Re: [kvm-unit-tests PATCH] s390x: Add strict mode to specification exception interpretation test

[Janis Schoetterl-Glausch <scgl@linux.vnet.ibm.com>]

On 12/3/21 12:15, Thomas Huth wrote:
> On 25/11/2021 15.47, Janis Schoetterl-Glausch wrote:
>> While specification exception interpretation is not required to occur,
>> it can be useful for automatic regression testing to fail the test if it
>> does not occur.
>> Add a `--strict` argument to enable this.
> 
> Thank you very much for adding this!

Sure :)
> 
> Some comments below...
> 
>> `--strict` takes a list of machine types (as reported by STIDP)
>> for which to enable strict mode, for example
>> `--strict 8562,8561,3907,3906,2965,2964`
>> will enable it for models z15 - z13.
>> Alternatively, strict mode can be enabled for all but the listed machine
>> types by prefixing the list with a `!`, for example
>> `--strict !1090,1091,2064,2066,2084,2086,2094,2096,2097,2098,2817,2818,2827,2828`
>> will enable it for z/Architecture models except those older than z13.
>>
>> Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
>> ---
>>
>> Apparently my message with inline patch did not make it to the mailing
>> list for some reason, so here's the patch again.
>>
>>   s390x/spec_ex-sie.c | 59 ++++++++++++++++++++++++++++++++++++++++-----
>>   1 file changed, 53 insertions(+), 6 deletions(-)
>>
>> diff --git a/s390x/spec_ex-sie.c b/s390x/spec_ex-sie.c
>> index 5dea411..9a063f9 100644
>> --- a/s390x/spec_ex-sie.c
>> +++ b/s390x/spec_ex-sie.c
>> @@ -7,6 +7,7 @@
>>    * specification exception interpretation is off/on.
>>    */
>>   #include <libcflat.h>
>> +#include <stdlib.h>
>>   #include <sclp.h>
>>   #include <asm/page.h>
>>   #include <asm/arch_def.h>
>> @@ -36,7 +37,7 @@ static void reset_guest(void)
>>       vm.sblk->icptcode = 0;
>>   }
>>   -static void test_spec_ex_sie(void)
>> +static void test_spec_ex_sie(bool strict)
>>   {
>>       setup_guest();
>>   @@ -61,14 +62,60 @@ static void test_spec_ex_sie(void)
>>       report(vm.sblk->icptcode == ICPT_PROGI
>>              && vm.sblk->iprcc == PGM_INT_CODE_SPECIFICATION,
>>              "Received specification exception intercept");
>> -    if (vm.sblk->gpsw.addr == 0xdeadbeee)
>> -        report_info("Interpreted initial exception, intercepted invalid program new PSW exception");
>> -    else
>> -        report_info("Did not interpret initial exception");
>> +    {
>> +        const char *msg;
> 
> Could you please move the variable declaration to the beginning of the function? Then you could get rid of the curly brackets and one level of indentation.

Yes.
> 
>> +        msg = "Interpreted initial exception, intercepted invalid program new PSW exception";
>> +        if (strict)
>> +            report(vm.sblk->gpsw.addr == 0xdeadbeee, msg);
>> +        else if (vm.sblk->gpsw.addr == 0xdeadbeee)
>> +            report_info(msg);
>> +        else
>> +            report_info("Did not interpret initial exception");
>> +    }
>>       report_prefix_pop();
>>       report_prefix_pop();
>>   }
>>   +static bool parse_strict(char **argv)
>> +{
>> +    uint16_t machine_id;
>> +    char *list;
>> +    bool ret;
>> +
>> +    if (!*argv)
>> +        return false;
> 
> I think this works ok with out current implementation of argv, but that's an "inofficial" implementation detail, so in case this ever gets changed, it might be better to check argc first before dereferencing argv here ... so could you please add a check for argc, too?

This is required by POSIX, isn't it? But then whether or not we comply with it is another question.
> 
>> +    if (strcmp("--strict", *argv))
>> +        return false;
>> +
>> +    machine_id = get_machine_id();
>> +    list = argv[1];

This needs to be covered by the argc check too, then.

>> +    if (!list) {
>> +        printf("No argument to --strict, ignoring\n");
>> +        return false;
> 
> You could also support --strict without arguments - that could turn on the strict mode unconditionally, I think.

--strict ! works. Granted it's not the best UI, but it is more consistent,
you could do --strict $(get list of ids from somewhere) and if the list is empty the semantics stay the same. 
> 
>> +    }
>> +    if (list[0] == '!') {
>> +        ret = true;
>> +        list++;
>> +    } else
>> +        ret = false;
>> +    while (true) {
>> +        long input = 0;
>> +
>> +        if (strlen(list) == 0)
>> +            return ret;
>> +        input = strtol(list, &list, 16);
>> +        if (*list == ',')
>> +            list++;
>> +        else if (*list != '\0')
>> +            break;
>> +        if (input == machine_id)
>> +            return !ret;
>> +    }
>> +    printf("Invalid --strict argument \"%s\", ignoring\n", list);
>> +    return ret;
>> +}
>> +
>>   int main(int argc, char **argv)
>>   {
>>       if (!sclp_facilities.has_sief2) {
>> @@ -76,7 +123,7 @@ int main(int argc, char **argv)
>>           goto out;
>>       }
>>   -    test_spec_ex_sie();
>> +    test_spec_ex_sie(parse_strict(argv + 1));
>>   out:
>>       return report_summary();
>>   }
>>
> 
>  Thomas
>