x/sys/unix: TestClockNanosleep failures on linux-mipsle-rtrk builder

[bcmills]

https://build.golang.org/log/f0d29a593ccba5704246987f49074bfca43dd1b9

--- FAIL: TestClockNanosleep (0.10s)
    syscall_linux_test.go:580: ClockNanosleep(CLOCK_REALTIME, TIMER_ABSTIME, &unix.Timespec{Sec:1573834610, Nsec:328568608} (=2019-11-15 17:16:50.328568608 +0100 CET m=+0.544561463), nil) = interrupted system call
FAIL
FAIL	golang.org/x/sys/unix	0.599s

CC @tklauser @bogojevic @milanknezevic @cherrymui

[ianlancetaylor]

Now that we use signals for preemption I guess all those calls need to check for EINTR.

[gopherbot]

Change https://golang.org/cl/207285 mentions this issue: unix: don't fail TestClockNanosleep on EINTR from ClockNanosleep

[bcmills]

@ianlancetaylor, aren't the runtime's handlers all registered with SA_RESTART set?

I would expect preemption to produce spurious EINTR return values in some cgo programs due to #20400, but I don't understand why it would affect a pure-Go package test.

[ianlancetaylor]

SA_RESTART does not apply to all system calls. For the Linux kernel, see http://man7.org/linux/man-pages/man7/signal.7.html, where it says that the clock_nanosleep system call, among others, always returns EINTR even if the signal was registered with SA_RESTART.

[gopherbot]

Change https://golang.org/cl/207290 mentions this issue: unix: fix EINTR check in TestClockNanosleep