WITNESS bug
Julian Elischer
julian at elischer.org
Tue Oct 19 16:16:19 PDT 2004
John Baldwin wrote:
>On Tuesday 19 October 2004 05:25 pm, Julian Elischer wrote:
>
>
>>John Baldwin wrote:
>>
>>
>>>On Tuesday 19 October 2004 12:01 pm, Kris Kennaway wrote:
>>>
>>>
>>>>On Tue, Oct 19, 2004 at 09:13:26AM -0400, Robert Huff wrote:
>>>>
>>>>
>>>>>Brian Fundakowski Feldman writes:
>>>>>
>>>>>
>>>>>>You should never not run with WITNESS_SKIPSPIN if you use
>>>>>>modules. Any spin mutexes not listed statically in the witness
>>>>>>code will cause your machine to immediately panic.
>>>>>>
>>>>>>
>>>>> If this is true (and I'm not disputing it), shouldn't it be
>>>>>noted in GENERIC and/or NOTES? For that matter, what's the penalty
>>>>>for not automatically including it as part of WITNESS?
>>>>>
>>>>>
>>>>Sometimes you don't want to use it, e.g. if you actually want to trace
>>>>spinlock operations with witness.
>>>>
>>>>
>>>True spin mutexes should be rarely used anyways, so I don't think modules
>>>needing spin mutexes is all that big of an issue. Almost all mutexes
>>>should just be regular mutexes.
>>>
>>>
>>netgraph uses a spin mutex for it's node locks
>>
>>
>
>This is likely a bug, esp. given that normal mutexes adaptively spin when it
>is advantageous to do so. :)
>
now that we have read-write locks it may be worth re looking at teh
netgraph version
of same to see if they can be used instead, but I doubt that the generic
ones would be as
lightweight.
>
>
>
More information about the freebsd-current
mailing list