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>
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>
"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>
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>
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>
//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>
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>
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
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
Otherwise both zgoos_linux.go and zgoos_android.go will be compiled
for GOOS=android.
LGTM=crawshaw, rsc
R=rsc, crawshaw
CC=golang-codereviews
https://golang.org/cl/178110043
We don't know what we need yet, so add them all.
Add them even on x86 architectures (as no-ops) so that
the GC can refer to them unconditionally.
Eventually we'll know what we want and probably
have just one 'prefetch' with an appropriate meaning
on each architecture.
LGTM=rlh
R=rlh
CC=golang-codereviews
https://golang.org/cl/179160043
Thanks to Aram Hăvărneanu, Nick Owens
and Russ Cox for the early reviews.
LGTM=aram, rsc
R=rsc, lucio.dere, aram, ality
CC=golang-codereviews, mischief
https://golang.org/cl/175370043
Race detector runtime does not tolerate operations on addresses
that was not previously declared with __tsan_map_shadow
(namely, data, bss and heap). The corresponding address
checks for atomic operations were removed in
https://golang.org/cl/111310044
Restore these checks.
It's tricker than just not calling into race runtime,
because it is the race runtime that makes the atomic
operations themselves (if we do not call into race runtime
we skip the atomic operation itself as well). So instead we call
__tsan_go_ignore_sync_start/end around the atomic operation.
This forces race runtime to skip all other processing
except than doing the atomic operation itself.
Fixes#9136.
LGTM=rsc
R=rsc
CC=golang-codereviews
https://golang.org/cl/179030043
