tree 12ac4c936f059be28088f7d3065317e6aca4019f
parent 0a90ecad071b3d1d5c6c789f6a6f1417b72862b2
author Michael Pratt <mpratt@google.com> 1588889601 -0400
committer Michael Pratt <mpratt@google.com> 1590094899 +0000

runtime: disable preemption in startTemplateThread

When a locked M wants to start a new M, it hands off to the template
thread to actually call clone and start the thread. The template thread
is lazily created the first time a thread is locked (or if cgo is in
use).

stoplockedm will release the P (_Pidle), then call handoffp to give the
P to another M. In the case of a pending STW, one of two things can
happen:

1. handoffp starts an M, which does acquirep followed by schedule, which
will finally enter _Pgcstop.

2. handoffp immediately enters _Pgcstop. This only occurs if the P has
no local work, GC work, and no spinning M is required.

If handoffp starts an M, and must create a new M to do so, then newm
will simply queue the M on newmHandoff for the template thread to do the
clone.

When a stop-the-world is required, stopTheWorldWithSema will start the
stop and then wait for all Ps to enter _Pgcstop. If the template thread
is not fully created because startTemplateThread gets stopped, then
another stoplockedm may queue an M that will never get created, and the
handoff P will never leave _Pidle. Thus stopTheWorldWithSema will wait
forever.

A sequence to trigger this hang when STW occurs can be visualized with
two threads:

  T1                                 T2
-------------------------------   -----------------------------

LockOSThread                      LockOSThread
  haveTemplateThread == 0
  startTemplateThread
    haveTemplateThread = 1
    newm                            haveTemplateThread == 1
      preempt -> schedule           g.m.lockedExt++
        gcstopm -> _Pgcstop         g.m.lockedg = ...
        park                        g.lockedm = ...
                                    return

                                 ... (any code)
                                   preempt -> schedule
                                     stoplockedm
                                       releasep -> _Pidle
                                       handoffp
                                         startm (first 3 handoffp cases)
                                          newm
                                            g.m.lockedExt != 0
                                            Add to newmHandoff, return
                                       park

Note that the P in T2 is stuck sitting in _Pidle. Since the template
thread isn't running, the new M will not be started complete the
transition to _Pgcstop.

To resolve this, we disable preemption around the assignment of
haveTemplateThread and the creation of the template thread in order to
guarantee that if handTemplateThread is set then the template thread
will eventually exist, in the presence of stops.

Fixes #38931

Change-Id: I50535fbbe2f328f47b18e24d9030136719274191
Reviewed-on: https://go-review.googlesource.com/c/go/+/232978
Run-TryBot: Michael Pratt <mpratt@google.com>
Reviewed-by: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>