[Skiboot] [PATCH] fast-reboot: improve integrity of fast reboots

Nicholas Piggin npiggin at gmail.com
Fri Feb 23 12:49:24 AEDT 2018


On Fri, 23 Feb 2018 11:04:51 +1100
Stewart Smith <stewart at linux.vnet.ibm.com> wrote:

> Nicholas Piggin <npiggin at gmail.com> writes:
> > This is a start on some basic checks that might detect firmware
> > problems that call for a full IPL. The patch is pretty rough at
> > the moment but it's just for comments.
> >
> > - Checksum text and most read-only data, and verify that against
> >   checksum taken at IPL.
> >
> > - Check integrity of memory allocation structures and freelists.  
> 
> Good idea.
> 
> > - Zeroes most OS memory (should be parallelised).  
> 
> Been meaning to do this for a while. I wonder how much time it really
> takes on big systems... Could be worth doing in parallel I guess.

Just tested in userpsace a single CPU can memset about 32GB/s, and
multiple on the same node can get 4x that. Multiple nodes should
scale pretty linearly.

A fully populated AC922 would be 64 seconds vs 8 seconds, so it will
get noticable.

> > There's probably lots more software (and harware) checks we should
> > do here. But this is a first hack at it. What do people think, is
> > this a good idea?  
> 
> Yeah, I think it is.

Okay cool, I'll keep working on it.

> For the checksum, we could use the sha512 code we have laying around
> rather than using another one?

Yep that would be better.

Thanks,
Nick


More information about the Skiboot mailing list