Is the L4Linux running as a Fiasco.OC server?

[Adam Lackorzynski]

Hi,

On Thu Mar 08, 2018 at 11:00:03 +0800, Zeyu Mi wrote:
> On Thu, Mar 8, 2018 at 8:14 AM, Adam Lackorzynski <adam at os.inf.tu-dresden.de
> > wrote:
> 
> >
> > On Wed Mar 07, 2018 at 15:50:24 +0800, Zeyu Mi wrote:
> > > Sorry for bothering you again...
> > >
> > > I have changed to snapshot-17.09 and the "s" command works normally.
> > >
> > > I also disabled the VCPU thread model in the L4Linux config and it can be
> > > compiled
> > > without any error :)
> > >
> > > However, the L4Linux keeps reporting the following error message if
> > running
> > > in QEMU.
> > >
> > > Non-resolvable page fault at b300b308, ip 6454e6.
> > > Page fault (non-resolved): pfa=b300b308 pc=6454e6
> > > Non-resolvable page fault at b300b308, ip 6454e6.
> > > Die message: Trap: 14
> > > Non-resolvable page fault at b300b308, ip 205390.
> > > Page fault (non-resolved): pfa=b300b308 pc=205390
> > > Non-resolvable page fault at b300b308, ip 205390.
> > > Die message: Trap: 14
> > >
> > > If running on the real hardware (a Skylake machine), it keeps reporting
> > the
> > > following message.
> > >
> > > L4x: Main thread running, waiting...
> > > Die message: Trap: 6
> > > Die message: Trap: 6
> > > Die message: Trap: 6
> > > Die message: Trap: 6
> > > Die message: Trap: 6
> > > Die message: Trap: 6
> >
> > That's strange. What are you running inside L4Linux?
> > What's also interesting is that on QEMU there are page-faults and on
> > hardware trap6.
> > For my setups, all looks fine.
> 
> Hi Adam,
> 
> I switched to snapshot-17.12, and turned off the VCPU thread model.
> While this version works normally on QEMU (with -cpu host option, Skylake
> machine), it triggers
> a general protection (trap 13) on the Skylake machine. The instruction
> triggering such fault is "XRSTORS64".
> It is quite strange to see such phenomenon since the QEMU uses the Skylake
> CPU to run the system, which is
> the same as the real hardware.
> 
> Therefore, I dived into the source code and found that the system chose
> different paths in "copy_kernel_to_xregs_booting"
> funtion. This function has the following code snippet.
> 
> if (state_cpu_has(X86_FEATURE_XSAVES))
>     XSTATE_OP(XRSTORS, xstate, lmask, hmask, err);
> else
>     XSTATE_OP(XRSTOR, xstate, lmask, hmask, err);
> 
> If running in QEMU, the system chose the else branch. But it chose the if
> branch if running in hardware.
> 
> Do you have any suggestion on fixing that?

Ok, so looks like an issue with recent FPU stuff, I need to get a
Skylake as a testbox.
Please try adding "noxsaves" on the L4Linux command line.



Adam