http://scale.eecs.berkeley.edu/mondriaan/.

IRC, freenode, #hurd, 2013-07-02

<xscript> in any case, what I wanted to check is if current hurd support
  PIE
<xscript`> I just saw samuel posted some fixes to have PIE working in hurd
<xscript`> are those included in the official image?
<youpi> sure
<youpi> it's just a trivial fixup in some address calculation code
<xscript> youpi: nice
<xscript> and does anyone know how complex would it be to implement some
  hackish support to have non-overlapping virtual addresses for
  applications supporting PIE?
<braunr> not too difficult
<xscript> really? I didn't expect such an answer XD
<xscript> I'd like to have something similar to a SASOS
<xscript> (single address space os)
<braunr> ?
<braunr> you mean an sasos on top of mach ?
<xscript> yes, but only for a few apps I want to evaluate
<braunr> i see
<xscript> the optimal would be to have all of hurd's servers on that mode
<braunr> you'l probably need to implement a small allocator but other than
  that it shouldn't be too hard, yes
<braunr> uh ??
<xscript> but running on 32 bits can be a problem here
<braunr> and not hurdish at all
<xscript> what's not hurdish?
<braunr> we do want address space separation
<xscript> well, you can have multiple address spaces (page tables), but
  without overlapping addresses between them
<xscript> that's exactly what I'm looking for
<braunr> sorry i don't see what you mean
<braunr> if you run several servers in the same address space, they can
  corrupt each other
<braunr> we don't want that
<braunr> it's that simple
<xscript> yes, sorry, I didn't explain myself
<xscript> I want a separate address space on each server
<xscript> but I want all memory allocations to be on addresses unique to
  the whole OS
<braunr> that still doesn't make sense
<xscript> well, it will still be secure
<xscript> but I know it does not make sense per se
<xscript> I want to do some experiments with a simulator
<braunr> why do you want them non overlapping if they're separate ?
<xscript> well, in my simulator I wouldn't need to change the page tables,
  protection is provided through other means
<braunr> segmentation ?
<xscript> that's one possibility
<xscript> (small address spaces)
<braunr> what do you have in mind ?
<braunr> it wouldn't be on top of mach anyway then
<braunr> mach implements paging
<xscript> what I'm simulating is something of the likes of Mondriaan
  (http://www.cs.utexas.edu/~witchel/pubs/mmp-asplos2002.pdf)
<xscript> paging is ok for me
<braunr> 19:06 < xscript> well, in my simulator I wouldn't need to change
  the page tables, protection is provided through other means
<braunr> it didn't sound so
<xscript> I meant switching page tables (cr3, etc)
<braunr> mach does that
<xscript> I know, I know, I can just ignore that part for the moment
<braunr> ok
<xscript> for now, I'd like to morph hurd into a SASOS using one page table
  per process
<xscript> I just wanted to know how hard that would be, without starting
  with a full dive into the code
<xscript> there are other options (OSes, microkernels), but none of them
  provides as many readily-available applications as hurd
<xscript> I suppose MINIX would also be easy to modify, but there's less
  apps there, and I also would like to tamper with MIG
<xscript> I just wonder how hard it would be to modify MIG