Date: Mon, 28 Jun 1999 13:38:07 -0700 (PDT) From: Matthew Dillon <dillon@apollo.backplane.com> To: Kirk McKusick <mckusick@flamingo.McKusick.COM> Cc: Peter Wemm <peter@netplex.com.au>, Julian Elischer <julian@whistle.com>, Alan Cox <alc@cs.rice.edu>, Mike Smith <mike@smith.net.au>, "John S. Dyson" <toor@dyson.iquest.net>, dg@root.com, dyson@iquest.net, current@freebsd.org, Greg Lehey <grog@lemis.com> Subject: Re: Found the startup panic - ccd ( patch included ) Message-ID: <199906282038.NAA24584@apollo.backplane.com> References: <199906281559.IAA02733@flamingo.McKusick.COM>
next in thread | previous in thread | raw e-mail | index | archive | help
:You are right that the buffer cache is very tricky. It is hard to
:change it in ways that preserves functionality and performance.
:The port of my buffer locking code from BSD/OS has caused problems
:largely because of a different buffer cache implementation and
:especially SMP models between the two systems (BSD/OS has no spl's
:for example). However, this is not a half baked idea that I have
:come up with and thrown in without thinking. Recursive locking is
:*required* to get stacking filesystems like union and loopback
:working reliably. As they currently stand they are full of code to
:try and track locks at layers all up and down the stack. This code
:is really nasty and in many cases broken (leading to deadlocks or
:crashes). The snapshot code also is most easily implemented with
:recursive locking, the alternative being to put lock monitoring
:all through bmap and block allocation code. So yes, it would be
:wonderful if the world could be simple, and I am not a huge fan of
:the complexity of the buffer locking code. But, I have considered
:the alternatives and they are worse.
:
: Kirk
Wow! Timeout! Miscommunication.
I am talking strictly about struct buf locking. I understand and agree
completely that recursive locking is absolutely necessary for the
VFS layering!
I am not throwing your work out - precisely the opposite, I want to keep
your work pretty much as is except for two things: (1) I want to replace
lockmgr locks ( and I don't mind implementing recursive exclusive locking
in the replacement, though I would rather avoid it! ), and (2) I want to
find a better way to implement the BUF_KERNPROC() stuff... basically I
want to try to find a way to localize the necessary effect in one place
in the code rather then have the BUF_KERNPROC() macros strewn all over the
code. Same effect, slightly different code organization, that's all.
There must be a choke point somewhere where we can put the code.
-Matt
Matthew Dillon
<dillon@backplane.com>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199906282038.NAA24584>