runtime: crash when receiving multiple signals on C threads on NetBSD

[minux]

run the following program on netbsd/386:
package main

/*
#include <signal.h>
#include <pthread.h>

static void *thread(void *p) {
    (void)p;
    const int N = 100;
    int i;
    for (i = 0; i < N; i++)
        pthread_kill(pthread_self(), SIGCHLD);
    return NULL;
}
void testSendSIG() {
    const int M = 10;
    pthread_t tid[M];
    int i;
    for (i = 0; i < M; i++)
        pthread_create(&tid[i], 0, thread, NULL);
    for (i = 0; i < M; i++)
        pthread_join(tid[i], 0);
}
*/
import "C"

import (
    "fmt"
    "os"
    "os/signal"
    "syscall"
)

func main() {
    ch, wait := make(chan os.Signal), make(chan bool)
    signal.Notify(ch, syscall.SIGCHLD)
    go func() {
        wait2 := wait
        for {
            select {
            case v := <-ch:
                fmt.Println("got", v)
                wait <- false
            case <-wait2:
                wait2 = nil
            }
        }
    }()
    wait <- true
    C.testSendSIG()
    <-wait
}


it will crash as follows:
SIGSEGV: segmentation violation
PC=0x3e8


goroutine 1 [syscall]:
main._Cfunc_testSendSIG(0x80b0ca0)
    command-line-arguments/_obj/_cgo_defun.c:43 +0x31
main.main()
    command-line-arguments/_obj/signal.cgo1.go:34 +0x14c

goroutine 3 [syscall]:
os/signal.loop()
    /home/builder/go.hg/src/pkg/os/signal/signal_unix.go:21 +0x1e
created by os/signal.init·1
    /home/builder/go.hg/src/pkg/os/signal/signal_unix.go:27 +0x31

goroutine 4 [select]:
main.func·001()
    command-line-arguments/_obj/signal.cgo1.go:23 +0x15c
created by main.main
    command-line-arguments/_obj/signal.cgo1.go:31 +0x12a

goroutine 5 [runnable]:
runtime.exitsyscall()
    /home/builder/go.hg/src/pkg/runtime/proc.c:1466 +0x109
runtime.cgocallbackg(0x18438c2c, 0x18438c3c, 0x4)
    /home/builder/go.hg/src/pkg/runtime/cgocall.c:244 +0x76
runtime.cgocallback_gofunc(0x0, 0x0)
    /home/builder/go.hg/src/pkg/runtime/asm_386.s:583 +0x5a
runtime.goexit()
    /home/builder/go.hg/src/pkg/runtime/proc.c:1289

eax     0x0
ebx     0x18439250
ecx     0x1
edx     0x0
edi     0x805c540
esi     0x18401780
ebp     0x813b160
esp     0x18438bd8
eip     0x3e8
eflags  0x10216
cs      0x1b
fs      0xab
gs      0xb3

[minux]

Comment 1:

i have reduced N to 1, and M to 2.
it seems we should be able to reproduce the problem even using export from multiple
C thread simultaneously.

[rsc]

Comment 2:

Labels changed: added go1.2.

[rsc]

Comment 3:

I can't reproduce this on Darwin/386. When I run the program it just says 'got child
exited' and then exits.
Are the pthread_creates all succeeding? Perhaps if one fails and leaves tid[i]
uninitialized, the later pthread_join will dereference a bad value.

[rsc]

Comment 4:

Also can you run it under gdb and see if you can get a stack trace for the C code that
is dying?

[rsc]

Comment 5:

It also works on linux/386. Starting to look like a NetBSD problem.

Labels changed: added go1.2maybe, removed go1.2.

[rsc]

Comment 6:

Labels changed: added go1.3, removed go1.2maybe.

[rsc]

Comment 7:

Labels changed: added release-go1.3.

[rsc]

Comment 8:

Labels changed: removed go1.3.

[rsc]

Comment 9:

Labels changed: added repo-main.

[rsc]

Comment 10:

It would be nice if someone found this but NetBSD is not one of the first class ports
(golang.org/wiki/PortingPolicy), so I am removing the release tag entirely. Adding
OS-NetBSD. If NetBSD is ever to become a first-class port, hopefully we'll notice this
issue and force someone to fix the bug then.

Labels changed: added os-netbsd, removed release-go1.3.

[rsc]

Comment 11:

Labels changed: added priority-later, release-none, removed priority-soon.

[mikioh]

Note that this still crashes w/ go1.6beta1 + the fix for #14067 on both netbsd/386 and netbsd/amd64.

[bradfitz]

All NetBSD bugs are on hold until there's a release later than NetBSD 7.1, which still had known issues fixed in NetBSD-current after 7.1.

[bradfitz]

Fixed by requiring NetBSD 8+ for Go 1.10.

Documentation bug is #22911