[PATCH 1/9] x86: refine TPAUSE and UMWAIT

[Jan Beulich]

On 05.03.2020 15:37, H.J. Lu wrote:
> On Thu, Mar 5, 2020 at 6:08 AM Jan Beulich <jbeulich@suse.com> wrote:
>>
>> On 05.03.2020 15:04, H.J. Lu wrote:
>>> On Thu, Mar 5, 2020 at 12:08 AM Jan Beulich <jbeulich@suse.com> wrote:
>>>>
>>>> On 04.03.2020 12:44, H.J. Lu wrote:
>>>>> On Wed, Mar 4, 2020 at 3:40 AM Jan Beulich <jbeulich@suse.com> wrote:
>>>>>> On 04.03.2020 12:36, H.J. Lu wrote:
>>>>>>> On Wed, Mar 4, 2020 at 1:37 AM Jan Beulich <jbeulich@suse.com> wrote:
>>>>>>>> Allowing 64-bit registers is misleading here: Elsewhere these get allowed
>>>>>>>> when there's no difference between either variant, because of 32-bit
>>>>>>>> destination registers having their upper halves zeroed in 64-bit mode.
>>>>>>>> Here, however, they're source registers, and hence specifying 64-bit
>>>>>>>> registers would lead to the ambiguity of whether the upper 32 bits
>>>>>>>> actually matter.
>>>>>>>>
>>>>>>>> Additionally, for proper code generation in 16-bit mode, IgnoreSize is
>>>>>>>> needed on both.
>>>>>>>
>>>>>>> Are there testcases to show IgnoreSize is needed on them?
>>>>>>
>>>>>> The situation with 16-bit test cases is rather poor anyway. I didn't
>>>>>> consider it reasonable to add such very special ones when far more
>>>>>> general ones don't exist. But if your question is to mean you demand
>>>>>
>>>>> Let's start from somewhere.
>>>>>
>>>>>> such to be added, then I'll (somewhat hesitantly) add/extend some.
>>>>>> Please clarify.
>>>>>
>>>>> Please add testcases.
>>>>
>>>> Actually they were there, in patch 2. I've moved them to this patch
>>>> and have just sent v1.1 for just this one patch.
>>>
>>> Do we need to adjust disassembler for 16-bit mode?
>>
>> I haven't checked, but (a) this would be an independent change
>> and (b) it would likely affect more than just the insns here
>> (see e.g. other parts of the series).
> 
> Disassembler should be adjusted when IgnoreSize is added
> to tpause and umwait.

But IgnoreSize has nothing to do with the disassembler. Right
now all I'm after is getting the assembler to produce correct
code. There are test cases to verify this is the case. Whether
the disassembler deals with any of this correctly (including
also the insns getting IgnoreSize added by "x86: add missing
IgnoreSize", which you've already approved) is an entirely
separate topic, which I may be willing to look into down the
road, but not now and here.

More generally, when anyone fixes one bug, why should they get
penalized to have their bug fix accepted only when they take
the - perhaps significant amount of - time to also fix another,
unrelated bug? In the case here, it is definitely not my fault
if 16-bit handling is in a bad state. And I can't afford fixing
all issues there are in one go.

Jan