jhibbits
7449055198
Add a memory barrier after taking a reference on the vnode holdcnt in _vhold ...
This is needed to avoid a race between the VNASSERT() below, and another
thread updating the VI_FREE flag, on weakly-ordered architectures.
On a 72-thread POWER9, without this barrier a 'make -j72 buildworld' would
panic on the assert regularly.
It may be possible to use a weaker barrier, and I'll investigate that once
all stability issues are worked out on POWER9.
2018-06-06 12:57:11 +00:00
2018-06-04 02:05:48 +00:00
2018-06-05 05:24:42 +00:00
2018-06-05 05:24:42 +00:00
2018-06-05 11:03:24 +00:00
2018-06-03 03:53:11 +00:00
2018-06-05 15:49:35 +00:00
2018-06-04 16:20:22 +00:00
2018-06-04 22:26:47 +00:00
2018-05-19 18:27:14 +00:00
2018-05-25 13:59:48 +00:00
2018-06-06 02:48:09 +00:00
2018-05-04 14:38:48 +00:00
2018-06-04 11:35:04 +00:00
2018-03-06 14:28:37 +00:00
2018-05-09 20:53:38 +00:00
2018-04-20 19:37:08 +00:00
2018-06-04 02:05:48 +00:00
2018-03-11 20:13:15 +00:00
2018-06-06 12:57:11 +00:00
2018-02-14 00:12:03 +00:00
2018-06-04 18:47:14 +00:00
2018-05-24 23:58:57 +00:00
2018-05-31 22:41:07 +00:00
2018-06-05 21:24:59 +00:00
2018-03-30 18:50:13 +00:00
2018-05-24 00:06:55 +00:00
2018-06-06 07:04:40 +00:00
2018-06-06 10:46:24 +00:00
2018-06-05 20:51:01 +00:00
2018-05-30 07:11:33 +00:00
2018-05-09 18:47:24 +00:00
2018-05-23 21:02:14 +00:00
2018-01-25 22:25:13 +00:00
2018-02-10 01:09:22 +00:00
2018-05-23 21:02:14 +00:00
2018-04-19 15:24:21 +00:00
2018-06-02 20:28:58 +00:00
2018-05-16 16:56:35 +00:00
2018-05-30 21:50:23 +00:00
2018-05-24 21:11:28 +00:00
2018-06-06 05:08:05 +00:00
2018-05-29 08:41:44 +00:00
2018-05-18 17:29:43 +00:00
2018-03-24 21:30:24 +00:00
2018-05-19 19:30:42 +00:00
2018-06-04 19:35:15 +00:00
2018-05-29 16:16:24 +00:00
2018-05-25 08:44:00 +00:00
7449055198