[Differential] [Commented On] D1711: Changes to the callout code to restore active semantics and also add a test-framework and test to validate thecallout code (and potentially for use by other tests).

[rrs (Randall Stewart)]

rrs added a comment.

Hans:

Let me explain to you how I think you are wrong, you are missing a
small subtle thing here....

When we do the callout_stop we set
cc_migration_cpu() = CPUBLOCK

*NOT* 

c->c_cpu = CPUBLOCK;

You are confusing the two things. The CPUBLOCK is used
in two different places. 

The loop in callout_lock() is
looking for

c->c_cpu != CPUBLOCK

This only gets set when callout_cpu_switch() is called *not* when callout_reset or callout_stop is
called.

So I think you have the wrong fish in the net.

What Hiren is seeing is something completely different. What is
strange is the cc_cpu[] structures he dumped has *no* locks present.. Look
back above and see the cc_lock structures of the cc_cpu's.. all of the
owners are 4.. i.e. MTX_UNLOCKED.

So how in the world are we getting a "thread holds spin lock too long".

Worse the callout that was being "inserted" is ACTIVE and PENDING, which
means it should be on the wheel, but it is *not* on the wheel .. at least
not that I can see... since Hiren has traced back on that callout and
the previous one on the list (which points to it) has a c_flags of 0.. 

This is very very odd.

REVISION DETAIL
  https://reviews.freebsd.org/D1711

To: rrs, gnn, rwatson, lstewart, jhb, kostikbel, sbruno, imp, adrian, hselasky
Cc: julian, hiren, jhb, kostikbel, emaste, delphij, neel, erj, freebsd-net