If we've already imported a named type, then there's no need to
process its associated methods except to validate that the signature
matches the existing known method.
However, the current import code still creates a new function node for
each method, saves its inline body (if any), and adds the node to the
global importlist. Because of this, the duplicate methods are never
garbage collected.
This CL changes the compiler to avoid amassing uncollectable garbage
or performing any unnecessary processing.
This is particularly noticeable for protobuf-heavy code. For the
motivating Go package, this CL reduced compile max-RSS from ~12GB to
~3GB and compile time from ~65s to ~50s.
Passes toolstash -cmp for std, cmd, and k8s.io/kubernetes/cmd/....
Matt Dee [Tue, 8 Aug 2017 19:58:27 +0000 (15:58 -0400)]
database/sql: fail on unsupported options when context is un-cancellable
Currently, the check for `ctx.Done() == context.Background().Done()`
comes before the check to see if we are ignoring any options. That
check should be done earlier, so that the options are not silently
ignored.
Fixes #21350
Change-Id: I3704e4209854c7d99f3f92498bae831cabc7e419
Reviewed-on: https://go-review.googlesource.com/53970 Reviewed-by: Daniel Theophanes <kardianos@gmail.com>
Run-TryBot: Daniel Theophanes <kardianos@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Michael McLoughlin [Sun, 16 Jul 2017 00:21:26 +0000 (18:21 -0600)]
crypto/rand: batch large calls to linux getrandom
The linux getrandom system call returns at most 33554431 = 2^25-1 bytes per
call. The existing behavior for larger reads is to report a failure, because
there appears to have been an unexpected short read. In this case the system
falls back to reading from "/dev/urandom".
This change performs reads of 2^25 bytes or more with multiple calls to
getrandom.
Fixes #20877
Change-Id: I618855bdedafd86cd11219fe453af1d6fa2c88a7
Reviewed-on: https://go-review.googlesource.com/49170 Reviewed-by: Adam Langley <agl@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Brian Kessler [Fri, 21 Jul 2017 08:19:42 +0000 (01:19 -0700)]
crypto/rsa: drop uneeded parameter in modInverse
The current modInverse implementation allocates a big.Int
for the second parameter of GCD, while only the first is needed.
This is unnecessary and can lead to a speed up for optimizations
of GCD where the second parameter is not calculated at all.
Dmitri Shuralyov [Tue, 8 Aug 2017 19:38:21 +0000 (15:38 -0400)]
net/http: log Readdir error to Server.ErrorLog
Now that issue #12438 is resolved, this TODO can be completed.
Create a logf helper, which is similar to Server.logf method,
but takes a *Request to infer the *Server and its ErrorLog from.
Update documentation of Server.ErrorLog to mention a new type
of errors that may be logged to it.
Also update a statement in documentation of Server.ErrorLog from:
// If nil, logging goes to os.Stderr via the log package's
// standard logger.
To:
// If nil, logging is done via the log package's standard logger.
The motivation for doing so is to avoid making inaccurate claims.
Logging may not go to os.Stderr if anyone overrides the log package's
default output via https://godoc.org/log#SetOutput. Saying that
the standard logger is used should be sufficient to explain the
behavior, and users can infer that os.Stderr is used by default,
unless it's changed.
Lynn Boger [Tue, 8 Aug 2017 17:45:41 +0000 (13:45 -0400)]
cmd/go,cmd/link: support buildmode c-shared on ppc64le
This change enables buildmode c-shared on ppc64le.
A bug was fixed in runtime/rt0_linux_ppc64le.s that was necessary to
make this work. In _rt0_ppc64le_linux_lib, there is code to store
the value of r2 onto the caller's stack. However, if this file
is compiled using a build mode that maintains the TOC address in
r2, then instructions will be inserted at the beginning of this
function to generate the r2 value for the callee, not the caller.
That means the r2 value for the callee is stored onto the caller's
stack. If caller and callee don't have the same r2 values, then
the caller will restore the wrong r2 value after it returns. This
situation can happen when using dlopen since the caller of this
function will be in ld64.so and will definitely have a different
TOC.
Updates #20756
Change-Id: I6e165e0d0716e73721bbbcc520e8302e4856e3ba
Reviewed-on: https://go-review.googlesource.com/53890 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Joe Kyo [Wed, 9 Aug 2017 02:26:45 +0000 (03:26 +0100)]
net/http: check If-Range header when request method is HEAD
When If-Range does not match and the requested resource is
available, server should return a "200 OK" response to client.
Currently server returns "200 OK" when the request method is
GET, but "206 Partial Content" when method is HEAD.
This change fixed this inconsistency.
Change-Id: I5ad979919f4f089baba54a4445b70ca38471a906
Reviewed-on: https://go-review.googlesource.com/54110
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Tom Bergan <tombergan@google.com>
Alberto Donizetti [Wed, 2 Aug 2017 16:22:21 +0000 (18:22 +0200)]
testing: explain how SkipNow and FailNow stop execution
SkipNow and FailNow must be called from the goroutine running the
test. This is already documented, but it's easy to call them by
mistake when writing subtests. In the following:
func TestPanic(t *testing.T) {
t.Run("", func(t2 *testing.T) {
t.FailNow() // BAD: should be t2.FailNow()
})
}
the FailNow call on the outer t *testing.T correctly triggers a panic
panic: test executed panic(nil) or runtime.Goexit
The error message confuses users (see issues #17421, #21175) because
there is no way to trace back the relevant part of the message ("test
executed ... runtime.Goexit") to a bad FailNow call without checking
the testing package source code and finding out that FailNow calls
runtime.Goexit.
To help users debug the panic message, mention in the SkipNow and
FailNow documentation that they stop execution by calling
runtime.Goexit.
Fixes #21175
Change-Id: I0a3e5f768e72b464474380cfffbf2b67396ac1b5
Reviewed-on: https://go-review.googlesource.com/52770 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Tom Bergan [Wed, 9 Aug 2017 01:01:08 +0000 (18:01 -0700)]
net/http: update bundled http2
Updates http2 to x/net/http2 git rev 1c05540f687 for:
http2: fix format argument warnings in tests
https://golang.org/cl/48090
http2: retry requests after receiving REFUSED STREAM
https://golang.org/cl/50471
http2: block RoundTrip when the Transport hits MaxConcurrentStreams
https://golang.org/cl/53250
Fixes #13774
Fixes #20985
Fixes #21229
Change-Id: Ie19b4a7cc395a0b7a25fac55f5051faaf94920bb
Reviewed-on: https://go-review.googlesource.com/54052
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
crypto/{aes,cipher,rc4}: fix out of bounds write in stream ciphers
Functions XORKeyStream should panic if len(dst) < len(src), but it
write to dst before bounds checking. In asm routines and fastXORBytes,
this is an out of bounds write.
cmd/compile: remove gc.Sysfunc calls from 387 backend
gc.Sysfunc must not be called concurrently.
We set up runtime routines used by the backend
prior to doing any backend compilation.
I missed the 387 ones; fix that.
Sysfunc should have been unexported during 1.9.
I will rectify that in a subsequent CL.
math/rand: fix uniform distribution stddev in tests
The standard deviation of a uniform distribution is size / √12.
The size of the interval [0, 255] is 256, not 255.
While we're here, simplify the expression.
The tests previously passed only because the error margin was large enough.
Sample observed standard deviations while running tests:
Lyle Franklin [Sat, 15 Jul 2017 20:33:56 +0000 (13:33 -0700)]
strings: add Examples for TrimFunc and variants during Gophercon!
Change-Id: I6bfe5b914cf11be1cd1f8e61d557cc718725f0be
Reviewed-on: https://go-review.googlesource.com/49013
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Regular HTTP is insecure, oeis.org supports HTTPS and it is actually
used in some other places in the codebase. This changes these final urls
to use HTTPS.
Change-Id: Ia46410a9c7ce67238a10cb6bfffaceca46112f58
Reviewed-on: https://go-review.googlesource.com/52072 Reviewed-by: Alberto Donizetti <alb.donizetti@gmail.com>
Martin Möhrmann [Thu, 11 May 2017 05:36:09 +0000 (07:36 +0200)]
runtime, internal/cpu: change cache line size for arm64 to 64 bytes
According to http://infocenter.arm.com:
* ARM Cortex-A53 (Raspberry Pi 3, Pine A64)
* ARM Cortex-A57 (Opteron A1100, Tegra X1)
* ARM Cortex-A72
all have a cache line size of 64 bytes.
Change-Id: I4b333e930792fb1a221b3ca6f395bfa1b7762afa
Reviewed-on: https://go-review.googlesource.com/43250
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Martin Möhrmann [Mon, 22 May 2017 05:42:42 +0000 (07:42 +0200)]
runtime: remove unused prefetch functions
The only non test user of the assembler prefetch functions is the
heapBits.prefetch function which is itself unused.
The runtime prefetch functions have no functionality on most platforms
and are not inlineable since they are written in assembler. The function
call overhead eliminates the performance gains that could be achieved with
prefetching and would degrade performance for platforms where the functions
are no-ops.
If prefetch functions are needed back again later they can be improved
by avoiding the function call overhead and implementing them as intrinsics.
Dmitri Shuralyov [Wed, 19 Jul 2017 03:59:40 +0000 (23:59 -0400)]
net/http: set Content-Type header in Redirect
Setting the Content-Type header explicitly allows browsers to know what
the type of the content is. Otherwise, they have to guess the type from
the content itself, which could lead to unpredictable behavior, and
increases CPU usage.
Not setting the Content-Type despite writing a body may also trigger
unwanted warnings in user middleware, and make it more difficult to
resolve valid issues where the user forgets to set Content-Type in
some situations where it should be set.
There is some precedent for doing this in http.FileServer, which
sets "Content-Type" to "text/html; charset=utf-8" before writing
<pre><a href=...></a></pre> HTML.
flag: make default Usage prefer CommandLine's output over just os.Stderr
CommandLine (exported in Go 1.2) has default output of os.Stderr.
Before it was exported, it made sense to have the global Usage func
(the implicit usage func if CommandLine.Usage is nil) hard-code
os.Stderr has its output. But once CommandLine was exported, Usage
should use it if provided.
Ian Lance Taylor [Fri, 4 Aug 2017 17:36:40 +0000 (10:36 -0700)]
runtime: mark activeModules nosplit/nowritebarrier
The activeModules function is called by the cgo pointer checking code,
which is called by the write barrier (when GODEBUG=cgocheck=2), and as
such must be nosplit/nowritebarrier.
doc: use better wording to explain type-aware completion
Some editors can filter the autocompletion suggestions based on
whether the code will compile once autocompleted. Explain this
feature with better wording.
Dmitry Savintsev [Fri, 4 Aug 2017 08:12:21 +0000 (10:12 +0200)]
doc/1.9: fix broken html link in CL 53030/53210
Change-Id: I7176becd10ad84cbfc3fb9427e190028626e5baf
Reviewed-on: https://go-review.googlesource.com/53291 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Alberto Donizetti [Fri, 4 Aug 2017 10:13:33 +0000 (12:13 +0200)]
doc: fix bad link in go1.9 release notes
Change-Id: I64ba37428f5cc560f0f20fe039feaecf5fcda93e
Reviewed-on: https://go-review.googlesource.com/53330 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Dmitry Savintsev [Thu, 3 Aug 2017 14:51:32 +0000 (16:51 +0200)]
doc/1.9: add CL 36696 for crypto/x509 to the release notes
add https://go-review.googlesource.com/c/36696
"crypto/x509: ignore CN if SAN extension present"
to the release notes.
Fixes #21289
Change-Id: Ifa184d3816806a8da3c67b68476c923329acf13e
Reviewed-on: https://go-review.googlesource.com/53030 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Andrew Bonventre [Thu, 3 Aug 2017 16:33:26 +0000 (12:33 -0400)]
doc: hide blog content for golang.google.cn
/blog redirects to blog.golang.org (currently blocked in China)
unless there is a local checkout of golang.org/x/blog, which is
not possible on App Engine Classic.
Change-Id: Ia695e663c9bebcc6c3bedea324c630299eaad4dc
Reviewed-on: https://go-review.googlesource.com/53051 Reviewed-by: Chris Broadfoot <cbro@golang.org>
Alberto Donizetti [Tue, 1 Aug 2017 15:41:51 +0000 (17:41 +0200)]
time: skip ZoneAbbr test in timezones with no abbreviation
The testZoneAbbr assumes that
Parse(RFC1123, t1.Format(RFC1123))
will always succeed. This is not true because Format will fall back to
the numeric zone (ex. -07) for timezones with no abbreviation, but
Parse won't accept the numeric zone when the layout specifies 'MST'
(an abbreviation).
Skip the zone abbreviation test in timezones with no abbreviation.
cmd/go: show examples with empty output in go test -list
Fixes #21205
Change-Id: I81b001eb42cbf2a5d5b7b82eb63548b22f501be5
Reviewed-on: https://go-review.googlesource.com/52110 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
cmd/compile: set/unset base register for better assembly print
For address of an auto or arg, on all non-x86 architectures
the assembler backend encodes the actual SP offset in the
instruction but leaves the offset in Prog unchanged. When the
assembly is printed in compile -S, it shows an offset
relative to pseudo FP/SP with an actual hardware SP base
register (e.g. R13 on ARM). This is confusing. Unset the
base register if it is indeed SP, so the assembly output is
consistent. If the base register isn't SP, it should be an
error and the error output contains the actual base register.
For address loading instructions, the base register isn't set
in the compiler on non-x86 architectures. Set it. Normally it
is SP and will be unset in the change mentioned above for
printing. If it is not, it will be an error and the error
output contains the actual base register.
No change in generated binary, only printed assembly. Passes
"go build -a -toolexec 'toolstash -cmp' std cmd" on all
architectures.
runtime: map bitmap and spans during heap initialization
We lazily map the bitmap and spans areas as the heap grows. However,
right now we're very slightly too lazy. Specifically, the following
can happen on 32-bit:
1. mallocinit fails to allocate any heap arena, so
arena_used == arena_alloc == arena_end == bitmap.
2. There's less than 256MB between the end of the bitmap mapping and
the next mapping.
3. On the first allocation, mheap.sysAlloc sees that there's not
enough room in [arena_alloc, arena_end) because there's no room at
all. It gets a 256MB mapping from somewhere *lower* in the address
space than arena_used and sets arena_alloc and arena_end to this
hole.
4. Since the new arena_alloc is lower than arena_used, mheap.sysAlloc
doesn't bother to call mheap.setArenaUsed, so we still don't have a
bitmap mapping or a spans array mapping.
5. mheap.grow, which called mheap.sysAlloc, attempts to fill in the
spans array and crashes.
Fix this by mapping the metadata regions for the initial arena_used
when the heap is initialized, rather than trying to wait for an
allocation. This maintains the intended invariant that the structures
are always mapped for [arena_start, arena_used).
Fixes #21044.
Change-Id: I4422375a6e234b9f979d22135fc63ae3395946b0
Reviewed-on: https://go-review.googlesource.com/51714
Run-TryBot: Austin Clements <austin@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
runtime: fall back to small mmaps if we fail to grow reservation
Right now, if it's possible to grow the arena reservation but
mheap.sysAlloc fails to get 256MB more of memory, it simply fails.
However, on 32-bit we have a fallback path that uses much smaller
mmaps that could take in this situation, but fail to.
This commit fixes mheap.sysAlloc to use a common failure path in case
it can't grow the reservation. On 32-bit, this path includes the
fallback.
Ideally, mheap.sysAlloc would attempt smaller reservation growths
first, but taking the fallback path is a simple change for Go 1.9.
Updates #21044 (fixes one of two issues).
Change-Id: I1e0035ffba986c3551479d5742809e43da5e7c73
Reviewed-on: https://go-review.googlesource.com/51713
Run-TryBot: Austin Clements <austin@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Gustav Westling [Sun, 30 Jul 2017 20:14:22 +0000 (22:14 +0200)]
.github: add .md extension to SUPPORT file
This makes GitHub render the markdown file automatically
on their web UI.
SUPPORT.md is the recommended file name according to the GitHub
documentation:
https://help.github.com/articles/adding-support-resources-to-your-project/
Vladimir Stefanovic [Thu, 4 May 2017 14:45:29 +0000 (16:45 +0200)]
runtime/pprof: prevent a deadlock that SIGPROF might create on mips{,le}
64bit atomics on mips/mipsle are implemented using spinlocks. If SIGPROF
is received while the program is in the critical section, it will try to
write the sample using the same spinlock, creating a deadloop.
Prevent it by creating a counter of SIGPROFs during atomic64 and
postpone writing the sample(s) until called from elsewhere, with
pc set to _LostSIGPROFDuringAtomic64.
Added a test case, per Cherry's suggestion. Works around #20146.
Ian Lance Taylor [Fri, 21 Jul 2017 23:37:40 +0000 (16:37 -0700)]
cmd/compile: consider exported flag in namedata
It is possible to have an unexported name with a nil package,
for an embedded field whose type is a pointer to an unexported type.
We must encode that fact in the type..namedata symbol name,
to avoid incorrectly merging an unexported name with an exported name.
Fixes #21120
Change-Id: I2e3879d77fa15c05ad92e0bf8e55f74082db5111
Reviewed-on: https://go-review.googlesource.com/50710
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Crawshaw <crawshaw@golang.org>
Joe Tsai [Fri, 21 Jul 2017 23:53:54 +0000 (16:53 -0700)]
encoding/json: ignore embedded fields of pointers to unexported non-structs
https://golang.org/cl/33773 fixes the JSON marshaler to avoid serializing
embedded fields on unexported types of non-struct types. However, Go allows
embedding pointer to types, so the check for whether the field is a non-struct
type must first dereference the pointer to get at the underlying type.
Furthermore, due to a edge-case in the behavior of StructField.PkgPath not
being a reliable indicator of whether the field is unexported (see #21122),
we use our own logic to determine whether the field is exported or not.
The logic in this CL may be simplified depending on what happens in #21122.
Fixes #21121
Updates #21122
Change-Id: I8dfd1cdfac8a87950df294a566fb96dfd04fd749
Reviewed-on: https://go-review.googlesource.com/50711 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
From Josh's comments on https://golang.org/cl/50310
Once I removed the "from the Go standard library" bit, the beginning
wasn't worth keeping. It also wasn't clear whether what it meant by
"cache contention". Processor caches, or user-level caches built with
sync.Map? It didn't seem worth clarifying and didn't convey any useful
information, so deleted.
net/http: document that after Hijack, Request.Body is invalid
We can make it panic with a more explicit and readable error message
during Go 1.10, but document it for now. This has always been the
case; it's not a new rule.
Updates #20933
Change-Id: I53c1fefb47a8f4aae0bb32fa742afa3a2ed20e8a
Reviewed-on: https://go-review.googlesource.com/50634 Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Dmitri Shuralyov [Wed, 19 Jul 2017 04:41:13 +0000 (00:41 -0400)]
net/http: improve signature of Redirect, NewRequest
In CL https://golang.org/cl/4893043 (6 years ago), a new package named
"url" was created (it is currently known as "net/url"). During that
change, some identifier name collisions were introduced, and two
parameters in net/http were renamed to "urlStr".
Since that time, Go has continued to put high emphasis on the quality
and readability of the documentation. Sometimes, that means making small
sacrifices in the implementation details of a package to ensure that
the godoc reads better, since that's what the majority of users interact
with. See https://golang.org/s/style#named-result-parameters:
> Clarity of docs is always more important than saving a line or two
> in your function.
I think the "urlStr" parameter name is suboptimal for godoc purposes,
and just "url" would be better.
During the review of https://golang.org/cl/4893043, it was also noted
by @rsc that having to rename parameters named "url" was suboptimal:
> It's unfortunate that naming the package url means
> you can't have a parameter or variable named url.
However, at the time, the name of the url package was still being
decided, and uri was an alternative name under consideration.
The reason urlStr was chosen is because it was a lesser evil
compared to naming the url package uri instead:
> Let's not get hung up on URI vs. URL, but I'd like s/uri/urlStr/ even for just
> that the "i" in "uri" looks very similar to the "l" in "url" in many fonts.
> Please let's go with urlStr instead of uri.
Now that we have the Go 1 compatibility guarantee, the name of the
net/url package is fixed. However, it's possible to improve the
signature of Redirect, NewRequest functions in net/http package
for godoc purposes by creating a package global alias to url.Parse,
and renaming urlStr parameter to url in the exported funcs. This CL
does so.
Bryan C. Mills [Tue, 9 May 2017 20:12:57 +0000 (16:12 -0400)]
sync: release m.mu during (*RWMutexMap).Range callbacks in sync_test
The mainline sync.Map has allowed mutations within Range callbacks
since https://golang.org/cl/37342. The reference implementations need
to do the same.
This change integrates https://go-review.googlesource.com/c/42956/
from x/sync.
cmd/go: fix test when go source tree has POSIX ACL
Fixes TestGoBuildUmask when the user has a POSIX ACL on the Go source tree.
Fixes #17909.
Change-Id: I5bc19099af8353afd41071258f4f317612b4c8c1
Reviewed-on: https://go-review.googlesource.com/50370 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
runtime: use SIGKILL if SIGQUIT is blocked; skip tests that need SIGQUIT
The runtime tests may be invoked from a parent that has SIGQUIT
blocked. For example, Java invokes subprocesses this way. In this
situation, TestCrashDumpsAllThreads and TestPanicSystemstack will fail
because they depend on SIGQUIT to get tracebacks, and any subprocess
test that times out will fail to kill the subprocess.
Fix this by detecting if SIGQUIT is blocked and, if so, skipping tests
that depend on it and using SIGKILL to kill timed-out subprocesses.
Based on a fix by Carl Henrik Lunde in
https://golang.org/issue/19196#issuecomment-316145733
Ian Lance Taylor [Thu, 20 Jul 2017 06:42:27 +0000 (23:42 -0700)]
runtime: don't call libc sigaction function in forked child
If we are using vfork, and if something (such as TSAN) is intercepting
the sigaction function, then we must call the system call, not the
libc function. Otherwise the intercepted sigaction call in the child
may trash the data structures in the parent.
The Go ecosystem provides many tools to make Go
development more productive and seamless. Document
the availability of the editor plugins and IDEs,
add an overview of feature support and screencasts.
Currently we trace mark assists even if they're satisfied entirely by
stealing. This means even if background marking is keeping up with
allocation, we'll still emit a trace event every N bytes of
allocation. The event will be a few microseconds, if that, but they're
frequent enough that, when zoomed out in the trace view, it looks like
all of the time is spent in mark assists even if almost none is.
Change this so we only emit a trace event if the assist actually has
to do assisting. This makes the traces of these events far more
useful.
Michael Munday [Mon, 17 Jul 2017 11:07:28 +0000 (07:07 -0400)]
cmd/compile: fix unaligned loads/stores to global variables on s390x
Load/store-merging and move optimizations can result in unaligned
memory accesses. This is fine so long as the load/store instruction
used does not take a relative offset. In the SSA rules this means we
must not merge (MOVDaddr (SB)) ops into loads/stores unless we can
guarantee the alignment of the target.
tSweepTerm and pauseStart are supposed to be when STW was triggered,
but right now they're captured a bit before STW. Move these down to
immediately before we trigger STW.
Currently, Windows stacks are either 128kB or 2MB depending on whether
the binary uses cgo. This is because we assume that Go system stacks
and the small amount of C code invoked by the standard library can
operate within smaller stacks, but general Windows C code assumes
larger stacks.
However, it's easy to call into arbitrary C code using the syscall
package on Windows without ever importing cgo into a binary. Such
binaries need larger system stacks even though they don't use cgo.
Fix this on 64-bit by increasing the system stack size to 2MB always.
This only costs address space, which is free enough on 64-bit to not
worry about. We keep (for now) the existing heuristic on 32-bit, where
address space comes at more of a premium.