]> Cypherpunks repositories - gostls13.git/log
gostls13.git
5 years ago[dev.link] cmd/oldlink: decouple from goobj2 package
Cherry Zhang [Wed, 1 Apr 2020 15:05:41 +0000 (11:05 -0400)]
[dev.link] cmd/oldlink: decouple from goobj2 package

The new object file support in the old linker should not be used.
This is a minimal change that removes stuff from the old linker's
loader package, so that it decouples from the goobj2 package,
allowing the latter to evolve.

Keep the change local in the loader package, so most of the old
linker doesn't need to change. At this point I don't think we
want to make significant changes to the old linker.

Change-Id: I078c4cbb35dc4627c4b82f512a4aceec9b594925
Reviewed-on: https://go-review.googlesource.com/c/go/+/226800
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/internal/goobj: use new style accessors
Cherry Zhang [Wed, 1 Apr 2020 02:31:25 +0000 (22:31 -0400)]
[dev.link] cmd/internal/goobj: use new style accessors

We already move to new style accessors in the linker. This will
allow us to get rid of the read side of old style ones.

Change-Id: Id0c171c5634a5977fe8a6f764cb0d48203993ab7
Reviewed-on: https://go-review.googlesource.com/c/go/+/226799
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/internal/goobj2, cmd/link: add and use new Sym accessors
Cherry Zhang [Wed, 1 Apr 2020 00:56:10 +0000 (20:56 -0400)]
[dev.link] cmd/internal/goobj2, cmd/link: add and use new Sym accessors

Along the line with Reloc and Aux, use new-style accessors for
the Sym type. They avoid reading unnecessary fields, and also
look nicer (to me).

Change-Id: Ie37c5149a6edb2184724b3dfa26952015e74c085
Reviewed-on: https://go-review.googlesource.com/c/go/+/226798
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/internal/goobj2: change StringRef encoding for better locality
Cherry Zhang [Wed, 1 Apr 2020 00:01:46 +0000 (20:01 -0400)]
[dev.link] cmd/internal/goobj2: change StringRef encoding for better locality

Previously, StringRef is encoded as an offset pointing to
{ len, [len]byte }. This CL changes it to { len, offset }, where
offset points the bytes.

With the new format, reading a string header is just reading two
adjacent uint32s, without accessing the string table. This should
improve locality of object file reading.

Change-Id: Iec30708f9d9adb2f0242db6c4767c0f8e730f4df
Reviewed-on: https://go-review.googlesource.com/c/go/+/226797
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: fix mmapping OutBuf on windows
Jeremy Faller [Mon, 30 Mar 2020 14:45:26 +0000 (10:45 -0400)]
[dev.link] cmd/link: fix mmapping OutBuf on windows

We got the permissions wrong on the mmapped region.

Change-Id: Ica6372fd9d9a787ab20a763e5785fb9fb34ff623
Reviewed-on: https://go-review.googlesource.com/c/go/+/226366
Run-TryBot: Jeremy Faller <jeremy@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: fix end-of-block padding (again)
Cherry Zhang [Tue, 31 Mar 2020 22:53:17 +0000 (18:53 -0400)]
[dev.link] cmd/link: fix end-of-block padding (again)

Make sure we never overrun the end address.

Should fix AIX build.

Change-Id: I9926387d4512ec8b4acc32b7f32cee2b2aca25b4
Reviewed-on: https://go-review.googlesource.com/c/go/+/226718
Reviewed-by: Jeremy Faller <jeremy@golang.org>
5 years ago[dev.link] cmd/link: unify Relocs.Count and len(rs)
Cherry Zhang [Mon, 30 Mar 2020 14:04:00 +0000 (10:04 -0400)]
[dev.link] cmd/link: unify Relocs.Count and len(rs)

The Count field in Relocs type is always equal to len(rs). Unify
them.

Change-Id: Ic77288ea58b61a98482b218e051d81047d0ddd88
Reviewed-on: https://go-review.googlesource.com/c/go/+/226717
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years ago[dev.link] cmd/link: store external relocations in Reloc2 format
Cherry Zhang [Sat, 28 Mar 2020 20:46:47 +0000 (16:46 -0400)]
[dev.link] cmd/link: store external relocations in Reloc2 format

Store external relocations in (almost) the same format as the Go
objects, so we can handle them more uniformly.

There is a small speedup:

(linking cmd/compile)
Deadcode        67.8ms ± 3%    61.1ms ± 3%   -9.94%  (p=0.008 n=5+5)
Dostkcheck      41.2ms ± 2%    38.8ms ± 3%   -5.99%  (p=0.008 n=5+5)

Change-Id: I8616e10b26235904201d6c9465f5ae32a49c9949
Reviewed-on: https://go-review.googlesource.com/c/go/+/226365
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years ago[dev.link] cmd/link: fix end-of-block padding
Cherry Zhang [Tue, 31 Mar 2020 18:36:19 +0000 (14:36 -0400)]
[dev.link] cmd/link: fix end-of-block padding

Make sure we write the entire address range we are asked to write,
with no holes between the blocks or at the end.

Should fix NetBSD build.

Change-Id: I13b1f551377cbc4bcde3650417ac95cba62ff807
Reviewed-on: https://go-review.googlesource.com/c/go/+/226617
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
5 years ago[dev.link] cmd/link: set attributes atomically
Cherry Zhang [Fri, 27 Mar 2020 20:32:22 +0000 (16:32 -0400)]
[dev.link] cmd/link: set attributes atomically

Now concurrent relocsym may access symbols attributes
concurrently, causing data race when using the race detector. I
think it is still safe as we read/write on different bits, and
not write the same symbol's attributes from multiple goroutines,
so it will always reads the right value regardless whether the
write happens before or after, as long as the memory model is not
so insane.

Use atomic accesses to appease the race detector. It doesn't seem
to cost much, at least on x86.

Change-Id: I2bfc3755ee59c87ed237d508f29d6172fa976392
Reviewed-on: https://go-review.googlesource.com/c/go/+/226368
Reviewed-by: Austin Clements <austin@google.com>
5 years ago[dev.link] cmd/link: make asmb parallel across all platforms
Jeremy Faller [Fri, 20 Mar 2020 16:24:35 +0000 (12:24 -0400)]
[dev.link] cmd/link: make asmb parallel across all platforms

Numbers for x86:

Asmb                      54.8ms ±15%    12.1ms ±28%   -78.02%  (p=0.000 n=10+10)
TotalTime                  545ms ± 3%     515ms ± 4%    -5.42%  (p=0.000 n=10+10)

Change-Id: I8fc17f7ccdf83fbd70cf7d27b44f41f9db9c9922
Reviewed-on: https://go-review.googlesource.com/c/go/+/224419
Run-TryBot: Jeremy Faller <jeremy@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: parallelize asmb on amd64
Jeremy Faller [Tue, 17 Mar 2020 14:24:40 +0000 (10:24 -0400)]
[dev.link] cmd/link: parallelize asmb on amd64

Introduces a parallel OutBuf implementation, and POC on amd64. Due to
some of the weird behaviors I saw on MacOS (SIGBUS while calling msync),
I will wait for feedback to port to other architectures.

On my mac, sped up Asmb by ~78% for cmd/compile (below). Will likely
have an appreciable speedup on kubelet benchmark.

Asmb                      39.1ms ±11%     8.5ms ±10%     -78.17%  (p=0.000 n=10+9)
TotalTime                  596ms ± 2%     577ms ± 8%      -3.07%  (p=0.034 n=8+10)

Change-Id: Id2a2577c3f4da155d8dccc862897f43b941877ac
Reviewed-on: https://go-review.googlesource.com/c/go/+/223742
Run-TryBot: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: move the wavefront past addexport()
Than McIntosh [Mon, 30 Mar 2020 14:06:54 +0000 (10:06 -0400)]
[dev.link] cmd/link: move the wavefront past addexport()

Reorganize the linker phase ordering so that addexport() runs before
loadlibfull. In previous CLs addexport() was changed to use loader
APIs but then copy back its work into sym.Symbol, so this change
removes the copying/shim code in question.

Change-Id: I17314a90007909e6242ee00e26393f3e4a02cf25
Reviewed-on: https://go-review.googlesource.com/c/go/+/226362
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: run setArchSyms earlier
Than McIntosh [Mon, 30 Mar 2020 14:05:23 +0000 (10:05 -0400)]
[dev.link] cmd/link: run setArchSyms earlier

Add some new loader.Sym equivalents to the archSyms struct so that we
can run setArchSyms earlier in the pipeline, and add a "mode" to
setArchSyms to control whether it should create loader.Sym symbols or
their *sym.Symbol equivalents.

These change needed for a subsequent patch in which addexport() is run
earlier as well

Change-Id: I0475c9388c39f13e045dd4aa9c90eaec42624810
Reviewed-on: https://go-review.googlesource.com/c/go/+/226361
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link/internal/loader: copy Dynid when migrating sym attrs
Than McIntosh [Mon, 30 Mar 2020 13:17:22 +0000 (09:17 -0400)]
[dev.link] cmd/link/internal/loader: copy Dynid when migrating sym attrs

Add support for migrating over the symbol Dynid property
when converting loader.Sym symbols to sym.Symbol.

Change-Id: Icc3b91b4adcae6f2ede7d915bb674cc206025217
Reviewed-on: https://go-review.googlesource.com/c/go/+/226360
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: use new-style Aux accessors in the loader
Cherry Zhang [Sat, 28 Mar 2020 19:26:05 +0000 (15:26 -0400)]
[dev.link] cmd/link: use new-style Aux accessors in the loader

And remove the old-style one.

Change-Id: Ia6c84448108dd1da68451df2221c7c1005c45a0d
Reviewed-on: https://go-review.googlesource.com/c/go/+/226364
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: use new-style relocation accessors in LoadFull
Cherry Zhang [Sat, 28 Mar 2020 18:58:28 +0000 (14:58 -0400)]
[dev.link] cmd/link: use new-style relocation accessors in LoadFull

This gives some small speedup:

(linking cmd/compile)
Loadlibfull     77.8ms ± 6%    68.6ms ± 5%  -11.88%  (p=0.008 n=5+5)

Also convert some Relocs.At to At2, which should have been done
earlier.

Change-Id: I2a66aeb5857234c6e645e1b23380149cffc8221f
Reviewed-on: https://go-review.googlesource.com/c/go/+/226363
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
5 years ago[dev.link] cmd/link, cmd/oldlink: detect object file format mismatch
Cherry Zhang [Wed, 25 Mar 2020 16:06:59 +0000 (12:06 -0400)]
[dev.link] cmd/link, cmd/oldlink: detect object file format mismatch

When using the new(old) linker but an old(new) object file is
found, give a better error message.

Change-Id: I94786f1a4b527c15c4f5b00457eab60d215a72a0
Reviewed-on: https://go-review.googlesource.com/c/go/+/225457
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: convert windynrelocsyms to loader APIs
Than McIntosh [Tue, 24 Mar 2020 13:20:01 +0000 (09:20 -0400)]
[dev.link] cmd/link: convert windynrelocsyms to loader APIs

Convert the linker's windynrelocsyms method to work with loader APIs.

Change-Id: I8c1fb153bfae8b5c6893876abc9534c34267a42c
Reviewed-on: https://go-review.googlesource.com/c/go/+/225200
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link/internal/loader: add Loader plt/got access methods
Than McIntosh [Tue, 24 Mar 2020 13:23:09 +0000 (09:23 -0400)]
[dev.link] cmd/link/internal/loader: add Loader plt/got access methods

Add accessor methods to get at the symbol {plt,got} value for
PE symbols. Fix a bug in the loaders SetPlt/SetGot methods.

Change-Id: I975bd6b86122622b206487c8798f8290ecd25a57
Reviewed-on: https://go-review.googlesource.com/c/go/+/225199
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: convert callgraph to new style
Cherry Zhang [Fri, 27 Mar 2020 16:16:07 +0000 (12:16 -0400)]
[dev.link] cmd/link: convert callgraph to new style

Change-Id: I24ca987d1264e5b568c60c0dee7334a117473e52
Reviewed-on: https://go-review.googlesource.com/c/go/+/226117
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] all: merge branch 'master' into dev.link
Cherry Zhang [Fri, 27 Mar 2020 17:39:19 +0000 (13:39 -0400)]
[dev.link] all: merge branch 'master' into dev.link

The only merge conflict is the addition of -spectre flag on
master and the addition of -go115newobj flag on dev.link.
Resolved trivially.

Change-Id: I5b46c2b25e140d6c3d8cb129acbd7a248ff03bb9

5 years agoruntime: converge duplicate calls to netpollBreak into one
Andy Pan [Fri, 27 Mar 2020 03:21:17 +0000 (03:21 +0000)]
runtime: converge duplicate calls to netpollBreak into one

There might be some concurrent (maybe not concurrent, just sequential but in a short time window) and duplicate calls to `netpollBreak`, trying to wake up a net-poller. If one has called `netpollBreak` and that waking event hasn't been received by epollwait/kevent/..., then the subsequent calls of `netpollBreak` ought to be ignored or in other words, these calls should be converged into one.

Benchmarks go1.13.5 darwin/amd64:

benchmark-func           time/op (old)  time/op (new)  delta
BenchmarkNetpollBreak-4  29668ns ±1%    3131ns ±2%     -89.45%

mem/B (old)  mem/B (new)  delta
154B ±13%    0B ±0%       -100%

Change-Id: I3cf757a5d6edc5a99adad7aea3baee4b7f2a8f5c
GitHub-Last-Rev: 15bcfbab8a5db51f65da01315a5880a5dbf9e028
GitHub-Pull-Request: golang/go#36294
Reviewed-on: https://go-review.googlesource.com/c/go/+/212737
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agocmd/compile: add rules to eliminate unnecessary signed shifts
Lynn Boger [Thu, 26 Mar 2020 20:01:40 +0000 (16:01 -0400)]
cmd/compile: add rules to eliminate unnecessary signed shifts

This change to the rules removes some unnecessary signed shifts
that appear in the math/rand functions. Existing rules did not
cover some of the signed cases.

A little improvement seen in math/rand due to removing 1 of 2
instructions generated for Int31n, which is inlined quite a bit.

Intn1000                 46.9ns ± 0%  45.5ns ± 0%   -2.99%  (p=1.000 n=1+1)
Int63n1000               33.5ns ± 0%  32.8ns ± 0%   -2.09%  (p=1.000 n=1+1)
Int31n1000               32.7ns ± 0%  32.6ns ± 0%   -0.31%  (p=1.000 n=1+1)
Float32                  32.7ns ± 0%  30.3ns ± 0%   -7.34%  (p=1.000 n=1+1)
Float64                  21.7ns ± 0%  20.9ns ± 0%   -3.69%  (p=1.000 n=1+1)
Perm3                     205ns ± 0%   202ns ± 0%   -1.46%  (p=1.000 n=1+1)
Perm30                   1.71µs ± 0%  1.68µs ± 0%   -1.35%  (p=1.000 n=1+1)
Perm30ViaShuffle         1.65µs ± 0%  1.65µs ± 0%   -0.30%  (p=1.000 n=1+1)
ShuffleOverhead          2.83µs ± 0%  2.83µs ± 0%   -0.07%  (p=1.000 n=1+1)
Read3                    18.7ns ± 0%  16.1ns ± 0%  -13.90%  (p=1.000 n=1+1)
Read64                    126ns ± 0%   124ns ± 0%   -1.59%  (p=1.000 n=1+1)
Read1000                 1.75µs ± 0%  1.63µs ± 0%   -7.08%  (p=1.000 n=1+1)

Change-Id: I11502dfca7d65aafc76749a8d713e9e50c24a858
Reviewed-on: https://go-review.googlesource.com/c/go/+/225917
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years agocmd/go/internal/modfetch/codehost: remove unused GitRepo function
Bryan C. Mills [Fri, 27 Mar 2020 15:26:37 +0000 (11:26 -0400)]
cmd/go/internal/modfetch/codehost: remove unused GitRepo function

For #37943

Change-Id: Ib8ba5d846f41afc0047c33b8145918d93326cdd0
Reviewed-on: https://go-review.googlesource.com/c/go/+/225937
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
5 years agocmd/go/internal/work: disallow testgo binary from installing to GOROOT
Bryan C. Mills [Fri, 27 Mar 2020 03:10:32 +0000 (23:10 -0400)]
cmd/go/internal/work: disallow testgo binary from installing to GOROOT

Installing to GOROOT makes tests non-parallelizable, since each test
depends on the installed contents of GOROOT already being up-to-date
and may reasonably assume that those contents do not change over the
course of the test.

Fixes #37573
Updates #30316

Change-Id: I2afe95ad11347bee3bb7c2d77a657db6d691cf05
Reviewed-on: https://go-review.googlesource.com/c/go/+/225897
Reviewed-by: Michael Matloob <matloob@golang.org>
5 years agorun.{bash,bat,rc}: use ../bin/go instead of the go binary in $PATH
Bryan C. Mills [Tue, 17 Mar 2020 14:22:39 +0000 (10:22 -0400)]
run.{bash,bat,rc}: use ../bin/go instead of the go binary in $PATH

https://golang.org/doc/contribute.html#quick_test currently suggests
running 'make.bash' and 'run.bash' separately, but 'run.bash'
potentially uses a 'go' command resolved from the wrong GOROOT,
which in turn sets the wrong GOROOT for further commands.

Updates #32674
Updates #17896

Change-Id: I4925d478d0fc7351c4f6d40830ab17d4d688348d
Reviewed-on: https://go-review.googlesource.com/c/go/+/223741
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
Reviewed-by: Katie Hockman <katie@golang.org>
5 years agoRevert "cmd/go: add support for GOPROXY fallback on unexpected errors"
Bryan C. Mills [Fri, 27 Mar 2020 03:07:34 +0000 (03:07 +0000)]
Revert "cmd/go: add support for GOPROXY fallback on unexpected errors"

This reverts CL 223257.

Reason for revert: broke TestScript/mod_gonoproxy on the longtest builders.

Change-Id: I8637c52c5a7d5333a37ed1e9998c49786525ecb1
Reviewed-on: https://go-review.googlesource.com/c/go/+/225757
Reviewed-by: Michael Matloob <matloob@golang.org>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/dist: skip API check on plan9 builders
Bryan C. Mills [Thu, 26 Mar 2020 16:00:07 +0000 (12:00 -0400)]
cmd/dist: skip API check on plan9 builders

The plan9-arm builder has a very slow filesystem and frequently times
out on this test. The api check verifies the API for all supported
GOOS/GOARCH/CGO_ENABLED combination anyway, so if we skip it on one
builder (or even most builders) there should be no loss of coverage.

Updates #37951

Change-Id: I86a93df2ec60a6af6d942e3954eef09ce67bb39e
Reviewed-on: https://go-review.googlesource.com/c/go/+/225662
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Carlos Amedee <carlos@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
5 years ago[dev.link] cmd/link: migrate symbol's Value and Align when converting to sym.Symbols
Cherry Zhang [Thu, 26 Mar 2020 18:43:15 +0000 (14:43 -0400)]
[dev.link] cmd/link: migrate symbol's Value and Align when converting to sym.Symbols

Currently, in LoadFull we migrate a symbol's Value to sym.Symbol
only for external symbols. And symbol's Align is not migrated at
all. As we move LoadFull forward, there are already places where
we set symbol's Value and Align (e.g. in doelf). Migrate them
correctly.

Currently I think we only set them on external symbols, but as
we move forward I think we'll need to set them on Go symbols as
well.

Change-Id: I63e97e38fc08b653ba9faefe15697944faf21bed
Reviewed-on: https://go-review.googlesource.com/c/go/+/225658
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
5 years ago[dev.link] cmd/link: convert textbuildid pass to new style
Cherry Zhang [Thu, 26 Mar 2020 16:54:21 +0000 (12:54 -0400)]
[dev.link] cmd/link: convert textbuildid pass to new style

Change-Id: Ic3a7bfc8b0290bd7bdc71e64cab74788328c41d1
Reviewed-on: https://go-review.googlesource.com/c/go/+/225657
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
5 years agocmd/go: add support for GOPROXY fallback on unexpected errors
Jay Conrod [Thu, 12 Mar 2020 22:56:40 +0000 (18:56 -0400)]
cmd/go: add support for GOPROXY fallback on unexpected errors

URLs in GOPROXY may now be separated with commas (,) or pipes (|). If
a request to a proxy fails with any error (including connection errors
and timeouts) and the proxy URL is followed by a pipe, the go command
will try the request with the next proxy in the list. If the proxy is
followed by a comma, the go command will only try the next proxy if
the error a 404 or 410 HTTP response.

The go command will determine how to connect to the checksum database
using the same logic. Before accessing the checksum database, the go
command sends a request to <proxyURL>/sumdb/<sumdb-name>/supported.
If a proxy responds with 404 or 410, or if any other error occurs and
the proxy URL in GOPROXY is followed by a pipe, the go command will
try the request with the next proxy. If all proxies respond with 404
or 410 or are configured to fall back on errors, the go command will
connect to the checksum database directly.

This CL does not change the default value or meaning of GOPROXY.

Fixes #37367

Change-Id: If53152ec1c3282c67d4909818b666af58884fb2c
Reviewed-on: https://go-review.googlesource.com/c/go/+/223257
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
5 years agocmd/dist: do not skip 'cmd' tests in race mode
Bryan C. Mills [Thu, 19 Mar 2020 17:00:44 +0000 (13:00 -0400)]
cmd/dist: do not skip 'cmd' tests in race mode

Fixes #37940

Change-Id: Ib869a4bf84296dac201cc7252431d7161b9c96f4
Reviewed-on: https://go-review.googlesource.com/c/go/+/224038
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alexander Rakoczy <alex@golang.org>
5 years agocmd/go/internal/base: rename EnvForDir to AppendPWD
Bryan C. Mills [Thu, 26 Mar 2020 02:17:04 +0000 (22:17 -0400)]
cmd/go/internal/base: rename EnvForDir to AppendPWD

EnvForDir does not immediately evoke “append”, and thus may not prompt
the reader to consider the possibility of aliasing bugs (as in
issue #38077). To make this behavior more obvious at the call site, rename
cmd/go/internal/base.EnvForDir to AppendPWD and swap the order of
arguments to a conventional “append” function (similar to those in the
strconv package).

For #38077

Change-Id: I16f09aa0fa8a269d51f0511eb402a44e2759eb94
Reviewed-on: https://go-review.googlesource.com/c/go/+/225578
Reviewed-by: Jay Conrod <jayconrod@google.com>
Reviewed-by: Michael Matloob <matloob@golang.org>
5 years agocmd/go: do not append to the global cfg.OrigEnv slice
Bryan C. Mills [Thu, 26 Mar 2020 02:24:44 +0000 (22:24 -0400)]
cmd/go: do not append to the global cfg.OrigEnv slice

Appending to a global slice is only safe if its length is already
equal to its capacity. That property is not guaranteed for slices in
general, and empirically does not hold for this one.

This is a minimal fix to make it easier to backport.
A more robust cleanup of the base.EnvForDir function will be sent in a
subsequent CL.

Fixes #38077
Updates #37940

Change-Id: I731d5bbd0e516642c2cf43e713eeea15402604e5
Reviewed-on: https://go-review.googlesource.com/c/go/+/225577
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
Reviewed-by: Michael Matloob <matloob@golang.org>
5 years agoruntime: improve MIPS64x memclr
Meng Zhuo [Tue, 24 Mar 2020 07:07:43 +0000 (15:07 +0800)]
runtime: improve MIPS64x memclr

Using MIPS MSA VLD/VST to improve mips64x large memclr.

name          old time/op    new time/op     delta
Memclr/5        23.2ns ± 0%     21.5ns ± 0%    -7.33%  (p=0.000 n=9+8)
Memclr/16       20.1ns ± 0%     17.1ns ± 0%   -14.93%  (p=0.000 n=10+10)
Memclr/64       27.2ns ± 0%     19.1ns ± 0%   -29.70%  (p=0.000 n=9+9)
Memclr/256      76.8ns ± 0%     24.1ns ± 0%   -68.66%  (p=0.000 n=10+10)
Memclr/4096     1.12µs ± 1%     0.18µs ± 0%   -84.32%  (p=0.000 n=10+8)
Memclr/65536    18.0µs ± 0%      2.8µs ± 0%   -84.29%  (p=0.000 n=10+10)
Memclr/1M        288µs ± 0%       45µs ± 0%   -84.20%  (p=0.000 n=10+10)
Memclr/4M       1.15ms ± 0%     0.18ms ± 0%   -84.21%  (p=0.000 n=9+10)
Memclr/8M       2.34ms ± 0%     1.39ms ± 0%   -40.55%  (p=0.000 n=10+8)
Memclr/16M      4.72ms ± 0%     4.74ms ± 0%    +0.52%  (p=0.000 n=9+10)
Memclr/64M      18.9ms ± 0%     18.9ms ± 0%      ~     (p=0.436 n=10+10)
GoMemclr/5      13.7ns ± 0%     16.9ns ± 0%   +23.36%  (p=0.000 n=10+10)
GoMemclr/16     14.3ns ± 0%      9.0ns ± 0%   -37.27%  (p=0.000 n=10+9)
GoMemclr/64     26.9ns ± 0%     13.7ns ± 0%   -49.07%  (p=0.000 n=10+10)
GoMemclr/256    77.8ns ± 0%     13.0ns ± 0%   -83.24%  (p=0.000 n=9+10)

name          old speed      new speed       delta
Memclr/5       215MB/s ± 0%    232MB/s ± 0%    +7.74%  (p=0.000 n=9+9)
Memclr/16      795MB/s ± 0%    935MB/s ± 0%   +17.60%  (p=0.000 n=10+10)
Memclr/64     2.35GB/s ± 0%   3.35GB/s ± 0%   +42.33%  (p=0.000 n=10+9)
Memclr/256    3.34GB/s ± 0%  10.65GB/s ± 0%  +219.16%  (p=0.000 n=10+10)
Memclr/4096   3.65GB/s ± 1%  23.30GB/s ± 0%  +538.36%  (p=0.000 n=10+10)
Memclr/65536  3.65GB/s ± 0%  23.21GB/s ± 0%  +536.59%  (p=0.000 n=10+10)
Memclr/1M     3.64GB/s ± 0%  23.07GB/s ± 0%  +532.96%  (p=0.000 n=10+10)
Memclr/4M     3.64GB/s ± 0%  23.08GB/s ± 0%  +533.36%  (p=0.000 n=9+10)
Memclr/8M     3.58GB/s ± 0%   6.02GB/s ± 0%   +68.20%  (p=0.000 n=10+8)
Memclr/16M    3.56GB/s ± 0%   3.54GB/s ± 0%    -0.51%  (p=0.000 n=9+10)
Memclr/64M    3.55GB/s ± 0%   3.55GB/s ± 0%      ~     (p=0.436 n=10+10)
GoMemclr/5     364MB/s ± 0%    296MB/s ± 0%   -18.76%  (p=0.000 n=9+10)
GoMemclr/16   1.12GB/s ± 0%   1.78GB/s ± 0%   +58.86%  (p=0.000 n=10+10)
GoMemclr/64   2.38GB/s ± 0%   4.66GB/s ± 0%   +96.27%  (p=0.000 n=10+9)
GoMemclr/256  3.29GB/s ± 0%  19.62GB/s ± 0%  +496.45%  (p=0.000 n=10+9)

Change-Id: I457858368f2875fd66818a41d2f0c190a850e8f1
Reviewed-on: https://go-review.googlesource.com/c/go/+/218177
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years agoruntime: ensure minTriggerRatio never exceeds maxTriggerRatio
Michael Anthony Knyszek [Wed, 18 Mar 2020 15:09:40 +0000 (15:09 +0000)]
runtime: ensure minTriggerRatio never exceeds maxTriggerRatio

Currently, the capping logic for the GC trigger ratio is such that if
gcpercent is low, we may end up setting the trigger ratio far too high,
breaking the promise of SetGCPercent and GOGC has a trade-off knob (we
won't start a GC early enough, and we will use more memory).

This change modifies the capping logic for the trigger ratio by scaling
the minTriggerRatio with gcpercent the same way we scale
maxTriggerRatio.

Fixes #37927.

Change-Id: I2a048c1808fb67186333d3d5a6bee328be2f35da
Reviewed-on: https://go-review.googlesource.com/c/go/+/223937
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
5 years agocrypto/rsa,crypto/ecdsa,crypto/ed25519: implement PublicKey.Equal
Filippo Valsorda [Wed, 18 Mar 2020 00:34:51 +0000 (20:34 -0400)]
crypto/rsa,crypto/ecdsa,crypto/ed25519: implement PublicKey.Equal

This makes all modern public keys in the standard library implement a
common interface (below) that can be used by applications for better
type safety and allows for checking that public (and private keys via
Public()) are equivalent.

interface {
    Equal(crypto.PublicKey) bool
}

Equality for ECDSA keys is complicated, we take a strict interpretation
that works for all secure applications (the ones not using the
unfortunate non-constant time CurveParams implementation) and fails
closed otherwise.

Tests in separate files to make them x_tests and avoid an import loop
with crypto/x509.

Re-landing of CL 223754. Dropped the test that was assuming named curves
are not implemented by CurveParams, because it's not true for all
curves, and anyway is not a property we need to test. There is still a
test to check that different curves make keys not Equal.

Fixes #21704
Fixes #38035

Reviewed-on: https://go-review.googlesource.com/c/go/+/223754
Reviewed-by: Katie Hockman <katie@golang.org>
Change-Id: I736759b145bfb4f7f8eecd78c324315d5a05385c
Reviewed-on: https://go-review.googlesource.com/c/go/+/225460
Run-TryBot: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years ago[dev.link] cmd/internal/goobj2: bump up version number
Cherry Zhang [Wed, 25 Mar 2020 16:26:32 +0000 (12:26 -0400)]
[dev.link] cmd/internal/goobj2: bump up version number

As we now have -go115newobj flag, it is better to use go115 in
the object file as well. And it already diverges from the go114
"new" object file format.

Change-Id: I315edf7524158b5c354393fe9a7ab9f6d7cc9808
Reviewed-on: https://go-review.googlesource.com/c/go/+/225458
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years ago[dev.link] cmd/link: fix buglet in setupdynexp
Cherry Zhang [Wed, 25 Mar 2020 21:22:57 +0000 (17:22 -0400)]
[dev.link] cmd/link: fix buglet in setupdynexp

This should restore deterministic order of dynexp, and fix
Solaris build.

Change-Id: Icb796babaa3238bff90fd8255ee9f023f2306c26
Reviewed-on: https://go-review.googlesource.com/c/go/+/225538
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: define _etext, etc. in the linker on Solaris
Cherry Zhang [Wed, 25 Mar 2020 21:10:16 +0000 (17:10 -0400)]
[dev.link] cmd/link: define _etext, etc. in the linker on Solaris

On Solaris, in the runtime it defines the external name of
runtime.etext as _etext (runtime/os3_solaris.go:13). In CL 224939
we changed to put external names in the ELF symbol table more
consistently. In this case it will contain _etext but not
runtime.etext.

To be conservative, this CL defines both runtime.etext and _text
in the linker.

Change-Id: I79f196e87b655042be97b0fbbab02d0ebc8db2fa
Reviewed-on: https://go-review.googlesource.com/c/go/+/225537
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years agoruntime: fix threshold calculation of TestPhysicalMemoryUtilization
Xiangdong Ji [Thu, 5 Dec 2019 03:22:34 +0000 (03:22 +0000)]
runtime: fix threshold calculation of TestPhysicalMemoryUtilization

Variable 'procs' used to calculate the threshold of overuse in
TestPhysicalMemoryUtilization should be updated if GOMAXPROCS
gets changed, otherwise the threshold could be a large number,
making the test meaningless.

Change-Id: I876cbf11457529f56bae77af1e35f4538a721f95
Reviewed-on: https://go-review.googlesource.com/c/go/+/210297
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
5 years agoscan: for style, adjust code for bad scan read counts
Rob Pike [Wed, 25 Mar 2020 22:32:27 +0000 (09:32 +1100)]
scan: for style, adjust code for bad scan read counts

Make the code more consistent with the rest of the file.
Should have caught this in review of CL 225357.

Change-Id: I12824cb436539c31604684e043ebb7587cc92471
Reviewed-on: https://go-review.googlesource.com/c/go/+/225557
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

5 years agotest/chan: fix trivial typos
sjnam [Wed, 25 Mar 2020 21:02:44 +0000 (21:02 +0000)]
test/chan: fix trivial typos

Substition -> Substitution

Change-Id: Iede578d733d1c041133742b61eb0573c3bd3b17c
GitHub-Last-Rev: 7815bd346dbe23a5412d0602a4f6480b893c447d
GitHub-Pull-Request: golang/go#38059
Reviewed-on: https://go-review.googlesource.com/c/go/+/225417
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agocmd/internal/obj/arm64: write test output to temp dir
Cherry Zhang [Wed, 25 Mar 2020 15:47:02 +0000 (11:47 -0400)]
cmd/internal/obj/arm64: write test output to temp dir

Write the test output to the temporary directory, not the current
directory.

May fix linux-mips64le-mengzhuo builder.

Change-Id: Ibfeb3d2879c11d498abc31df4efe776fc09a6ad6
Reviewed-on: https://go-review.googlesource.com/c/go/+/225440
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Bonventre <andybons@golang.org>
5 years agoruntime: prevent preemption while timer is in timerModifying
Michael Anthony Knyszek [Wed, 25 Mar 2020 16:41:20 +0000 (16:41 +0000)]
runtime: prevent preemption while timer is in timerModifying

Currently if a goroutine is preempted while owning a timer in the
timerModifying state, it could self-deadlock. When the goroutine is
preempted and calls into the scheduler, it could call checkTimers. If
checkTimers encounters the timerModifying timer and calls runtimer on
it, then runtimer will spin, waiting for that timer to leave the
timerModifying state, which it never will.

So far we got lucky that for the most part that there were no preemption
points while timerModifying is happening, however CL 221077 seems to
have introduced one, leading to sporadic self-deadlocks.

This change disables preemption explicitly while a goroutines holds a
timer in timerModifying. Since only checkTimers (and thus runtimer) is
called from the scheduler, this is sufficient to prevent
preemption-based self-deadlocks.

Fixes #38070.
Updates #37894.

Change-Id: Idbfac310889c92773023733ff7e2ff87e9896f0c
Reviewed-on: https://go-review.googlesource.com/c/go/+/225497
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agobufio: don't panic when Scanner sees an impossible Read count
Ian Lance Taylor [Wed, 25 Mar 2020 03:09:11 +0000 (20:09 -0700)]
bufio: don't panic when Scanner sees an impossible Read count

Fixes #38053

Change-Id: Ib0f9777f37eeaa07eb8ecb6df3e97e9d4b46dcd8
Reviewed-on: https://go-review.googlesource.com/c/go/+/225357
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
5 years agocmd/api: make NewWatcher populate its own package and import metadata
Bryan C. Mills [Fri, 20 Mar 2020 19:15:35 +0000 (15:15 -0400)]
cmd/api: make NewWatcher populate its own package and import metadata

This partially undoes the optimizations of CL 177597, but makes up
some of the difference by caching the package list and import metadata
and making the initial calls concurrently, including in TestMain.
That reduces the critical path from two sequential 'go list'
invocations to just one (run many times concurrently), and eliminates
the need for assumptions about the consistency of the 'std' dependency
graph across platforms (and hard-coded special cases for packages that
violate those assumptions).

In the process, this simplifies and fixes TestBenchmark (which has
been silently broken since CL 164623).

This increases 'time go tool dist test api' on my workstation from
0m8.4s / 0m13.8s / 0m1.7s to 0m10.5s / 0m23.1s / 0m5.1s,
compared to 0m12.4s / 0m23.2s / 0m4.7s before CL 177597.

(That is, this change retains about half of the wall-time speedup, but
almost none of the user-time speedup.)

Tested manually using 'go test -race -bench=. cmd/api'.

Fixes #37951

Change-Id: Icd537e035e725e1ee7c41d97da5c6651233b927e
Reviewed-on: https://go-review.googlesource.com/c/go/+/224619
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alexander Rakoczy <alex@golang.org>
5 years agocmd/link/internal/arm64: increase the function alignment to 16
fanzha02 [Wed, 25 Mar 2020 03:11:24 +0000 (11:11 +0800)]
cmd/link/internal/arm64: increase the function alignment to 16

On arm64, a function's address is 16 bytes aligned, and
the assembler aligns the size of function symbol to 16 bytes,
so to keep the consistent, this patch changes the function
alignment in the linker to 16 bytes.

Change-Id: I4d1e89a56200453b7b586fe3f4656bada7544214
Reviewed-on: https://go-review.googlesource.com/c/go/+/225397
Reviewed-by: eric fang <eric.fang@arm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years ago[dev.link] cmd/link: remove lookup function from relocsym
Cherry Zhang [Tue, 24 Mar 2020 18:35:45 +0000 (14:35 -0400)]
[dev.link] cmd/link: remove lookup function from relocsym

The lookup is only used for DWARF section symbols, for which we
can just link the symbols to the sections.

Change-Id: Id8426fbf59bab2528f57e28e2043e0b405656a9f
Reviewed-on: https://go-review.googlesource.com/c/go/+/225204
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: pass loader to relocation functions
Cherry Zhang [Tue, 24 Mar 2020 16:55:14 +0000 (12:55 -0400)]
[dev.link] cmd/link: pass loader to relocation functions

It is needed in Xcoffadddynrel. And it will be needed once we
move more things to new style (the sym.Symbol parameters will
also need to change).

Change-Id: Ie12683f9b44e21f1e6ea711bf2f4c5c32282e5b3
Reviewed-on: https://go-review.googlesource.com/c/go/+/225203
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: convert doxcoff to new style
Cherry Zhang [Tue, 24 Mar 2020 16:08:36 +0000 (12:08 -0400)]
[dev.link] cmd/link: convert doxcoff to new style

Change-Id: Ic1e4ed6c14e049b1ba2f7c00f986433ab7ebe932
Reviewed-on: https://go-review.googlesource.com/c/go/+/225202
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years agocmd/internal/obj/arm64: add support of PCALIGN directive
diaxu01 [Fri, 6 Sep 2019 09:55:54 +0000 (09:55 +0000)]
cmd/internal/obj/arm64: add support of PCALIGN directive

Recently, we get requirements of instructions and functions alignment
from the gVisor project. To fit the alignment requirement of interrupt
table, they require an instruction's address to be aligned 128 bytes
and a function's entry address to be aligned 2K bytes. Thus we add
support for PCALIGN directive first. Below is a discussion about this
topic. https://groups.google.com/forum/m/#!topic/golang-dev/RPj90l5x86I

Functions in Go are aligned to 16 bytes on arm64, thus now we only
support 8 and 16 bytes alignment.

This patch adds support for PCALIGN directive. This directive can be
used within Go asm to align instruction by padding NOOP directives.

This patch also adds a test to verify the correnctness of the PCALIGN
directive. The test is contributed by Fannie Zhang <Fannie.Zhang@arm.com>.

Change-Id: I709e6b94847fe9e1824f42f4155355f90c63d523
Reviewed-on: https://go-review.googlesource.com/c/go/+/207117
Reviewed-by: eric fang <eric.fang@arm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agotest: make runindir tests pass regardless of whether module mode is in use
Bryan C. Mills [Tue, 24 Mar 2020 20:18:02 +0000 (16:18 -0400)]
test: make runindir tests pass regardless of whether module mode is in use

The "runindir" tests used "go run", but relied on relative imports
(which are not supported by "go run" in module mode). Instead, such
tests must use fully-qualified imports, which require either a go.mod
file (in module mode) or that the package be in an appropriate
subdirectory of GOPATH/src (in GOPATH mode).

To set up such a directory, we use yet another copy of the same
overlayDir function currently found in the misc subdirectory of this
repository.

Fixes #33912
Updates #30228

Change-Id: If3d7ea2f7942ba496d98aaaf24a90bcdcf4df9f7
Reviewed-on: https://go-review.googlesource.com/c/go/+/225205
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agocmd/compile: use load and test instructions on s390x
Ruixin(Peter) Bao [Tue, 26 Nov 2019 20:33:37 +0000 (15:33 -0500)]
cmd/compile: use load and test instructions on s390x

The load and test instructions compare the given value
against zero and will produce a condition code indicating
one of the following scenarios:

0: Result is zero
1: Result is less than zero
2: Result is greater than zero
3: Result is not a number (NaN)

The instruction can be used to simplify floating point comparisons
against zero, which can enable further optimizations.

This CL also reduces the size of .text section of math.test binary by around
0.7 KB (in hexadecimal, from 1358f0 to 135620).

Change-Id: I33cb714f0c6feebac7a1c46dfcc735e7daceff9c
Reviewed-on: https://go-review.googlesource.com/c/go/+/209159
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years ago[dev.link] cmd/link: convert addexport to loader interfaces
Than McIntosh [Tue, 24 Mar 2020 13:05:19 +0000 (09:05 -0400)]
[dev.link] cmd/link: convert addexport to loader interfaces

Convert the linker's "addexport" function to use loader interfaces
for symbol manipulation instead of *sym.Symbol.

At the moment "addexport" happens after loadlibfull (there are other
chunks of functionality in the way that haven't been converted), so
this implementation contains temporary shim code to copy back the
contents of updated loader.Sym's into the corresponding sym.Symbol.

Change-Id: I867b08e66562a2bed51560fd0be2cb64d344709c
Reviewed-on: https://go-review.googlesource.com/c/go/+/224384
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: add loader.Sym equivalents to ArchSyms
Than McIntosh [Wed, 18 Mar 2020 17:26:38 +0000 (13:26 -0400)]
[dev.link] cmd/link: add loader.Sym equivalents to ArchSyms

Populate ArchSyms with loader.Sym versions of important symbols,
so as to be able to convert the code that uses ArchSyms to the
new loader interfaces.

Change-Id: If6766f7164642a9dc2e31fcf7f9280a6e95e5d23
Reviewed-on: https://go-review.googlesource.com/c/go/+/224383
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years ago[dev.link] cmd/link: move setupdynexp before loadlibfull
Than McIntosh [Fri, 20 Mar 2020 12:31:38 +0000 (08:31 -0400)]
[dev.link] cmd/link: move setupdynexp before loadlibfull

Rewrite setupdynexp to work with loader.Sym, and move the call to it
before the body of loadlibfull. After loadlibfull is complete,
construct the old *sym.Symbol version of dynexp, since not all all
clients that access this list are converted to the loader APIs.

Change-Id: I347d24958e2f3e2332fbe33f2eb6ec25cc126bdb
Reviewed-on: https://go-review.googlesource.com/c/go/+/224382
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years ago[dev.link] cmd/link: minor tweaks to PropagateLoaderChangesToSymbols
Than McIntosh [Thu, 19 Mar 2020 15:04:37 +0000 (11:04 -0400)]
[dev.link] cmd/link: minor tweaks to PropagateLoaderChangesToSymbols

Update PropagateLoaderChangesToSymbols so that it no longer requires
a sym.Symbols pointer. The intent is to generalize it a little to
allow it to be used in more than just linker Dwarf generation.

Change-Id: I9bddc5d39839eacd9113c945bb59d2873c0b088c
Reviewed-on: https://go-review.googlesource.com/c/go/+/224381
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
5 years ago[dev.link] cmd/link: relocating Errorf() to ErrorReporter
Than McIntosh [Thu, 19 Mar 2020 12:17:54 +0000 (08:17 -0400)]
[dev.link] cmd/link: relocating Errorf() to ErrorReporter

Add an Errorf method to ErrorReporter. The hope is that we can
consolidate error handling/reporting in this helper, and eventually
do away with Link.Errorf and the global Errorf function (which
can be removed once we've eliminated enough uses of *sym.Symbol).

Change-Id: Ie1147020b8409b9c57acfd71c942b287b214afca
Reviewed-on: https://go-review.googlesource.com/c/go/+/224380
Reviewed-by: Jeremy Faller <jeremy@golang.org>
5 years agomath: correct Atan2(±y,+∞) = ±0 on s390x
Brian Kessler [Sat, 14 Mar 2020 05:24:38 +0000 (23:24 -0600)]
math: correct Atan2(±y,+∞) = ±0 on s390x

The s390x assembly implementation was previously only handling this
case correctly for x = -Pi.  Update the special case handling for
any y.

Fixes #35446

Change-Id: I355575e9ec8c7ce8bd9db10d74f42a22f39a2f38
Reviewed-on: https://go-review.googlesource.com/c/go/+/223420
Run-TryBot: Brian Kessler <brian.m.kessler@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
5 years agocmd/compile: fold constants into immediate instructions on riscv64
Joel Sing [Mon, 9 Mar 2020 16:31:22 +0000 (03:31 +1100)]
cmd/compile: fold constants into immediate instructions on riscv64

Where possible, fold constants into versions of instructions that take
an immediate. This avoids the need to allocate a register and load the
immediate into it.

Change-Id: If911ca41235e218490679aed2ce5f48bf807a2b3
Reviewed-on: https://go-review.googlesource.com/c/go/+/222639
Run-TryBot: Joel Sing <joel@sing.id.au>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agotest: re-enable atomic intrinsic related tests on riscv64
Joel Sing [Sun, 15 Mar 2020 15:58:23 +0000 (02:58 +1100)]
test: re-enable atomic intrinsic related tests on riscv64

riscv64 now has atomic intrinsics, so re-enable the atomic intrinsic tests.

Fixes #36765

Change-Id: I838f27570a94d7fa5774c43f1ca5f4df2ca104cf
Reviewed-on: https://go-review.googlesource.com/c/go/+/223560
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years agocmd/compile: provide Add/Cas/Exchange atomic intrinsics on riscv64
Joel Sing [Sun, 15 Mar 2020 15:51:54 +0000 (02:51 +1100)]
cmd/compile: provide Add/Cas/Exchange atomic intrinsics on riscv64

Provide Add32, Add64, Cas32, Cas64, Exchange32 and Exchange64 atomic
intrinsics on riscv64.

Updates #36765

Change-Id: I9a3b7d2ce3d49f699171fd76a0fed891d149a6bb
Reviewed-on: https://go-review.googlesource.com/c/go/+/223559
Run-TryBot: Joel Sing <joel@sing.id.au>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agoos/exec: for TestExtraFiles failure, print readlink of unexpected fd
Ian Lance Taylor [Tue, 24 Mar 2020 20:13:38 +0000 (13:13 -0700)]
os/exec: for TestExtraFiles failure, print readlink of unexpected fd

For #25628

Change-Id: If1dce7ba9310e1418e67b9954c989471b775a28e
Reviewed-on: https://go-review.googlesource.com/c/go/+/225278
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/internal/obj/ppc64: fix PCALIGN on ppc64le
Lynn Boger [Tue, 17 Mar 2020 13:24:47 +0000 (09:24 -0400)]
cmd/internal/obj/ppc64: fix PCALIGN on ppc64le

This fixes a potential issue with the previous implementation
of PCALIGN on ppc64. Previously PCALIGN was processed inside of
asmout and indicated the padding size by setting the value in
the optab, changing it back after the alignment instructions
were added. Now PCALIGN is processed outside of asmout, and optab
is not changed.

Change-Id: I8b0093a0e2b7e06176af27e05150d04ae2c55d60
Reviewed-on: https://go-review.googlesource.com/c/go/+/225198
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agointernal/poll: assume we have CancelIoEX on Windows
Ian Lance Taylor [Tue, 24 Mar 2020 04:11:01 +0000 (21:11 -0700)]
internal/poll: assume we have CancelIoEX on Windows

As of the Go 1.11 release we require at least Windows 7, so CancelIoEx
is always available.  This lets us simplify the code to not require
dedicated threads to handle I/O requests.

Fixes #37956

Change-Id: If1dc4ac4acb61c43e4f2a9f26f225869050262a5
Reviewed-on: https://go-review.googlesource.com/c/go/+/225060
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
5 years agoruntime: fix code so defer record is not added to g0 defer list during panic
Dan Scales [Fri, 20 Mar 2020 16:31:20 +0000 (09:31 -0700)]
runtime: fix code so defer record is not added to g0 defer list during panic

newdefer() actually adds the new defer to the current g's defer chain. That
happens even if we are on the system stack, in which case the g will be the g0
stack. For open-coded defers, we call newdefer() (only during panic processing)
while on the system stack, so the new defer is unintentionally added to the
g0._defer defer list. The code later correctly adds the defer to the user g's
defer list.

The g0._defer list is never used. However, that pointer on the g0._defer list can
keep a defer struct alive that is intended to be garbage-collected (smaller defers
use a defer pool, but larger-sized defer records are just GC'ed). freedefer() does
not zero out pointers when it intends that a defer become garbage-collected. So,
we can have the pointers in a defer that is held alive by g0._defer become invalid
(in particular d.link). This is the cause of the bad pointer bug in this issue

The fix is to change newdefer (only used in two places) to not add the new defer
to the gp._defer list. We just do it after the call with the correct gp pointer.
(As mentioned above, this code was already there after the newdefer in
addOneOpenDeferFrame.) That ensures that defers will be correctly
garbage-collected and eliminate the bad pointer.

This fix definitely fixes the original repro. I added a test and tried hard to
reproduce the bug (based on the original repro code), but awasn't actually able to
cause the bug. However, the test is still an interesting mix of heap-allocated,
stack-allocated, and open-coded defers.

Fixes #37688

Change-Id: I1a481b9d9e9b9ba4e8726ef718a1f4512a2d6faf
Reviewed-on: https://go-review.googlesource.com/c/go/+/224581
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agoruntime: always use GetQueuedCompletionStatusEx on Windows
Ian Lance Taylor [Tue, 24 Mar 2020 03:42:29 +0000 (20:42 -0700)]
runtime: always use GetQueuedCompletionStatusEx on Windows

We used to fall back to GetQueuedCompletionStatus if
GetQueuedCompletionStatus was not available, but as of Go 1.11 we
require Windows 7 or later, so GetQueuedCompletionStatusEx is always
available.

Fixes #37957

Change-Id: I7d8d49a92ab7b1f5afdc54a442f696aaf4a5168e
Reviewed-on: https://go-review.googlesource.com/c/go/+/225059
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
5 years agoruntime: don't call wakeNetPoller during timerModifying
Ian Lance Taylor [Mon, 23 Mar 2020 18:38:49 +0000 (11:38 -0700)]
runtime: don't call wakeNetPoller during timerModifying

Reduce the length of time that other timer functions can see timerModifying.
In particular avoid system calls.

Fixes #38023

Change-Id: I1b61229c668e6085d9ee6dca9488a90055386c36
Reviewed-on: https://go-review.googlesource.com/c/go/+/224902
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
5 years agocmd/go: clarify behavior with no arguments in 'go help mod download'
Jay Conrod [Tue, 24 Mar 2020 17:33:46 +0000 (13:33 -0400)]
cmd/go: clarify behavior with no arguments in 'go help mod download'

'go mod download' is equivalent to 'go mod download all'.

Fixes #38031

Change-Id: I7aec7e5a1370a3e248eba6daad9a75ec21f33a83
Reviewed-on: https://go-review.googlesource.com/c/go/+/225201
Run-TryBot: Jay Conrod <jayconrod@google.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agoRevert "cmd/compile: convert 386 port to use addressing modes pass"
Keith Randall [Tue, 24 Mar 2020 16:48:05 +0000 (16:48 +0000)]
Revert "cmd/compile: convert 386 port to use addressing modes pass"

This reverts commit CL 222782.

Reason for revert: Reverting to see if 386 errors go away

Update #37881

Change-Id: I74f287404c52414db1b6ff1649effa4ed9e5cc0c
Reviewed-on: https://go-review.googlesource.com/c/go/+/225218
Reviewed-by: Bryan C. Mills <bcmills@google.com>
5 years agoRevert "cmd/compile: disable mem+op operations on 386"
Keith Randall [Tue, 24 Mar 2020 16:46:49 +0000 (16:46 +0000)]
Revert "cmd/compile: disable mem+op operations on 386"

This reverts commit CL 224837.

Reason for revert: Reverting partial reverts of 222782.

Update #37881

Change-Id: Ie9bf84d6e17ed214abe538965e5ff03936886826
Reviewed-on: https://go-review.googlesource.com/c/go/+/225217
Reviewed-by: Bryan C. Mills <bcmills@google.com>
5 years agoRevert "cmd/compile: disable addressingmodes pass for 386"
Keith Randall [Tue, 24 Mar 2020 16:45:38 +0000 (16:45 +0000)]
Revert "cmd/compile: disable addressingmodes pass for 386"

This reverts commit CL 225057.

Reason for revert: Undoing partial reverts of CL 222782

Update #37881

Change-Id: Iee024cab2a580a37a0fc355e0e3c5ad3d8fdaf7d
Reviewed-on: https://go-review.googlesource.com/c/go/+/225197
Reviewed-by: Bryan C. Mills <bcmills@google.com>
5 years agocrypto: implement Hash.String
Katie Hockman [Mon, 23 Mar 2020 14:11:43 +0000 (10:11 -0400)]
crypto: implement Hash.String

Fixes #33430

Change-Id: I323323b3136dd7b408005c3bb5ea05e3b566bd38
Reviewed-on: https://go-review.googlesource.com/c/go/+/224937
Run-TryBot: Katie Hockman <katie@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
5 years ago[dev.link] cmd/link: use Extname in doxcoff
Cherry Zhang [Sat, 21 Mar 2020 20:06:57 +0000 (16:06 -0400)]
[dev.link] cmd/link: use Extname in doxcoff

On AIX, when external linking, we need to change the function
names to start with a dot and make function descriptors with the
names without the dot. Currently this is done through symbol
renaming, which is not friendly for switching to the loader.

In this CL we use symbol's external name for this. This allows us
to get rid of symbol renaming.

Change-Id: If72602d17e96f0339fdac2e2321f1edfb292b5f9
Reviewed-on: https://go-review.googlesource.com/c/go/+/224940
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: convert mangleTypeSym to new style
Cherry Zhang [Sat, 21 Mar 2020 17:50:46 +0000 (13:50 -0400)]
[dev.link] cmd/link: convert mangleTypeSym to new style

Use symbol's Extname, instead of symbol renaming, for the mangled
names.

The old symbol Rename has an interesting logic of "merging"
symbols, when a symbol is renamed to the name of an existing
symbol. It turns out that this is needed for linking against
shared libraries, where the Go object has a reference to a symbol
with the original name, and the shared libary provides a symbol
under the mangled name. Implement this logic with the loader.

Change-Id: Ib95d7a9c93a52f8e02f4a51ac67240d6ebfc1c6a
Reviewed-on: https://go-review.googlesource.com/c/go/+/224939
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/link: remove global datap, moved to *Link
Jeremy Faller [Mon, 9 Mar 2020 15:08:04 +0000 (11:08 -0400)]
[dev.link] cmd/link: remove global datap, moved to *Link

This change moves datap from global space into the link context. Rather
than having it exist in context, we could have it returned from dodata,
and pass it as a parameter, but it is used in awkward places in the
Arch functions. Easiest for now is just keeping it in the context, until
we more formally move it to slices of loader.Sym.

This is a largely non-functional change.

Change-Id: Ica93bd857c39913ad470a61c63bc8d21704d6308
Reviewed-on: https://go-review.googlesource.com/c/go/+/222664
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: refactor some of dodata
Jeremy Faller [Mon, 9 Mar 2020 14:35:31 +0000 (10:35 -0400)]
[dev.link] cmd/link: refactor some of dodata

This is a non-functional change, just moving things around, making
dodata a more approachable piece of code.

Change-Id: I06fa047cbef2313040a31998fa8d242ccb2fedea
Reviewed-on: https://go-review.googlesource.com/c/go/+/222662
Run-TryBot: Jeremy Faller <jeremy@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
5 years agocmd/compile: provide Load32/Load64/Store32/Store64 atomic intrinsics on riscv64
Joel Sing [Sun, 15 Mar 2020 15:47:40 +0000 (02:47 +1100)]
cmd/compile: provide Load32/Load64/Store32/Store64 atomic intrinsics on riscv64

Updates #36765

Change-Id: Id5ce5c5f60112e4f4cf9eec1b1ec120994934950
Reviewed-on: https://go-review.googlesource.com/c/go/+/223558
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agoruntime: negate errno value for mips pipe/pipe2
Ian Lance Taylor [Sun, 22 Mar 2020 19:57:36 +0000 (12:57 -0700)]
runtime: negate errno value for mips pipe/pipe2

The callers expect negative errno values, so negate them when necessary.

No test because there is no reasonable way to make pipe/pipe2 fail.
This was reported on a system on which pipe2 returned ENOSYS.

Fixes #37997

Change-Id: I3ad6cbbc2521cf495f8df6ec991a3f781122b508
Reviewed-on: https://go-review.googlesource.com/c/go/+/224592
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com>
5 years agosyscall/js: make wasm_exec.js compatible with Webpack
nao20010128nao [Sat, 21 Mar 2020 06:52:58 +0000 (06:52 +0000)]
syscall/js: make wasm_exec.js compatible with Webpack

In Webpack, require("fs") will always be empty. This behavior throws an error: "fs.writeSync is not function". It happens when you did "fmt.Println".
This PR avoids such problem and use polyfill in wasm_exec.js on Webpack.

Change-Id: I55f2c75ce86b7f84d2d92e8e217b5decfbe3c8a1
GitHub-Last-Rev: aecc847e3f9d5617ea4b00196ef2810c2458f085
GitHub-Pull-Request: golang/go#35805
Reviewed-on: https://go-review.googlesource.com/c/go/+/208600
Reviewed-by: Richard Musiol <neelance@gmail.com>
5 years agosyscall/js: allow copyBytesTo(Go|JS) to use Uint8ClampedArray
Aurélio A. Heckert [Sun, 22 Mar 2020 23:38:42 +0000 (20:38 -0300)]
syscall/js: allow copyBytesTo(Go|JS) to use Uint8ClampedArray

closes #38011

Change-Id: Ic50f2f27456dccdc3fca1bda076871af1eb81705
Reviewed-on: https://go-review.googlesource.com/c/go/+/224638
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Richard Musiol <neelance@gmail.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/go: list test packages even when the main package has an error
Michael Matloob [Mon, 23 Mar 2020 17:52:37 +0000 (13:52 -0400)]
cmd/go: list test packages even when the main package has an error

List test packages (when list is run with -e) even when the main package
has an error. This is useful to get complete data for go/packages.

Fixes #37971

Change-Id: If6ba0270a319ea5e003d1ed8b1ad39e479e95509
Reviewed-on: https://go-review.googlesource.com/c/go/+/224944
Run-TryBot: Michael Matloob <matloob@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
5 years agoRevert "crypto/rsa,crypto/ecdsa,crypto/ed25519: implement PublicKey.Equal"
Bryan C. Mills [Mon, 23 Mar 2020 20:38:00 +0000 (20:38 +0000)]
Revert "crypto/rsa,crypto/ecdsa,crypto/ed25519: implement PublicKey.Equal"

This reverts CL 223754.

Reason for revert: new tests are failing on all longtest builders.

Change-Id: I2257d106c132f3a02c0af6b20061d4f9a8093c4f
Reviewed-on: https://go-review.googlesource.com/c/go/+/225077
Run-TryBot: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
Reviewed-by: Katie Hockman <katie@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/compile: disable addressingmodes pass for 386
Keith Randall [Mon, 23 Mar 2020 20:07:06 +0000 (13:07 -0700)]
cmd/compile: disable addressingmodes pass for 386

Update #37881

Change-Id: I1f9a3f57f6215a19c31765c257ee78715eab36b7
Reviewed-on: https://go-review.googlesource.com/c/go/+/225057
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/go: define a build tag for any GOEXPERIMENT which is enabled.
Dan Scales [Wed, 11 Mar 2020 19:08:03 +0000 (12:08 -0700)]
cmd/go: define a build tag for any GOEXPERIMENT which is enabled.

For each experiment that has been enabled in the toolchain, define a build tag
with the same name (but prefixed by "goexperiment.") which can be used for
compiling alternative files for the experiment. This allows changes for the
experiment, like extra struct fields in the runtime, without affecting the base
non-experiment code at all.

I use this capability in my CL for static lock ranking
(https://go-review.googlesource.com/c/go/+/207619), so that static lock ranking
can be fully enabled as a GOEXPERIMENT, but there is no overhead in the runtime
when the experiment is not enabled.

I added a test in cmd/go/testdata/scripts to make sure the build tags are being
defined properly. In order to implement the test, I needed to provide environment
variable GOEXPSTRING to the test scripts (with its value set from
objabi.Expstring(), so that it can determine the experiments baked into the
toolchain.

I filed https://github.com/golang/go/issues/37937 to make a builder with
GOEXPERIMENT set to 'staticlockranking'. This builder will ensure another variant
of GOEXPERIMENT is being tested regularly for this change, as well as checking
static lock ranking in the runtime.

Change-Id: Ieb4b86107238febd105558c1e639d30cfe57ab5c
Reviewed-on: https://go-review.googlesource.com/c/go/+/222925
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
5 years agostrings: narrow the search range of IndexByte in Index
Andy Pan [Mon, 23 Mar 2020 04:53:40 +0000 (04:53 +0000)]
strings: narrow the search range of IndexByte in Index

Same as https://golang.org/cl/224589.

Change-Id: I6a9e2ea60bf6e1888a95bad0331c92079a7eff99
GitHub-Last-Rev: 81c13c0f5b2ee6ae7842a7b73799b7821f78be59
GitHub-Pull-Request: golang/go#38016
Reviewed-on: https://go-review.googlesource.com/c/go/+/224593
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
5 years agocmd/compile: disable mem+op operations on 386
Keith Randall [Sun, 22 Mar 2020 20:27:02 +0000 (13:27 -0700)]
cmd/compile: disable mem+op operations on 386

Rolling back portions of CL 222782 to see if that helps
issue #37881 any.

Update #37881

Change-Id: I9cc3ff8c469fa5e4b22daec715d04148033f46f7
Reviewed-on: https://go-review.googlesource.com/c/go/+/224837
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
5 years agocrypto/rsa,crypto/ecdsa,crypto/ed25519: implement PublicKey.Equal
Filippo Valsorda [Wed, 18 Mar 2020 00:34:51 +0000 (20:34 -0400)]
crypto/rsa,crypto/ecdsa,crypto/ed25519: implement PublicKey.Equal

This makes all modern public keys in the standard library implement a
common interface (below) that can be used by applications for better
type safety and allows for checking that public (and private keys via
Public()) are equivalent.

interface {
    Equal(crypto.PublicKey) bool
}

Equality for ECDSA keys is complicated, we take a strict interpretation
that works for all secure applications (the ones not using the
unfortunate non-constant time CurveParams implementation) and fails
closed otherwise.

Tests in separate files to make them x_tests and avoid an import loop
with crypto/x509.

Fixes #21704

Change-Id: Id5379c96384a11c5afde0614955360e7470bb1c4
Reviewed-on: https://go-review.googlesource.com/c/go/+/223754
Reviewed-by: Katie Hockman <katie@golang.org>
5 years agocmd/compile: fold LEAQ with constant scale into LEA
Giovanni Bajo [Mon, 23 Mar 2020 08:21:44 +0000 (09:21 +0100)]
cmd/compile: fold LEAQ with constant scale into LEA

Discovered this after rebasing CL196679 (use poset bounds in prove).
Some tests fail with that CL applied:

codegen/smallintiface.go:11: linux/amd64/: opcode not found: "^LEAQ\\truntime.staticuint64s\\+8\\(SB\\)"
codegen/smallintiface.go:16: linux/amd64/: opcode not found: "^LEAQ\\truntime.staticuint64s\\+2024\\(SB\\)"
codegen/smallintiface.go:21: linux/amd64/: opcode not found: "^LEAQ\\truntime.staticuint64s\\+24\\(SB\\)"

The only difference in prove SSA dumps is that a single Lsh64x64
op with constant shift (<< 3) is marked as bounded. This triggers
a different rule matching sequence in lower, which manages to generate
worse code for the above testcases.

This CL fixes the above test after CL196679 is applied. Right now,
these rules never trigger (this CL passes toolstash -cmp), so I can't
write a test.

Change-Id: I353f1c79c1875cac1da82cd8afa1e05e42684f1c
Reviewed-on: https://go-review.googlesource.com/c/go/+/224877
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agocrypto/x509: add RevocationList and CreateRevocationList
Roland Shoemaker [Fri, 20 Mar 2020 20:28:43 +0000 (20:28 +0000)]
crypto/x509: add RevocationList and CreateRevocationList

The existing Certificate.CreateCRL method generates non-conformant CRLs and
as such cannot be used for implementations that require standards
compliance. This change implements a new top level method, CreateCRL, which
generates compliant CRLs, and offers an extensible API if any
extensions/fields need to be supported in the future.

Here is an example Issuer/CRL generated using this change:
-----BEGIN CERTIFICATE-----
MIIBNjCB3aADAgECAgEWMAoGCCqGSM49BAMCMBIxEDAOBgNVBAMTB3Rlc3Rpbmcw
IhgPMDAwMTAxMDEwMDAwMDBaGA8wMDAxMDEwMTAwMDAwMFowEjEQMA4GA1UEAxMH
dGVzdGluZzBZMBMGByqGSM49AgEGCCqGSM49AwEHA0IABLHrudbSM36sn1VBrmm/
OfQTyEsI4tIUV1VmneOKHL9ENBGCiec4GhQm2SGnDT/sZy2bB3c3yozh/roS6cZJ
UZqjIDAeMA4GA1UdDwEB/wQEAwIBAjAMBgNVHQ4EBQQDAQIDMAoGCCqGSM49BAMC
A0gAMEUCIQCoAYN6CGZPgd5Sw5a1rd5VexciT5MCxTfXj+ZfJNfoiAIgQVCTB8AE
Nm2xset7+HOgtQYlKNw/rGd8cFcv5Y9aUzo=
-----END CERTIFICATE-----
-----BEGIN X509 CRL-----
MIHWMH0CAQEwCgYIKoZIzj0EAwIwEjEQMA4GA1UEAxMHdGVzdGluZxgPMDAwMTAx
MDIwMDAwMDBaGA8wMDAxMDEwMzAwMDAwMFowFjAUAgECGA8wMDAxMDEwMTAxMDAw
MFqgHjAcMA4GA1UdIwQHMAWAAwECAzAKBgNVHRQEAwIBBTAKBggqhkjOPQQDAgNJ
ADBGAiEAjqfj/IG4ys5WkjrbTNpDbr+saHGO/NujLJotlLL9KzgCIQDm8VZPzj0f
NYEQgAW4nsiUzlvEUCoHMw0141VCZXv67A==
-----END X509 CRL-----

Fixes #35428

Change-Id: Id96b6f47698d0bed39d586b46bd12374ee6ff88f
GitHub-Last-Rev: c83a6017164e71df3989fe57322b3b4869a09f37
GitHub-Pull-Request: golang/go#36945
Reviewed-on: https://go-review.googlesource.com/c/go/+/217298
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agoruntime: use OnesCount64 to count allocated objects in a span
Michael Anthony Knyszek [Mon, 27 Jan 2020 20:41:38 +0000 (20:41 +0000)]
runtime: use OnesCount64 to count allocated objects in a span

This change modifies the implementation of (*mspan).countAlloc by
using OnesCount64 (which on many systems is intrinsified). It does so by
using an unsafe pointer cast, but in this case we don't care about
endianness because we're just counting bits set.

This change means we no longer need the popcnt table which was redundant
in the runtime anyway. We can also simplify the logic here significantly
by observing that mark bits allocations are always 8-byte aligned, so we
don't need to handle any edge-cases due to the fact that OnesCount64
operates on 64 bits at a time: all irrelevant bits will be zero.

Overall, this implementation is significantly faster than the old one on
amd64, and should be similarly faster (or better!) on other systems
which support the intrinsic. On systems which do not, it should be
roughly the same performance because OnesCount64 is implemented using a
table in the general case.

Results on linux/amd64:

name                         old time/op  new time/op  delta
MSpanCountAlloc/bits=64-4    16.8ns ± 0%  12.7ns ± 0%  -24.40%  (p=0.000 n=5+4)
MSpanCountAlloc/bits=128-4   23.5ns ± 0%  12.8ns ± 0%  -45.70%  (p=0.000 n=4+5)
MSpanCountAlloc/bits=256-4   43.5ns ± 0%  12.8ns ± 0%  -70.67%  (p=0.000 n=4+5)
MSpanCountAlloc/bits=512-4   59.5ns ± 0%  15.4ns ± 0%  -74.12%  (p=0.008 n=5+5)
MSpanCountAlloc/bits=1024-4   116ns ± 1%    23ns ± 0%  -79.84%  (p=0.000 n=5+4)

Change-Id: Id4c994be22224653af5333683a69b0937130ed04
Reviewed-on: https://go-review.googlesource.com/c/go/+/216558
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
5 years agoruntime: add countAlloc benchmark
Michael Anthony Knyszek [Wed, 18 Mar 2020 18:46:04 +0000 (18:46 +0000)]
runtime: add countAlloc benchmark

This change adds a small microbenchmark for (*mspan).countAlloc, which
we're about to replace. Admittedly this isn't a critical piece of code,
but the benchmark was useful in understanding the performance change.

Change-Id: Iea93c00f571ee95534a42f2ef2ab026b382242b3
Reviewed-on: https://go-review.googlesource.com/c/go/+/224438
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

5 years agocmd/compile: panic if trying to alias an intrinsic with no definitions
Michael Anthony Knyszek [Fri, 20 Mar 2020 19:08:13 +0000 (19:08 +0000)]
cmd/compile: panic if trying to alias an intrinsic with no definitions

Currently if we try to alias an intrinsic which hasn't been defined for
any architecture (such as by accidentally creating the alias before the
intrinsic is created with addF), then we'll just silently not apply any
intrinsics to those aliases.

Catch this particular case by panicking in alias if we try to apply the
alias and it did nothing.

Change-Id: I98e75fc3f7206b08fc9267cedb8db3e109ec4f5d
Reviewed-on: https://go-review.googlesource.com/c/go/+/224637
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years agocmd/compile: declare runtime bit func aliases after math/bits intrinsics
Michael Anthony Knyszek [Fri, 20 Mar 2020 15:26:46 +0000 (15:26 +0000)]
cmd/compile: declare runtime bit func aliases after math/bits intrinsics

Currently runtime/internal/sys bit-manipulation functions are aliased to
math/bits functions, which are intrinsified. Unfortunately these aliases
are declared before the intrinsified versions are generated, resulting
in the generic version of the code being copied over.

This change moves the aliases for bit operations in runtime/internal/sys
after the addF calls to generate those intrinsics in SSA, so that the
intrinsified SSA representation of those functions actually get copied
over.

This should improve the overall performance of the runtime (especially
the page allocator) since these bit operations will actually be
intrinsified now.

Change-Id: I4377da13f9a7bb6aee608e50df0297148bf8f806
Reviewed-on: https://go-review.googlesource.com/c/go/+/224437
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: set "local" attribute on symbols during preload
Than McIntosh [Thu, 19 Mar 2020 16:55:43 +0000 (12:55 -0400)]
[dev.link] cmd/link: set "local" attribute on symbols during preload

Set the "local" attribute on object file loader symbols during preload,
as opposed to the current strategy of only propagating the local attr
when we convert to sym.Symbols.

Change-Id: I18965e2fdaa9a0b9315d95466d17d9f41aa40a45
Reviewed-on: https://go-review.googlesource.com/c/go/+/224379
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: add loader.Loader apis for symbol dynamic id
Than McIntosh [Wed, 18 Mar 2020 18:32:17 +0000 (14:32 -0400)]
[dev.link] cmd/link: add loader.Loader apis for symbol dynamic id

Add SymDynid and SetSymDynid methods to the loader. This symbol
property is currently backed by a map.

Change-Id: Iaf86b1d8aaa775fa102fadea30394eb8a670e0e2
Reviewed-on: https://go-review.googlesource.com/c/go/+/224378
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
5 years ago[dev.link] cmd/link: invoke oldlink if old object format is chosen
Cherry Zhang [Fri, 20 Mar 2020 21:02:47 +0000 (17:02 -0400)]
[dev.link] cmd/link: invoke oldlink if old object format is chosen

Now we can choose the old object file format by setting
-gcflags=all=-go115newobj=false -asmflags=all=-go115newobj=false -ldflags=all=-go115newobj=false

Tested that setting all three to default false and it still works.

Change-Id: I9514b62a676916cc383b8afa389489fe7b8fa2bf
Reviewed-on: https://go-review.googlesource.com/c/go/+/224625
Reviewed-by: Than McIntosh <thanm@google.com>
5 years ago[dev.link] cmd/internal/obj: restore old DWARF symbol handling
Cherry Zhang [Fri, 20 Mar 2020 20:05:24 +0000 (16:05 -0400)]
[dev.link] cmd/internal/obj: restore old DWARF symbol handling

When old object file format is used, serialize DWARF symbols in
the old way.

Change-Id: I73a97f10bba367ac29c52f8f3d0f8f3b34a42523
Reviewed-on: https://go-review.googlesource.com/c/go/+/224624
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>