Hifn driver in SMP (was Re: GELI - disk encryption GEOM class committed.)

Sam Leffler sam at errno.com
Tue Aug 9 04:19:28 GMT 2005


Mike Tancsa wrote:
> At 11:12 AM 08/08/2005, Poul-Henning Kamp wrote:
> 
> 
>> I belive there is a bug in the HiFn chips that makes them do a soft reset
>> under some set of circumstances which we have never been able to nail
>> down.
> 
> 
> Actually,
>         I think this is something different.  I know the issue you are 
> referring to, and it seems to happen on many (but not all) motherboards. 
> Note, this problem does not happen in UP mode on this box, only on SMP.  
> Also, I just booted RELENG_4_11 on the box and installed an SMP kernel.
> 
> hippo# hifnstats
> input 7648447680 bytes 2338053 packets
> output 7648431264 bytes 2338052 packets
> invalid 0 nomem 0 abort 0
> noirq 1263291 unaligned 0
> totbatch 0 maxbatch 0
> nomem: map 0 load 0 mbuf 0 mcl 0 cr 0 sd 0
> hippo#
> 
> I am able to run
> /usr/bin/openssl aes-128-cbc -in big -k pass | ssh -c aes128-cbc 
> mdtancsa at 127.0.0.1 "cat - >  /dev/null"
> until the cows come home without issue, even with an SMP kernel built. 
> So it seems like its something with this box and RELENG_6 that causes 
> the box to totally lock up

I much prefer cryptotest for exercising the hardware.  If you increase 
the number of concurrent threads (-t I think) you can really load the 
device.

I wouldn't be surprised if there were an smp locking bug in the crypto 
code as I'm not sure it's been well-exercised recently and with more of 
the kernel coming out from under Giant the locking within the subsystem 
is getting more testing.

	Sam


More information about the freebsd-current mailing list