runtime: crash in GC on 386

[dvyukov]

Current tip (13699:21130d62eeb0), linux/386, run enough times on a parallel machine:

$ GOARCH=386 GOMAXPROCS=32 go test std

panic: runtime error: invalid memory address or nil pointer dereference [recovered]
    panic: runtime error: invalid memory address or nil pointer dereference
[signal 0xb code=0x1 addr=0x0 pc=0x8073d94]

goroutine 50 [running]:
testing.func·003(0xf7756fd4, 0xf7756100)
    /usr/local/google/home/dvyukov/go_vanilla/src/pkg/testing/testing.go:268 +0x123
----- stack segment boundary -----
runtime_test.TestStopTheWorldDeadlock(0x18855200, 0xe)
    /usr/local/google/home/dvyukov/go_vanilla/src/pkg/runtime/proc_test.go:25 +0x24
testing.tRunner(0x18855200, 0x82b3dd4, 0x0)
    /usr/local/google/home/dvyukov/go_vanilla/src/pkg/testing/testing.go:273 +0x71
created by testing.RunTests
    /usr/local/google/home/dvyukov/go_vanilla/src/pkg/testing/testing.go:349 +0x711

The culprit is:

// mgc0.c
static struct {
    uint64  full;  // lock-free list of full blocks
    uint64  empty; // lock-free list of empty blocks
...
} work;

8c does not align uint64 on 8-bytes, occasionally full/empty crosses cache-line boundary
and then

TEXT runtime·atomicload64(SB), 7, $0
    MOVL    4(SP), BX
    MOVL    8(SP), AX
    // MOVQ (%EAX), %MM0
    BYTE $0x0f; BYTE $0x6f; BYTE $0x00
    // MOVQ %MM0, 0(%EBX)
    BYTE $0x0f; BYTE $0x7f; BYTE $0x03
    // EMMS
    BYTE $0x0F; BYTE $0x77
    RET

becomes not particularly atomic.

There are 2 ways to fix it:
1. Fix the compilers to properly align uint64 (I am not sure why they do not align it
now).
2. Just patch the GC to manually align the vars.

[dvyukov]

Comment 1:

I think sync/atomic is also broken.
There is another option - atomic ops can check whether the adds is aligned and fallback
to LOCK CMPXCHG8B. Some C/C++ libraries use the trick, but in C/C++ is not expected to
happen, while in Go it can happen with fair probability, so I think it's a bad option.

[minux]

Comment 2:

related issue #3799: 5c, 5g, 8c, 8g: make 64-bit fields 64-bit aligned.

[rsc]

Comment 3:

Status changed to Duplicate.

Merged into issue #3799.