Date: Thu, 3 Oct 2019 23:28:07 +0000 From: Rick Macklem <rmacklem@uoguelph.ca> To: Peter Jeremy <peter@rulingia.com>, "freebsd-current@FreeBSD.org" <freebsd-current@FreeBSD.org> Subject: re: Reproducable deadlock in NFS client Message-ID: <YTOPR0101MB1163001233FBD6E7DFEE0790DD9F0@YTOPR0101MB1163.CANPRD01.PROD.OUTLOOK.COM>
next in thread | raw e-mail | index | archive | help
Hi Peter,=0A=
=0A=
You could try a couple of things:=0A=
1 - kib@ just put a patch up on phabricator that reorganizes the handling=
=0A=
of vnode_pager_setsize().=0A=
D21883=0A=
(If you could test this patch, that might be the best approach.)=0A=
or=0A=
2 - The only differences between the post r352392 code and the older stuff=
=0A=
is that it calls vnode_pager_setsize() when the size hasn't changed.=
=0A=
I can't think of why that might cause a problem, but??=0A=
I have a patch in phabricator D21814 that doesn't do the=0A=
vnode_pager_setsize() call when the size doesn't change.=0A=
--> If this patch were to avoid the hang, it could help diagnose the=0A=
problem.=0A=
The other difference is that it called vnode_pager_setsize() when ther=
e=0A=
was a small change, but not enough to affect a page boundary. I can't =
think=0A=
of how this would affect things either, but..=0A=
or=0A=
If you can't test either of the above patches, you could try reverting both=
=0A=
r352393 and r352457, which would put things back the way they've been=0A=
for years, to see if that works ok.=0A=
=0A=
Good luck with it, rick=0A=
ps: Btw, capturing "procstat -kk" and "ps axHl" would give you/us more info=
.=0A=
(The "H" on "ps" shows the iod threads.)=0A=
If you can drop into the debugger when it is hung as above, you could=
=0A=
capture the stuff listed here:=0A=
https://www.freebsd.org/doc/en_US.ISO8859-1/books/developers-handbook/kerne=
ldebug-deadlocks.html=0A=
=0A=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?YTOPR0101MB1163001233FBD6E7DFEE0790DD9F0>