Re: [Xen-devel] [PATCH XTF] Functional: Add a UMIP test

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 21/07/17 02:42, Boqun Feng wrote:
> On Thu, Jul 20, 2017 at 10:38:59AM +0100, Andrew Cooper wrote:
> > On 20/07/17 06:29, Boqun Feng (Intel) wrote:
> > > Add a "umip" test for the User-Model Instruction Prevention. The test
> > > simply tries to run sgdt/sidt/sldt/str/smsw in guest user-mode with
> > > CR4_UMIP = 1.
> > >
> > > Signed-off-by: Boqun Feng (Intel) <boqun.feng@xxxxxxxxx>
> > Thankyou very much for providing a test.
> >
> > As a general remark, how have you found XTF to use?
> Great tool! Especially when you need to run Xen in a simulated
> environment like simics and want to test something, bringing up even a
> simple Linux domainU would be a lot of pain. ;-) While XTF just works
> like a charm and it's easy to write a test case, though according to
> your comments I'm now very good at it now ;-)
I'm glad to hear this.

> > > +void test_main(void)
> > > +{
> > > +    unsigned long exp;
> > > +    unsigned long cr4 = read_cr4();
> > This is all good.  However, it is insufficient to properly test the UMIP
> > behaviour.  Please look at the cpuid-faulting to see how I structured
> > things.
> >
> > In particular, you should:
> >
> > 1) Test the regular behaviour of the instructions.
> > 2) Search for UMIP, skipping if it isn't available.
> > 3) Enable UMIP.
> Maybe I also need to provide a write_cr4_safe() similar as wrmsr_safe(),
> in case that cpuid indicates UMIP supported while UMIP CR4 bit is not
> allowed to set, which means a bug?
Yes.  You are entirely correct.  Feel free to put write_cr4_safe() in 
lib.h along with the other *_safe() variants.
> > 4) Test the instructions again, this time checking for #GP in userspace.
> > 5) Disable UMIP.
> > 6) Check again for regular behaviour.
> >
> > This way, you also check that turning it off works as well as turning it on.
> >
> > In addition, each test needs to check more than just the block of tests
> > below.
> >
> > 1) The tests should run the instructions natively, and forced through the
> > instruction emulator.  See the FPU Exception Emulation test which is along
> > the same lines.  One thing to be aware of though is that in older versions
> > of Xen, the s??? instructions weren't implemented in the instruction
> > emulator, so the test should tolerate and skip if it gets #UD back.
> Rogar that.
:)

Roger.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel