[CRIU] [PATCH 0/2] uffd: do not handle VDSO pages in lazy daemon

[Mike Rapoport]

Hi Adrian,

On Thu, Mar 24, 2016 at 11:31:34AM +0100, Adrian Reber wrote:
> On Wed, Mar 23, 2016 at 03:48:41PM +0200, Mike Rapoport wrote:
> > The VDSO is just a few pages and there's no actual cost at injecting those
> > page into a process address space at the very begining.
> > On the other hand, taking care of VDSO on the userfaultfd side creates
> > undesired complexity which may be avoided.
> 
> Being clueless about VDSO I would like to understand if it is just not
> possible/doesn't make much sense to restore VDSO pages, or if it just
> adds undesired complexity which is not worth it as it really is a small
> number of pages.

vDSO is a syscall handler that kernel loads to a program address space
automatically, at least on some architectures.
On x86, arm64 and ppc64 vdso is 2 pages, I don't know about others.
And, having 2 pages pre-copied is not worth the pain of dealing with vdso in
lazy-pages daemon, IMHO.
 
> I am in no way against these patches, I would just like to understand it
> better. Thanks.
> 
> 		Adrian

--
Sincerely yours,
Mike.