runtime: GC with gctrace=2 fails

[rsc]

This crashes, and seems to have been crashing for a while:

go test -c strings && GODEBUG=gctrace=2,gccheckmark=0 ./strings.test

My cleanup means that now in gccheckmark=1 mode it will also crash, but that's not really my fault. (My cleanup fixed the bug that in gccheckmark=1 mode the gctrace=2 was not doing a full mark sweep in the second GC.)

@aclements @RLH

[rsc]

g% go test -c strings && GODEBUG=gctrace=2,gccheckmark=0 ./strings.test
gc1(1): 0+1+103+0 us, 2 -> 4 MB, 310 (310-0) objects, 5 goroutines, 39/0/0 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc2(1): 152+1+331+0 us, 2 -> 2 MB, 235 (311-76) objects, 6 goroutines, 39/0/31 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc3(1): 1+1+129+0 us, 2 -> 4 MB, 31207 (31294-87) objects, 5 goroutines, 399/1/0 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc4(1): 995+2+336+1 us, 2 -> 2 MB, 233 (31294-31061) objects, 5 goroutines, 399/0/392 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc5(1): 2+2+210+1 us, 2 -> 6 MB, 3350 (34420-31070) objects, 6 goroutines, 399/0/0 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc6(1): 145+1+287+0 us, 6 -> 6 MB, 247 (34420-34173) objects, 6 goroutines, 399/0/62 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc7(1): 1+1+95+0 us, 2 -> 10 MB, 237 (34422-34185) objects, 6 goroutines, 399/1/0 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc8(1): 54+1+305+0 us, 10 -> 10 MB, 237 (34422-34185) objects, 6 goroutines, 399/0/29 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc9(1): 1+2+190+1 us, 2 -> 14 MB, 237 (34424-34187) objects, 6 goroutines, 399/6/0 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
gc10(1): 56+1+340+0 us, 14 -> 14 MB, 237 (34424-34187) objects, 6 goroutines, 399/0/29 sweeps, 0(0) handoff, 0(0) steal, 0/0/0 yields
fatal error: runtime: sudog with non-nil elem

goroutine 1 [running]:
runtime.throw(0x1eeff0, 0x20)
    /Users/rsc/g/go/src/runtime/panic.go:511 +0xa0 fp=0x208423bd0 sp=0x208423bb8
runtime.releaseSudog(0x2083525c0)
    /Users/rsc/g/go/src/runtime/proc.go:201 +0x32 fp=0x208423bf0 sp=0x208423bd0
runtime.chanrecv(0x13c8c0, 0x20842a2a0, 0x208423dc8, 0x208349c01, 0x60000)
    /Users/rsc/g/go/src/runtime/chan.go:429 +0x4a6 fp=0x208423c78 sp=0x208423bf0
runtime.chanrecv1(0x13c8c0, 0x20842a2a0, 0x208423dc8)
    /Users/rsc/g/go/src/runtime/chan.go:318 +0x2b fp=0x208423ca8 sp=0x208423c78
testing.RunTests(0x213050, 0x2aca00, 0x2e, 0x2e, 0x1)
    /Users/rsc/g/go/src/testing/testing.go:557 +0xa8c fp=0x208423e50 sp=0x208423ca8
testing.(*M).Run(0x2083765a0, 0x2ace60)
    /Users/rsc/g/go/src/testing/testing.go:486 +0x79 fp=0x208423ed8 sp=0x208423e50
main.main()
    strings/_test/_testmain.go:278 +0x1d5 fp=0x208423fa0 sp=0x208423ed8
runtime.main()
    /Users/rsc/g/go/src/runtime/proc.go:88 +0x1d2 fp=0x208423fe0 sp=0x208423fa0
runtime.goexit()
    /Users/rsc/g/go/src/runtime/asm_amd64.s:2446 +0x1 fp=0x208423fe8 sp=0x208423fe0
g% 

[dvyukov]

This seems like an issue with GC (heap corruption) rather than with sudoges. At least I don't see how sg.elem can be non nil in chan code.

[aclements]

Fixed by e31e35a