三色标记与写屏障:并发标记期间新指针怎么不丢对象

  共 460 次 GC;前 4 行:
    gc 1 @0.002s 4%: 0.028+1.0+0.020 ms clock, 0.28+0/1.3/0.10+0.20 ms cpu, 3->4->1 MB, 4 MB goal, 0 MB stacks, 0 MB globals, 10 P
    gc 2 @0.004s 5%: 0.020+0.66+0.021 ms clock, 0.20+0.088/0.74/0.001+0.21 ms cpu, 3->4->1 MB, 4 MB goal, 0 MB stacks, 0 MB globals, 10 P
    gc 3 @0.005s 5%: 0.016+0.75+0.007 ms clock, 0.16+0.13/0.83/0.029+0.073 ms cpu, 3->4->1 MB, 4 MB goal, 0 MB stacks, 0 MB globals, 10 P
    gc 4 @0.006s 6%: 0.024+0.86+0.009 ms clock, 0.24+0.080/1.1/0.28+0.092 ms cpu, 3->4->1 MB, 4 MB goal, 0 MB stacks, 0 MB globals, 10 P
  逐段解释第 1 行:@0.002s 时开始;GC 占总 CPU 的 4%;墙钟时间 三段 = 0.028 + 1.0 + 0.020 ms(清扫终止 STW + 并发标记 + 标记终止 STW);CPU 时间 = 0.28 + 0/1.3/0.10 + 0.20 ms(STW + 辅助/后台/空闲标记 + STW);堆 3→4→1 MB(GC 开始时 → 标记结束时 → 存活),目标 4 MB,10 个 P
  460 次 GC 的两次 STW 耗时(ms):清扫终止最大 0.045、标记终止最大 0.120;并发标记墙钟时间中位 1.400

  mgc.go:8-12
    // GC threads to run in parallel. It is a concurrent mark and sweep that uses a write barrier. It is
    // non-generational and non-compacting. Allocation is done using size segregated per P allocation
    // areas to minimize fragmentation while eliminating locks in the common case.
    //
    // The algorithm decomposes into several steps.
  mgc.go:25-27
    //
    //    a. Stop the world. This causes all Ps to reach a GC safe-point.
    //
  mgc.go:36-46
    //    scanned until all Ps have enabled the write barrier, which is
    //    accomplished using STW.
    //
    //    b. Start the world. From this point, GC work is done by mark
    //    workers started by the scheduler and by assists performed as
    //    part of allocation. The write barrier shades both the
    //    overwritten pointer and the new pointer value for any pointer
    //    writes (see mbarrier.go for details). Newly allocated objects
    //    are immediately marked black.
    //
    //    c. GC performs root marking jobs. This includes scanning all
  mbarrier.go:24-34
    // Go uses a hybrid barrier that combines a Yuasa-style deletion
    // barrier—which shades the object whose reference is being
    // overwritten—with Dijkstra insertion barrier—which shades the object
    // whose reference is being written. The insertion part of the barrier
    // is necessary while the calling goroutine's stack is grey. In
    // pseudocode, the barrier is:
    //
    //     writePointer(slot, ptr):
    //         shade(*slot)
    //         if current stack is grey:
    //             shade(ptr)

  main.storePtr 的汇编(与写屏障有关的指令):
    (<gtmp>/gc2/main.go:9)	MOVD	16(g), R16
    (<gtmp>/gc2/main.go:9)	MOVD.W	R30, -16(RSP)
    (<gtmp>/gc2/main.go:9)	MOVD	R29, -8(RSP)
    (<gtmp>/gc2/main.go:9)	MOVWU	runtime.writeBarrier(SB), R2
    (<gtmp>/gc2/main.go:9)	MOVD	(R0), R2
    (<gtmp>/gc2/main.go:9)	CALL	runtime.gcWriteBarrier2(SB)
    (<gtmp>/gc2/main.go:9)	STP	(R1, R2), (R25)
    (<gtmp>/gc2/main.go:9)	MOVD	R1, (R0)
    (<gtmp>/gc2/main.go:9)	MOVD	-8(RSP), R29
  main.storeInt 的汇编(与写屏障有关的指令):
    (<gtmp>/gc2/main.go:12)	MOVD	R1, 8(R0)
  storePtr 里对 runtime.writeBarrier 标志的检查与 runtime.gcWriteBarrier 调用的个数:2

  GC 空闲(没有后台 GC)                     指针写 a.next = b 中位 2.37 ns/次;整数写 a.val[0] = i 中位 1.01 ns/次
  后台不停地 runtime.GC()                 指针写 a.next = b 中位 17.59 ns/次;整数写 a.val[0] = i 中位 1.11 ns/次;这段时间后台 runtime.GC() 完成了 143 轮

  go env GOEXPERIMENT = 
  go list 的 ToolTags 里与 GC 有关的:goexperiment.greenteagc
  默认(greenteagc 开):3 次运行的 GC 次数 [459, 457, 463];每次运行 GC 的 CPU 时间合计(ms)[1677.6, 1691.0, 1666.2];两次 STW 之和的最大值(ms)[0.191, 0.192, 0.188]
  GOEXPERIMENT=nogreenteagc:3 次运行的 GC 次数 [553, 554, 559];每次运行 GC 的 CPU 时间合计(ms)[2170.4, 2163.0, 2240.2];两次 STW 之和的最大值(ms)[0.191, 0.189, 0.185]