Re: Proposal for change to kernel linker for fixing a VNET and DPCPU problem.

From: John Baldwin <jhb_at_freebsd.org>
Date: Tue, 25 Sep 2012 11:07:02 -0400
On Friday, September 21, 2012 12:56:56 pm Julian Elischer wrote:
> On 9/21/12 2:22 AM, Mikolaj Golub wrote:
> > On Fri, Sep 07, 2012 at 01:28:16AM -0700, Anuranjan Shukla wrote:
> >> Hi George,
> >> Thanks for taking a look. Some answers/comments below.
> >>
> >>>> Building FreeBSD without the network stack (network stack as a module)
> >>>> ----------------------------------------------------------------------
> >>>>
> >>> This would be interesting for many reasons, and I think it would be a good
> >>> contribution.  Does the work you've done in this area handle the VNET
> >>> stuff that is in the stack as well?  That is, how well does the network
> >>> stack
> >>> as a module play with the vnet architecture?
> >> I'll follow up on this one separately.
> > FYI, there is at least this issue with virtualized global variables in modules:
> >
> > http://lists.freebsd.org/pipermail/freebsd-virtualization/2011-July/000737.html
> >
> > On archs that use link_elf.c (i.e. all except amd64, which uses
> > link_elf_obj.c) virtualized global variables in modules can not be
> > accessed from another modules, because link_elf on a module load does
> > relocation only for VNET variables defined in this module.
> >
> > As it was pointed by Marko Zec, the same issue is with DPCPU.
> >
> > The latest patch I have (both for VNET and DPCPU):
> >
> > http://people.freebsd.org/~trociny/link_elf.c.pcpu_vnet.patch
> >
> > The fix is to make the linker on a module load recognize "external"
> > VNET/DPCPU variables defined in the previously loaded modules and
> > relocate them accordingly. For this set_pcpu_list and set_vnet_list
> > are used, where the addresses of modules 'set_pcpu' and 'set_vnet'
> > linker sets are stored in.
> 
> it makes sense to me, but I really am not a linker person..
> I think it woul be good to get Doug Rabson  to weigh in on it, and
> maybe john Baldwin..
> 
> moving to -current as it's not a "net" issue really..

I think the proposed patch is ok.

-- 
John Baldwin
Received on Tue Sep 25 2012 - 13:17:13 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:30 UTC