Found with GODEBUG=wbshadow=2 mode.
Eventually that will run automatically, but right now
it still detects other missing write barriers.
Change-Id: I1320d5340a9e421c779f24f3b170e33974e56e4f
Reviewed-on: https://go-review.googlesource.com/2278
Reviewed-by: Rick Hudson <rlh@golang.org>
Found with GODEBUG=wbshadow=2 mode.
Eventually that will run automatically, but right now
it still detects other missing write barriers.
Change-Id: Iea83d693480c2f3008b4e80d55821acff65970a6
Reviewed-on: https://go-review.googlesource.com/2277
Reviewed-by: Keith Randall <khr@golang.org>
Preparation for replacing many memmove calls in runtime
with typedmemmove, which is a clearer description of what
the routine is doing.
For the same reason, rename writebarriercopy to typedslicecopy.
Change-Id: I6f23bef2c2215509fefba175b16908f76dc7538c
Reviewed-on: https://go-review.googlesource.com/2276
Reviewed-by: Rick Hudson <rlh@golang.org>
Add write barrier to atomic operations manipulating pointers.
In general an atomic write of a pointer word may indicate racy accesses,
so there is no strictly safe way to attempt to keep the shadow copy
in sync with the real one. Instead, mark the shadow copy as not used.
Redirect sync/atomic pointer routines back to the runtime ones,
so that there is only one copy of the write barrier and shadow logic.
In time we might consider doing this for most of the sync/atomic
functions, but for now only the pointer routines need that treatment.
Found with GODEBUG=wbshadow=1 mode.
Eventually that will run automatically, but right now
it still detects other missing write barriers.
Change-Id: I852936b9a111a6cb9079cfaf6bd78b43016c0242
Reviewed-on: https://go-review.googlesource.com/2066
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
The Gobuf.g goroutine pointer is almost always updated by assembly code.
In one of the few places it is updated by Go code - func save - it must be
treated as a uintptr to avoid a write barrier being emitted at a bad time.
Instead of figuring out how to emit the write barriers missing in the
assembly manipulation, change the type of the field to uintptr, so that
it does not require write barriers at all.
Goroutine structs are published in the allg list and never freed.
That will keep the goroutine structs from being collected.
There is never a time that Gobuf.g's contain the only references
to a goroutine: the publishing of the goroutine in allg comes first.
Goroutine pointers are also kept in non-GC-visible places like TLS,
so I can't see them ever moving. If we did want to start moving data
in the GC, we'd need to allocate the goroutine structs from an
alternate arena. This CL doesn't make that problem any worse.
Found with GODEBUG=wbshadow=1 mode.
Eventually that will run automatically, but right now
it still detects other missing write barriers.
Change-Id: I85f91312ec3e0ef69ead0fff1a560b0cfb095e1a
Reviewed-on: https://go-review.googlesource.com/2065
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Found with GODEBUG=wbshadow=1 mode.
Eventually that will run automatically, but right now
it still detects other missing write barriers.
Change-Id: Ic8624401d7c8225a935f719f96f2675c6f5c0d7c
Reviewed-on: https://go-review.googlesource.com/2064
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
This is the detection code. It works well enough that I know of
a handful of missing write barriers. However, those are subtle
enough that I'll address them in separate followup CLs.
GODEBUG=wbshadow=1 checks for a write that bypassed the
write barrier at the next write barrier of the same word.
If a bug can be detected in this mode it is typically easy to
understand, since the crash says quite clearly what kind of
word has missed a write barrier.
GODEBUG=wbshadow=2 adds a check of the write barrier
shadow copy during garbage collection. Bugs detected at
garbage collection can be difficult to understand, because
there is no context for what the found word means.
Typically you have to reproduce the problem with allocfreetrace=1
in order to understand the type of the badly updated word.
Change-Id: If863837308e7c50d96b5bdc7d65af4969bf53a6e
Reviewed-on: https://go-review.googlesource.com/2061
Reviewed-by: Austin Clements <austin@google.com>
When constants were declared using unexported constants,
the type information was lost when those constants were filtered out.
This CL propagates the type information of unexported constants
so that it is available for display.
This is a follow-up to CL 144110044, which fixed this problem
specifically for _ constants.
Updates #5397.
Change-Id: I3f0c767a4007d88169a5634ab2870deea4e6a740
Reviewed-on: https://go-review.googlesource.com/2091
Reviewed-by: Robert Griesemer <gri@golang.org>
Noticed while investigating the speed of the runtime tests, as part
of debugging while Plan 9's runtime tests are timing out on GCE.
Change-Id: I95f5a3d967a0b45ec1ebf10067e193f51db84e26
Reviewed-on: https://go-review.googlesource.com/2283
Reviewed-by: Ian Lance Taylor <iant@golang.org>
This reverts commit ab0535ae3f.
I think it will remain useful to distinguish code that must
run on a system stack from code that can run on either stack,
even if that distinction is no
longer based on the implementation language.
That is, I expect to add a //go:systemstack comment that,
in terms of the old implementation, tells the compiler,
to pretend this function was written in C.
Change-Id: I33d2ebb2f99ae12496484c6ec8ed07233d693275
Reviewed-on: https://go-review.googlesource.com/2275
Reviewed-by: Russ Cox <rsc@golang.org>
This CL splits the (ever growing) list of ca cert locations by major unix
platforms (darwin, windows and plan9 are already handled seperately).
Although it is clear the unix variants cannot manage to agree on some standard
locations, we can avoid to some extent an artificial ranking of priority
amongst the supported GOOSs.
* Split certFiles definition by GOOS
* Include NetBSD ca cert location
Fixes#9285
Change-Id: I6df2a3fddf3866e71033e01fce43c31e51b48a9e
Reviewed-on: https://go-review.googlesource.com/2208
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
This ensures that changing an image.YCbCr's Y values can't change its
chroma values, even after re-slicing up to capacity.
Change-Id: Icb626561522e336a3220e10f456c95330ae7db9e
Reviewed-on: https://go-review.googlesource.com/2209
Reviewed-by: Rob Pike <r@golang.org>
Previously, we ended up passing two compiled objects for the package
being tested when linking the test executable. Somewhat by luck, this
worked most of the time but occasionally it did not. This changes the
linking code to not pass two objects for the same ImportPath and to
always pass the object for the test version of the package and removes
some unecessary nil checks.
Change-Id: I7bbd3fc708f14672ee2cc6aed3397421fceb8a38
Reviewed-on: https://go-review.googlesource.com/1840
Reviewed-by: Ian Lance Taylor <iant@golang.org>
liblink used to encode both SETEQ BP and SETEQ CH as 0f 94 c5,
however, SETEQ BP should have used a REX prefix.
Fixes#8545.
Change-Id: Ie59c990cdd0ec506cffe4318e9ad1b48db5e57dd
Reviewed-on: https://go-review.googlesource.com/2270
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
This CL adds missing ipv4-mapped ipv6 address test cases to TestParseIP.
Change-Id: I3144d2a88d409bd515cf52f8711d407bfa81ed68
Reviewed-on: https://go-review.googlesource.com/2205
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Shell out to `uname -r` this time, so that the test will compile
even if the platform doesn't have syscall.Sysctl.
Change-Id: I3a19ab5d820bdb94586a97f4507b3837d7040525
Reviewed-on: https://go-review.googlesource.com/2271
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The test program requires static constructor, which in turn needs
external linking to work, but external linking never works on 10.6.
This should fix the darwin-{386,amd64} builders.
Change-Id: I714fdd3e35f9a7e5f5659cf26367feec9412444f
Reviewed-on: https://go-review.googlesource.com/2235
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
If the user provided a key but no value via -ldflag -X,
another linker flag was used as the value.
Placing the user's flags at the end avoids this problem.
It also provides the user the opportunity to
override existing linker flags.
Fixes#8810.
Change-Id: I96f4190713dc9a9c29142e56658446fba7fb6bc8
Reviewed-on: https://go-review.googlesource.com/2242
Reviewed-by: Minux Ma <minux@golang.org>
Remove use of itod on posix systems and replace with call to itoa.
Build and use same itoa function on all systems.
Fix infinite recursion in iota function for the case -1<<63.
Change-Id: I89d7e742383c5c4aeef8780501c78a3e1af87a6f
Reviewed-on: https://go-review.googlesource.com/2213
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Updated the issue tracker link the compiler prints out
when asking for a bug report after an internal error.
Change-Id: I092b118130f131c6344d9d058bea4ad6379032b8
Reviewed-on: https://go-review.googlesource.com/2218
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Preventing returning io.EOF on non-connection oriented sockets is
already applied to Unix variants. This CL applies it to Windows.
Update #4856.
Change-Id: I82071d40f617e2962d0540b9d1d6a10ea4cdb2ec
Reviewed-on: https://go-review.googlesource.com/2203
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
There is no reason to have the redundant test case TestDNSThreadLimt
because TestLookupIPDeadline does cover what we need to test with
-dnsflood flag and more.
Also this CL moves TestLookupIPDeadline into lookup_test.go to avoid
abusing to control the order of test case execution by using file name.
Change-Id: Ib417d7d3411c59d9352c03c996704d584368dc62
Reviewed-on: https://go-review.googlesource.com/2204
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Fixes build on plan9 and windows.
Change-Id: Ic9b02c641ab84e4f6d8149de71b9eb495e3343b2
Reviewed-on: https://go-review.googlesource.com/2233
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
I missed this one in golang.org/cl/2232 and only tested the patch
on openbsd/amd64.
Change-Id: I4ff437ae0bfc61c989896c01904b6d33f9bdf0ec
Reviewed-on: https://go-review.googlesource.com/2234
Reviewed-by: Minux Ma <minux@golang.org>
This is a genuine bug exposed by our test for issue 9456: our wrapper
for pthread_create is not initialized until we initialize cgo itself,
but it is possible that a static constructor could call pthread_create,
and in that case, it will be calling a nil function pointer.
Fix that by also initializing the sys_pthread_create function pointer
inside our pthread_create wrapper function, and use a pthread_once to
make sure it is only initialized once.
Fix build for openbsd.
Change-Id: Ica4da2c21fcaec186fdd3379128ef46f0e767ed7
Reviewed-on: https://go-review.googlesource.com/2232
Reviewed-by: David Crawshaw <crawshaw@golang.org>
%lL will prepend the current directory to the filename, which is not
what we want here (as the file name is already absolute).
Fixes#9150.
Change-Id: I4c9386be6baf421393b92d9401a264b4692986d0
Reviewed-on: https://go-review.googlesource.com/2231
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Some libraries, for example, OpenBLAS, create work threads in a global constructor.
If we're doing cpu profiling, it's possible that SIGPROF might come to some of the
worker threads before we make our first cgo call. Cgocallback used to terminate the
process when that happens, but it's better to miss a couple profiling signals than
to abort in this case.
Fixes#9456.
Change-Id: I112b8e1a6e10e6cc8ac695a4b518c0f577309b6b
Reviewed-on: https://go-review.googlesource.com/2141
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Avoid the decimal lookup in digits array and compute the decimal character value directly.
Reduce calls to 64bit division on 32bit plattforms by splitting conversion into smaller blocks.
Convert value to uintptr type when it can be represented by uintptr.
on darwin/386
benchmark old ns/op new ns/op delta
BenchmarkFormatInt 8352 7466 -10.61%
BenchmarkAppendInt 4281 3401 -20.56%
BenchmarkFormatUint 2785 2251 -19.17%
BenchmarkAppendUint 1770 1223 -30.90%
on darwin/amd64
benchmark old ns/op new ns/op delta
BenchmarkFormatInt 5531 5492 -0.71%
BenchmarkAppendInt 2435 2295 -5.75%
BenchmarkFormatUint 1628 1569 -3.62%
BenchmarkAppendUint 726 750 +3.31%
Change-Id: Ifca281cbdd62ab7d7bd4b077a96da99eb12cf209
Reviewed-on: https://go-review.googlesource.com/2105
Reviewed-by: Robert Griesemer <gri@golang.org>
We already had client support for trailers, but no way for a server to
set them short of hijacking the connection.
Fixes#7759
Change-Id: Ic83976437739ec6c1acad5f209ed45e501dbb93a
Reviewed-on: https://go-review.googlesource.com/2157
Reviewed-by: Andrew Gerrand <adg@golang.org>
Following change 2154, the goatoi function
was renamed atoi.
However, this definition conflicts with the
atoi function defined in the Plan 9 runtime,
which takes a []byte instead of a string.
This change fixes the build on Plan 9.
Change-Id: Ia0f7ca2f965bd5e3cce3177bba9c806f64db05eb
Reviewed-on: https://go-review.googlesource.com/2165
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
They are no longer needed now that C is gone.
goatoi -> atoi
gofuncname/funcname -> funcname/cfuncname
goroundupsize -> already existing roundupsize
Change-Id: I278bc33d279e1fdc5e8a2a04e961c4c1573b28c7
Reviewed-on: https://go-review.googlesource.com/2154
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
Now that we've removed all the C code in runtime and the C compilers,
there is no need to have a separate stackguard field to check for C
code on Go stack.
Remove field g.stackguard1 and rename g.stackguard0 to g.stackguard.
Adjust liblink and cmd/ld as necessary.
Change-Id: I54e75db5a93d783e86af5ff1a6cd497d669d8d33
Reviewed-on: https://go-review.googlesource.com/2144
Reviewed-by: Keith Randall <khr@golang.org>
Use Fatalf for formatting directive rather than plain Fatal.
Change-Id: Iebd30cd6326890e9501746113a6d97480949e3d2
Reviewed-on: https://go-review.googlesource.com/2161
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The goalg function was a holdover from when we had algorithm
tables in both C and Go. It is no longer needed.
Change-Id: Ia0c1af35bef3497a899f22084a1a7b42daae72a0
Reviewed-on: https://go-review.googlesource.com/2099
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The error message for decoding a unquoted value into a struct field with
the ,string option specified has two arguments when one is needed.
Make the error message take one argument and add a test in order to cover
the case when a unquoted value is specified.
Also add error value as the missing argument for Fatalf call in test.
Fixes the following go vet reports:
decode.go:602: wrong number of args for format in Errorf call: 1 needed but 2 args
decode_test.go:1088: missing argument for Fatalf("%v"): format reads arg 1, have only 0 args
Change-Id: Id036e10c54c4a7c1ee9952f6910858ecc2b84134
Reviewed-on: https://go-review.googlesource.com/2109
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Use log.Fatalf for formatting directives instead of log.Fatal
Change-Id: Ia207b320f5795c63cdfa71f92c19ca6d05cc833f
Reviewed-on: https://go-review.googlesource.com/2160
Reviewed-by: Minux Ma <minux@golang.org>
Rename "gothrow" to "throw" now that the C version of "throw"
is no longer needed.
This change is purely mechanical except in panic.go where the
old version of "throw" has been deleted.
sed -i "" 's/[[:<:]]gothrow[[:>:]]/throw/g' runtime/*.go
Change-Id: Icf0752299c35958b92870a97111c67bcd9159dc3
Reviewed-on: https://go-review.googlesource.com/2150
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
The new test case produces the longest string representation possible and thereby uses
all of the 65 bytes in the buffer array used by the formatBits function.
Change-Id: I11320c4de56ced5ff098b7e37f1be08e456573e2
Reviewed-on: https://go-review.googlesource.com/2108
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Specify what will happen if len(dst) != len(src).
Change-Id: I66afa3730f637753b825189687418f14ddec3629
Reviewed-on: https://go-review.googlesource.com/1754
Reviewed-by: Adam Langley <agl@golang.org>
Currently we do very a complex rebalancing of runnable goroutines
between queues, which tries to preserve scheduling fairness.
Besides being complex and error-prone, it also destroys all locality
of scheduling.
This change uses simpler scheme: leave runnable goroutines where
they are, during starttheworld start all Ps with local work,
plus start one additional P in case we have excessive runnable
goroutines in local queues or in the global queue.
The schedler must be able to operate efficiently w/o the rebalancing,
because garbage collections do not have to happen frequently.
The immediate need is execution tracing support: handling of
garabage collection which does stoptheworld/starttheworld several
times becomes exceedingly complex if the current execution can
jump between Ps during starttheworld.
Change-Id: I4fdb7a6d80ca4bd08900d0c6a0a252a95b1a2c90
Reviewed-on: https://go-review.googlesource.com/1951
Reviewed-by: Rick Hudson <rlh@golang.org>
This test code is ugly. There must be a better way.
But for now, fix the build.
Change-Id: I33064145ea37f11abf040ec97caa87669be1a9fa
Reviewed-on: https://go-review.googlesource.com/2114
Reviewed-by: Ian Lance Taylor <iant@golang.org>
numCanOpen will never be larger than 0 in maybeOpenNewConnections() when this
code path is taken, so no new connections can ever be opened.
Change-Id: Id1302e8d9afb3a67be61b5e738fe07ef81d20fe0
Reviewed-on: https://go-review.googlesource.com/1550
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Move the checks for empty rotate changes
from the beginning of rotate to the callers.
Remove additional variable p used instead of existing m with same value.
Remove special casing of equal ranges (i==j) to exit early as no
work is saved vs checking (i!=j) and making a single
swapRange call if this is false.
benchmark old ns/op new ns/op delta
BenchmarkStableString1K 417195 425218 +1.92%
BenchmarkStableInt1K 126661 124498 -1.71%
BenchmarkStableInt64K 10365014 10417438 +0.51%
BenchmarkStable1e2 132151 130648 -1.14%
BenchmarkStable1e4 42027428 40812649 -2.89%
BenchmarkStable1e6 8524772364 8430192391 -1.11%
Change-Id: Ia7642e9d31408496970c700f5843d53cc3ebe817
Reviewed-on: https://go-review.googlesource.com/2100
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
When go parses #cgo lines, expand ${SRCDIR} into the path to the
source directory. This allows options to be passed to the
compiler and linker that involve file paths relative to the
source code directory. Without the expansion the paths would be
invalid when the current working directory changes.
Fixes#7891Fixes#5428
Change-Id: I343a145a9771a5ccbaa958e4a1ecd1716fcae52d
Reviewed-on: https://go-review.googlesource.com/1756
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Test more stuff:
1) flagNoPointers, an incorrect value was the cause of #9425
2) Total function layout size
3) gc program
Change-Id: I73f65fe740215938fa930d2f096febd9db0a0021
Reviewed-on: https://go-review.googlesource.com/2090
Reviewed-by: Ian Lance Taylor <iant@golang.org>
The code concerning quoted-printable encoding (RFC 2045) and its
variant for MIME headers (RFC 2047) is currently spread in
mime/multipart and net/mail. It is also not exported.
This commit is the first step to fix that issue. It moves the
quoted-printable decoding code from mime/multipart to
mime/internal/quotedprintable. The exposed API is unchanged.
Concerns #4943.
Change-Id: I11352afbb2edb4d6ef62870b9bc5c87c639eff12
Reviewed-on: https://go-review.googlesource.com/1810
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Add a nil byte at the end of the itoa buffer,
before calling gostringnocopy. This prevents
gostringnocopy to read past the buffer size.
Change-Id: I87494a8dd6ea45263882536bf6c0f294eda6866d
Reviewed-on: https://go-review.googlesource.com/2033
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Originally it used r.Int63() to show "Uint32", and now we use the correct r.Uint32() method.
Fixes#9429
Change-Id: I8a1228f1ca1af93b0e3104676fc99000257c456f
Reviewed-on: https://go-review.googlesource.com/2069
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The freebsd-386 and freebsd-amd64 builders are timing out sometimes.
This will give them some more breathing room.
Change-Id: Ib65bd172cca046a52861759a4232d7b4b6514fa8
Reviewed-on: https://go-review.googlesource.com/1994
Reviewed-by: Ian Lance Taylor <iant@golang.org>
malloc checks kindNoPointers and if it is not set and the object
is one pointer in size, it assumes it contains a pointer. So we
must set kindNoPointers correctly; it isn't just a hint.
Fixes#9425
Change-Id: Ia43da23cc3298d6e3d6dbdf66d32e9678f0aedcf
Reviewed-on: https://go-review.googlesource.com/2055
Reviewed-by: Russ Cox <rsc@golang.org>
Replace with uses of //go:linkname in Go files, direct use of name in .s files.
The only one that really truly needs a jump is reflect.call; the jump is now
next to the runtime.reflectcall assembly implementations.
Change-Id: Ie7ff3020a8f60a8e4c8645fe236e7883a3f23f46
Reviewed-on: https://go-review.googlesource.com/1962
Reviewed-by: Austin Clements <austin@google.com>
These signals are used by glibc to broadcast setuid/setgid to all
threads and to send pthread cancellations. Unlike other signals, the
Go runtime does not intercept these because they must invoke the libc
handlers (see issues #3871 and #6997). However, because 1) these
signals may be issued asynchronously by a thread running C code to
another thread running Go code and 2) glibc does not set SA_ONSTACK
for its handlers, glibc's signal handler may be run on a Go stack.
Signal frames range from 1.5K on amd64 to many kilobytes on ppc64, so
this may overflow the Go stack and corrupt heap (or other stack) data.
Fix this by ensuring that these signal handlers have the SA_ONSTACK
flag (but not otherwise taking over the handler).
This has been a problem since Go 1.1, but it's likely that people
haven't encountered it because it only affects setuid/setgid and
pthread_cancel.
Fixes#9600.
Change-Id: I6cf5f5c2d3aa48998d632f61f1ddc2778dcfd300
Reviewed-on: https://go-review.googlesource.com/1887
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Currently when we get a CGI or FCGI request, the remote port of the client
is hard coded to zero, despite nearly every webserver passing down the
REMOTE_PORT variable.
This was likely originally excluded because the CGI RFC (rfc3875) does not
mention anything about the remote port of the client. However every webserver
tested does pass REMOTE_PORT down. This includes Apache 2.2, Apache 2.4,
nginx and lighttpd.
Fixes#8351
Change-Id: I4c6366cb39f0ccc05e038bd31d85f93b76e8d0c8
Reviewed-on: https://go-review.googlesource.com/1750
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Calls to goproc/deferproc used to push & pop two extra arguments,
the argument size and the function to call. Now, we allocate space
for those arguments in the outargs section so we don't have to
modify the SP.
Defers now use the stack pointer (instead of the argument pointer)
to identify which frame they are associated with.
A followon CL might simplify funcspdelta and some of the stack
walking code.
Fixes issue #8641
Change-Id: I835ec2f42f0392c5dec7cb0fe6bba6f2aed1dad8
Reviewed-on: https://go-review.googlesource.com/1601
Reviewed-by: Russ Cox <rsc@golang.org>
OpenBSD 5.5 changed its kernel ABI and OpenBSD 5.6 enabled it.
This CL works on both 5.5 and 5.6.
Fixes#9102.
Change-Id: I4a295be9ab8acbc99e550d8cb7e8f8dacf3a03c5
Reviewed-on: https://go-review.googlesource.com/1932
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Previously, this code generated bogus section indexes for dynamic
symbols. It turns out this didn't matter, since we only emit these
when generating an executable and in an executable it only matters
whether a symbol is defined or undefined, but it leads to perplexing
code full of mysterious constants.
Unfortunately, this happens too early to put in real section indexes,
so just use section index 1 to distinguish the symbol from an
undefined symbol.
Change-Id: I0e514604bf31f21683598ebd3e020b66acf767ef
Reviewed-on: https://go-review.googlesource.com/1720
Reviewed-by: Russ Cox <rsc@golang.org>
This will be used by ppc64 to add call stubs to the .text section.
ARM needs a similar pass to generate veneers for arm->thumb
transitions.
Change-Id: Iaee74036e60643a56fab15b564718f359c5910eb
Reviewed-on: https://go-review.googlesource.com/2004
Reviewed-by: Russ Cox <rsc@golang.org>
Disabled by default, but invaluable when you need it.
Change-Id: If4a75d11d14f70b6840d339aaec4b940dc406493
Reviewed-on: https://go-review.googlesource.com/2012
Reviewed-by: Russ Cox <rsc@golang.org>
For arm and powerpc, as well as x86 without aes instructions.
Contains a mixture of ideas from cityhash and xxhash.
Compared to our old fallback on ARM, it's ~no slower on
small objects and up to ~50% faster on large objects. More
importantly, it is a much better hash function and thus has
less chance of bad behavior.
Fixes#8737
benchmark old ns/op new ns/op delta
BenchmarkHash5 173 181 +4.62%
BenchmarkHash16 252 212 -15.87%
BenchmarkHash64 575 419 -27.13%
BenchmarkHash1024 7173 3995 -44.31%
BenchmarkHash65536 516940 313173 -39.42%
BenchmarkHashStringSpeed 300 279 -7.00%
BenchmarkHashBytesSpeed 478 424 -11.30%
BenchmarkHashInt32Speed 217 207 -4.61%
BenchmarkHashInt64Speed 262 231 -11.83%
BenchmarkHashStringArraySpeed 609 631 +3.61%
Change-Id: I0a9335028f32b10ad484966e3019987973afd3eb
Reviewed-on: https://go-review.googlesource.com/1360
Reviewed-by: Russ Cox <rsc@golang.org>
Pointers to zero-sized values may end up pointing to the next
object in memory, and possibly off the end of a span. This
can cause memory leaks and/or confuse the garbage collector.
By putting the overflow pointer at the end of the bucket, we
make sure that pointers to any zero-sized keys or values don't
accidentally point to the next object in memory.
fixes#9384
Change-Id: I5d434df176984cb0210b4d0195dd106d6eb28f73
Reviewed-on: https://go-review.googlesource.com/1869
Reviewed-by: Russ Cox <rsc@golang.org>
Handles the case where the parent is pid 1 (common in docker
containers).
Attempted and failed to write a test for this.
Fixes#9263.
Change-Id: I5c6036446c99e66259a4fab1660b6a594f875020
Reviewed-on: https://go-review.googlesource.com/1372
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Move the symMerge recursion stopping condition
from the beginning of symMerge to the callers.
This halves the number of calls to symMerge
while running 'go test sort'.
benchmark old ns/op new ns/op delta
BenchmarkStable1e6 8358117060 7954143849 -4.83%
BenchmarkStable1e4 40116117 38583285 -3.82%
BenchmarkStableInt1K 119150 115182 -3.33%
BenchmarkStableInt64K 9799845 9515475 -2.90%
BenchmarkStableString1K 388901 393516 +1.19%
BenchmarkStable1e2 124917 123618 -1.04%
Change-Id: I7ba2ca277f213b076fe6830e1139edb47ac53800
Reviewed-on: https://go-review.googlesource.com/1820
Reviewed-by: Robert Griesemer <gri@golang.org>
intDataSize ignores unsigned integers, forcing reads/writes to miss the fast path.
Fixes#8956
Change-Id: Ie79b565b037db3c469aa1dc6d0a8a5a9252d5f0a
Reviewed-on: https://go-review.googlesource.com/1777
Reviewed-by: Russ Cox <rsc@golang.org>
This was a copy-paste error from 9l. Besides incorrectly referring to
cnames9, 6l doesn't even use a->class, so simply remove this.
Fixes#9320
Change-Id: I0e3440c9dae1c3408eb795b3645f9f1dd8f50aed
Reviewed-on: https://go-review.googlesource.com/1516
Reviewed-by: Russ Cox <rsc@golang.org>
Some applications use unpadded base64 format, omitting the trailing
'=' padding characters from the standard base64 format, either to
minimize size or (more justifiably) to avoid use of the '=' character.
Unpadded flavors are standard and documented in section 3.2 of RFC 4648.
To support these unpadded flavors, this change adds two predefined
encoding variables, RawStdEncoding and RawURLEncoding, for unpadded
encodings using the standard and URL character set, respectively.
The change also adds a function WithPadding() to customize the padding
character or disable padding in a custom Encoding.
Finally, I noticed that the existing base64 test-suite was only
exercising the StdEncoding, and not referencing URLEncoding at all.
This change adds test-suite functionality to exercise all four encodings
(the two existing ones and the two new unpadded flavors),
although it still doesn't run *every* test on all four encodings.
Naming: I used the "Raw" prefix because it's more concise than "Unpadded"
and seemed just as expressive, but I have no strong preferences here.
Another short alternative prefix would be "Min" ("minimal" encoding).
Change-Id: Ic0423e02589b39a6b2bb7d0763bd073fd244f469
Reviewed-on: https://go-review.googlesource.com/1511
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Couldn't handle a hex string terminated by anything
other than spaces. Easy to fix.
Fixes#9124.
Change-Id: I18f89a0bd99a105c9110e1ede641873bf9daf3af
Reviewed-on: https://go-review.googlesource.com/1538
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Check for Set error when a boolean flag isn't explicitly given a value.
Fixes#9345
Change-Id: I97a1289f8cf27567d1a726ebe5ef167c800f357c
Reviewed-on: https://go-review.googlesource.com/1897
Reviewed-by: Andrew Gerrand <adg@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Make golint a bit happier
Change-Id: I8a14342f3e492e92bf5efa611f9ef91176624031
Reviewed-on: https://go-review.googlesource.com/1891
Reviewed-by: Minux Ma <minux@golang.org>
Given a file of size N, a request for "Range: bytes=N-*" should
return a 416 [1]. Currently, it returns a 206 and a body of 0
bytes, with the illegal Content-Range of "bytes N-(N-1)/N" [2].
[1]: RFC 7233, sec 2.1: "If a valid byte-range-set includes at least one
byte-range-spec with a first-byte-pos that is less than the current
length of the representation, [...]". sec 3.1: "If all of the
preconditions are true, the server supports the Range header field for
the target resource, and the specified range(s) are invalid or
unsatisfiable, the server SHOULD send a 416 (Range Not Satisfiable)
response."
[2]: RFC 7233, sec 4.2: "A Content-Range field value is invalid if it
contains a byte-range-resp that has a last-byte-pos value less than its
first-byte-pos value, [...]"
Fixes#8988
Change-Id: If3e1134e7815f5d361efea01873b29aafe3de817
Reviewed-on: https://go-review.googlesource.com/1862
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
with uintptr, the check for < 0 will never succeed in mem_plan9.go's
sbrk() because the brk_ syscall returns -1 on failure. fixes the plan9/amd64 build.
this failed on plan9/amd64 because of the attempt to allocate 136GB in mallocinit(),
which failed. it was just by chance that on plan9/386 allocations never failed.
Change-Id: Ia3059cf5eb752e20d9e60c9619e591b80e8fb03c
Reviewed-on: https://go-review.googlesource.com/1590
Reviewed-by: Anthony Martin <ality@pbrane.org>
Reviewed-by: David du Colombier <0intro@gmail.com>
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Co-hacking with Dave Cheney.
Fixes#9405
Change-Id: I14fc3b6a47dcdb5e514e93d062b804bb24e89f47
Reviewed-on: https://go-review.googlesource.com/1875
Reviewed-by: Dave Cheney <dave@cheney.net>
Instead of relying on the asm names declared in the gccgo version of
cgo_export.h, just emit a dummy symbol with the right asm name. This
is enough to let the _cgo_main link succeed, which is all that matters
here.
Fixes#9294.
Change-Id: I803990705b6b226ed0adf17dc57b58a9f501b213
Reviewed-on: https://go-review.googlesource.com/1901
Reviewed-by: Minux Ma <minux@golang.org>
This was brought to my attention because a user thought that because
the file was named "example.go" it served as an example of good coding
practice. It's not an example, of course, but may as well use a more
idiomatic style anyhow.
Change-Id: I7aa720f603f09f7d597fb7536dbf46ef09144e28
Reviewed-on: https://go-review.googlesource.com/1902
Reviewed-by: Minux Ma <minux@golang.org>
"x*41" computes the same value as "x*31 + x*7 + x*3" and (when
compiled by gc) requires just one multiply instruction instead of
three.
Alternatively, the expression could be written as "(x<<2+x)<<3 + x" to
use shifts instead of multiplies (which is how GCC optimizes "x*41").
But gc currently emits suboptimal instructions for this expression
anyway (e.g., separate SHL+ADD instructions rather than LEA on
386/amd64). Also, if such an optimization was worthwhile, it would
seem better to implement it as part of gc's strength reduction logic.
Change-Id: I7156b793229d723bbc9a52aa9ed6111291335277
Reviewed-on: https://go-review.googlesource.com/1830
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Previously, liblink would silently truncate register offset constants
to 32 bits. For example,
MOVD $0x200000004(R2),R3
would assemble to
addis r31,r2,0
addi r3,r31,4
To fix this, limit C_LACON to 32 bit (signed) offsets and introduce a
new C_DACON operand type for larger register offsets. We don't
implement this currently, but at least liblink will now give an error
if it encounters an address like this.
Change-Id: I8e87def8cc4cc5b75498b0fb543ac7666cf2964e
Reviewed-on: https://go-review.googlesource.com/1758
Reviewed-by: Minux Ma <minux@golang.org>
On ppc64, there are three ELF ABI versions an ELF file can request.
Previously, we used 0, which means "unspecified". On our test
machines, this meant to use the default (v1 for big endian and v2 for
little endian), but apparently some systems can pick the wrong ABI if
neither is requested. Leaving this as 0 also confuses libbfd, which
confuses gdb, objdump, etc.
Fix these problems by specifying ABI v1 for big endian and v2 for
little endian.
Change-Id: I4d3d5478f37f11baab3681a07daff3da55802322
Reviewed-on: https://go-review.googlesource.com/1800
Reviewed-by: Minux Ma <minux@golang.org>
When we do y = &x for global variables x and y, y gets initialized
at link time. Do the same for y = &x.f if x is a struct and y=&x[5]
if x is an array.
fixes#9217fixes#9355
Change-Id: Iea3c0ce2ce1b309e2b760e345608fd95460b5713
Reviewed-on: https://go-review.googlesource.com/1691
Reviewed-by: Minux Ma <minux@golang.org>
SSLv3 (the old minimum) is still supported and can be enabled via the
tls.Config, but this change increases the default minimum version to TLS
1.0. This is now common practice in light of the POODLE[1] attack
against SSLv3's CBC padding format.
[1] https://www.imperialviolet.org/2014/10/14/poodle.htmlFixes#9364.
Change-Id: Ibae6666ee038ceee0cb18c339c393155928c6510
Reviewed-on: https://go-review.googlesource.com/1791
Reviewed-by: Minux Ma <minux@golang.org>
Fix TLS_FALLBACK_SCSV check when comparing the client version to the
default max version. This enables the TLS_FALLBACK_SCSV check by default
in servers that do not explicitly set a max version in the tls config.
Change-Id: I5a51f9da6d71b79bc6c2ba45032be51d0f704b5e
Reviewed-on: https://go-review.googlesource.com/1776
Reviewed-by: Adam Langley <agl@golang.org>
Fix style by removing unnecessary named result parameter.
Fix doc comment while here.
Change-Id: If8394e696ab37e00a95484d5137955aa06c59520
Reviewed-on: https://go-review.googlesource.com/1781
Reviewed-by: Yasuhiro MATSUMOTO <mattn.jp@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
On ppc64, liblink rewrites MOVD's of >32-bit constants by putting the
constant in memory and rewriting the MOVD to load from that memory
address. However, there were two bugs in the condition:
a) owing to an incorrect sign extension, it triggered for all negative
constants, and
b) it could trigger for constant offsets from registers (addresses of
the form $n(Rm) in assembly)
Together, these meant instructions of the form MOVD $-n(Rm), x were
compiled by putting -n in memory and rewriting the MOVD to load this
constant from memory (completely dropping Rm).
Change-Id: I1f6cc980efa3e3d6f164b46c985b2c3b55971cca
Reviewed-on: https://go-review.googlesource.com/1752
Reviewed-by: Minux Ma <minux@golang.org>
People are probably not making this mistake anymore.
Fixes#9164
Change-Id: I86b440ed63d09b4ca676bba7034838860f1a5d8b
Reviewed-on: https://go-review.googlesource.com/1782
Reviewed-by: Minux Ma <minux@golang.org>
It shouldn't semacquire() inside an acquirem(), the runtime
thinks that means deadlock. It actually isn't a deadlock, but it
looks like it because acquirem() does m.locks++.
Candidate for inclusion in 1.4.1. runtime.Stack with all=true
is pretty unuseable in GOMAXPROCS>1 environment.
fixes#9321
Change-Id: Iac6b664217d24763b9878c20e49229a1ecffc805
Reviewed-on: https://go-review.googlesource.com/1600
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
I broke the build in https://golang.org/change/207950a when I made
http.Transport send "Connection: close" request headers when
DisableKeepAlives was set true because I didn't run all the tests
before submitting.
httputil.DumpRequestOut used Transport to get its output, and used it
with DisableKeepAlives, so this changed the output.
Rather than updating golden data in our tests (because surely others
depend on the exact bytes from these in their tests), switch to not
using DisableKeepAlives in DumpRequestOut instead, so the output is
the same as before.
Change-Id: I9fad190be8032e55872e6947802055a6d65244df
Reviewed-on: https://go-review.googlesource.com/1632
Reviewed-by: Andrew Gerrand <adg@golang.org>
No bug was open, but I found an old email to myself to investigate
when I suspected this was happening.
Change-Id: Icedefec6f15a000eaabb2693b0640b3b6c8bf82c
Reviewed-on: https://go-review.googlesource.com/1578
Reviewed-by: Andrew Gerrand <adg@golang.org>
Actually fixing this "bug" would be weird, since io.LimitReader already
does what we need, as demonstrated by net/http's use.
Thanks to @davidfstr for pointing this out.
Change-Id: If707bcc698d1666a369b39ddfa9770685fbe3879
Reviewed-on: https://go-review.googlesource.com/1579
Reviewed-by: Rob Pike <r@golang.org>
Most types are reflexive (k == k for all k of type t), so don't
bother calling equal(k, k) when the key type is reflexive.
Change-Id: Ia716b4198b8b298687843b94b878dbc5e8fc2c65
Reviewed-on: https://go-review.googlesource.com/1480
Reviewed-by: Russ Cox <rsc@golang.org>
Remove carriage returns from //go:generate lines.
Carriage returns are the predecessor of BOMs and still
live on Windows.
Fixes#9264
Change-Id: I637748c74335c696b3630f52f2100061153fcdb4
Reviewed-on: https://go-review.googlesource.com/1564
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
People keep pasting all.bash output into GitHub bugs, which turns
the # lines into <h1> headlines. Add some more #s so that the
bug reports are more readable. Not ideal but seems like the best
of a few bad options.
Change-Id: I4c69930ec304b2d504d7cd66221281a8577b87ae
Reviewed-on: https://go-review.googlesource.com/1286
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
//go:nowritebarrier can only be used in package runtime.
It does not disable write barriers; it is an assertion, checked
by the compiler, that the following function needs no write
barriers.
Change-Id: Id7978b779b66dc1feea39ee6bda9fd4d80280b7c
Reviewed-on: https://go-review.googlesource.com/1224
Reviewed-by: Rick Hudson <rlh@golang.org>
Still using the ancient go/types API. Updating that to the modern API
should be a separate effort in a separate change.
Change-Id: Ic1c5ae3c13711d34fe757507ecfc00ee883810bf
Reviewed-on: https://go-review.googlesource.com/1404
Reviewed-by: David Symonds <dsymonds@golang.org>
We forgot to do the usual API review.
Make that not possible in the future.
I'll pull this change over to the main
branch too, but it's more important
(and only testable) here.
LGTM=bradfitz
R=bradfitz
CC=golang-codereviews
https://golang.org/cl/185050043
I read through and vetted these but others should look too.
LGTM=bradfitz, adg
R=r, minux, bradfitz, adg
CC=adg, golang-codereviews, gri, iant
https://golang.org/cl/182560043
This flag no longer exists. It has been replaced with -unit=byte.
Change-Id: Iff9fc501f2c666067c9b1948c4623c8e3adddb8d
Reviewed-on: https://go-review.googlesource.com/1287
Reviewed-by: Russ Cox <rsc@golang.org>
I tried to submit this in Go 1.4 as cl/107540044 but tripped over the
changes for getting C off the G stack. This is a rewritten version that
avoids cgo and works directly with the underlying log device.
Change-Id: I14c227dbb4202690c2c67c5a613d6c6689a6662a
Reviewed-on: https://go-review.googlesource.com/1285
Reviewed-by: Keith Randall <khr@golang.org>
It could only handle one finalizer before it raised an out-of-bounds error.
Fixes issue #9172
Change-Id: Ibb4d0c8aff2d78a1396e248c7129a631176ab427
Reviewed-on: https://go-review.googlesource.com/1201
Reviewed-by: Russ Cox <rsc@golang.org>
needm used to print an error before exiting when it was called too
early, but this error was lost in the transition to Go. Bring back
the error so we don't silently exit(1) when this happens.
Change-Id: I8086932783fd29a337d7dea31b9d6facb64cb5c1
Reviewed-on: https://go-review.googlesource.com/1226
Reviewed-by: Russ Cox <rsc@golang.org>
Avoids a potential O(n^2) performance problem when dequeueing
from very popular channels.
benchmark old ns/op new ns/op delta
BenchmarkChanPopular 2563782 627201 -75.54%
Change-Id: I231aaeafea0ecd93d27b268a0b2128530df3ddd6
Reviewed-on: https://go-review.googlesource.com/1200
Reviewed-by: Russ Cox <rsc@golang.org>
If the symbol table isn't sorted, we print it and abort. However, we
were missing the line break after each symbol, resulting in one
gigantic line instead of a nicely formatted table.
Change-Id: Ie5c6f3c256d0e648277cb3db4496512a79d266dd
Reviewed-on: https://go-review.googlesource.com/1182
Reviewed-by: Russ Cox <rsc@golang.org>
It is unused. It was introduced in the CL that added InputOffset.
I suspect it was an editing mistake.
LGTM=bradfitz
R=bradfitz
CC=golang-codereviews
https://golang.org/cl/182580043
When we start work on Gerrit, ppc64 and garbage collection
work will continue in the master branch, not the dev branches.
(We may still use dev branches for other things later, but
these are ready to be merged, and doing it now, before moving
to Git means we don't have to have dev branches working
in the Gerrit workflow on day one.)
TBR=rlh
CC=golang-codereviews
https://golang.org/cl/183140043
640 bytes ought to be enough for anybody.
We'll bring this back down before Go 1.5. That's issue 9214.
TBR=rlh
CC=golang-codereviews
https://golang.org/cl/188730043
This is going to hurt a bit but we'll make it better later.
Now the race detector can be run again.
I added the write barrier optimizations from
CL 183020043 to try to make it hurt a little less.
TBR=rlh
CC=golang-codereviews
https://golang.org/cl/185070043
This is the last system-dependent file written by cmd/dist.
They are all now written by go generate.
cmd/dist is not needed to start building package runtime
for a different system anymore.
Now all the generated files can be assumed generated, so
delete the clumsy hacks in cmd/api.
Re-enable api check in run.bash.
LGTM=bradfitz
R=bradfitz
CC=golang-codereviews
https://golang.org/cl/185040044
When liblink sees something like
JMP x
...
x: JMP y
it rewrites the first jump to jump directly to y. This is
fine if y is a resolved label. However, it *also* does this
if y is a function symbol, but fails to carry over the
relocation that would later patch in that symbol's value. As
a result, the original jump becomes either a self-jump (if
relative) or a jump to PC 0 (if absolute).
Fix this by disabling this optimization if the jump being
patched in is a jump to a symbol.
LGTM=minux
R=rsc, minux
CC=golang-codereviews
https://golang.org/cl/185890044
Frankly, I don't understand how the current code could possibly work except
when every android program is using cgo. Discovered this while working on
the iOS port.
LGTM=crawshaw, rsc
R=rsc, crawshaw
CC=golang-codereviews
https://golang.org/cl/177470043
The new semantics of split require the newline be present.
The test was stale.
LGTM=adg
R=golang-codereviews, adg
CC=golang-codereviews
https://golang.org/cl/182480043
Scanner can't handle stupid long lines and there are
reports of stupid long lines in production.
Note the issue isn't long "//go:generate" lines, but
any long line in any Go source file.
To be fair, if you're going to have a stupid long line
it's not a bad bet you'll want to run it through go
generate, because it's some embeddable asset that
has been machine generated. (One could ask why
that generation process didn't add a newline or two,
but we should cope anyway.)
Rewrite the file scanner in "go generate" so it can
handle arbitrarily long lines, and only stores in memory
those lines that start "//go:generate".
Also: Adjust the documentation to make clear that it
does not parse the file.
Fixes#9143.
Fixes#9196.
LGTM=rsc, dominik.honnef
R=rsc, cespare, minux, dominik.honnef
CC=golang-codereviews
https://golang.org/cl/182970043
While we're at there, also add a message to prompt the user to install
Graphviz if "dot" command is not found.
Fixes#9178.
LGTM=adg, alex.brainman, cookieo9, rsc
R=rsc, adg, bradfitz, alex.brainman, cookieo9, smyrman
CC=golang-codereviews
https://golang.org/cl/180380043
Move change from CL 170770043 to correct file and regenerate docs
for changes from CL 164120043.
LGTM=adg
R=golang-codereviews, adg, bradfitz
CC=golang-codereviews
https://golang.org/cl/183000043
During garbage collection, after scanning a stack, we think about
shrinking it to reclaim some memory. The shrinking code (called
while the world is stopped) checked that the status was Gwaiting
or Grunnable and then changed the state to Gcopystack, to essentially
lock the stack so that no other GC thread is scanning it.
The same locking happens for stack growth (and is more necessary there).
oldstatus = runtime·readgstatus(gp);
oldstatus &= ~Gscan;
if(oldstatus == Gwaiting || oldstatus == Grunnable)
runtime·casgstatus(gp, oldstatus, Gcopystack); // oldstatus is Gwaiting or Grunnable
else
runtime·throw("copystack: bad status, not Gwaiting or Grunnable");
Unfortunately, "stop the world" doesn't stop everything. It stops all
normal goroutine execution, but the network polling thread is still
blocked in epoll and may wake up. If it does, and it chooses a goroutine
to mark runnable, and that goroutine is the one whose stack is shrinking,
then it can happen that between readgstatus and casgstatus, the status
changes from Gwaiting to Grunnable.
casgstatus assumes that if the status is not what is expected, it is a
transient change (like from Gwaiting to Gscanwaiting and back, or like
from Gwaiting to Gcopystack and back), and it loops until the status
has been restored to the expected value. In this case, the status has
changed semi-permanently from Gwaiting to Grunnable - it won't
change again until the GC is done and the world can continue, but the
GC is waiting for the status to change back. This wedges the program.
To fix, call a special variant of casgstatus that accepts either Gwaiting
or Grunnable as valid statuses.
Without the fix bug with the extra check+throw in casgstatus, the
program below dies in a few seconds (2-10) with GOMAXPROCS=8
on a 2012 Retina MacBook Pro. With the fix, it runs for minutes
and minutes.
package main
import (
"io"
"log"
"net"
"runtime"
)
func main() {
const N = 100
for i := 0; i < N; i++ {
l, err := net.Listen("tcp", "127.0.0.1:0")
if err != nil {
log.Fatal(err)
}
ch := make(chan net.Conn, 1)
go func() {
var err error
c1, err := net.Dial("tcp", l.Addr().String())
if err != nil {
log.Fatal(err)
}
ch <- c1
}()
c2, err := l.Accept()
if err != nil {
log.Fatal(err)
}
c1 := <-ch
l.Close()
go netguy(c1, c2)
go netguy(c2, c1)
c1.Write(make([]byte, 100))
}
for {
runtime.GC()
}
}
func netguy(r, w net.Conn) {
buf := make([]byte, 100)
for {
bigstack(1000)
_, err := io.ReadFull(r, buf)
if err != nil {
log.Fatal(err)
}
w.Write(buf)
}
}
var g int
func bigstack(n int) {
var buf [100]byte
if n > 0 {
bigstack(n - 1)
}
g = int(buf[0]) + int(buf[99])
}
Fixes#9186.
LGTM=rlh
R=austin, rlh
CC=dvyukov, golang-codereviews, iant, khr, r
https://golang.org/cl/179680043
These accomplished the same thing, but R_CALLPOWER expected
the whole instruction to be in the addend (and completely
overwrote what was in the text section), while R_PPC64_REL24
overwrites only bits 6 through 24 of whatever was in the text
section. Make R_CALLPOWER work like R_PPC64_REL24 to ease the
implementation of dynamic linking.
LGTM=rsc
R=rsc
CC=golang-codereviews, minux
https://golang.org/cl/177430043
warning: src/cmd/5g/reg.c:461 format mismatch d VLONG, arg 5
warning: src/cmd/6g/reg.c:396 format mismatch d VLONG, arg 5
warning: src/cmd/9g/reg.c:440 format mismatch d VLONG, arg 5
LGTM=minux
R=rsc, minux
CC=golang-codereviews
https://golang.org/cl/179300043
This was based on the 9c peephole optimizer, modified to work
with code generated by gc and use the proginfo infrastructure
in gc.
LGTM=rsc
R=rsc, bradfitz, minux
CC=golang-codereviews
https://golang.org/cl/179190043
This adds some utilities for converting between the CC, V, and
VCC variants of operations and uses these to derive the
ProgInfo entries for these variants (which are identical to
the ProgInfo for the base operations).
The 9g peephole optimizer will also use these conversion
utilities.
LGTM=minux, rsc
R=rsc, dave, minux
CC=golang-codereviews
https://golang.org/cl/180110044
