Re: Fatal trap 12: page fault panic with recent kernel with ZFS

On May 18, 2009, at 11:12 PM, Kip Macy wrote:
>>  http://www.wanderview.com/svn/public/misc/zfs/zfs_kmem_limit.diff
>>
>> But I trigger it based on kmem thresholds.  See arc_reclaim_pages().
>>
>> I can try to put together a smaller patch tomorrow evening that  
>> signals the
>> pager based on size vs. c target.  The main reason I didn't  
>> implement it in
>> my previous patch was because I was concerned with the arc being  
>> prevented
>> from growing at all once its been shrunk.  It only grows when size  
>> exceeds
>> its current target by a certain amount.  This may require some  
>> careful
>> balancing or hysteresis or something.
>>
>
> I was actually referring to your comment about telling the ARC to tell
> ZFS to reclaim cached vnodes. Not that the kmem changes aren't good,
> but I don't think they're necessary on amd64.

I thought I was referring to that too.  :-)  I guess I thought  
signaling the pageout daemon to move vnodes from active to inactive  
was the way to do that.  Are you saying there is another caching layer  
of vnodes going on in zfs?  I'll look around for that this evening.

Thanks.

- Ben