Some iOS environments may support exec. wasip1 and js do not, but
trying to exec on those platforms is inexpensive anyway and gives
better test coverage for the ios path.
Change-Id: I4baffb2ef5dc7d81e6a260f69033bfb229f13d92
Reviewed-on: https://go-review.googlesource.com/c/go/+/486275
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Bryan Mills <bcmills@google.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Auto-Submit: Bryan Mills <bcmills@google.com>
The modification of these rules is optimization to load/store global
variables. If there are a sequence of loads/stores nearby a global
variable address, the address can only be loaded from GOT once instead
of every time.
For #43264
Change-Id: Idedaf6c81f085955371320f51bca148ffb42a2d8
Reviewed-on: https://go-review.googlesource.com/c/go/+/348732
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Ian Lance Taylor <iant@google.com>
Auto-Submit: Ian Lance Taylor <iant@google.com>
As part of developing #57411, we ran into cases where a flag was
defined in one package init and Set in another package init, and there
was no init ordering implied by the spec between those two
packages. Changes in initialization ordering as part of #57411 caused
a Set to happen before the definition, which makes the Set silently
fail.
This CL makes the Set fail loudly in that situation.
Currently Set *does* fail kinda quietly in that situation, in that it
returns an error. (It seems that no one checks the error from Set,
at least for string flags.) Ian suggsted that instead we panic at
the definition site if there was previously a Set called on that
(at the time undefined) flag.
So Set on an undefined flag is ok and returns an error (as before),
but defining a flag which has already been Set causes a panic. (The
API for flag definition has no way to return an error, and does
already panic in some situations like a duplicate definition.)
Update #57411
Change-Id: I39b5a49006f9469de0b7f3fe092afe3a352e4fcb
Reviewed-on: https://go-review.googlesource.com/c/go/+/480215
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Ian Lance Taylor <iant@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Allow switching to wasmtime through the GOWASIRUNTIME variable. This
will allow builders to run the wasip1 standard library tests against
the wasmtime WASI runtime.
For #59583
Change-Id: I4d5200df7bb27b66e041f00e89d4c2e585f5da7c
Reviewed-on: https://go-review.googlesource.com/c/go/+/485615
Reviewed-by: Bryan Mills <bcmills@google.com>
TryBot-Bypass: Johan Brandhorst-Satzkorn <johan.brandhorst@gmail.com>
Run-TryBot: Johan Brandhorst-Satzkorn <johan.brandhorst@gmail.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
__tsan_fini will call exit which will call destructors which
may in principle call back into Go functions. Prepare the scheduler
by calling entersyscall before __tsan_fini.
Fixes#59711
Change-Id: Ic4df8fba3014bafa516739408ccfc30aba4f22ad
Reviewed-on: https://go-review.googlesource.com/c/go/+/486615
Reviewed-by: Michael Pratt <mpratt@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Auto-Submit: Ian Lance Taylor <iant@google.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Ian Lance Taylor <iant@google.com>
Memory op combining is currently done using arch-specific rewrite rules.
Instead, do them as a arch-independent rewrite pass. This ensures that
all architectures (with unaligned loads & stores) get equal treatment.
This removes a lot of rewrite rules.
The new pass is a bit more comprehensive. It handles things like out-of-order
writes and is careful not to apply partial optimizations that then block
further optimizations.
Change-Id: I780ff3bb052475cd725a923309616882d25b8d9e
Reviewed-on: https://go-review.googlesource.com/c/go/+/478475
Reviewed-by: Keith Randall <khr@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
This test fails when run on ios. (Although ios does not normally
support "exec", in the corellium environment it does.)
For #26061.
Change-Id: Idfdc53758aaabf0cb87ae50f9a4666deebf57fd6
Reviewed-on: https://go-review.googlesource.com/c/go/+/487355
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Bryan Mills <bcmills@google.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
This change resolves an issue where checkdead could result in a double lock when shedtrace is enabled. This fix involves adding unlocks before all throws in the checkdead function to ensure the scheduler lock is properly released.
Fixes#59758
Change-Id: If3ddf9969f4582c3c88dee9b9ecc355a63958103
Reviewed-on: https://go-review.googlesource.com/c/go/+/487375
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Auto-Submit: Michael Pratt <mpratt@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
Leaked goroutines are the only explanation I can think of for excess
allocs in TestDiscard, and TestOutputRace is the only place I can see
where the log package leaks goroutines. Let's fix that leak and see if
it eliminates the TestDiscard flakes.
Fixes#58797 (maybe).
Change-Id: I2d54dcba3eb52bd10a62cd1c380131add6a2f651
Reviewed-on: https://go-review.googlesource.com/c/go/+/487356
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
The current definitions of StackLimit and StackGuard only indirectly
specify the NOSPLIT stack limit and duplicate a literal constant
(928). Currently, they define the stack guard delta, and from there
compute the NOSPLIT limit.
Rationalize these by defining a new constant, abi.StackNosplitBase,
which consolidates and directly specifies the NOSPLIT stack limit (in
the default case). From this we then compute the stack guard delta,
inverting the relationship between these two constants. While we're
here, we rename StackLimit to StackNosplit to make it clearer what's
being limited.
This change does not affect the values of these constants in the
default configuration. It does slightly change how
StackGuardMultiplier values other than 1 affect the constants, but
this multiplier is a pretty rough heuristic anyway.
before after
stackNosplit 800 800
_StackGuard 928 928
stackNosplit -race 1728 1600
_StackGuard -race 1856 1728
For #59670.
Change-Id: Ia94094c5e47897e7c088d24b4a5e33f5c2768db5
Reviewed-on: https://go-review.googlesource.com/c/go/+/486976
Auto-Submit: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
We also rename the constants related to unsafe-points: currently, they
follow the same naming scheme as the PCDATA table indexes, but are not
PCDATA table indexes.
For #59670.
Change-Id: I06529fecfae535be5fe7d9ac56c886b9106c74fd
Reviewed-on: https://go-review.googlesource.com/c/go/+/485497
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Austin Clements <austin@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
The previous assert triggers whenever the 40-character git commit
contains the substring "999", which happens with a probability
decidedly greater than zero.
For #57001.
Change-Id: If0f1bc1a3dd0e6b7e66768d0cf3a79545ee4e5ed
Reviewed-on: https://go-review.googlesource.com/c/go/+/486399
Reviewed-by: Michael Pratt <mpratt@google.com>
Auto-Submit: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
The initial purpose of PCALIGN was to identify code
where it would be beneficial to align code for performance,
but avoid cases where too many NOPs were added. On p10, it
is now necessary to enforce a certain alignment in some
cases, so the behavior of PCALIGN needs to be slightly
different. Code will now be aligned to the value specified
on the PCALIGN instruction regardless of number of NOPs added,
which is more intuitive and consistent with power assembler
alignment directives.
This also adds 64 as a possible alignment value.
The existing values used in PCALIGN were modified according to
the new behavior.
A testcase was updated and performance testing was done to
verify that this does not adversely affect performance.
Change-Id: Iad1cf5ff112e5bfc0514f0805be90e24095e932b
Reviewed-on: https://go-review.googlesource.com/c/go/+/485056
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Archana Ravindar <aravind5@in.ibm.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
Reviewed-by: Paul Murphy <murp@ibm.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
A CI machine has been set up to verify GOPPC64=power10 on ppc64/linux.
This should be sufficient to verify the PCrel relocation support works
for BE.
Note, power10/ppc64/linux is an oddball case. Today, it can only link
internally. Furthermore, all PCrel relocs are resolved at link time,
so it works despite ELFv1 having no official support for PCrel relocs
today.
Change-Id: Ibf79df69406ec6f9352c9d7d941ad946dba74e73
Reviewed-on: https://go-review.googlesource.com/c/go/+/485075
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
Today Current may fail if the binary is not built with cgo
and USER and/or HOME is not set in the environment.
That should not cause the test to fail.
After this change,
GOCACHE=$(go env GOCACHE) CGO_ENABLED=0 USER= HOME= go test os/user
now passes on linux/amd64.
For #59583.
Change-Id: Id290cd1088051e930d73f0dd554177124796e8f2
Reviewed-on: https://go-review.googlesource.com/c/go/+/487015
Reviewed-by: Michael Pratt <mpratt@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
Reviewed-by: Johan Brandhorst-Satzkorn <johan.brandhorst@gmail.com>
Auto-Submit: Bryan Mills <bcmills@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Forgot to 'git add' this test case in the CL.
Change-Id: Idde1d3d4525a99bdab0d3d69ac635a96a7cd5d73
Reviewed-on: https://go-review.googlesource.com/c/go/+/487335
Auto-Submit: Russ Cox <rsc@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
ssagen.ssafn already holds the ir.Func, and ssa.Frontend.SetWBPos and
ssa.Frontend.Lsym are simple wrappers around parts of the ir.Func.
Expose the ir.Func through ssa.Frontend, allowing us to remove these
wrapper methods and allowing future access to additional features of the
ir.Func if needed.
While we're here, drop ssa.Frontend.Line, which is unused.
For #58298.
Change-Id: I30c4cbd2743e9ad991d8c6b388484a7d1e95f3ae
Reviewed-on: https://go-review.googlesource.com/c/go/+/484215
Auto-Submit: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
Frame pointer unwinding during execution tracing sometimes crashes.
Until this is diagnosed and fixed, it should be turned off by default.
Updates #59692
Change-Id: I0f2ca24b6d48435b0acfd3da8e4f25b9cfa4ec19
Reviewed-on: https://go-review.googlesource.com/c/go/+/486382
Reviewed-by: Felix Geisendörfer <felix.geisendoerfer@datadoghq.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Nick Ripley <nick.ripley@datadoghq.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Michael Pratt <mpratt@google.com>
The scavenge index currently doesn't guard against overflow, and CL
436395 removed the minHeapIdx optimization that allows the chunk scan to
skip scanning chunks that haven't been mapped for the heap, and are only
available as a consequence of chunks' mapped region being rounded out to
a page on both ends.
Because the 0'th chunk is never mapped, minHeapIdx effectively prevents
overflow, fixing the iOS breakage.
This change also refactors growth and initialization a little bit to
decouple it from pageAlloc a bit and share code across platforms.
Change-Id: If7fc3245aa81cf99451bf8468458da31986a9b0a
Reviewed-on: https://go-review.googlesource.com/c/go/+/486695
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Add a struct called Source that holds the function, file and line
of a location in the program's source code.
When HandleOptions.AddSource is true, the ReplaceAttr function will
get an Attr whose key is SourceKey and whose value is a *Source.
We use *Source instead of Source to save an allocation. The pointer
and the value each cause one allocation up front: the pointer when it
is created, and the value when it is assigned to the `any` field of a
slog.Value (handle.go:283). If a ReplaceAttr function wanted to modify
a Source value, it would have to create a new slog.Value to return,
causing a second allocation, but the function can modify a *Source in
place.
TextHandler displays a Source as "file:line".
JSONHandler displays a Source as a group of its non-zero fields.
This replaces the previous design, where source location was always a
string with the format "file:line". The new design gives users more
control over how to output and consume source locations.
Fixes#59280.
Change-Id: I84475abd5ed83fc354b50e34325c7b246cf327c7
Reviewed-on: https://go-review.googlesource.com/c/go/+/486376
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Jonathan Amsterdam <jba@google.com>
Reviewed-by: Alan Donovan <adonovan@google.com>
During bootstrapping, cmd/dist writes a file indicating which
GOOS/GOARCH combinations are valid, and which support cgo-enabled
builds. That information previously went into the go/build package,
but today it fits in more naturally in the internal/platform package
(which already has a number of functions indicating feature support
for GOOS/GOARCH combinations).
Moreover, as of CL 450739 the cmd/go logic for determining whether to
use cgo is somewhat more nuanced than the go/build logic: cmd/go
checks for the presence of a C compiler, whereas go/build does not
(mostly because it determines its configuration at package-init time,
and checking $PATH for a C compiler is somewhat expensive).
To simplify this situation, this change:
- consolidates the “cgo supported” check in internal/platform
(alongside many other platform-support checks) instead of making
it a one-off in go/build,
- and updates a couple of tests to use testenv.HasCGO instead of
build.Default.CgoEnabled to decide whether to test a cgo-specific
behavior.
For #58884.
For #59500.
Change-Id: I0bb2502dba4545a3d98c9e915727382ce536a0f3
Reviewed-on: https://go-review.googlesource.com/c/go/+/483695
Auto-Submit: Bryan Mills <bcmills@google.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Bryan Mills <bcmills@google.com>
Refactor io.ReadAll to check for realloc of the buffer only after the
first read.
Fixes: #59702
Change-Id: I93b99139e6756f21738d47e7d9ad08e1d167258e
Reviewed-on: https://go-review.googlesource.com/c/go/+/486236
Auto-Submit: Emmanuel Odeke <emmanuel@orijtech.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@google.com>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
This reverts commit CL 486379.
Submitted out of order and breaks bootstrap.
Change-Id: Ie20a61cc56efc79a365841293ca4e7352b02d86b
Reviewed-on: https://go-review.googlesource.com/c/go/+/486917
TryBot-Bypass: Austin Clements <austin@google.com>
Reviewed-by: David Chase <drchase@google.com>
This reverts commit CL 486380.
Submitted out of order and breaks bootstrap.
Change-Id: I67bd225094b5c9713b97f70feba04d2c99b7da76
Reviewed-on: https://go-review.googlesource.com/c/go/+/486916
Reviewed-by: David Chase <drchase@google.com>
TryBot-Bypass: Austin Clements <austin@google.com>
This reverts commit CL 486381.
Submitted out of order and breaks bootstrap.
Change-Id: Ia472111cb966e884a48f8ee3893b3bf4b4f4f875
Reviewed-on: https://go-review.googlesource.com/c/go/+/486915
Reviewed-by: David Chase <drchase@google.com>
TryBot-Bypass: Austin Clements <austin@google.com>
The current definitions of StackLimit and StackGuard only indirectly
specify the NOSPLIT stack limit and duplicate a literal constant
(928). Currently, they define the stack guard delta, and from there
compute the NOSPLIT limit.
Rationalize these by defining a new constant, abi.StackNosplitBase,
which consolidates and directly specifies the NOSPLIT stack limit (in
the default case). From this we then compute the stack guard delta,
inverting the relationship between these two constants. While we're
here, we rename StackLimit to StackNosplit to make it clearer what's
being limited.
This change does not affect the values of these constants in the
default configuration. It does slightly change how
StackGuardMultiplier values other than 1 affect the constants, but
this multiplier is a pretty rough heuristic anyway.
before after
stackNosplit 800 800
_StackGuard 928 928
stackNosplit -race 1728 1600
_StackGuard -race 1856 1728
For #59670.
Change-Id: Ibe20825ebe0076bbd7b0b7501177b16c9dbcb79e
Reviewed-on: https://go-review.googlesource.com/c/go/+/486380
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
We really need 3 mix steps between the data being hashed and the output.
One mix can only spread a 1 bit change to 32 bits. The second mix
can spread to all 128 bits, but the spread is not complete. A third mix
spreads out ~evenly to all 128 bits.
The amd64 version has 3 mix steps.
Fixes#59643
Change-Id: I54ad8686ca42bcffb6d0ec3779d27af682cc96e9
Reviewed-on: https://go-review.googlesource.com/c/go/+/486616
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Currently these fields are uninitialized causing failures on aix-ppc64,
which has a slightly oddly-defined address space compared to the rest.
Change-Id: I2aa46731174154dce86c2074bd0b00eef955d86d
Reviewed-on: https://go-review.googlesource.com/c/go/+/486655
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Bypass: Michael Knyszek <mknyszek@google.com>
The gVisor team reported a regression in their checkers,
which don't set Config.GoVersion, processing files that say
//go:build go1.13 but still use 'any' (which happened in Go 1.18).
That situation should continue to work, since it worked before,
so add a special case for not knowing the GoVersion.
Change-Id: I8820d8ccbdf76d304e2c7e45f6aaa993ff3d16a6
Reviewed-on: https://go-review.googlesource.com/c/go/+/486398
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@google.com>
Remove calls to Value.Resolve from Record.AddAttrs, Record.Add and Logger.With.
Handlers must resolve values themselves; document that in Handler.
Call Value.Resolve in the built-in handlers.
Updates #59292.
Change-Id: I00ba2131be0b16e3b1a22741249fd6f81c3efde1
Reviewed-on: https://go-review.googlesource.com/c/go/+/486375
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Jonathan Amsterdam <jba@google.com>
Reviewed-by: Alan Donovan <adonovan@google.com>
Found by running
$ go run golang.org/x/tools/go/analysis/passes/nilness/cmd/nilness@latest std
No actual bugs--other than one panic(nil)--but a
few places where error nilness was unclear.
Change-Id: Ia916ba30f46f29c1bcf928cc62280169b922463a
Reviewed-on: https://go-review.googlesource.com/c/go/+/486675
Reviewed-by: Ian Lance Taylor <iant@google.com>
Run-TryBot: Alan Donovan <adonovan@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Alan Donovan <adonovan@google.com>
The point of DialWithTimeout seems to be to test what happens when the
connection times out during handshake. However, the test wasn't
actually verifying that the connection made it into the handshake at
all. That would not only fail to test the intended behavior, but also
leak the Accept goroutine until arbitrarily later, at which point it
may call t.Error after the test t is already done.
Instead, we now:
- retry the test with a longer timeout if we didn't accept a
connection, and
- wait for the Accept goroutine to actually complete when the test
finishes.
Fixes#59646.
Change-Id: Ie56ce3297e2c183c02e67b8f6b26a71e50964558
Reviewed-on: https://go-review.googlesource.com/c/go/+/485115
Run-TryBot: Bryan Mills <bcmills@google.com>
Reviewed-by: Roland Shoemaker <roland@golang.org>
Auto-Submit: Bryan Mills <bcmills@google.com>
Commit-Queue: Bryan Mills <bcmills@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Add an AfterFunc function, which registers a function to run after
a context has been canceled.
Add support for contexts that implement an AfterFunc method, which
can be used to avoid the need to start a new goroutine watching
the Done channel when propagating cancellation signals.
Fixes#57928
Change-Id: If0b2cdcc4332961276a1ff57311338e74916259c
Reviewed-on: https://go-review.googlesource.com/c/go/+/482695
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Damien Neil <dneil@google.com>
Reviewed-by: Sameer Ajmani <sameer@golang.org>
Previously, all.bash and all.bat restored the original $PATH before
calling 'dist banner', so that it would do its job of checking whether
the user still needs to add $GOROOT/bin to their $PATH. That worked for
those scripts, but had no effect on make.bash nor make.bat.
Instead of trying to extend that logic to more scripts, change the
approach to provide dist an unmodified copy of $PATH via an internal
to dist environment variable $DIST_UNMODIFIED_PATH. The make.bash and
make.bat scripts happen to use dist env -p to modify $PATH, making it
viable to add the internal variable there instead of in each script.
It currently works by adding semicolon terminators to dist env output
so that make.bash's 'eval $(dist env -p)' works as before but is able to
export DIST_UNMODIFIED_PATH for following dist invocations to observe.
Nothing needs to be done for Windows since its 'set ENV=val' format
already has that effect.
Plan 9 doesn't use the -p flag of dist env, and checks that GOROOT/bin
is bound before /bin rather than looking at the $PATH env var like other
OSes, so it may not have this bug. I don't have easy access to Plan 9
and haven't tried to confirm.
Fixes#42563.
Change-Id: I74691931167e974a930f7589d22a48bb6b931163
Reviewed-on: https://go-review.googlesource.com/c/go/+/485896
Reviewed-by: Bryan Mills <bcmills@google.com>
Run-TryBot: Dmitri Shuralyov <dmitshur@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Auto-Submit: Dmitri Shuralyov <dmitshur@golang.org>
This change makes it so that on Linux the Go runtime explicitly marks
page heap memory as either available to be backed by hugepages or not
using heuristics based on density.
The motivation behind this change is twofold:
1. In default Linux configurations, khugepaged can recoalesce hugepages
even after the scavenger breaks them up, resulting in significant
overheads for small heaps when their heaps shrink.
2. The Go runtime already has some heuristics about this, but those
heuristics appear to have bit-rotted and result in haphazard
hugepage management. Unlucky (but otherwise fairly dense) regions of
memory end up not backed by huge pages while sparse regions end up
accidentally marked MADV_HUGEPAGE and are not later broken up by the
scavenger, because it already got the memory it needed from more
dense sections (this is more likely to happen with small heaps that
go idle).
In this change, the runtime uses a new policy:
1. Mark all new memory MADV_HUGEPAGE.
2. Track whether each page chunk (4 MiB) became dense during the GC
cycle. Mark those MADV_HUGEPAGE, and hide them from the scavenger.
3. If a chunk is not dense for 1 full GC cycle, make it visible to the
scavenger.
4. The scavenger marks a chunk MADV_NOHUGEPAGE before it scavenges it.
This policy is intended to try and back memory that is a good candidate
for huge pages (high occupancy) with huge pages, and give memory that is
not (low occupancy) to the scavenger. Occupancy is defined not just by
occupancy at any instant of time, but also occupancy in the near future.
It's generally true that by the end of a GC cycle the heap gets quite
dense (from the perspective of the page allocator).
Because we want scavenging and huge page management to happen together
(the right time to MADV_NOHUGEPAGE is just before scavenging in order to
break up huge pages and keep them that way) and the cost of applying
MADV_HUGEPAGE and MADV_NOHUGEPAGE is somewhat high, the scavenger avoids
releasing memory in dense page chunks. All this together means the
scavenger will now more generally release memory on a ~1 GC cycle delay.
Notably this has implications for scavenging to maintain the memory
limit and the runtime/debug.FreeOSMemory API. This change makes it so
that in these cases all memory is visible to the scavenger regardless of
sparseness and delays the page allocator in re-marking this memory with
MADV_NOHUGEPAGE for around 1 GC cycle to mitigate churn.
The end result of this change should be little-to-no performance
difference for dense heaps (MADV_HUGEPAGE works a lot like the default
unmarked state) but should allow the scavenger to more effectively take
back fragments of huge pages. The main risk here is churn, because
MADV_HUGEPAGE usually forces the kernel to immediately back memory with
a huge page. That's the reason for the large amount of hysteresis (1
full GC cycle) and why the definition of high density is 96% occupancy.
Fixes#55328.
Change-Id: I8da7998f1a31b498a9cc9bc662c1ae1a6bf64630
Reviewed-on: https://go-review.googlesource.com/c/go/+/436395
Reviewed-by: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
