Ronald G. Minnich [Wed, 22 Mar 2017 21:40:55 +0000 (14:40 -0700)]
os/exec: handle Unshareflags with CLONE_NEWNS
In some newer Linux distros, systemd forces
all mount namespaces to be shared, starting
at /. This disables the CLONE_NEWNS
flag in unshare(2) and clone(2).
While this problem is most commonly seen
on systems with systemd, it can happen anywhere,
due to how Linux namespaces now work.
Hence, to create a private mount namespace,
it is not sufficient to just set
CLONE_NEWS; you have to call mount(2) to change
the behavior of namespaces, i.e.
mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)
This is tested and working and we can now correctly
start child process with private namespaces on Linux
distros that use systemd.
The new test works correctly on Ubuntu 16.04.2 LTS.
It fails if I comment out the new Mount, and
succeeds otherwise. In each case it correctly
cleans up after itself.
Fixes #19661
Change-Id: I52240b59628e3772b529d9bbef7166606b0c157d
Reviewed-on: https://go-review.googlesource.com/38471 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Keith Randall [Sat, 18 Mar 2017 18:16:30 +0000 (11:16 -0700)]
cmd/compile: don't merge load+op if other op arg is still live
We want to merge a load and op into a single instruction
l = LOAD ptr mem
y = OP x l
into
y = OPload x ptr mem
However, all of our OPload instructions require that y uses
the same register as x. If x is needed past this instruction, then
we must copy x somewhere else, losing the whole benefit of merging
the instructions in the first place.
Disable this optimization if x is live past the OP.
Also disable this optimization if the OP is in a deeper loop than the load.
Dave Cheney [Wed, 22 Mar 2017 23:56:58 +0000 (10:56 +1100)]
cmd/internal/obj/mips: standardize on sys.MIPS and sys.MIPS64 constants
CL 38446 introduced the use of the sys.ArchFamily type into the
cmd/internal/obj/mips package and redefined the mips.Mips32 and
mips.Mips64 constants in terms of their sys.ArchFamily counterparts.
This CL removes these local declarations and consolidates on sys.MIPS
and sys.MIPS64 respectively.
Martin Möhrmann [Sat, 4 Mar 2017 06:18:26 +0000 (07:18 +0100)]
regexp: add ASCII fast path for context methods
The step method implementations check directly if the next rune
only needs one byte to be decoded and avoid calling utf8.DecodeRune
for such ASCII characters.
Introduce the same fast path optimization for rune decoding
for the context methods.
Results for regexp benchmarks that use the context methods:
Richard Musiol [Sat, 25 Feb 2017 17:37:17 +0000 (18:37 +0100)]
syscall: use CLONE_VFORK and CLONE_VM
This greatly improves the latency of starting a child process when
the Go process is using a lot of memory. Even though the kernel uses
copy-on-write, preparation for that can take up to several 100ms under
certain conditions. All other goroutines are suspended while starting
a subprocess so this latency directly affects total throughput.
With CLONE_VM the child process shares the same memory with the parent
process. On its own this would lead to conflicting use of the same
memory, so CLONE_VFORK is used to suspend the parent process until the
child releases the memory when switching to to the new program binary
via the exec syscall. When the parent process continues to run, one
has to consider the changes to memory that the child process did,
namely the return address of the syscall function needs to be restored
from a register.
A simple benchmark has shown a difference in latency of 16ms vs. 0.5ms
at 10GB memory usage. However, much higher latencies of several 100ms
have been observed in real world scenarios. For more information see
comments on #5838.
Fixes #5838
Change-Id: I6377d7bd8dcd00c85ca0c52b6683e70ce2174ba6
Reviewed-on: https://go-review.googlesource.com/37439 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
When unmarshaling, if an element is empty, eg. '<tag></tag>', and
destination type is int, uint, float or bool, do not attempt to parse
value (""). Set to its zero value instead.
Fixes #13417
Change-Id: I2d79f6d8f39192bb277b1a9129727d5abbb2dd1f
Reviewed-on: https://go-review.googlesource.com/38386 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Robert Griesemer [Wed, 22 Mar 2017 05:23:15 +0000 (22:23 -0700)]
cmd/compile/internal/syntax: replace inlined statement lists with syntax.BlockStmt
This simplifies the code and removes a premature optimization.
It increases the amount of allocated syntax.Node space by ~0.4%
for parsing all of std lib, which is negligible.
Before the change (best of 5 runs):
$ go test -run StdLib -fast
parsed 1517022 lines (3394 files) in 793.487886ms (1911840 lines/s)
allocated 387.086Mb (267B/line, 487.828Mb/s)
After the change (best of 5 runs):
$ go test -run StdLib -fast
parsed 1516911 lines (3392 files) in 805.028655ms (1884294 lines/s)
allocated 388.466Mb (268B/line, 482.549Mb/s)
Change-Id: Id19d6210fdc62393862ba3b04913352d95c599be
Reviewed-on: https://go-review.googlesource.com/38439
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Robert Griesemer [Tue, 21 Mar 2017 22:22:13 +0000 (15:22 -0700)]
cmd/compile/internal/syntax: add position info for { and } braces
This change adds position information for { and } braces in the
source. There's a 1.9% increase in memory use for syntax.Nodes,
which is negligible relative to overall compiler memory consumption.
Parsing the std library (using syntax package only) and memory
consumption before this change (fastest of 5 runs):
$ go test -run StdLib -fast
parsed 1516827 lines (3392 files) in 780.612335ms (1943124 lines/s)
allocated 379.903Mb (486.673Mb/s)
After this change (fastest of 5 runs):
$ go test -run StdLib -fast
parsed 1517022 lines (3394 files) in 793.487886ms (1911840 lines/s)
allocated 387.086Mb (267B/line, 487.828Mb/s)
While not an exact apples-to-apples comparison (the syntax package
has changed and is also parsed), the overall impact is small.
Also: Small improvements to nodes_test.go.
Change-Id: Ib8a7f90bbe79de33d83684e33b1bf8dbc32e644a
Reviewed-on: https://go-review.googlesource.com/38435 Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Sam Whited [Wed, 8 Mar 2017 22:12:58 +0000 (16:12 -0600)]
encoding/xml: format test output using subtests
Change-Id: I2d155c838935cd8427abd142a462ff4c56829715
Reviewed-on: https://go-review.googlesource.com/37948 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Starting in go1.9, the minimum processor requirement for ppc64 is POWER8.
Therefore, the checks for OldArch and the code enabled by it are not necessary
anymore.
Carlos Eduardo Seo [Wed, 22 Mar 2017 17:25:30 +0000 (14:25 -0300)]
runtime/internal/atomic: Remove unnecessary checks for GOARCH_ppc64
Starting in go1.9, the minimum processor requirement for ppc64 is POWER8. This
means the checks for GOARCH_ppc64 in asm_ppc64x.s can be removed, since we can
assume LBAR and STBCCC instructions (both from ISA 2.06) will always be
available.
Mikio Hara [Thu, 9 Feb 2017 21:22:39 +0000 (06:22 +0900)]
net/http: fix possible nil pointer dereference in TestOnlyWriteTimeout
TestOnlyWriteTimeout assumes wrongly that:
- the Accept method of trackLastConnListener is called only once
- the shared variable conn never becomes nil
and crashes on some circumstances.
cmd/compile: disable typPtr caching in the backend
The only new Types that the backend introduces
are pointers to Types generated by the frontend.
Usually, when we generate a *T,
we cache the resulting Type in T,
to avoid recreating it later.
However, that caching is not concurrency safe.
Rather than add mutexes, this CL disables that
caching before starting the backend.
The backend generates few enough new *Ts that the
performance impact of this is small, particularly
if we pre-create some commonly used *Ts.
Brad Fitzpatrick [Tue, 21 Mar 2017 17:23:08 +0000 (10:23 -0700)]
syscall, os: use pipe2 syscall on FreeBSD instead of pipe
The pipe2 syscall exists in all officially supported FreeBSD
versions: 10, 11 and future 12.
The pipe syscall no longer exists in 11 and 12. To build and
run Go on these versions, kernel needs COMPAT_FREEBSD10 option.
Based on Gleb Smirnoff's https://golang.org/cl/38422
Matthew Dempsky [Tue, 21 Mar 2017 17:15:14 +0000 (10:15 -0700)]
cmd/compile/internal/gc: fix liveness regression
During AllocFrame, we drop unused variables from Curfn.Func.Dcl, but
there might still be OpVarFoo instructions that reference those
variables. This wasn't a problem though because gvardefx used to emit
ANOP for unused variables instead of AVARFOO.
As an easy fix, if we see OpVarFoo (or OpKeepAlive) referencing an
unused variable, we can ignore it.
Vladimir Stefanovic [Tue, 21 Mar 2017 13:57:58 +0000 (14:57 +0100)]
cmd/compile,runtime: fix atomic And8 for mipsle
Removing stray xori that came from big endian copy/paste.
Adding atomicand8 check to runtime.check() that would have revealed
this error.
Might fix #19396.
Volker Dobler [Mon, 6 Mar 2017 09:07:25 +0000 (10:07 +0100)]
net/http/cookiejar: fix out-of-bounds errors on malformed domains
The old implementation of Jar made the assumption that the host names
in the URLs given to SetCookies() and Cookies() methods are well-formed.
This is not an unreasonable assumption as malformed host names do not
trigger calls to SetCookies or Cookies (at least not from net/http)
as the HTTP request themselves are not executed. But there can be other
invocations of these methods and at least on Linux it was possible to
make DNS lookup to domain names with two trailing dots (see issue #7122).
This is an old bug and this CL revives an old change (see
https://codereview.appspot.com/52100043) to fix the issue. The discussion
around 52100043 focused on the interplay between the jar and the
public suffix list and who is responsible for which type if domain name
canonicalization. The new bug report in issue #19384 used a nil public
suffix list which demonstrates that the package cookiejar alone exhibits
this problem and any solution cannot be fully delegated to the
implementation of the used PublicSuffixList: Package cookiejar itself
needs to protect against host names of the form ".." which triggered an
out-of-bounds error.
This CL does not address the issue of host name canonicalization and
the question who is responsible for it. This CL just prevents the
out-of-bounds error: It is a very conservative change, i.e. one might
still set and retrieve cookies for host names like "weird.stuf...".
Several more test cases document how the current code works.
cmd/compile: fix pos of typenames created during SSA construction
Prior to this CL, the function's position was used.
The dottype Node's position is clearly better.
I'm not thrilled about introducing a reference to
lineno in the middle of SSA construction;
I will have to remove it later.
My immediate goal is stability and correctness of positions,
though, since that aids refactoring, so this is an improvement.
An example from package io:
func (t *multiWriter) WriteString(s string) (n int, err error) {
var p []byte // lazily initialized if/when needed
for _, w := range t.writers {
if sw, ok := w.(stringWriter); ok {
n, err = sw.WriteString(s)
The w.(stringWriter) type assertion includes loading
the address of static type data for stringWriter:
LEAQ type."".stringWriter(SB), R10
Prior to this CL, this instruction was given the line number
of the function declaration.
After this CL, this instruction is given the line number
of the type assertion itself.
Rob Pike [Tue, 21 Mar 2017 03:24:26 +0000 (20:24 -0700)]
encoding/gob: document the extra byte after a singleton
This paragraph has been added, as the notion was missing from the
documentation.
If a value is passed to Encode and the type is not a struct (or pointer to struct,
etc.), for simplicity of processing it is represented as a struct of one field.
The only visible effect of this is to encode a zero byte after the value, just as
after the last field of an encoded struct, so that the decode algorithm knows when
the top-level value is complete.
Fixes #16978
Change-Id: I5f008e792d1b6fe80d2e026a7ff716608889db32
Reviewed-on: https://go-review.googlesource.com/38414 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Rob Pike [Tue, 21 Mar 2017 03:18:02 +0000 (20:18 -0700)]
text/template,html/template: state that Funcs must happen before parsing
Any method that affects the parse must happen before parsing.
This obvious point is clear, but it's not clear to some that the
set of defined functions affect the parse.
Fixes #18971
Change-Id: I8b7f8c8cf85b028c18e5ca3b9797de92ea910669
Reviewed-on: https://go-review.googlesource.com/38413 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Previously, we handled recursive interfaces by deferring typechecking
of interface methods, while eagerly expanding interface embeddings.
This CL switches to eagerly evaluating interface methods, and
deferring expanding interface embeddings to dowidth. This allows us to
detect recursive interface embeddings with the same mechanism used for
detecting recursive struct embeddings.
Updates #16369.
Change-Id: If4c0320058047f8a2d9b52b9a79de47eb9887f95
Reviewed-on: https://go-review.googlesource.com/38391
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Jakob Borg [Mon, 13 Mar 2017 23:21:51 +0000 (08:21 +0900)]
net/http: fix ProxyFromEnvironment panic on invalid $NO_PROXY value
Given an entry in $no_proxy like ":1" we would interpret it as an empty
host name and a port number, then check the first character of the host
name for dots. This would then cause an index out of range panic. This
change simply skips these entries, as the following checks would anyway
have returned false.
Pascal S. de Kloe [Wed, 16 Nov 2016 15:35:42 +0000 (16:35 +0100)]
encoding/json: reduce unmarshal mallocs for unmapped fields
JSON decoding performs poorly for unmapped and ignored fields. We noticed better
performance when unmarshalling unused fields. The loss comes mostly from calls
to scanner.error as described at #17914.
benchmark old ns/op new ns/op delta
BenchmarkIssue10335-8 431 408 -5.34%
BenchmarkUnmapped-8 1744 1314 -24.66%
benchmark old allocs new allocs delta
BenchmarkIssue10335-8 4 3 -25.00%
BenchmarkUnmapped-8 18 4 -77.78%
benchmark old bytes new bytes delta
BenchmarkIssue10335-8 320 312 -2.50%
BenchmarkUnmapped-8 568 344 -39.44%
Matthew Dempsky [Sat, 11 Mar 2017 00:03:56 +0000 (16:03 -0800)]
cmd/compile: sort CFG blocks in PC order during liveness
This CL changes the order that liveness analysis visits CFG blocks to
PC order, rather than RPO. This doesn't meaningfully change anything
except that the PCDATA_StackMapIndex values will be assigned in PC
order too.
However, this does have the benefit that the subsequent CL to port
liveness analysis to the SSA CFG (which has blocks in PC order) will
now pass toolstash-check.
Michael Munday [Mon, 20 Mar 2017 21:31:35 +0000 (17:31 -0400)]
cmd/internal/obj/s390x: cleanup objz.go
This CL deletes some unnecessary code in objz.go that existed to
support instruction scheduling. It's likely instruction scheduling
will never be done in this part of the backend so this code can
just be deleted.
This file can probably be cleaned up a bit more, but I think this
is a good start.
Passes: go build -toolexec 'toolstash -cmp' -a std.
cmd/go, cmd/compile: always optimize when building runtime
When optimizations are disabled, the compiler
cannot eliminate enough write barriers to satisfy
the runtime's nowritebarrier and nowritebarrierrec
annotations.
Enforce that requirement, and for convenience,
have cmd/go elide -N when compiling the runtime.
This came up in practice for me when running
toolstash -cmp. When toolstash -cmp detected
mismatches, it recompiled with -N, which caused
runtime compilation failures.
CL 27254 changed a constant string to a byte array
in encoding/hex and got significant performance
improvements.
hex.Encode used the string twice in a single function.
The rewrite rules lower constant strings into components.
The pointer component requires an aux symbol.
The existing implementation created a new aux symbol every time.
As a result, constant string pointers were never CSE'd.
Tighten then moved the pointer calculation next to the uses, i.e.
into the loop.
The re-use of aux syms enabled by this CL
occurs 3691 times during make.bash.
This CL should not go in without CL 38338
or something like it.
Robert Griesemer [Mon, 20 Mar 2017 20:08:26 +0000 (13:08 -0700)]
cmd/gofmt: clarify doc string even more
Since "columns of alignment" are terminated whenever indentation
changes from one line to the next, alignment with spaces will work
independent of the actually chosen tab width. Don't mention tab width
anymore.
Follow-up on https://golang.org/cl/38374/.
For #19618.
Change-Id: I58e47dfde57834f56a98d9119670757a12fb9c41
Reviewed-on: https://go-review.googlesource.com/38379 Reviewed-by: Rob Pike <r@golang.org>
Brad Fitzpatrick [Mon, 20 Mar 2017 16:44:34 +0000 (16:44 +0000)]
net/http: deflake TestServerAllowsBlockingRemoteAddr more
As noted in https://github.com/golang/go/issues/19161#issuecomment-287554171,
CL 37771 (adding use of the new httptest.Server.Client to all net/http
tests) accidentally reverted DisableKeepAlives for this test. For
many tests, DisableKeepAlives was just present to prevent goroutines
from staying active after the test exited. In this case it might
actually be important. (We'll see)
Updates #19161
Change-Id: I11f889f86c932b51b11846560b68dbe5993cdfc3
Reviewed-on: https://go-review.googlesource.com/38373 Reviewed-by: Michael Munday <munday@ca.ibm.com>
Austin Clements [Thu, 16 Mar 2017 21:02:24 +0000 (17:02 -0400)]
runtime: disallow malloc or panic in scavenge
Mallocs and panics in the scavenge path are particularly nasty because
they're likely to silently self-deadlock on the mheap.lock. Avoid
sinking lots of time into debugging these issues in the future by
turning these into immediate throws.
Austin Clements [Tue, 7 Mar 2017 21:38:29 +0000 (16:38 -0500)]
runtime: introduce a type for lfstacks
The lfstack API is still a C-style API: lfstacks all have unhelpful
type uint64 and the APIs are package-level functions. Make the code
more readable and Go-style by creating an lfstack type with methods
for push, pop, and empty.
I noted in CL 38327 that the SSA test API felt a bit
clunky after the ssa.Func/ssa.Cache/ssa.Config refactoring,
and promised to clean it up once the dust settled.
The dust has settled.
Along the way, this CL fixes a potential latent bug,
in which the amd64 test context was used for all dummy Syslook calls.
The lone SSA test using the s390x context did not depend on the
Syslook context being correct, so the bug did not arise in practice.
This reduces the number of calls back into the
gc Type routines, which will help performance
in a concurrent backend.
It also reduces the number of callsites
that must be considered in making the transition.
Passes toolstash-check -all. No compiler performance changes.
cmd/compile: separate ssa.Frontend and ssa.TypeSource
Prior to this CL, the ssa.Frontend field was responsible
for providing types to the backend during compilation.
However, the types needed by the backend are few and static.
It makes more sense to use a struct for them
and to hang that struct off the ssa.Config,
which is the correct home for readonly data.
Now that Types is a struct, we can clean up the names a bit as well.
This has the added benefit of allowing early construction
of all types needed by the backend.
This will be useful for concurrent backend compilation.
Passes toolstash-check -all. No compiler performance change.
Updates #15756
Change-Id: I021658c8cf2836d6a22bbc20cc828ac38c7da08a
Reviewed-on: https://go-review.googlesource.com/38336 Reviewed-by: Matthew Dempsky <mdempsky@google.com>