5.1 PAE testing results: 2 panics

[Cagle, John (ISS-Houston)]

Interesting new datapoint on my PAE testing -- I built a kernel without
SMP and APIC (just the stock PAE config file) and now I _cannot_ crash
it.  Of course, it's only using one processor... :-(

Is it possible that the memory allocation algorithms aren't taking into
consideration that there may be multiple processors with SMP enabled?
Are there any large per-processor data structures that aren't being
accounted for when you have more than one processor?

Regards,
John

> -----Original Message-----
> From: Cagle, John (ISS-Houston) 
> Sent: Friday, May 23, 2003 10:03 AM
> To: Jake Burkholder
> Cc: current at freebsd.org
> Subject: RE: 5.1 PAE testing results: 2 panics
> 
> > -----Original Message-----
> > From: Jake Burkholder [mailto:jake at locore.ca]
> > >  panic: kmem_malloc(4096): kmem_map too small: 96624640 total
> > > allocated
> > > 
> > > I tried lowering kern.maxvnodes from 536130 (default) to
> > 200000, but
> > > that did not alleviate the panic.
> > 
> > Can you try reducing this by half again to 100,000 and see if
> > that helps things?  Increasing the amount of available kva 
> > with options KVA_PAGES may also work, but finding the right 
> > limits for vnodes is a better solution.  Note that the values 
> > for KVA_PAGES are doubled with PAE, eg 512 gives you the 
> > standard 3G user/1G kernel split.
> 
> I have gone as low as 10,000 for kern.maxvnodes, and it still 
> Panics with "kmem_map too small".  How low should I go?
> 
> I also built a kernel with KVA_PAGES set to 512 and it 
> paniced the same way (without changing kern.maxvnodes).  When 
> I also changed maxvnodes to 100,000, then the system hung 
> instead of panicing...
> 
> Anything else I can try?  Keep in mind I'm running SMP.  
> Should I try it with a UP kernel?
> 
> Thanks,
> John
> --------------------------------
> John Cagle     john.cagle at hp.com
> Principal Member Technical Staff
>    Industry Standard Servers
>     Hewlett-Packard Company