Re: [kvm-unit-tests PATCH 4/5] s390x: smp: Allocate memory in DMA31 space

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

On 1/14/22 14:16, Claudio Imbrenda wrote:
> On Fri, 14 Jan 2022 14:13:01 +0100
> Janosch Frank <frankja@linux.ibm.com> wrote:
> 
>> On 1/14/22 14:01, Claudio Imbrenda wrote:
>>> On Fri, 14 Jan 2022 13:50:52 +0100
>>> Nico Boehr <nrb@linux.ibm.com> wrote:
>>>    
>>>> On Fri, 2022-01-14 at 10:02 +0000, Janosch Frank wrote:
>>>>> The store status at address order works with 31 bit addresses so
>>>>> let's
>>>>> use them.
>>>>>
>>>>> Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
>>>>> ---
>>>>>    s390x/smp.c | 4 ++--
>>>>>    1 file changed, 2 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/s390x/smp.c b/s390x/smp.c
>>>>> index 32f128b3..c91f170b 100644
>>>>> --- a/s390x/smp.c
>>>>> +++ b/s390x/smp.c
>>>>
>>>> [...]
>>>>   
>>>>> @@ -244,7 +244,7 @@ static void test_func_initial(void)
>>>>>    
>>>>>    static void test_reset_initial(void)
>>>>>    {
>>>>> -       struct cpu_status *status = alloc_pages(0);
>>>>> +       struct cpu_status *status = alloc_pages_flags(1, AREA_DMA31);
>>>>
>>>> Why do we need two pages now?
>>>
>>> actually, wait.....
>>>
>>>           struct cpu_status *status = alloc_pages_flags(1, AREA_DMA31);
>>>           uint32_t r;
>>>
>>>           report_prefix_push("store status at address");
>>>           memset(status, 0, PAGE_SIZE * 2);
>>>
>>> we were allocating one page, and using 2!
>>>
>>> @Janosch do we need 1 or 2 pages?
>>>    
>>
>> Have a look at the memcmp() below those lines.
>>
>> I test if the status page has changed by doing a memcmp against the
>> second page.
> 
> so we do need 2 pages, and using 1 was a bug
> 

We need two for the store status at address tests but only one for the 
initial reset test Nico is pointing out here.