runtime/pprof: SIGPROF crashes process on netbsd/amd64 #10166
Description
This is on
$ uname -a
NetBSD edamame.mirbsd.org 7.0_BETA NetBSD 7.0_BETA (GENERIC.201501181820Z) amd64
$ go version
go version devel +3b1d692 Sat Mar 14 10:47:19 2015 +0000 netbsd/amd64
At tip, the runtime/pprof test fails, which also makes all.bash fail:
=== RUN TestMemoryProfiler
--- PASS: TestMemoryProfiler (0.02s)
=== RUN TestCPUProfile
exit status 255
FAIL runtime/pprof 0.038s
Looking at ktrace output, it looks like the SIGPROF is what kills the process:
16246 3 pprof.test RET compat_60__lwp_park -1 errno 4
Interrupted system call
16246 3 pprof.test PSIG SIGPROF caught handler=0x456980
mask=(): code=SI_TIMER sent by pid=0, uid=0 with sigval 0x2)
16246 3 pprof.test CALL setcontext(0x40)
16246 3 pprof.test RET setcontext -1 errno 14 Bad address
16246 3 pprof.test CALL exit(0xffffffffffffffff)
According to Minux:
has NetBSD changed kernel signal handler ABI?
or there might some stack/register corruption in runtime.sigtramp. The implicit
setcontext(2) is picking up a wrong address for ucontext. Should we preserve
callee-saved registers in runtime.sigtramp? I'd start by setting a break on
runtime.sigtramp, and then disassemble the function it should return to and try
understand where does 0x40 come from.