On 8/20/19 2:55 PM, Thomas Huth wrote:
> On 8/20/19 2:25 PM, Janosch Frank wrote:
>> On 8/20/19 1:59 PM, Thomas Huth wrote:
>>> On 8/20/19 12:55 PM, Janosch Frank wrote:
[...]
>>> ... maybe we could also introduce such a variable as a global variable
>>> in lib/s390x/ since this is already the third or fourth time that we use
>>> it in the kvm-unit-tests...
>>
>> Sure I also thought about that, any particular place?
> 
> No clue. Maybe lib/s390x/mmu.c ? Or a new file called lowcore.c ?
> 
>>>> +static inline void diag288_uneven(void)
>>>> +{
>>>> +	register unsigned long fc asm("1") = 0;
>>>> +	register unsigned long time asm("1") = 15;
>>>
>>> So you're setting register 1 twice? And "time" is not really used in the
>>> inline assembly below? How's that supposed to work? Looks like a bug to
>>> me... if not, please explain with a comment in the code here.
>>
>> Well I'm waiting for a spec exception here, so it doesn't have to work.> I'll probably just remove the register variables and do a:
>>
>> "diag %r1,%r2,0x288"
> 
> Yes, I think that's easier to understand.
> 
> BTW, is there another documentation of diag 288 beside the "CP
> programming services" manual? At least my version of that specification
> does not say that the fc register has to be even...

I used the non-public lpar documentation...

> 
>>>> +static void test_bite(void)
>>>> +{
>>>> +	if (lc->restart_old_psw.addr) {
>>>> +		report("restart", true);
>>>> +		return;
>>>> +	}
>>>> +	lc->restart_new_psw.addr = (uint64_t)test_bite;
>>>> +	diag288(CODE_INIT, 15, ACTION_RESTART);
>>>> +	while(1) {};
>>>
>>> Should this maybe timeout after a minute or so?
>>
>> Well run_tests.sh does timeout externally.
>> Do you need it backed into the test?
> 
> I sometimes also run the tests without the wrapper script, so in that
> case it would be convenient ... but I can also quit QEMU manually in
> that case, so it's not a big issue.

How about setting the clock comparator, that should trigger an
unexpected external interrupt?

> 
>  Thomas
>