(Thanks to ken and rsc for pointing this out)
rsc:
ken pointed out that there's a race in the new
one-lock-per-channel code. the issue is that
if one goroutine has gone to sleep doing
select {
case <-c1:
case <-c2:
}
and then two more goroutines try to send
on c1 and c2 simultaneously, the way that
the code makes sure only one wins is the
selgen field manipulation in dequeue:
// if sgp is stale, ignore it
if(sgp->selgen != sgp->g->selgen) {
//prints("INVALID PSEUDOG POINTER\n");
freesg(c, sgp);
goto loop;
}
// invalidate any others
sgp->g->selgen++;
but because the global lock is gone both
goroutines will be fiddling with sgp->g->selgen
at the same time.
This results in a 7% slowdown in the single threaded case for a
ping-pong microbenchmark.
Since the cas predominantly succeeds, adding a simple check first
didn't make any difference.
R=rsc
CC=golang-dev
https://golang.org/cl/180068
* add runtime sliceslice1 for x[lo:]
* remove runtime arraytoslice, rewriting &arr into arr[0:len(arr)].
* port cgen_inline into 8g, 5g.
* use native memmove in maps
R=ken2
https://golang.org/cl/157106
In thread.c, we need to cast to whatever the native
size of intptr is on the system, but we only have
uintptr available. They're the same size, but can't
do signed casts without this one :).
R=rsc
CC=golang-dev
https://golang.org/cl/156073
SELinux will cause mmap to fail when we request w+x memory unless the
user has configured their policies. We have a warning in make.bash,
but it's quite likely that the policy will be reset at some point and
then all their binaries start failing.
This patch prints a warning on Linux when mmap fails with EACCES.
R=rsc
CC=golang-dev
https://golang.org/cl/152086
now that all arguments are passed on the stack.
go/test: passes 89% (310/345)
R=rsc
APPROVED=rsc
DELTA=33 (13 added, 14 deleted, 6 changed)
OCL=36009
CL=36022
because they are in package runtime.
another step to enforcing package boundaries.
R=r
DELTA=732 (114 added, 93 deleted, 525 changed)
OCL=35811
CL=35824
* in 6l, -K already meant check for stack underflow.
add -KK to mean double-check stack overflows
even in nosplit functions.
* comment out print locks; they deadlock too easily
but are still useful to put back for special occasions.
* let runcgo assembly switch to scheduler stack
without involving scheduler directly. because runcgo
gets called from matchmg, it is too hard to keep it
from being called on other stacks.
R=r
DELTA=94 (65 added, 18 deleted, 11 changed)
OCL=35591
CL=35604
better mach binaries.
cgo working on darwin+linux amd64+386.
eliminated context switches - pi is 30x faster.
add libcgo to build.
on snow leopard:
- non-cgo binaries work; all tests pass.
- cgo binaries work on amd64 but not 386.
R=r
DELTA=2031 (1316 added, 626 deleted, 89 changed)
OCL=35264
CL=35304
* change ldt0setup to set GS itself; nacl won't let us do it.
* change breakpoint to INT $3 so 8l can translate to HLT for nacl.
* panic if closure is needed on nacl.
* do not try to access symbol table on nacl.
* mmap in 64kB chunks.
nacl support:
* system calls, threading, locks.
R=r
DELTA=365 (357 added, 5 deleted, 3 changed)
OCL=34880
CL=34906
runtime package for use by debugger,
which needs to make sure that all ptrace calls
about a given pid come from the same thread.
R=r
DELTA=175 (90 added, 63 deleted, 22 changed)
OCL=31546
CL=31558
* keep coherent SP/PC in gobuf
(i.e., SP that would be in use at that PC)
* gogocall replaces setspgoto,
should work better in presence of link registers
* delete unused system calls
only amd64; 386 is now broken
R=r
DELTA=548 (183 added, 183 deleted, 182 changed)
OCL=30381
CL=30442
not observed: do not use malloc to allocate stacks
during garbage collection, because it would make the
malloc data structures change underfoot.
R=r
DELTA=6 (3 added, 0 deleted, 3 changed)
OCL=30323
CL=30326
