The syscall.AllThreadsSyscall() fixup mechanism needs to cooperate
with signal handling to ensure a notetsleepg() thread can wake up
to run the mDoFixup() function.
Fixes#43149
Change-Id: I6651b25bc44a4de47d3fb71d0293d51aef8b79c7
Reviewed-on: https://go-review.googlesource.com/c/go/+/277434
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Austin Clements <austin@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
We want almost all cmd subdirectories anyway, and relative to the cost
of the rest of toolchain bootstrapping, copying/rewriting a few extra
source files is way cheaper than the engineering cost of forgetting to
maintain these lists as we split out new packages.
While here, also add cmd/internal/archive (and make it compile with Go
1.4) because it'll be needed in subsequent refactorings anyway; and
skip files starting with # (emacs temporary files) and test files
ending with _test.go.
Change-Id: Ic86e680a5fdfaecd617c36d5d04413293b2d6f52
Reviewed-on: https://go-review.googlesource.com/c/go/+/279832
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Under linux+cgo, OS threads are launched via pthread_create().
This abstraction, under linux, requires we avoid blocking
signals 32,33 and 34 indefinitely because they are needed to
reliably execute POSIX-semantics threading in glibc and/or musl.
When blocking signals the go runtime generally re-enables them
quickly. However, when a thread exits (under cgo, this is
via a return from mstart()), we avoid a deadlock in C-code by
not blocking these three signals.
Fixes#42494
Change-Id: I02dfb2480a1f97d11679e0c4b132b51bddbe4c14
Reviewed-on: https://go-review.googlesource.com/c/go/+/269799
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Adjust the new regabi code a bit to make the rewrites apply cleanly.
Change-Id: Ice5378e94d94ab45ca0572f44ab8c94b847271b8
Reviewed-on: https://go-review.googlesource.com/c/go/+/279530
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
By default (and with -mod=readonly), the go command imports an error
if a package provided by an implicitly required module is
imported by a package in the main module. This import requires an
update to go.mod: the module must be required explicitly.
The package loader now provides a hint that 'go get' should be run on
the importing package. This is preferred to 'go get' on the imported
package, since that would add an "// indirect" requirement.
For #43131
Change-Id: I0b353ce8ac8c4ddf1a9863544dfaf6c1964daf42
Reviewed-on: https://go-review.googlesource.com/c/go/+/279528
Trust: Jay Conrod <jayconrod@google.com>
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
The Go PE linker does not support enough generalized PE logic to
properly handle .rsrc sections gracefully. Instead a few things are
special cased for these. The linker also does not support PE's "grouped
sections" features, in which input objects have several named sections
that are sorted, merged, and renamed in the output file. In the past,
more sophisticated support for resources or for PE features like grouped
sections have not been necessary, as Go's own object formats are pretty
vanilla, and GNU binutils also produces pretty vanilla objects where all
sections are already merged.
However, GNU binutils is lagging with arm support, and here LLVM has
picked up the slack. In particular, LLVM has its own rc/cvtres combo,
which are glued together in mingw LLVM distributions as windres, a
command line compatible tool with binutils' windres, which supports arm
and arm64. But there's a key difference between binutils' windres and
LLVM's windres: the LLVM one uses proper grouped sections.
So, this commit adds grouped sections support for resource sections to
the linker. We don't attempt to plumb generic support for grouped
sections, just as there isn't generic support already for what resources
require. Instead we augment the resource handling logic to deal with
standard two-section resource objects.
We also add a test for this, akin to the current test for more vanilla
binutils resource objects, and make sure that the rsrc tests are always
performed.
Fixes#42866.
Fixes#43182.
Change-Id: I059450021405cdf2ef1c195ddbab3960764ad711
Reviewed-on: https://go-review.googlesource.com/c/go/+/268337
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Trust: Alex Brainman <alex.brainman@gmail.com>
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
The mid-case n := n.(*ir.AssignExpr) does not lend itself
well to pulling the code into a new function, because n will
be a function argument and will not be redeclarable.
Change-Id: I673f2aa37eea64b083725326ed3fa36447bcc7af
Reviewed-on: https://go-review.googlesource.com/c/go/+/279426
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Sets up for removing Func from Node interface.
That means that once the Name reorg is done,
which will let us remove Name, Sym, and Val,
Node will be basically a minimal interface.
Passes buildall w/ toolstash -cmp.
Change-Id: I6e87897572debd7f8e29b4f5167763dc2792b408
Reviewed-on: https://go-review.googlesource.com/c/go/+/279484
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Misc cleanup for splitting package gc: API tweaks
and boundary adjustments.
The change in ir.NewBlockStmt makes it a drop-in
replacement for liststmt.
Change-Id: I9455fe8ccae7d71fe8ccf390ac96672389bf4f3d
Reviewed-on: https://go-review.googlesource.com/c/go/+/279305
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Move various code out of Main itself and into helper functions
that can be moved into other packages as package gc splits up.
Similarly, move order and instrument inside walk to reduce the amount
of API surface needed from the eventual package walk.
Change-Id: I7849258038c6e39625a0385af9c0edd6a3b654a1
Reviewed-on: https://go-review.googlesource.com/c/go/+/279304
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Abstract the typecheck API a bit more so that it is
easier to move into a new package.
Change-Id: Ia0a0146151fa7f6073113e68a2c3f6e42a5d0ad8
Reviewed-on: https://go-review.googlesource.com/c/go/+/279303
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Having a global MaxWidth lets us avoid needing to
refer to thearch from split-out packages when all
they need is thearch.MAXWIDTH.
And make a couple interface changes to let dowidth
avoid importing package ir directly.
Then it can move into package types.
Change-Id: I2c12e8e22252597530e648848320e19bdd490a01
Reviewed-on: https://go-review.googlesource.com/c/go/+/279302
Trust: Russ Cox <rsc@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
In issue11656.go, it tests that if the runtime can get a
reasonable traceback when it faults at a non-function PC. It does
it by jumping to an address that contains an illegal or trap
instruction. When it traps, the SIGTRAP crashes the runtime.
This CL changes it to use an instruction that triggers SIGSEGV.
This is due to two reasons:
- currently, the handling of bad PC is done by preparePanic,
which is only used for a panicking signal (SIGSEGV, SIGBUS,
SIGFPE), not a fatal signal (e.g. SIGTRAP).
- the test uses defer+recover to get a traceback, which only
works for panicking signals, not fatal signals.
Ideally, we should handle all kinds of faults (SIGSEGV, SIGBUS,
SIGILL, SIGTRAP, etc.) with a nice traceback. I'll leave this
for the future.
This CL also adds RISCV64 support.
Fixes#43283.
Change-Id: I5e0fbf8530cc89d16e05c3257d282bc1d4d03405
Reviewed-on: https://go-review.googlesource.com/c/go/+/279423
Trust: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Go 1.15 pack's r command creates the output file if it does not
exist. The system "ar" command does this as well. Do the same.
For bazelbuild/rules_go#2762.
Change-Id: Icd88396b5c714b735c859a29ab29851e4301f4d2
Reviewed-on: https://go-review.googlesource.com/c/go/+/279516
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Add compiler support for emitting ABI wrappers by creating real IR as
opposed to introducing ABI aliases. At the moment these are "no-op"
wrappers in the sense that they make a simple call (using the existing
ABI) to their target. The assumption here is that once late call
expansion can handle both ABI0 and the "new" ABIInternal (register
version), it can expand the call to do the right thing.
Note that the runtime contains functions that do not strictly follow
the rules of the current Go ABI0; this has been handled in most cases
by treating these as ABIInternal instead (these changes have been made
in previous patches).
Generation of ABI wrappers (as opposed to ABI aliases) is currently
gated by GOEXPERIMENT=regabi -- wrapper generation is on by default if
GOEXPERIMENT=regabi is set and off otherwise (but can be turned on
using "-gcflags=all=-abiwrap -ldflags=-abiwrap"). Wrapper generation
currently only workd on AMD64; explicitly enabling wrapper for other
architectures (via the command line) is not supported.
Also in this patch are a few other command line options for debugging
(tracing and/or limiting wrapper creation). These will presumably go
away at some point.
Updates #27539, #40724.
Change-Id: I1ee3226fc15a3c32ca2087b8ef8e41dbe6df4a75
Reviewed-on: https://go-review.googlesource.com/c/go/+/270863
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Trust: Than McIntosh <thanm@google.com>
This CL changes so that all literals are represented with a new,
smaller ir.BasicLit type, so that ir.Name is only used to represent
declared constants.
Passes buildall w/ toolstash -cmp.
Change-Id: I4702b8e3fa945617bd05881d7a2be1205f229633
Reviewed-on: https://go-review.googlesource.com/c/go/+/279153
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
For OSLICELITs, we were reusing the Ntype field after type checking to
hold the length of the OSLICELIT's backing array. However, Ntype is
only meant for nodes that can represent types. Today, this works only
because we currently use Name for all OLITERAL constants (whether
declared or not), whereas we should be able to represent them more
compactly with a dedicated type that doesn't implement Ntype.
Passes buildall w/ toolstash -cmp.
Change-Id: I385f1d787c41b016f507a5bad9489d59ccfde7f2
Reviewed-on: https://go-review.googlesource.com/c/go/+/279152
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
declNames always returns a slice of *ir.Names, so return that directly
rather than as []ir.Node.
While here, also change iimport to directly create ir.ODCL/ir.OAS
statements, rather than calling variter. Allows eliminating a use of
ir.TypeNode.
Passes buildall w/ toolstash -cmp.
Change-Id: Icb75e993c4957b6050c797ba64ee71cfb7a19644
Reviewed-on: https://go-review.googlesource.com/c/go/+/279315
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
Adjust the pattern matching in one of the race output test to allow
for the possible introduction of an ABI wrapper. Normally for tests
that match traceback output wrappers are not an issue since they
are screened out by Go's traceback mechanism, but in this case the
race runtime is doing the unwinding, so the wrapper may be visible.
Change-Id: I45413b5c4701d4c28cc760fccc8203493dbe2874
Reviewed-on: https://go-review.googlesource.com/c/go/+/278756
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Trust: Than McIntosh <thanm@google.com>
The FreeBSD syscall convention uses the carry flag to indicate whether
an error has occured. The sys_umtx_op, thr_new, and pipe2 syscall
wrappers were failing to account for this convention and silently
suppressing errors as a result. This commit corrects these wrappers
by copying the pattern used by the other fallible syscall wrappers.
Note that futexsleep1 must now explicitly ignore the ETIMEDOUT error
from sys_umtx_op. Previously ETIMEDOUT was implicitly ignored because
sys_umtx_op never returned an error.
Fixes#43106.
Change-Id: I9c422b87cf4c6d308003bf42c3b419f785578b5d
Reviewed-on: https://go-review.googlesource.com/c/go/+/276892
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Trust: Than McIntosh <thanm@google.com>
Treat the compiler's -linkobj output as "compiler object, which
means "pack c" will "see through" the file and add individual
entry to the new archive, instead of the object as a whole.
This is somewhat peculiar. But Go 1.15's cmd/pack does this,
although seemingly accidental. We just do the same. FWIW, it
does make things more consistent with/without -linkobj flag.
Fixes#43271.
Change-Id: I6b2d99256db7ebf0fa430f85afa7464e334f6bcb
Reviewed-on: https://go-review.googlesource.com/c/go/+/279483
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Fix up the assembly shim routines runtime.panic{Index,Slice} and
friends so that their tail calls target ABIInternal and not ABI0
functions. This is so as to ensure that these calls don't go through
an ABI0->ABIInternal wrapper (which would throw off the machinery in
the called routines designed to detect whether the violation happened
in the runtime).
Note that when the compiler starts emitting real register calls to
these routines, we'll need to rewrite them to update the arg size and
ensure that args are in the correct registers. For example, the
current shim
TEXT runtime·panicIndex<ABIInternal>(SB),NOSPLIT,$0-16
MOVQ AX, x+0(FP)
MOVQ CX, y+8(FP)
JMP runtime·goPanicIndex<ABIInternal>(SB)
will need to change to
TEXT runtime·panicIndex<ABIInternal>(SB),NOSPLIT,$0
// AX already set up properly
MOVQ CX, BX // second argument expected in BX
JMP runtime·goPanicIndex<ABIInternal>(SB)
Change-Id: I48d1b5138fb4d229380ad12735cfaca5c50e6cc3
Reviewed-on: https://go-review.googlesource.com/c/go/+/278755
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Trust: Than McIntosh <thanm@google.com>
May fix#43302.
Change-Id: I6b7ddf94495c4fa80cf8a50a38eef5f8b2872669
Reviewed-on: https://go-review.googlesource.com/c/go/+/279481
Trust: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
This CL changes NewDeclNameAt to take an Op argument to set the Op up
front, and updates all callers to provide the appropriate Op. This
allows dropping the Name.SetOp method.
Passes buildall w/ toolstash -cmp.
Change-Id: I20e580f62d3c8a81223d1c162327c11b37bbf3f0
Reviewed-on: https://go-review.googlesource.com/c/go/+/279314
Trust: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
Rather than creating Names w/ ONONAME earlier and later adding in the
details, this CL changes the import logic to create and add details at
the same time.
Passes buildall w/ toolstash -cmp.
Change-Id: Ifaabade3cef8cd80ddd6644bff79393b934255d9
Reviewed-on: https://go-review.googlesource.com/c/go/+/279313
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
We only need to call itabname when actually creating the OCONVIFACE
ops, not any time we test whether a type implements an
interface. Additionally, by moving this call out of implements, we
make it purely based on types, which makes it safe to move to package
types.
Does not pass toolstash -cmp, because it shuffles symbol creation
order.
Change-Id: Iea8e0c9374218f4d97b4339020ebd758d051bd03
Reviewed-on: https://go-review.googlesource.com/c/go/+/279333
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
This change adds two examples of using the Read function: one that reads
one metric and one that reads all metrics.
Change-Id: I4940a44c9b1d65f3f7a1554e3145ff07e6492fc1
Reviewed-on: https://go-review.googlesource.com/c/go/+/275855
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
Trust: Michael Knyszek <mknyszek@google.com>
Main knows a bit too much about nowritebarrierrec.
Abstract the API a little bit to make the package split easier.
Change-Id: I4b76bdb1fed73dfb0d44e1a6c86de8c2d29a9488
Reviewed-on: https://go-review.googlesource.com/c/go/+/279301
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
isIntrinsicCall and ssaDumpInline are the only two "forward references"
to ssa by earlier phases. Make them a bit more explicit so that the
uses and the definitions can end up in different packages.
Change-Id: I02c7a27464fbedef9fee43c0e4094fa08b4d7a5c
Reviewed-on: https://go-review.googlesource.com/c/go/+/279300
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Separate embed, cgo pragmas, and Main trackScopes variable
from noder more cleanly.
This lets us split embed and noder into new packages.
It also assumes that the local embedded variables will be
removed and deletes them now for simplicity.
Change-Id: I9638bcc2c5f0e76440de056c6285b6aa2f73a00d
Reviewed-on: https://go-review.googlesource.com/c/go/+/279299
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Clean up a TODO (and make the package gc split easier)
by moving the exportsym walk out of iexport proper.
Also move exportsym call out of fninit.
Change-Id: Ie5887a68d325f7154201f4a35b9b4be4bf4b48dd
Reviewed-on: https://go-review.googlesource.com/c/go/+/279298
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
There are various global variables tracking the state of the
compilation. Collect them in a single global struct instead.
The struct definition is in package ir, but the struct itself is
still in package gc. It may eventually be threaded through the
code, but in the short term will end up in package typecheck.
Change-Id: I019db07aaedaed2c9b67dd45a4e138dc6028e54c
Reviewed-on: https://go-review.googlesource.com/c/go/+/279297
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Deleting the Pc assignment from Patch is safe because the actual PCs
are not assigned until well after the compiler is done patching jumps.
And it proves that replacing uses of Patch with SetTarget will be safe later.
Change-Id: Iffcbe03f0b5949ccd4c91e79c1272cd06be0f434
Reviewed-on: https://go-review.googlesource.com/c/go/+/279296
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Accumulated fixes to recent changes, to make the code safe
for automated deinterfacing.
Change-Id: I200737046cea88f3356b2402f09e2ca477fb8456
Reviewed-on: https://go-review.googlesource.com/c/go/+/279232
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
The list of conflicted files for this merge is:
src/cmd/compile/internal/gc/inl.go
src/cmd/compile/internal/gc/order.go
src/cmd/compile/internal/gc/ssa.go
test/fixedbugs/issue20415.go
test/fixedbugs/issue22822.go
test/fixedbugs/issue28079b.go
inl.go was updated for changes on dev.regabi: namely that OSELRECV has
been removed, and that OSELRECV2 now only uses List, rather than both
Left and List.
order.go was updated IsAutoTmp is now a standalone function, rather
than a method on Node.
ssa.go was similarly updated for new APIs involving package ir.
The tests are all merging upstream additions for gccgo error messages
with changes to cmd/compile's error messages on the dev.regabi branch.
Change-Id: Icaaf186d69da791b5994dbb6688ec989caabec42
On darwin, where we use libc for syscalls, when the runtime exits,
it calls libc exit function, which may call back into user code,
e.g. invoking functions registered with atexit. In particular, it
may call back into Go. But at this point, the Go runtime is
already exiting, so this wouldn't work.
On non-libc platforms we use exit syscall directly, which doesn't
invoke any callbacks. Use _exit on darwin to achieve the same
behavior.
No test for now, as it doesn't pass on all platforms (see trybot
run of PS2).
May fix#42465.
May fix#43294.
Change-Id: Ia1ada22b5da8cb64fdd598d0541eb90e195367eb
Reviewed-on: https://go-review.googlesource.com/c/go/+/269378
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
The netbsd kernel has a bug [1] that occassionally prevents netpoll from
waking with netpollBreak, which could result in missing timers for an
unbounded amount of time, as netpoll can't restart with a shorter delay
when an earlier timer is added.
Prior to CL 232298, sysmon could detect these overrun timers and
manually start an M to run them. With this fallback gone, the bug
actually prevents timer execution indefinitely.
As a workaround, we add back sysmon detection only for netbsd.
[1] https://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=50094
Updates #42515
Change-Id: I8391f5b9dabef03dd1d94c50b3b4b3bd4f889e66
Reviewed-on: https://go-review.googlesource.com/c/go/+/277332
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Austin Clements <austin@google.com>
Trust: Michael Pratt <mpratt@google.com>
This was part of a performance improvement made by CL 232298 to
reduce timer latency. On multiprocessor Plan 9 machines, it triggers
memory faults often enough that the builder test suite never completes
successfully. See issue #42303 for discussion. As shown by the benchmark
result below, worst case latency on plan9_arm is very bad even with the
wakep call in place - in the tickers-per-P=1 case, a 3ms timer is 270ms late.
Skipping the wakep call and running the benchmark again shows some cases
worse, some better. The performance cost doesn't seem excessive for this
temporary workaround which makes the plan9_arm builders usable again.
With wakep call:
cpu% go test -bench Latency time
goos: plan9
goarch: arm
pkg: time
BenchmarkParallelTimerLatency-4 100 10985859 avg-late-ns 18630963 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=1-4 195 270294688 avg-late-ns 542057670 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=2-4 234 182452000 avg-late-ns 423933688 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=3-4 280 193003004 avg-late-ns 408034405 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=4-4 282 132819086 avg-late-ns 313624570 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=5-4 339 71152187 avg-late-ns 189014519 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=6-4 315 26860484 avg-late-ns 101759844 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=7-4 357 19106739 avg-late-ns 59435620 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=8-4 376 7246933 avg-late-ns 38888461 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=9-4 267 40476892 avg-late-ns 205851926 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=10-4 294 87836303 avg-late-ns 252059695 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=2ms/tickers-per-P=1-4 379 4127144 avg-late-ns 10494927 max-late-ns
Without wakep call:
BenchmarkParallelTimerLatency-4 61 10775151 avg-late-ns 18668517 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=1-4 199 299587535 avg-late-ns 597182307 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=2-4 272 184561831 avg-late-ns 449739837 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=3-4 235 154983257 avg-late-ns 370940553 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=4-4 290 150034689 avg-late-ns 332399843 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=5-4 298 47540764 avg-late-ns 133709031 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=6-4 350 20379394 avg-late-ns 81742809 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=7-4 363 14403223 avg-late-ns 98901212 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=8-4 375 12293090 avg-late-ns 50266552 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=9-4 336 40628820 avg-late-ns 150946099 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=300µs/tickers-per-P=10-4 289 88265539 avg-late-ns 280770418 max-late-ns
BenchmarkStaggeredTickerLatency/work-dur=2ms/tickers-per-P=1-4 375 8364937 avg-late-ns 22598421 max-late-ns
Fixes#42303
Change-Id: I70c63cb2a2bad46950a7cd9dfc7bb32943710d32
Reviewed-on: https://go-review.googlesource.com/c/go/+/275672
Reviewed-by: David du Colombier <0intro@gmail.com>
Trust: Michael Pratt <mpratt@google.com>
For #11656
For #43283
Change-Id: I1fcf2b24800f421e36201af43130b487abe605b1
Reviewed-on: https://go-review.googlesource.com/c/go/+/279312
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Previously, reassigned was failing to detect reassignments due to
channel receives in select statements (OSELRECV, OSELRECV2), or due to
standalone 2-value receive assignments (OAS2RECV). This was reported
as a devirtualization panic, but could have caused mis-inlining as
well.
Fixes#43292.
Change-Id: Ic8079c20c0587aeacff9596697fdeba80a697b12
Reviewed-on: https://go-review.googlesource.com/c/go/+/279352
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
The issue11656 code was using the trap instruction as a PC value,
but it is intended to call a PC value that contains the trap instruction.
It doesn't matter too much as in practice the address is not
executable anyhow. But may as well have the code act the way it
is documented to act.
Also, don't run the test with gccgo/GoLLVM, as it can't work.
The illegal instruction will have no unwind data, so the unwinder
won't be able to get past it. In other words, gccgo/GoLLVM suffer
from the exact problem that the issue describes, but it seems insoluble.
For golang/go#11656
Change-Id: Ib2e50ffc91d215fd50e78f742fafe476c92d704e
Reviewed-on: https://go-review.googlesource.com/c/go/+/278473
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
