Re: Resume Issues :Exec of NX page, Synaptics Botchup

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Monday, May 23, 2011, Yu, Fenghua wrote:
> > On Monday, May 23, 2011, Parag Warudkar wrote:
> > >
> > > On Sun, 22 May 2011, Linus Torvalds wrote:
> > >
> > > >
> > > > Hmm. The "Code: " line is just full of complete garbage, so I think
> > > > the real issue is that you really are trying to execute data.
> > > >
> > > > And that in turn seems to be because "setup_disablecpuid()" has
> > > > actually been free'd, because it is marked as __init.
> > > >
> > > > Which is fine at the initial bootup, but not so fine at resume
> > time,
> > > > since it was free'd long long ago by then.
> > > >
> > > > And it definitely shouldn' t be called at resume time. There's
> > > > something wrong there. That call trace is odd:
> > > >
> > > >   Call Trace:
> > > >    [<ffffffff8148a119>] ? identify_cpu+0xd8/0x2d8
> > > >    [<ffffffff8148a32d>] identify_secondary_cpu+0x14/0x1b
> > > >    [<ffffffff8148bf0f>] smp_store_cpu_info+0x3c/0x3e
> > > >    [<ffffffff8148c2ef>] start_secondary+0xf7/0x1d2
> > > >
> > > > because none of those should be calling "setup_disablecpuid()" at
> > all.
> > > >
> > > > Hmm. In fact, RIP is "setup_disablecpuid+0x40/0x40", ie it is one
> > past
> > > > the _end_ of setup_disablecpuid.
> > > >
> > > > I suspect that is actually "setup_smep()" that got called, an dthat
> > > > there was some garbage data in there that caused it to jump back a
> > > > bit.
> > > >
> > > > Does the attached patch fix it?
> > > >
> > >
> > > Now I get a different call trace for the same NX error. (And
> > Synaptics is
> > > completely dead this time - not attributable to the patch, it was
> > dumb
> > > luck perhaps that it wored last time.)
> > 
> > Any chance to try with commit de5397ad5b9ad22e2401c4dacdf1bb3b19c05679
> > (x86, cpu: Enable/disable Supervisor Mode Execution Protection)
> > reverted?
> > 
> > Rafael
> 
> de5397ad5b9ad22e2401c4dacdf1bb3b19c05679 shouldn't affect a system unless the CPU has SMEP feature which is unlikely available out there.
> 
> You can check if your CPU has "smep" feature in /proc/cpuinfo. Unlikely you will see the feature on your machine. Then the commit shouldn't cause the problem.
> 
> You can unconditionally disable SMEP by kernel option "nosmep".

Still, it is known to cause problems to people (see for example
https://bugzilla.kernel.org/show_bug.cgi?id=35602), so it might be a
good idea to set "nosmep" by default.

Thanks,
Rafael