Junyang Shao [Fri, 10 Oct 2025 17:42:59 +0000 (17:42 +0000)]
[dev.simd] cmd/compile, simd: remove mask load and stores
We have convert mask to bits already, the API of mask load and stores
are inconsistent with them, also mask load and stores could just be
hidden behind peepholes. So this CL removes them, the next CL will add
the peephole for them.
Change-Id: Ifa7d23fb52bb0efd1785935ead4d703927f16d2b
Reviewed-on: https://go-review.googlesource.com/c/go/+/710915 Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Junyang Shao [Mon, 13 Oct 2025 18:30:05 +0000 (18:30 +0000)]
[dev.simd] simd/_gen: add mem peephole with feat mismatches
This CL attempts to add peepholes for Op -> Opload where the Opload has
a different CPU feature than Op. However the new simdgen changes doesn't
do anything because such peepholes do not exist.
Change-Id: I20c3e4b43bb7414c3a309d77786218372ca1b5b8
Reviewed-on: https://go-review.googlesource.com/c/go/+/711380 Reviewed-by: David Chase <drchase@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
David Chase [Fri, 5 Sep 2025 23:05:18 +0000 (19:05 -0400)]
[dev.simd] cmd/compile: track which CPU features are in scope
analysis for
- is this block only reached through feature checks?
- does the function signature imply AVX-something?
- is there an instruction in this block which implies AVX-something?
and keep track of which features those are. Features =
AVX, AVX2, AVX512, etc.
Cherry Mui [Thu, 2 Oct 2025 22:24:12 +0000 (18:24 -0400)]
cmd/link: support .def file with MSVC clang toolchain
lld-link supports .def file, but requires a "-def:" (or "/def:")
flag. (MinGW linker, on the other hand, requires no flag.) Pass
the flag when using MSVC-based toolchain.
The Director function has been superseded by Rewrite.
Rewrite avoids fundamental security issues with hop-by-hop header
handling in the Director API and has better default handling
of X-Forwarded-* headers.
Fixes #73161
Change-Id: Iadaf3070e0082458f79fb892ade51cb7ce832802
Reviewed-on: https://go-review.googlesource.com/c/go/+/708615 Reviewed-by: Nicholas Husin <husin@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Nicholas Husin <nsh@golang.org>
Nicholas S. Husin [Thu, 2 Oct 2025 16:41:25 +0000 (12:41 -0400)]
net/http: update bundled x/net/http2 and delete obsolete http2inTests
http2inTests is no longer needed after go.dev/cl/708135 and should be
deleted. To prevent errors in future vendored dependency updates,
h2_bundle.go is also updated together in this change.
Change-Id: I7b8c3f6854203fab4ec639a2a268df0cd2b1dee7
Reviewed-on: https://go-review.googlesource.com/c/go/+/708595 Reviewed-by: Damien Neil <dneil@google.com> Reviewed-by: Nicholas Husin <husin@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Damien Neil [Thu, 2 Oct 2025 17:45:01 +0000 (10:45 -0700)]
io/fs: move path name documentation up to the package doc comment
Perhaps surprisingly to users, io/fs path names are slash-separated.
Move the documentation for path names up to the top of the package
rather than burying it in the ValidPath documentation.
Change-Id: Id338df07c74a16be74c687ac4c45e0513ee40a8c
Reviewed-on: https://go-review.googlesource.com/c/go/+/708616
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Alan Donovan <adonovan@google.com>
Auto-Submit: Damien Neil <dneil@google.com>
Michael Anthony Knyszek [Thu, 2 Oct 2025 19:15:34 +0000 (19:15 +0000)]
runtime: skip TestGoroutineLeakProfile under mayMoreStackPreempt
This may be the long-term fix, but we first need to understand if this
just makes the tests flaky, or if it's revealing an actual underlying
issue. I'm leaning toward the former. If it is the former, ideally we
just make the tests robust (wait longer, maybe?).
For now, this change will make the longtest builders OK again.
For #75729.
Change-Id: If9b30107d04a8e5af5670850add3a53f9471eec6
Reviewed-on: https://go-review.googlesource.com/c/go/+/708715 Reviewed-by: Roland Shoemaker <roland@golang.org> Reviewed-by: Michael Pratt <mpratt@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Carlos Amedee <carlos@golang.org>
Daniel Martí [Sat, 30 Aug 2025 20:13:12 +0000 (21:13 +0100)]
cmd/go: update `go help mod edit` with the tool and ignore sections
The types were added to the docs, but not the fields in GoMod.
While here, I was initially confused about what is the top-level type
given that `type Module` comes first. Move `type GoMod` to the top
as it is the actual top-level type.
Change-Id: I1270154837501f5c7f5b21959b2841fd4ac808d0
Reviewed-on: https://go-review.googlesource.com/c/go/+/700116 Reviewed-by: Sean Liao <sean@liao.dev> Reviewed-by: Carlos Amedee <carlos@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Michael Matloob <matloob@golang.org> Reviewed-by: Junyang Shao <shaojunyang@google.com>
cmd/compile: propagate len([]T{}) to make builtin to allow stack allocation
Updates #75620
Change-Id: I6a6a6964af4512e30eb4806e1dc7b0fd0835744f
Reviewed-on: https://go-review.googlesource.com/c/go/+/707255 Reviewed-by: Keith Randall <khr@google.com> Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Keith Randall <khr@golang.org> Reviewed-by: Carlos Amedee <carlos@golang.org>
Daniel McCarney [Mon, 29 Sep 2025 18:31:38 +0000 (14:31 -0400)]
crypto/tls: rm marshalEncryptedClientHelloConfigList dead code
This package internal function has no call sites.
Change-Id: I262058199fd2f387ef3b5e21099421720cc5413e
Reviewed-on: https://go-review.googlesource.com/c/go/+/707815
TryBot-Bypass: Daniel McCarney <daniel@binaryparadox.net>
Auto-Submit: Roland Shoemaker <roland@golang.org>
Auto-Submit: Daniel McCarney <daniel@binaryparadox.net> Reviewed-by: Carlos Amedee <carlos@golang.org> Reviewed-by: Roland Shoemaker <roland@golang.org>
Joe Tsai [Mon, 15 Sep 2025 19:56:11 +0000 (12:56 -0700)]
encoding/json: fix Decoder.InputOffset regression in goexperiment.jsonv2
The Decoder.InputOffset method was always ambiguous about the exact
offset returned since anything between the end of the previous token
to the start of the next token could be a valid result.
Empirically, it seems that the behavior was to report
the end of the previous token unless Decoder.More is called,
in which case it reports the start of the next token.
This is an odd semantic since a relatively side-effect free method
like More is not quite so side-effect free.
However, our goal is to preserve historical v1 semantic when possible
regardless of whether it made sense.
Note that jsontext.Decoder.InputOffset consistently always reports
the end of the previous token. Users can explicitly choose the
exact position they want by inspecting the UnreadBuffer.
Fixes #75468
Change-Id: I1e946e83c9d29dfc09f2913ff8d6b2b80632f292
Reviewed-on: https://go-review.googlesource.com/c/go/+/703856 Reviewed-by: Damien Neil <dneil@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Junyang Shao <shaojunyang@google.com>
This test features a 5s timeout, which is far too close
to the natural variance in scheduling on an overloaded
CI builder machine to make a reliable test. Skipping.
Updates #72104
Change-Id: I52133a2d101808c923e316e0c7fdce9edbb31b10
Reviewed-on: https://go-review.googlesource.com/c/go/+/708075
Auto-Submit: Alan Donovan <adonovan@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Michael Matloob <matloob@golang.org> Reviewed-by: Michael Matloob <matloob@google.com>
Joel Sing [Mon, 15 Sep 2025 10:10:05 +0000 (20:10 +1000)]
test/codegen: codify handling of floating point constants on arm64
While here, reorder Float32ConstantStore/Float64ConstantStore for
consistency.
Change-Id: Ic1b3e9f9474965d15bc94518d78d1a4a7bda93f3
Reviewed-on: https://go-review.googlesource.com/c/go/+/703756 Reviewed-by: Keith Randall <khr@golang.org>
Auto-Submit: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Carlos Amedee <carlos@golang.org>
Auto-Submit: Joel Sing <joel@sing.id.au> Reviewed-by: Keith Randall <khr@google.com>
Remove redundant "type:" prefix check on symbol names in isFixedLoad,
also refactor some duplicate code into methods.
Change-Id: I8358422596eea8c39d1a30a554bd0aae8b570038
Reviewed-on: https://go-review.googlesource.com/c/go/+/701275 Reviewed-by: Keith Randall <khr@google.com> Reviewed-by: Keith Randall <khr@golang.org>
Auto-Submit: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Carlos Amedee <carlos@golang.org>
mime: extend "builtinTypes" to include a more complete list of common types
Implement all agreed upon types, using IANA's listed media types to decide
when there is a disagreement in type. Except in the case of `.wav` where
`audio/wav` is used.
Fixes #69530
Change-Id: Iec99a6ceb534073be83c8390f48799bec3e4cfc7
GitHub-Last-Rev: e314c5ec6d9aba753dca5f6dbb9d1741bac43227
GitHub-Pull-Request: golang/go#69533
Reviewed-on: https://go-review.googlesource.com/c/go/+/614376 Reviewed-by: Damien Neil <dneil@google.com> Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Auto-Submit: Sean Liao <sean@liao.dev> Reviewed-by: Sean Liao <sean@liao.dev>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Carlos Amedee <carlos@golang.org>
The first arg of a generic function is the dictionary. This dictionary
is never nil, but it gets a nil check becuase the dict arg is treated as
a slice during construction.
Note how the function begins with a TESTB that loads the dict to perform
the nil check.
This CL eliminates that nil check.
For most generic functions, this doesn't matter too much, but not
infrequently are generic functions written which never actually use the
dictionary (like cmp.Compare), so I suspect this might help in hot code
to avoid repeatedly touching the dictionary in memory, and in cases
where the generic function is not inlined (and thus the dict dropped).
Change-Id: I43dcea5f2aa37372f773e5edc6a2ef1dee0a8db7
Reviewed-on: https://go-review.googlesource.com/c/go/+/706655
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: David Chase <drchase@google.com> Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Keith Randall <khr@google.com>
Auto-Submit: Keith Randall <khr@golang.org>
Steve Muir [Thu, 18 Sep 2025 14:54:57 +0000 (07:54 -0700)]
runtime: initialise debug settings much earlier in startup process
This is necessary specifically to set the value of `debug.decoratemappings`
sufficiently early in the startup sequence that all memory ranges allocated
can be named appropriately using the new Linux-specific naming API
introduced in #71546.
Ian Lance Taylor [Mon, 29 Sep 2025 04:22:09 +0000 (21:22 -0700)]
runtime: correct reference to getStackMap in comment
Change-Id: I9b1fa390434dbda7d49a36b0114c68f942c11d3f
Reviewed-on: https://go-review.googlesource.com/c/go/+/707575
Auto-Submit: Ian Lance Taylor <iant@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com> Reviewed-by: Carlos Amedee <carlos@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
cmd/link: use a .def file to mark exported symbols on Windows
Binutils defaults to exporting all symbols when building a Windows DLL.
To avoid that we were marking symbols with __declspec(dllexport) in
the cgo-generated headers, which instructs ld to export only those
symbols. However, that approach makes the headers hard to reuse when
importing the resulting DLL into other projects, as imported symbols
should be marked with __declspec(dllimport).
A better approach is to generate a .def file listing the symbols to
export, which gets the same effect without having to modify the headers.
Updates #30674
Fixes #56994
Change-Id: I22bd0aa079e2be4ae43b13d893f6b804eaeddabf
Reviewed-on: https://go-review.googlesource.com/c/go/+/705776 Reviewed-by: Michael Knyszek <mknyszek@google.com> Reviewed-by: Junyang Shao <shaojunyang@google.com> Reviewed-by: Than McIntosh <thanm@golang.org> Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
GetFileSizeEx was generated before mkwinsyscall was updated to use
SyscallN. Regenerate to use the new style.
Fixes #75642
Change-Id: Ia473a167633b67fb75b5762d693848ecee425a7e
Reviewed-on: https://go-review.googlesource.com/c/go/+/707615 Reviewed-by: Roland Shoemaker <roland@golang.org>
Auto-Submit: Dmitri Shuralyov <dmitshur@golang.org> Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
Ian Lance Taylor [Mon, 29 Sep 2025 04:25:24 +0000 (21:25 -0700)]
cmd/link: use correct length for pcln.cutab
The pcln.cutab slice holds uint32 elements, as can be seen in the
runtime.moduledata type. The slice was being created with the len
(and cap) set to the size of the slice, which means that the count
was four times too large. This patch sets the correct len/cap.
This doesn't matter for the runtime because nothing looks at
the len of cutab. Since the incorrect len is larger, all valid
indexes remain valid. Using the correct length means that more
invalid indexes will be caught at run time, but such cases are unlikely.
Still, using the correct len is less confusing.
While we're here use the simpler sliceSym for pcln.pclntab.
Change-Id: I09f680b3287467120d994b171c86c784085e3d27
Reviewed-on: https://go-review.googlesource.com/c/go/+/707595
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Cherry Mui <cherryyz@google.com> Reviewed-by: Carlos Amedee <carlos@golang.org>
Auto-Submit: Ian Lance Taylor <iant@golang.org>
David Chase [Tue, 2 Sep 2025 16:10:26 +0000 (12:10 -0400)]
[dev.simd] simd: use new data movement instructions to do "fast" transposes
This is a test/example/performance-comparison. Looking at the
generated code shows that there is still a lot of checking that
perhaps we can figure out how to optimize away.
runtime: jump instead of branch in netbsd/arm64 entry point
CL 706175 removed the NOFRAME directive from _rt0_arm64_netbsd but
did not change the BL instruction to a JMP instruction. This causes the
frame pointer to be stored on the stack, this making direct load from
RSP to be off by 8 bytes.
Cq-Include-Trybots: luci.golang.try:gotip-netbsd-arm64
Change-Id: I0c212fbaba74cfce508f961090dc6e66154c3054
Reviewed-on: https://go-review.googlesource.com/c/go/+/707675
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Michael Pratt <mpratt@google.com> Reviewed-by: Cherry Mui <cherryyz@google.com>
Jes Cok [Thu, 25 Sep 2025 04:41:12 +0000 (04:41 +0000)]
debug/elf: prevent offset overflow
When applying relocations, a malformed ELF file can provide an offset
that, when added to the relocation size, overflows. This wrapped-around
value could then incorrectly pass the bounds check, leading to a panic
when the slice is accessed with the original large offset.
This change eliminates the manual bounds and overflow checks
and writes a relocation to slice by calling putUint.
The putUint helper function centralizes the logic for validating slice
access, correctly handling both out-of-bounds and integer overflow conditions.
This simplifies the relocation code and improves robustness when parsing
malformed ELF files.
Fixes #75516
Change-Id: I00d806bf5501a9bf70200585ba4fd0475d7b2ddc
GitHub-Last-Rev: 49144311d31fecc63cb81b6c31bf9a206acb0596
GitHub-Pull-Request: golang/go#75522
Reviewed-on: https://go-review.googlesource.com/c/go/+/705075 Reviewed-by: Florian Lehner <lehner.florian86@gmail.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Junyang Shao <shaojunyang@google.com>
Auto-Submit: Ian Lance Taylor <iant@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Commit-Queue: Ian Lance Taylor <iant@golang.org>
AN Long [Thu, 25 Sep 2025 14:38:19 +0000 (23:38 +0900)]
all: fix typos
Change-Id: I290812905b6b5c52f289f7f8524f93aef19e6efe
Reviewed-on: https://go-review.googlesource.com/c/go/+/706775
Auto-Submit: Sean Liao <sean@liao.dev> Reviewed-by: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Roland Shoemaker <roland@golang.org> Reviewed-by: Sean Liao <sean@liao.dev>
Kir Kolyshkin [Thu, 11 Sep 2025 22:45:37 +0000 (15:45 -0700)]
os/user: simplify go:build
The go:build constraint for these files is way too complicated,
and can be simplified by using unix tag.
Change-Id: Id8278db0ba799a4e951d3c976f77c7402bebb332
Reviewed-on: https://go-review.googlesource.com/c/go/+/703155 Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Mark Freeman <markfreeman@google.com> Reviewed-by: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Cherry Mui [Sun, 22 Jun 2025 19:40:02 +0000 (15:40 -0400)]
runtime: use a smaller arena size on Wasm
On Wasm, some programs have very small heap. Currently,
we use 4 MB arena size (like all other 32-bit platforms).
For a very small program, it needs to allocate one heap
arena, 4 MB size at a 4 MB aligned address. So we'll
need 8 MB of linear memory, whereas only a smaller
portion is actually used by the program. On Wasm, samll
programs are not uncommon (e.g. WASI plugins), and
users are concerned about the memory usage.
This CL switches to a smaller arena size, as well as a
smaller page allocator chunk size (both are now 512 KB).
So the heap will be grown in 512 KB granularity. For a
helloworld program, it now uses less than 3 MB of
linear memory, instead of 8 MB.
Change-Id: Ibd66c1fa6e794a12c00906cbacc8f2e410f196c4
Reviewed-on: https://go-review.googlesource.com/c/go/+/683296 Reviewed-by: David Chase <drchase@google.com> Reviewed-by: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Add a field to HTTP2Config controlling how we behave when an HTTP/2
connection reaches its concurrency limit.
This field will have no effect until golang.org/x/net/http2 is
updated to make use of it, and h2_bundle.go is updated with the
new http2 package.
For #67813
Change-Id: Ic72a0986528abb21649f28e9fe7cf6e1236b388d
Reviewed-on: https://go-review.googlesource.com/c/go/+/615875
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Damien Neil <dneil@google.com> Reviewed-by: Nicholas Husin <nsh@golang.org> Reviewed-by: Nicholas Husin <husin@google.com>
Damien Neil [Wed, 25 Sep 2024 18:48:17 +0000 (11:48 -0700)]
net/http: add more tests of transport connection pool
Add a variety of addtional tests exercising client connection pooling,
in particular HTTP/2 connection behavior.
Change-Id: I7609d36db5865f1b95c903cfadb0c3233e046c09
Reviewed-on: https://go-review.googlesource.com/c/go/+/615896 Reviewed-by: Nicholas Husin <husin@google.com> Reviewed-by: Nicholas Husin <nsh@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Damien Neil <dneil@google.com>
these are in the same style as the 32-bit select-from-pair,
including the grouped variant. This does not quite capture
the full awesome power of VSHUFPD where it can select
differently in each group; that will be some other method,
that is more complex.
David Chase [Fri, 19 Sep 2025 17:07:59 +0000 (13:07 -0400)]
[dev.simd] simd, cmd/compile: add 128 bit select-from-pair
Using this name until a better one appears:
x.Select128FromPair(3, 2, y)
Includes test for constant and variable case.
Checks for unexpected immediates (using the zeroing flag,
which is not supported for this intrinsic) and panics.
We can assume it is available for users now. Let's switch to using it
when cgo is enabled.
Since neither LUCY nor legacy trybots provide illumos, I tested this
locally in a OpenIndiana VM, with and without osusergo, with cgo enabled
and disabled.
This is a continuation of CL 315278.
Fixes #14709
Change-Id: I922049e7ea5f450f6900914b30967e522e56cfc9
Reviewed-on: https://go-review.googlesource.com/c/go/+/702975 Reviewed-by: Andrew Stormont <andyjstormont@gmail.com>
Auto-Submit: Damien Neil <dneil@google.com> Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com> Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Cherry Mui <cherryyz@google.com> Reviewed-by: Damien Neil <dneil@google.com>
Auto-Submit: Kirill Kolyshkin <kolyshkin@gmail.com>
TryBot-Bypass: Kirill Kolyshkin <kolyshkin@gmail.com>
Cherry Mui [Fri, 26 Sep 2025 18:38:22 +0000 (14:38 -0400)]
[dev.simd] cmd/compile: don't require single use for SIMD load/store folding
For load and store on scalar values, we fold the address to the
load/stoer instruction without requiring the address having one
use. Do the same for SIMD, and remove the single use check.
Change-Id: Ie7d1bbae1b32bb8c069548197632edae36b419b9
Reviewed-on: https://go-review.googlesource.com/c/go/+/707137 Reviewed-by: Junyang Shao <shaojunyang@google.com> Reviewed-by: David Chase <drchase@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Michael Anthony Knyszek [Fri, 26 Sep 2025 17:05:43 +0000 (17:05 +0000)]
runtime: move TestReadMetricsSched to testprog
There are just too many flakes resulting from background pollution by
the testing package and other tests. Run in a subprocess where at least
the environment can be more tightly controlled.
Fixes #75049.
Change-Id: Iad59edaaf31268f1fcb77273f01317d963708fa6
Reviewed-on: https://go-review.googlesource.com/c/go/+/707155 Reviewed-by: Michael Pratt <mpratt@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
This adds methods SelectFromPair for {Int,Uint,Float}32x4
and SelectFromPairGrouped for {Int,Uint,Float}32x8.
Each of these has the signature
```
func(x T32xK.Method(a,b,c,d uint8, y T32xK) T32xK)
```
where a, b, c, d can be 0-7 and each one specifies an
element from the concatenated elements of x (0-3) and
y (4-7). When a, b, c, d are constants, 1 or 2
instructions are generated, otherwise, it's done the
harder-slower way with a function call.
qmuntal [Thu, 21 Aug 2025 08:45:23 +0000 (10:45 +0200)]
internal/poll: don't call SetFilePointerEx in Seek for overlapped handles
Overlapped handles don't have the file pointer updated when performing
I/O operations, so there is no need to call syscall.SetFilePointerEx in
FD.Seek. Updating the in-memory offset is sufficient.
Updates #74951 (provides a more complete fix)
Change-Id: Ibede6625cdbd501fc92cfdf8ce2782ec291af2b6
Reviewed-on: https://go-review.googlesource.com/c/go/+/698035
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Dmitri Shuralyov <dmitshur@google.com> Reviewed-by: Damien Neil <dneil@google.com>
Damien Neil [Thu, 25 Sep 2025 20:24:01 +0000 (13:24 -0700)]
mime/multipart: percent-encode CR and LF in header values to avoid CRLF injection
When provided with a field or file name containing newlines,
multipart.FileContentDisposition and other header-producing functions
could create an invalid header value.
In some scenarios, this could permit a malicious input to perform
a CRLF injection attack:
Percent-endode \r and \n characters in headers, as recommended by
https://html.spec.whatwg.org/multipage/form-control-infrastructure.html#multipart/form-data-encoding-algorithm
The above algorithm also recommends using percent-encoding for quotes,
but preserve the existing backslash-escape behavior for now.
Empirically, browsers understand backslash-escape in attribute values.
Fixes #75557
Change-Id: Ia203df6ef45a098070f3ebb17f9b6cf80c520ed4
Reviewed-on: https://go-review.googlesource.com/c/go/+/706677
Auto-Submit: Damien Neil <dneil@google.com> Reviewed-by: Nicholas Husin <nsh@golang.org> Reviewed-by: Nicholas Husin <husin@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Exceptionally, we decided to make a compliance-related change following
CMVP's updated Implementation Guidance on September 2nd.
The Security Policy will be updated to reflect the new zip hash.
mkzip.go has been modified to accept versions of the form vX.Y.Z-hash,
where the -hash suffix is ignored for fips140.Version() but used to
name the zip file and the unpacked cache directory.
The new zip is generated with
go run ../../src/cmd/go/internal/fips140/mkzip.go -b c2097c7c v1.0.0-c2097c7c
from c2097c7c which is the current release-branch.go1.24 head.
The full diff between the zip file contents is included below.
// PCT runs the named Pairwise Consistency Test (if operated in FIPS mode) and
-// returns any errors. If an error is returned, the key must not be used.
+// aborts the program (stopping the module input/output and entering the "error
+// state") if the test fails.
//
-// PCTs are mandatory for every key pair that is generated/imported, including
+// PCTs are mandatory for every generated (but not imported) key pair, including
// ephemeral keys (which effectively doubles the cost of key establishment). See
// Implementation Guidance 10.3.A Additional Comment 1.
//
@@ -66,17 +67,23 @@
//
// If a package p calls PCT during key generation, an invocation of that
// function should be added to fipstest.TestConditionals.
-func PCT(name string, f func() error) error {
+func PCT(name string, f func() error) {
if strings.ContainsAny(name, ",#=:") {
panic("fips: invalid self-test name: " + name)
}
if !Enabled {
- return nil
+ return
}
err := f()
if name == failfipscast {
err = errors.New("simulated PCT failure")
}
- return err
+ if err != nil {
+ fatal("FIPS 140-3 self-test failed: " + name + ": " + err.Error())
+ panic("unreachable")
+ }
+ if debug {
+ println("FIPS 140-3 PCT passed:", name)
+ }
}
diff -ru golang.org/fips140@v1.0.0/fips140/v1.0.0/ecdh/ecdh.go golang.org/fips140@v1.0.0-c2097c7c/fips140/v1.0.0-c2097c7c/ecdh/ecdh.go
--- golang.org/fips140@v1.0.0/fips140/v1.0.0/ecdh/ecdh.go 1980-01-10 00:00:00.000000000 +0100
+++ golang.org/fips140@v1.0.0-c2097c7c/fips140/v1.0.0-c2097c7c/ecdh/ecdh.go 1980-01-10 00:00:00.000000000 +0100
@@ -161,6 +161,27 @@
if err != nil {
continue
}
+
+ // A "Pairwise Consistency Test" makes no sense if we just generated the
+ // public key from an ephemeral private key. Moreover, there is no way to
+ // check it aside from redoing the exact same computation again. SP 800-56A
+ // Rev. 3, Section 5.6.2.1.4 acknowledges that, and doesn't require it.
+ // However, ISO 19790:2012, Section 7.10.3.3 has a blanket requirement for a
+ // PCT for all generated keys (AS10.35) and FIPS 140-3 IG 10.3.A, Additional
+ // Comment 1 goes out of its way to say that "the PCT shall be performed
+ // consistent [...], even if the underlying standard does not require a
+ // PCT". So we do it. And make ECDH nearly 50% slower (only) in FIPS mode.
+ fips140.PCT("ECDH PCT", func() error {
+ p1, err := c.newPoint().ScalarBaseMult(privateKey.d)
+ if err != nil {
+ return err
+ }
+ if !bytes.Equal(p1.Bytes(), privateKey.pub.q) {
+ return errors.New("crypto/ecdh: public key does not match private key")
+ }
+ return nil
+ })
+
return privateKey, nil
}
}
@@ -188,28 +209,6 @@
panic("crypto/ecdh: internal error: public key is the identity element")
}
- // A "Pairwise Consistency Test" makes no sense if we just generated the
- // public key from an ephemeral private key. Moreover, there is no way to
- // check it aside from redoing the exact same computation again. SP 800-56A
- // Rev. 3, Section 5.6.2.1.4 acknowledges that, and doesn't require it.
- // However, ISO 19790:2012, Section 7.10.3.3 has a blanket requirement for a
- // PCT for all generated keys (AS10.35) and FIPS 140-3 IG 10.3.A, Additional
- // Comment 1 goes out of its way to say that "the PCT shall be performed
- // consistent [...], even if the underlying standard does not require a
- // PCT". So we do it. And make ECDH nearly 50% slower (only) in FIPS mode.
- if err := fips140.PCT("ECDH PCT", func() error {
- p1, err := c.newPoint().ScalarBaseMult(key)
- if err != nil {
- return err
- }
- if !bytes.Equal(p1.Bytes(), publicKey) {
- return errors.New("crypto/ecdh: public key does not match private key")
- }
- return nil
- }); err != nil {
- panic(err)
- }
-
k := &PrivateKey{d: bytes.Clone(key), pub: PublicKey{curve: c.curve, q: publicKey}}
return k, nil
}
diff -ru golang.org/fips140@v1.0.0/fips140/v1.0.0/ecdsa/cast.go golang.org/fips140@v1.0.0-c2097c7c/fips140/v1.0.0-c2097c7c/ecdsa/cast.go
--- golang.org/fips140@v1.0.0/fips140/v1.0.0/ecdsa/cast.go 1980-01-10 00:00:00.000000000 +0100
+++ golang.org/fips140@v1.0.0-c2097c7c/fips140/v1.0.0-c2097c7c/ecdsa/cast.go 1980-01-10 00:00:00.000000000 +0100
@@ -51,8 +51,8 @@
}
}
-func fipsPCT[P Point[P]](c *Curve[P], k *PrivateKey) error {
- return fips140.PCT("ECDSA PCT", func() error {
+func fipsPCT[P Point[P]](c *Curve[P], k *PrivateKey) {
+ fips140.PCT("ECDSA PCT", func() error {
hash := testHash()
drbg := newDRBG(sha512.New, k.d, bits2octets(P256(), hash), nil)
sig, err := sign(c, k, drbg, hash)
diff -ru golang.org/fips140@v1.0.0/fips140/v1.0.0/ecdsa/ecdsa.go golang.org/fips140@v1.0.0-c2097c7c/fips140/v1.0.0-c2097c7c/ecdsa/ecdsa.go
--- golang.org/fips140@v1.0.0/fips140/v1.0.0/ecdsa/ecdsa.go 1980-01-10 00:00:00.000000000 +0100
+++ golang.org/fips140@v1.0.0-c2097c7c/fips140/v1.0.0-c2097c7c/ecdsa/ecdsa.go 1980-01-10 00:00:00.000000000 +0100
@@ -166,11 +166,6 @@
return nil, err
}
priv := &PrivateKey{pub: *pub, d: d.Bytes(c.N)}
- if err := fipsPCT(c, priv); err != nil {
- // This can happen if the application went out of its way to make an
- // ecdsa.PrivateKey with a mismatching PublicKey.
- return nil, err
- }
return priv, nil
}
@@ -203,10 +198,7 @@
},
d: k.Bytes(c.N),
}
- if err := fipsPCT(c, priv); err != nil {
- // This clearly can't happen, but FIPS 140-3 mandates that we check it.
- panic(err)
- }
+ fipsPCT(c, priv)
return priv, nil
}
diff -ru golang.org/fips140@v1.0.0/fips140/v1.0.0/ecdsa/hmacdrbg.go golang.org/fips140@v1.0.0-c2097c7c/fips140/v1.0.0-c2097c7c/ecdsa/hmacdrbg.go
--- golang.org/fips140@v1.0.0/fips140/v1.0.0/ecdsa/hmacdrbg.go 1980-01-10 00:00:00.000000000 +0100
+++ golang.org/fips140@v1.0.0-c2097c7c/fips140/v1.0.0-c2097c7c/ecdsa/hmacdrbg.go 1980-01-10 00:00:00.000000000 +0100
@@ -121,7 +121,7 @@
//
// This should only be used for ACVP testing. hmacDRBG is not intended to be
// used directly.
-func TestingOnlyNewDRBG(hash func() fips140.Hash, entropy, nonce []byte, s []byte) *hmacDRBG {
+func TestingOnlyNewDRBG[H fips140.Hash](hash func() H, entropy, nonce []byte, s []byte) *hmacDRBG {
return newDRBG(hash, entropy, nonce, plainPersonalizationString(s))
}
@@ -88,10 +85,6 @@
}
copy(priv.seed[:], seed)
precomputePrivateKey(priv)
- if err := fipsPCT(priv); err != nil {
- // This clearly can't happen, but FIPS 140-3 requires that we check.
- panic(err)
- }
return priv, nil
}
@@ -137,12 +130,6 @@
copy(priv.prefix[:], h[32:])
- if err := fipsPCT(priv); err != nil {
- // This can happen if the application messed with the private key
- // encoding, and the public key doesn't match the seed anymore.
- return nil, err
- }
-
return priv, nil
}
Alan Donovan [Fri, 12 Sep 2025 20:04:24 +0000 (16:04 -0400)]
std: remove unused declarations
One function was unused by mistake, and is now used as intended.
Change-Id: I42ae7481c6f794b310bdac656ea525bd882f146e
Reviewed-on: https://go-review.googlesource.com/c/go/+/706815
Auto-Submit: Alan Donovan <adonovan@google.com> Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org> Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Cherry Mui [Mon, 22 Sep 2025 15:57:19 +0000 (11:57 -0400)]
[dev.simd] simd, cmd/compile: mark simd vectors uncomparable
SIMD vector types are opqaue, and are expected to be operated with
methods. It is not always possible to compare the two vectors
efficiently. Instead of adding more magic to the compiler to
handle the == operator, mark the vector types uncomparable.
Change-Id: I4ca5d5e80ca7d8992dffa7b3c0386b75eb19cfa8
Reviewed-on: https://go-review.googlesource.com/c/go/+/705855 Reviewed-by: Junyang Shao <shaojunyang@google.com>
TryBot-Bypass: Cherry Mui <cherryyz@google.com> Reviewed-by: David Chase <drchase@google.com>
Michael Pratt [Thu, 25 Sep 2025 17:12:12 +0000 (10:12 -0700)]
sync/atomic: correct Uintptr.Or return doc
Uintptr.Or returns the old value, just like all of the other Or
functions. This was a typo in the original CL 544455.
Fixes #75607.
Change-Id: I260959e7e32e51f1152b5271df6cc51adfa02a4d
Reviewed-on: https://go-review.googlesource.com/c/go/+/706816 Reviewed-by: Michael Knyszek <mknyszek@google.com> Reviewed-by: Mauri de Souza Meneguzzo <mauri870@gmail.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Michael Pratt <mpratt@google.com>
Michael Pratt [Fri, 30 May 2025 21:05:41 +0000 (17:05 -0400)]
runtime: acquire/release C TSAN lock when calling cgo symbolizer/tracebacker
When calling into C via cmd/cgo, the generated code calls
_cgo_tsan_acquire / _cgo_tsan_release around the C call to report a
dummy lock to the C/C++ TSAN runtime. This is necessary because the
C/C++ TSAN runtime does not understand synchronization within Go and
would otherwise report false positive race reports. See the comment in
cmd/cgo/out.go for more details.
Various C functions in runtime/cgo also contain manual calls to
_cgo_tsan_acquire/release where necessary to suppress race reports.
However, the cgo symbolizer and cgo traceback functions called from
callCgoSymbolizer and cgoContextPCs, respectively, do not have any
instrumentation [1]. They call directly into user C functions with no
TSAN instrumentation.
This means they have an opportunity to report false race conditions. The
most direct way is via their argument. Both are passed a pointer to a
struct stored on the Go stack, and both write to fields of the struct.
If two calls are passed the same pointer from different threads, the C
TSAN runtime will think this is a race.
This is simple to achieve for the cgo symbolizer function, which the
new regression test does. callCgoSymbolizer is called on the standard
goroutine stack, so the argument is a pointer into the goroutine stack.
If the goroutine moves Ms between two calls, it will look like a race.
On the other hand, cgoContextPCs is called on the system stack. Each M
has a unique system stack, so for it to pass the same argument pointer
on different threads would require the first M to exit, free its stack,
and the same region of address space to be used as the stack for a new
M. Theoretically possible, but quite unlikely.
Both of these are addressed by providing a C wrapper in runtime/cgo that
calls _cgo_tsan_acquire/_cgo_tsan_release around calls to the symbolizer
and traceback functions.
There is a lot of room for future cleanup here. Most runtime/cgo
functions have manual instrumentation in their C implementation. That
could be removed in favor of instrumentation in the runtime. We could
even theoretically remove the instrumentation from cmd/cgo and move it
to cgocall. None of these are necessary, but may make things more
consistent and easier to follow.
[1] Note that the cgo traceback function called from the signal handler
via x_cgo_callers _does_ have manual instrumentation.
Fixes #73949.
Cq-Include-Trybots: luci.golang.try:gotip-freebsd-amd64,gotip-linux-amd64-longtest,gotip-windows-amd64-longtest
Change-Id: I6a6a636c9daa38f7fd00694af76b75cb93ba1886
Reviewed-on: https://go-review.googlesource.com/c/go/+/677955 Reviewed-by: Michael Knyszek <mknyszek@google.com>
Auto-Submit: Michael Pratt <mpratt@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Alan Donovan [Thu, 25 Sep 2025 16:41:08 +0000 (12:41 -0400)]
cmd/fix: remove all functionality
The buildtag fixer has been incorporated into the vet analyzer
of the same name; all other fixers were already no-ops since
CL 695855.
Fixes #73605
Updates #71859
Change-Id: I90b6c730849a5ecbac3e6fb6fc0e062b5de74831
Reviewed-on: https://go-review.googlesource.com/c/go/+/706758 Reviewed-by: Michael Matloob <matloob@golang.org> Reviewed-by: Michael Matloob <matloob@google.com>
Auto-Submit: Alan Donovan <adonovan@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Cherry Mui [Tue, 16 Sep 2025 17:34:19 +0000 (13:34 -0400)]
cmd/link: handle -w flag in external linking mode
Currently, when the -w flag is set, it doesn't actually disable
the debug info generation with in external linking mode. (It does
make the Go object have no debug info, but C objects may still
have.) Pass "-Wl,-S" to let the external linker disable debug info
generation.
Change-Id: I0fce56b9f23a45546b69b9e6dd027c5527b1bc87
Reviewed-on: https://go-review.googlesource.com/c/go/+/705857 Reviewed-by: David Chase <drchase@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Joel Sing [Tue, 9 Sep 2025 15:00:22 +0000 (01:00 +1000)]
cmd/internal/obj/riscv: remove ACFLWSP/ACFSWSP and ACFLW/ACFSW
These are RV32-only instructions that will not be implemented.
Updates #71105
Change-Id: Ie386fe36e56f1151bb8756088dd79804584317c0
Reviewed-on: https://go-review.googlesource.com/c/go/+/702395
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Meng Zhuo <mengzhuo1203@gmail.com> Reviewed-by: Mark Ryan <markdryan@rivosinc.com> Reviewed-by: Mark Freeman <markfreeman@google.com> Reviewed-by: Michael Pratt <mpratt@google.com>
image/jpeg: replace fdct.go and idct.go with new implementation in dct.go
The fdct.go and idct.go files were derived from the MPEG-SSG and
JPEG-IJG reference code and therefore carry licenses specific to those
groups. Various license checkers flag these files as potentially
problematic. The code is also not terribly well documented.
This CL fixes the license problem by adding a new, from-scratch
implementation using a different algorithm. As a bonus, the new code
is both faster and more accurate than the old encumbered code.
On speed, the new code is up to 20% faster; benchmarks below.
On accuracy, in the set of blocks used in the test, we can measure
the number of output values that are off-by-one from the exact rounded answer.
The old FDCT was off in 8.6% of values; the new one is off in 2.5%.
The old IDCT was off in 1.4% of values; the new one is off in 1.2%.
goos: darwin
goarch: arm64
pkg: image/jpeg
cpu: Apple M3 Pro
│ old │ new │
│ sec/op │ sec/op vs base │
FDCT-12 619.6n ± 3% 586.5n ± 1% -5.34% (p=0.000 n=10)
IDCT-12 752.4n ± 4% 628.0n ± 1% -16.54% (p=0.000 n=10)
goos: linux
goarch: amd64
cpu: Intel(R) Xeon(R) CPU @ 2.30GHz
│ old │ new │
│ sec/op │ sec/op vs base │
FDCT-16 1.817µ ± 0% 1.542µ ± 0% -15.11% (p=0.000 n=10)
IDCT-16 1.897µ ± 0% 1.514µ ± 0% -20.22% (p=0.000 n=10)
goos: linux
goarch: arm64
cpu: whatever gotip-linux-arm64 has
│ old │ new │
│ sec/op │ sec/op vs base │
FDCT-8 1.844µ ± 0% 1.847µ ± 0% +0.14% (p=0.000 n=10)
IDCT-8 2.127µ ± 0% 1.973µ ± 0% -7.26% (p=0.000 n=10)
Change-Id: Ie6d14103c8478ba5a779f234da84f345828eb925
Reviewed-on: https://go-review.googlesource.com/c/go/+/705518
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Alan Donovan <adonovan@google.com>
Auto-Submit: Russ Cox <rsc@golang.org> Reviewed-by: Nigel Tao <nigeltao@google.com>
image/jpeg: correct and test reference slowFDCT and slowIDCT
The reference implementations slowFDCT and slowIDCT were not
rounding correctly, making the test not as good as it could be.
Before, the real implementations were required to always produce
values within ±2 of the reference; now, with no changes,
the real implementations produce values within ±1 of the (corrected)
reference.
Also tighten the test to return an error not just on a single value
exceeding tolerance but also on too many values at exactly that
tolerance.
Change-Id: I3dd6ca7582178fef972fb812d848f7a0158a6ed8
Reviewed-on: https://go-review.googlesource.com/c/go/+/705517
Auto-Submit: Russ Cox <rsc@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Alan Donovan <adonovan@google.com>
Keith Randall [Tue, 23 Sep 2025 23:31:26 +0000 (16:31 -0700)]
cmd/compile: don't rely on loop info when there are irreducible loops
Loop information is sketchy when there are irreducible loops.
Sometimes blocks inside 2 loops can be recorded as only being part of
the outer loop. That causes tighten to move values that want to move
into such a block to move out of the loop altogether, breaking the
invariant that operations have to be scheduled after their args.
Fixes #75569
Change-Id: Idd80e6d2268094b8ae6387563081fdc1e211856a
Reviewed-on: https://go-review.googlesource.com/c/go/+/706355 Reviewed-by: David Chase <drchase@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Keith Randall <khr@google.com>
Michael Matloob [Mon, 17 Mar 2025 15:45:52 +0000 (11:45 -0400)]
runtime: add specialized malloc functions for sizes up to 512 bytes
This CL adds a generator function in runtime/_mkmalloc to generate
specialized mallocgc functions for sizes up throuht 512 bytes. (That's
the limit where it's possible to end up in the no header case when there
are scan bits, and where the benefits of the specialized functions
significantly diminish according to microbenchmarks). If the
specializedmalloc GOEXPERIMENT is turned on, mallocgc will call one of
these functions in the no header case.
malloc_generated.go is the generated file containing the specialized
malloc functions.
malloc_stubs.go contains the templates that will be stamped to create
the specialized malloc functions.
malloc_tables_generated contains the tables that mallocgc will use to
select the specialized function to call.
I've had to update the two stdlib_test.go files to account for the new
submodule mkmalloc is in. mprof_test accounts for the changes in the
stacks since different functions can be called in some cases.
I still need to investigate heapsampling.go.
Change-Id: Ia0f68dccdf1c6a200554ae88657cf4d686ace819
Reviewed-on: https://go-review.googlesource.com/c/go/+/665835 Reviewed-by: Michael Knyszek <mknyszek@google.com> Reviewed-by: Michael Matloob <matloob@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Michael Anthony Knyszek [Fri, 15 Aug 2025 17:09:05 +0000 (17:09 +0000)]
runtime: eliminate global span queue [green tea]
This change removes the locked global span queue and replaces the
fixed-size local span queue with a variable-sized local span queue. The
variable-sized local span queue grows as needed to accomodate local
work. With no global span queue either, GC workers balance work amongst
themselves by stealing from each other.
The new variable-sized local span queues are inspired by the P-local
deque underlying sync.Pool. Unlike the sync.Pool deque, however, both
the owning P and stealing Ps take spans from the tail, making this
incarnation a strict queue, not a deque. This is intentional, since we
want a queue-like order to encourage objects to accumulate on each span.
These variable-sized local span queues are crucial to mark termination,
just like the global span queue was. To avoid hitting the ragged barrier
too often, we must check whether any Ps have any spans on their
variable-sized local span queues. We maintain a per-P atomic bitmask
(another pMask) that contains this state. We can also use this to speed
up stealing by skipping Ps that don't have any local spans.
The variable-sized local span queues are slower than the old fixed-size
local span queues because of the additional indirection, so this change
adds a non-atomic local fixed-size queue. This risks getting work stuck
on it, so, similarly to how workbufs work, each worker will occasionally
dump some spans onto its local variable-sized queue. This scales much
more nicely than dumping to a global queue, but is still visible to all
other Ps.
For #73581.
Change-Id: I814f54d9c3cc7fa7896167746e9823f50943ac22
Reviewed-on: https://go-review.googlesource.com/c/go/+/700496 Reviewed-by: Michael Pratt <mpratt@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>