Re: [kvm-unit-tests PATCH] s390x: Fix uv_call() exception behavior

[Janosch Frank <frankja@linux.ibm.com>]

On 1/18/21 3:40 PM, Thomas Huth wrote:
> On 18/01/2021 15.03, Janosch Frank wrote:
>> On a program exception we usually skip the instruction that caused the
>> exception and continue. That won't work for UV calls since a "brc 3,0b"
>> will retry the instruction if the CC is > 1.
>>
>> Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
>> ---
>>
>> I know this isn't very pretty.
>> I'm open for suggestions.
>>
>> ---
>>   lib/s390x/asm/uv.h | 14 ++++++++------
>>   1 file changed, 8 insertions(+), 6 deletions(-)
>>
>> diff --git a/lib/s390x/asm/uv.h b/lib/s390x/asm/uv.h
>> index 4c2fc48..252f1a3 100644
>> --- a/lib/s390x/asm/uv.h
>> +++ b/lib/s390x/asm/uv.h
>> @@ -53,21 +53,23 @@ struct uv_cb_share {
>>   static inline int uv_call(unsigned long r1, unsigned long r2)
>>   {
>>   	int cc;
>> +	struct lowcore *lc = 0x0;
>>   
>>   	/*
>> -	 * The brc instruction will take care of the cc 2/3 case where
>> -	 * we need to continue the execution because we were
>> -	 * interrupted. The inline assembly will only return on
>> -	 * success/error i.e. cc 0/1.
>> -	*/
>> +	 * CC 2 and 3 tell us to re-execute because the instruction
>> +	 * hasn't yet finished.
>> +	 */
>> +	lc->pgm_int_code = 0;
>> +retry:
>>   	asm volatile(
>>   		"0:	.insn rrf,0xB9A40000,%[r1],%[r2],0,0\n"
>> -		"		brc	3,0b\n"
>>   		"		ipm	%[cc]\n"
>>   		"		srl	%[cc],28\n"
>>   		: [cc] "=d" (cc)
>>   		: [r1] "a" (r1), [r2] "a" (r2)
>>   		: "memory", "cc");
>> +	if (!lc->pgm_int_code && cc > 1)
>> +		goto retry;
> 
> Why not simply:
> 
> 	do {
> 		asm volatile(...);
> 	} while (!lc->pgm_int_code && cc > 1)
> 
> ?

That would also be an option but it would basically be the same horrible
looking quick fix.

Claudio proposed implementing a one shot uv_call that doesn't branch
back. We should be able to use that purely for privilege checks.

> 
>    Thomas
>