runtime: TestNetpollDeadlock failures

[gopherbot]

#!watchflakes
post <- (pkg == "runtime" || pkg == "runtime:mayMoreStackPreempt") && test == "TestNetpollDeadlock"

Issue created automatically to collect these failures.

Example (log):

--- FAIL: TestNetpollDeadlock (2640.70s)
    exec.go:193: test timed out while running command: /tmp/workdir/tmp/go-build229666404/testprognet.exe NetpollDeadlock
    crash_test.go:506: /tmp/workdir/tmp/go-build229666404/testprognet.exe NetpollDeadlock: exit status 2
    crash_test.go:509: output does not start with "done\n":
        dialing
        SIGQUIT: quit
        PC=0x8d2fc m=0 sigcode=0

        goroutine 0 gp=0x23de40 m=0 mp=0x23e6c0 [idle]:
        runtime.futex(0x23e800, 0x80, 0x0, 0x0, 0x0, 0x0)
...
        r24     0x0
        r25     0x0
        r26     0x151bf0
        r27     0x237000
        r28     0x23de40
        r29     0xfffff9abfd58
        lr      0x4a40c
        sp      0xfffff9abfd60
        pc      0x8d2fc
        fault   0x0

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "runtime" && test == "TestNetpollDeadlock"

2024-03-08 22:14 linux-arm64-longtest go@c220fbab runtime.TestNetpollDeadlock (log)

--- FAIL: TestNetpollDeadlock (2640.70s)
    exec.go:193: test timed out while running command: /tmp/workdir/tmp/go-build229666404/testprognet.exe NetpollDeadlock
    crash_test.go:506: /tmp/workdir/tmp/go-build229666404/testprognet.exe NetpollDeadlock: exit status 2
    crash_test.go:509: output does not start with "done\n":
        dialing
        SIGQUIT: quit
        PC=0x8d2fc m=0 sigcode=0

        goroutine 0 gp=0x23de40 m=0 mp=0x23e6c0 [idle]:
        runtime.futex(0x23e800, 0x80, 0x0, 0x0, 0x0, 0x0)
...
        r24     0x0
        r25     0x0
        r26     0x151bf0
        r27     0x237000
        r28     0x23de40
        r29     0xfffff9abfd58
        lr      0x4a40c
        sp      0xfffff9abfd60
        pc      0x8d2fc
        fault   0x0

2024-03-09 14:19 windows-amd64-longtest go@74726def runtime.TestNetpollDeadlock (log)

--- FAIL: TestNetpollDeadlock (2587.85s)
    exec.go:193: test timed out while running command: C:\Users\gopher\AppData\Local\Temp\1\go-build3501004536\testprognet.exe NetpollDeadlock
    crash_test.go:506: C:\Users\gopher\AppData\Local\Temp\1\go-build3501004536\testprognet.exe NetpollDeadlock: exit status 1
    crash_test.go:509: output does not start with "done\n":
        dialing

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "runtime" && test == "TestNetpollDeadlock"

2024-03-12 17:24 linux-386-longtest go@34d28ba9 runtime.TestNetpollDeadlock (log)

--- FAIL: TestNetpollDeadlock (2641.35s)
    exec.go:193: test timed out while running command: /workdir/tmp/go-build2841960519/testprognet.exe NetpollDeadlock
    crash_test.go:506: /workdir/tmp/go-build2841960519/testprognet.exe NetpollDeadlock: exit status 2
    crash_test.go:509: output does not start with "done\n":
        dialing
        SIGQUIT: quit
        PC=0x80bc49d m=0 sigcode=0

        goroutine 0 gp=0x823b700 m=0 mp=0x823bd40 [idle]:
        runtime.futex(0x823be08, 0x80, 0x0, 0x0, 0x0, 0x0)
...
        edx    0x0
        edi    0x0
        esi    0x0
        ebp    0x0
        esp    0xff987e44
        eip    0x80bc49d
        eflags 0x286
        cs     0x23
        fs     0x0
        gs     0x63

— watchflakes

[mknyszek]

@rsc, in triage, we think this may somehow be related to the timer refactor. The first failure happened immediately after the last CL of the refactor landed. (This is also plausible because timers affect netpoll deadlines.)

However, what's weird is that the test is incredibly simple, basically just calling net.Dial in an init function, so I dunno. Hopefully it's not hard to reproduce.

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- (pkg == "runtime" || pkg == "runtime:mayMoreStackPreempt") && test == "TestNetpollDeadlock"

2024-03-08 22:14 gotip-linux-386-longtest go@3c39b2ed runtime:mayMoreStackPreempt.TestNetpollDeadlock (log)

=== RUN   TestNetpollDeadlock
=== PAUSE TestNetpollDeadlock
=== CONT  TestNetpollDeadlock
    exec.go:193: test timed out while running command: /home/swarming/.swarming/w/ir/x/t/go-build3587028201/testprognet.exe NetpollDeadlock
    crash_test.go:506: /home/swarming/.swarming/w/ir/x/t/go-build3587028201/testprognet.exe NetpollDeadlock: exit status 2
    crash_test.go:509: output does not start with "done\n":
        dialing
        SIGQUIT: quit
        PC=0x80bc35d m=0 sigcode=0
        
...
        edi    0x0
        esi    0x0
        ebp    0x0
        esp    0xffa308e4
        eip    0x80bc35d
        eflags 0x286
        cs     0x23
        fs     0x0
        gs     0x63
--- FAIL: TestNetpollDeadlock (2633.08s)

2024-03-11 18:35 gotip-windows-amd64-longtest go@edbb5a13 runtime:mayMoreStackPreempt.TestNetpollDeadlock (log)

=== RUN   TestNetpollDeadlock
=== PAUSE TestNetpollDeadlock
=== CONT  TestNetpollDeadlock
    exec.go:193: test timed out while running command: C:\b\s\w\ir\x\t\go-build2767286069\testprognet.exe NetpollDeadlock
    crash_test.go:506: C:\b\s\w\ir\x\t\go-build2767286069\testprognet.exe NetpollDeadlock: exit status 1
    crash_test.go:509: output does not start with "done\n":
        dialing
--- FAIL: TestNetpollDeadlock (2586.49s)

2024-03-12 17:04 gotip-windows-amd64-longtest go@5eb87c39 runtime:mayMoreStackPreempt.TestNetpollDeadlock (log)

=== RUN   TestNetpollDeadlock
=== PAUSE TestNetpollDeadlock
=== CONT  TestNetpollDeadlock
    crash_test.go:170: running C:\b\s\w\ir\x\w\goroot\bin\go.exe build -o C:\b\s\w\ir\x\t\go-build1475190012\testprognet.exe
    crash_test.go:176: built testprognet in 2.1601831s
    exec.go:193: test timed out while running command: C:\b\s\w\ir\x\t\go-build1475190012\testprognet.exe NetpollDeadlock
    crash_test.go:506: C:\b\s\w\ir\x\t\go-build1475190012\testprognet.exe NetpollDeadlock: exit status 1
    crash_test.go:509: output does not start with "done\n":
        dialing
--- FAIL: TestNetpollDeadlock (2590.73s)

2024-03-13 18:22 gotip-windows-amd64-longtest go@38723f2b runtime:mayMoreStackPreempt.TestNetpollDeadlock (log)

=== RUN   TestNetpollDeadlock
=== PAUSE TestNetpollDeadlock
=== CONT  TestNetpollDeadlock
    exec.go:193: test timed out while running command: C:\b\s\w\ir\x\t\go-build1389385579\testprognet.exe NetpollDeadlock
    crash_test.go:506: C:\b\s\w\ir\x\t\go-build1389385579\testprognet.exe NetpollDeadlock: exit status 1
    crash_test.go:509: output does not start with "done\n":
        dialing
--- FAIL: TestNetpollDeadlock (2590.85s)

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- (pkg == "runtime" || pkg == "runtime:mayMoreStackPreempt") && test == "TestNetpollDeadlock"

2025-12-30 22:04 gotip-plan9-386 go@d64add4d runtime.TestNetpollDeadlock (log)

=== RUN   TestNetpollDeadlock
=== PAUSE TestNetpollDeadlock
=== CONT  TestNetpollDeadlock
    crash_test.go:597: /tmp/go-build2818447297/testprognet.exe NetpollDeadlock: exit status: 'testprognet.exe 14537: 2'
    crash_test.go:600: output does not start with "done\n":
        dialing
        panic: runtime error: invalid memory address or nil pointer dereference
        [signal sys: trap: fault write code=0x0 addr=0x0 pc=0x60f0]
        
        goroutine 1 [running]:
...
        	/home/swarming/luci/386/.swarming/w/ir/x/w/goroot/src/net/dial.go:505
        net.Dial({0x110784, 0x3}, {0x112371, 0xf})
        	/home/swarming/luci/386/.swarming/w/ir/x/w/goroot/src/net/dial.go:472 +0x76
        main.NetpollDeadlockInit()
        	/home/swarming/luci/386/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprognet/net.go:19 +0x84
        main.registerInit(...)
        	/home/swarming/luci/386/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprognet/main.go:20
        main.init.0()
        	/home/swarming/luci/386/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprognet/net.go:13 +0x66
--- FAIL: TestNetpollDeadlock (0.04s)

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- (pkg == "runtime" || pkg == "runtime:mayMoreStackPreempt") && test == "TestNetpollDeadlock"

2026-04-29 17:23 gotip-linux-amd64-msan-clang15 go@a2214422 runtime.TestNetpollDeadlock (log)

=== RUN   TestNetpollDeadlock
=== PAUSE TestNetpollDeadlock
=== CONT  TestNetpollDeadlock
    crash_test.go:597: /home/swarming/.swarming/w/ir/x/t/go-build2056329708/testprognet.exe NetpollDeadlock: exec: WaitDelay expired before I/O complete
--- FAIL: TestNetpollDeadlock (5.16s)

— watchflakes