On 11.03.20 07:26, Andrew Jones wrote:
> On Tue, Mar 10, 2020 at 09:18:16PM +0100, Christian Borntraeger wrote:
>>
>>
>> On 10.03.20 18:27, Andrew Jones wrote:
>>> On Tue, Mar 10, 2020 at 05:54:59PM +0100, Christian Borntraeger wrote:
>>>> For s390 the guest memory size must be 1M aligned. I need something like the following to make this work:
>>>>
>>>> diff --git a/tools/testing/selftests/kvm/demand_paging_test.c b/tools/testing/selftests/kvm/demand_paging_test.c
>>>> index c1e326d3ed7f..f85ec3f01a35 100644
>>>> --- a/tools/testing/selftests/kvm/demand_paging_test.c
>>>> +++ b/tools/testing/selftests/kvm/demand_paging_test.c
>>>> @@ -164,6 +164,10 @@ static struct kvm_vm *create_vm(enum vm_guest_mode mode, int vcpus,
>>>> pages += ((2 * vcpus * vcpu_memory_bytes) >> PAGE_SHIFT_4K) /
>>>> PTES_PER_4K_PT;
>>>> pages = vm_adjust_num_guest_pages(mode, pages);
>>>> +#ifdef __s390x__
>>>> + /* s390 requires 1M aligned guest sizes */
>>>> + pages = (pages + 255) & ~0xff;
>>>> +#endif
>>>>
>>>> pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
>>>>
>>>>
>>>> any better idea how to do that?
>>>>
>>>
>>> For this one we could patch[*] vm_adjust_num_guest_pages(). That would
>>> also allow the one on line 382, and another one at dirty_log_test.c:300
>>> to be hidden.
>>
>> I tried that first but then I ran into several other asserts that checked for
>> num_pages = vm_adjust_num_guest_pages(num_pages)
>>
>> See kvm_util.c: TEST_ASSERT(vm_adjust_num_guest_pages(vm->mode, npages) == npages
>>
>> So it seems like a bigger rework is necessary to avoid this little hack :-/
>
> There's just this one other assert, and it'll only fire if the number of
> guest pages aren't selectect correctly. One must just be sure they
> always select the number correctly or do
>
> adjusted_num_pages = vm_adjust_num_guest_pages(mode, guessed_num_pages);
> vm_userspace_mem_region_add(..., adjusted_num_pages, ...);
>
> to ensure it. If we patch vm_adjust_num_guest_pages() as suggested below
> then the assert should never fire when the number is already correct,
> because vm_adjust_num_guest_pages() doesn't change an already correct
> number, i.e.
>
> adjusted_num_pages == vm_adjust_num_guest_pages(mode, adjusted_num_pages)
>
> If an assert is firing after making that change, then I wonder if not
> all s390 memregions are 1M aligned?
I just checked your patch and it seems to work fine. No idea what I did wrong
in my test. Can you respin this as a proper patch against kvm/queue?
