Re: Fiasco.oc & trustzone

On Fri Sep 14, 2012 at 12:37:41 +0400, Nikolay Golikov wrote:
> 2012/9/14 Adam Lackorzynski <[email protected]>:
> >
> > On Thu Sep 13, 2012 at 13:09:52 +0400, Nikolay Golikov wrote:
> >> Yes, I found it. Is that tested on real hw? If I understand correctly,
> >> it shouldn't work. Because it use l4sigma0_map_iomem for mapping
> >> memory region to vmm's address space.
> >
> > It assumes a different approach for getting memory to the ns side.
> > Initially bootstrap is configured to only uses a part of the memory so
> > that the other part can be used for the ns side. From the point of view
> > of the vmm this is 'iomem' because it's not ram. However, getting ram
> > for the vmm shall also work via a dataspace.
> >
> 
> Now I get strange error:
> In bootstrap I define region for ns ram as Region::Arch.
> 
> SIGMA0: Dump of all resource maps
> RAM:------------------------
> [0:70000000;70000fff]
> [0:70061000;7008ffff]
> [0:70097000;70097fff]
> [0:7009f000;7013ffff]
> [4:70140000;70172fff]
> [0:70173000;70177fff]
> [4:70178000;7018efff]
> [0:7018f000;71010fff]
> [4:71011000;71011fff]
> [0:71012000;710fffff]
> [4:71100000;7127dfff]
> [0:7127e000;7effffff]
> IOMEM:----------------------
> [0:0;6fffffff]
> [0:80000000;ffffffff]
> 
> But on
> 
>   int err = l4sigma0_map_iomem(_sigma0.cap(), Ram_base, Ram_base, size, 1);
> (Ram_base = 0x80000000, size = 0x4000000)
> 
> I have an error:
> 
> TZ-VMM  | L4Re[rm]: unhandled read page fault @8858a820 pc=8858a820
> TZ-VMM  | L4Re: unhandled exception: pc=0x8858a820
> 
> Before it mapping for aisptz works normally:
> 
> l4sigma0_map_iomem(_sigma0.cap(), _base, _base, AISPTZ_SIZE, 1)
> (_base = 0x53F00000)

In your l4sigma0_map_iomem calls you are doing identity mappings, i.e.
you must make sure that the address space in your task is free from
0x80000000 - 0xbffffff and 0x53F00000 - 0x53F00000 + AISPTZ_SIZE - 1
The latter is likely to be free by luck, however, from 2gb to 3gb
there's for example the stack of your program, which you destroy by
doing the sigma0-map call. Consequently the program misbehaves, here
with a page-fault.
For getting access to device registers io should be used, for example via
l4io_request_iomem(). Mapping device registers cached is probably also
not wanted. In your case doing some special with l4sigma0_map_iomem, you
must reserve free space in your virtual address space via
l4re_rm_reserve_area / L4Re::Rm::reserve_area and map the memory to the
location returned by reserve_area.



Adam
-- 
Adam                 [email protected]
  Lackorzynski         http://os.inf.tu-dresden.de/~adam/