stopped
at an async safe point. This is not yet lit up because asyncPreempt is
not yet injected, but prepares us for that.
This works by conservatively scanning the registers dumped in the
frame of asyncPreempt and its parent frame, which was stopped at an
asynchronous safe point.
Conservative scanning works by only marking words that are pointers to
valid, allocated heap objects. One complication is pointers to stack
objects. In this case, we can't determine if the stack object is still
"allocated" or if it was freed by an earlier GC. Hence, we need to
propagate the conservative-ness of scanning stack objects: if all
pointers found to a stack object were found via conservative scanning,
then the stack object itself needs to be scanned conservatively, since
its pointers may point to dead objects.
For #10958, #24543.
Change-Id: I7ff84b058c37cde3de8a982da07002eaba126fd6
Reviewed-on: https://go-review.googlesource.com/c/go/+/201761
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This adds asynchronous preemption function for amd64 and 386. These
functions spill and restore all register state that can be used by
user Go code.
For the moment we stub out the other arches.
For #10958, #24543.
Change-Id: I6f93fabe9875f4834922a5712362e79045c00aca
Reviewed-on: https://go-review.googlesource.com/c/go/+/201759
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This adds a sigctxt.pushCall method that pushes a call at the signaled
site. We'll use this to inject asynchronous preemptions and in some
places we use it to clean up preparePanic.
For the moment this only works on 386 and amd64. We stub it out on
other platforms and will avoid calling the stubbed version.
For #10958, #24543.
Change-Id: I49e0e853f935d32dd67a70c6cafbae44ee68af8e
Reviewed-on: https://go-review.googlesource.com/c/go/+/201758
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Currently, the compiler fails to mark any unsafe-points in the initial
instructions of a function as unsafe points. This happens because
unsafe points are encoded as a stack map index of -2 and the compiler
emits PCDATA instructions when there's a change in the stack map
index, but I had set the initial stack map index to -2. The actual
initial PCDATA value assumed by the PCDATA encoder and the runtime is
-1. Hence, if the first instructions had a stack map index of -2, no
PCDATA was emitted, which cause the runtime to assume the index was -1
instead.
This was particularly problematic in the runtime, where the compiler
was supposed to mark only calls as safe-points and everything else as
unsafe-points. Runtime leaf functions, for example, should have been
marked as entirely unsafe-points, but were instead marked entirely as
safe-points.
Fix this by making the PCDATA instruction generator assume the initial
PCDATA value is -1 instead of -2, so it will emit a PCDATA instruction
right away if the first real instruction is an unsafe-point.
This increases the size of the cmd/go binary by 0.02% since we now
emit slightly more PCDATA than before.
For #10958, #24543.
Change-Id: I92222107f799130072b36d49098d2686f1543699
Reviewed-on: https://go-review.googlesource.com/c/go/+/202084
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This doesn't do anything yet, but it will provide a way to disable
non-cooperative preemption.
For #10958, #24543.
Change-Id: Ifdef303f103eabd0922ced8d9bebbd5f0aa2cda4
Reviewed-on: https://go-review.googlesource.com/c/go/+/201757
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Current implementation of httputil.DumpRequestOut
incorrectly resets the Request.Body prematurely
before Content-Length/Transfer-Encoding detection
in newTransferWriter()
This fix avoids resetting the Request.Body when
Request.ContentLength is set to '0' by the caller
and Request.Body is set to a custom reader. To allow
newTransferWriter() to treat this situation as
'Transfer-Encoding: chunked'.
Fixes#34504
Change-Id: Ieab6bf876ced28c32c084e0f4c8c4432964181f5
Reviewed-on: https://go-review.googlesource.com/c/go/+/197898
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This change updates the GOARCH/GOOS discussion at the top of the
"Installing Go from source" document to better reflect the current
status. In particular:
- The GOARCH list now focuses on simply listing the supported
architectures, with no notes about their supposed "maturity", since
the same GOARCH can be mature on a GOOS and not so mature on another.
- Outdated notes about some archs being new and "not well-exercised"
have been removed in favour of a following list of which ports are
first class.
- The list of supported OS has been updated (added: AIX, Illumos),
and sorted in alphabetical order.
- A note about the runtime support being the same for all ARCHS,
"including garbage collection and efficient array slicing and" etc etc
has been removed, since it doesn't seem particularly relevant in a
"install from source" instruction page, and it's likely a leftover
from the time this doc page was the landing place for new people and
it felt the need to "sell" Go.
Updates #27689Fixes#35009
Change-Id: Ic4eca91dca3135adc7bed4fe00b4f157768f0e81
Reviewed-on: https://go-review.googlesource.com/c/go/+/202197
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
In the deadcode pass, we use a work queue for the flood algorithm.
Currently this is a simple LIFO queue. In this order, there is
poor locality in accessing object files.
Since the global indices are assigned in package DAG order, edges
are mostly either within a package or from a smaller index to a
larger one. (With named symbols, there can be backward edges, but
shouldn't be too many.) Using a min-heap for the work queue, we
access all symbols in one object, then move to next one. It
rarely needs to revisit an object that is already visted. This
should result in better locality.
Benchmark result from Than (thanks!):
name old time/op new time/op delta
LinkCompiler 1.74s ±11% 1.61s ± 9% -7.80% (p=0.000 n=20+19)
LinkWithoutDebugCompiler 1.27s ±11% 1.15s ± 9% -9.02% (p=0.000 n=20+20)
Currently this uses the container/heap package, which uses
interface elements. If this allocates too much, we may consider
to hand-code the min heap.
Change-Id: I216d5291c432fe1f40b0b8f4f1b9d388807bf6c5
Reviewed-on: https://go-review.googlesource.com/c/go/+/204438
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Currently in the linker we load internal packges first, then the
main package, and then load imported packages following the
dependency graph. As a result, packages are loaded mostly in the
dependency order, except the internal packages. The global symbol
indices are assigned the same way.
By loading the internal packages at the end, the packages are
loaded in the dependency order, so are the global indices. This
way, a relocation edge is mostly either within a packge or a
forward edge from a smaller index to a larger one. This allows
us to use a min-heap work queue in the deadcode pass, to achieve
better spatial locality (in the next CL).
Change-Id: I01fa9b3cf0c9e9e66006040f6378a51fd78f0f39
Reviewed-on: https://go-review.googlesource.com/c/go/+/204437
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Cgo internal linking was disabled when switching to new object
files. Reeanble the tests, as they get supported.
Change-Id: I11d2ac8785cce73f3a42f6935c10d9f067bc90a4
Reviewed-on: https://go-review.googlesource.com/c/go/+/204520
Reviewed-by: Than McIntosh <thanm@google.com>
Even though bitwise operations may be slightly more
performant, the readability improvement of a mod
operation is worth the tradeoff.
Change-Id: I352c92ad355c6eb6ef99e3da00e1eff2d2ea5812
Reviewed-on: https://go-review.googlesource.com/c/go/+/204739
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
On illumos systems, and at least historically on Solaris systems, it is
possible for port_getn(3C) calls to return some number of events and
then fail with error ETIME.
Generally we expect this to happen if the caller passes an nget value
larger than 1 and calls with a timeout; if less than the requested
number of events accumulate the system will still return them after
timeout failure so the caller must check the updated nget value in the
ETIME case. Note that although less likely this can still happen even
when requesting just 1 event, especially with a short timeout value or
on a busy system.
Fixes#35261
Change-Id: I0d83251b69a2fadc64c4e8e280aa596e2e1548ba
Reviewed-on: https://go-review.googlesource.com/c/go/+/204801
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
In newobj mode, cgo attributes are typically set later, as we
create sym.Symbols later. But when internal cgo linking, the
host object loaders still work with sym.Symbols, and the cgo
attributes need to be set for them to work properly. Therefore,
set them early. This will cause creating some Symbols eagerly,
but they are mostly host object symbols and will need to be
created anyway.
Now all cgo internal linking tests pass on ELF systems.
Change-Id: I023a4df4429acc8ebf5e185f62e6809198497a78
Reviewed-on: https://go-review.googlesource.com/c/go/+/204857
Reviewed-by: Than McIntosh <thanm@google.com>
If multiple goroutines call time.(*Timer).Reset then the timer will go
from timerWaiting to timerDeleted to timerModifying to timerModifiedLater.
The timer can be on a different P, meaning that simultaneously cleantimers
could change it from timerDeleted to timerRemoving to timerRemoved.
If Reset sees timerRemoved, it was doing an atomic.Store of timerWaiting,
meaning that it did not necessarily see the other values set in the timer,
so the timer could appear to be in an inconsistent state. Use atomic.Cas
to avoid that possibility.
Updates #6239
Updates #27707Fixes#35272
Change-Id: I1d59a13dc4f2ff4af110fc6e032c8c9d59cfc270
Reviewed-on: https://go-review.googlesource.com/c/go/+/204717
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
In CL 204617, I intend to make "bound" parameter to have special meaning
in typecheckarraylit, so we can distinguish between type-checks array
literal and slice literal. But we end up with other solution. The CL was
submitted without reverting the "bound" parameter in case of slice
literal.
Technically, it's not harmful, but causes the code harder to read and maintain.
Change-Id: Ia522ccc9a6b8e25d7eaad4aa4957cb4fa18edc60
Reviewed-on: https://go-review.googlesource.com/c/go/+/204618
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
User's program was mutating time.Local variable and crashing
itself as a consequence. Instead of documenting that time.Local
variable should not be mutated, recommended way of setting the
system's time zone has been documented.
Fixes#34814
Change-Id: I7781189855c3bf2ea979dfa07f86c283eed27091
Reviewed-on: https://go-review.googlesource.com/c/go/+/200457
Reviewed-by: Ian Lance Taylor <iant@golang.org>
I was doing some testing with GODEBUG=schedtrace=1,scheddetail=1 and I
noticed that the program hung after a throw with "all goroutines are
asleep". This is because when doing a throw or fatal panic with schedtrace
the panic code does a final schedtrace, which needs to acquire the
scheduler lock. The checkdead function is always called with the scheduler
lock held. So checkdead would throw with the scheduler lock held, then
the panic code would call schedtrace, which would block trying to acquire
the scheduler lock.
This problem will only happen for people debugging the runtime, but
it's easy to avoid by having checkdead unlock the scheduler lock before
it throws. I only did this for the throws that can happen for a normal
program, not for throws that indicate some corruption in the scheduler data.
Change-Id: Ic62277b3ca6bee6f0fca8d5eb516c59cb67855cb
Reviewed-on: https://go-review.googlesource.com/c/go/+/204778
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
