Re: [PATCH] Add hwcap2 bits for POWER9

Re: [PATCH] Add hwcap2 bits for POWER9

[Tulio Magno Quites Machado Filho]

Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:

> On 08-01-2016 13:36, Peter Bergner wrote:
>> On Fri, 2016-01-08 at 11:25 -0200, Tulio Magno Quites Machado Filho wrote:
>>> Peter, this solves the issue you reported previously [1].
>>>
>>> [1] https://sourceware.org/ml/libc-alpha/2015-12/msg00522.html
>> 
>> Agreed, thanks.  I'll also add the POWER9 support to the GCC side
>> of the patch now that the glibc code is upstream.
>
> I do not see these bits being added in kernel side yet and GLIBC usual
> only sync these kind of bits *after* they are included in kernel side.
> So I would advise to either get these pieces (kernel support and hwcap
> advertise) in kernel before 2.23 release, otherwise revert the patches.

Ack.
It has just been sent to the correspondent Linux mailing list:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2016-January/137763.html

-- 
Tulio Magno

Re: [PATCH] Add hwcap2 bits for POWER9

[Carlos O'Donell]

On 01/11/2016 10:16 AM, Tulio Magno Quites Machado Filho wrote:
> Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:
> 
>> On 08-01-2016 13:36, Peter Bergner wrote:
>>> On Fri, 2016-01-08 at 11:25 -0200, Tulio Magno Quites Machado Filho wrote:
>>>> Peter, this solves the issue you reported previously [1].
>>>>
>>>> [1] https://sourceware.org/ml/libc-alpha/2015-12/msg00522.html
>>>
>>> Agreed, thanks.  I'll also add the POWER9 support to the GCC side
>>> of the patch now that the glibc code is upstream.
>>
>> I do not see these bits being added in kernel side yet and GLIBC usual
>> only sync these kind of bits *after* they are included in kernel side.
>> So I would advise to either get these pieces (kernel support and hwcap
>> advertise) in kernel before 2.23 release, otherwise revert the patches.
> 
> Ack.
> It has just been sent to the correspondent Linux mailing list:
> https://lists.ozlabs.org/pipermail/linuxppc-dev/2016-January/137763.html

Please revert the changes from glibc until you checkin support to linux
kernel mainline.

Leaving these bits in increases the risk that someone uses to deploy a glibc
that then may have the wrong value.

Cheers,
Carlos.
 

Re: [PATCH] Add hwcap2 bits for POWER9

[Tulio Magno Quites Machado Filho]

"Carlos O'Donell" <carlos@redhat.com> writes:

> On 01/11/2016 10:16 AM, Tulio Magno Quites Machado Filho wrote:
>> Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:
>> 
>>> On 08-01-2016 13:36, Peter Bergner wrote:
>>>> On Fri, 2016-01-08 at 11:25 -0200, Tulio Magno Quites Machado Filho wrote:
>>>>> Peter, this solves the issue you reported previously [1].
>>>>>
>>>>> [1] https://sourceware.org/ml/libc-alpha/2015-12/msg00522.html
>>>>
>>>> Agreed, thanks.  I'll also add the POWER9 support to the GCC side
>>>> of the patch now that the glibc code is upstream.
>>>
>>> I do not see these bits being added in kernel side yet and GLIBC usual
>>> only sync these kind of bits *after* they are included in kernel side.
>>> So I would advise to either get these pieces (kernel support and hwcap
>>> advertise) in kernel before 2.23 release, otherwise revert the patches.
>> 
>> Ack.
>> It has just been sent to the correspondent Linux mailing list:
>> https://lists.ozlabs.org/pipermail/linuxppc-dev/2016-January/137763.html
>
> Please revert the changes from glibc until you checkin support to linux
> kernel mainline.
>
> Leaving these bits in increases the risk that someone uses to deploy a glibc
> that then may have the wrong value.

Could you clarify this statement, please?
I fail to see how they could have the wrong value.

These bits are enabled if you're running a kernel on a processor that
supports ISA 3.0 and/or VSX IEEE Float 128-bit.  The kernel must also support
this processor.

I see 3 possible scenarios:
 - Processor that doesn't support these features: as these bits are 0 by
   default, they have the correct value.
 - Both processor and kernel support these features: both bits are set,
   which is the correct value.
 - Processor support these features, but an old kernel doesn't support: these
   bits would be 0.  But, that's a valid scenario.

Am I missing another scenario?

However, I do agree with the concerns raised by Peter and Adhemerval: glibc
should be in sync with the kernel by the time of the release in order to
guarantee both bits are reserved for the exact same goal and we should have
both AT_HWCAP and AT_PLATFORM supporting the new processor.
With that said, I was planning to revert both commits d2de9ef7 and b1f19b8e
if we don't get the kernel patch accepted into the powerpc tree in time for
the release 2.23.

-- 
Tulio Magno

Re: [PATCH] Add hwcap2 bits for POWER9

[Carlos O'Donell]

On 01/11/2016 02:55 PM, Tulio Magno Quites Machado Filho wrote:
> "Carlos O'Donell" <carlos@redhat.com> writes:
> 
>> On 01/11/2016 10:16 AM, Tulio Magno Quites Machado Filho wrote:
>>> Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:
>>>
>>>> On 08-01-2016 13:36, Peter Bergner wrote:
>>>>> On Fri, 2016-01-08 at 11:25 -0200, Tulio Magno Quites Machado Filho wrote:
>>>>>> Peter, this solves the issue you reported previously [1].
>>>>>>
>>>>>> [1] https://sourceware.org/ml/libc-alpha/2015-12/msg00522.html
>>>>>
>>>>> Agreed, thanks.  I'll also add the POWER9 support to the GCC side
>>>>> of the patch now that the glibc code is upstream.
>>>>
>>>> I do not see these bits being added in kernel side yet and GLIBC usual
>>>> only sync these kind of bits *after* they are included in kernel side.
>>>> So I would advise to either get these pieces (kernel support and hwcap
>>>> advertise) in kernel before 2.23 release, otherwise revert the patches.
>>>
>>> Ack.
>>> It has just been sent to the correspondent Linux mailing list:
>>> https://lists.ozlabs.org/pipermail/linuxppc-dev/2016-January/137763.html
>>
>> Please revert the changes from glibc until you checkin support to linux
>> kernel mainline.
>>
>> Leaving these bits in increases the risk that someone uses to deploy a glibc
>> that then may have the wrong value.
> 
> Could you clarify this statement, please?
> I fail to see how they could have the wrong value.

Until it is checked into the mainline kernel it is not canonical.

That's the rule. There are no other discussions to be had.

The single rule avoids discussions like "it can never be wrong because that's
what our ABI says it is."
 
> However, I do agree with the concerns raised by Peter and Adhemerval: glibc
> should be in sync with the kernel by the time of the release in order to
> guarantee both bits are reserved for the exact same goal and we should have
> both AT_HWCAP and AT_PLATFORM supporting the new processor.
> With that said, I was planning to revert both commits d2de9ef7 and b1f19b8e
> if we don't get the kernel patch accepted into the powerpc tree in time for
> the release 2.23.

Exactly. That's perfect. We can backport them to 2.23.1 if you get in later.

Cheers,
Carlos.
 

Re: [PATCH] Add hwcap2 bits for POWER9

[Steven Munroe]

On Mon, 2016-01-11 at 15:48 -0500, Carlos O'Donell wrote:
> On 01/11/2016 02:55 PM, Tulio Magno Quites Machado Filho wrote:
> > "Carlos O'Donell" <carlos@redhat.com> writes:
> > 
> >> On 01/11/2016 10:16 AM, Tulio Magno Quites Machado Filho wrote:
> >>> Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:
> >>>
> >>>> On 08-01-2016 13:36, Peter Bergner wrote:
> >>>>> On Fri, 2016-01-08 at 11:25 -0200, Tulio Magno Quites Machado Filho wrote:
> >>>>>> Peter, this solves the issue you reported previously [1].
> >>>>>>
> >>>>>> [1] https://sourceware.org/ml/libc-alpha/2015-12/msg00522.html
> >>>>>
> >>>>> Agreed, thanks.  I'll also add the POWER9 support to the GCC side
> >>>>> of the patch now that the glibc code is upstream.
> >>>>
> >>>> I do not see these bits being added in kernel side yet and GLIBC usual
> >>>> only sync these kind of bits *after* they are included in kernel side.
> >>>> So I would advise to either get these pieces (kernel support and hwcap
> >>>> advertise) in kernel before 2.23 release, otherwise revert the patches.
> >>>
> >>> Ack.
> >>> It has just been sent to the correspondent Linux mailing list:
> >>> https://lists.ozlabs.org/pipermail/linuxppc-dev/2016-January/137763.html
> >>
> >> Please revert the changes from glibc until you checkin support to linux
> >> kernel mainline.
> >>
> >> Leaving these bits in increases the risk that someone uses to deploy a glibc
> >> that then may have the wrong value.
> > 
> > Could you clarify this statement, please?
> > I fail to see how they could have the wrong value.
> 
> Until it is checked into the mainline kernel it is not canonical.
> 
> That's the rule. There are no other discussions to be had.
> 
Well is was posted to to powerpc next:
https://git.kernel.org/powerpc/c/e708c24cd01ce80b1609d8bacc

We have agreement between the kernel and GLIBC (and the ABI). 

The issue is just coordination across communities and individuals that
may not being paying attention to other communities dead lines.

Have you ever tried to push a string, up hill. That is open source
development in nutshell. ;)

So it is in flight and glibc is soft/slush freeze. I would hate to
revert this one day just to add it back to the next. Especially if those
days straddle the hard freeze ...

So can we let this ride a day or too?

> The single rule avoids discussions like "it can never be wrong because that's
> what our ABI says it is."
> 

Re: [PATCH] Add hwcap2 bits for POWER9

[Carlos O'Donell]

On 01/12/2016 11:39 AM, Steven Munroe wrote:
>> That's the rule. There are no other discussions to be had.
>>
> Well is was posted to to powerpc next:
> https://git.kernel.org/powerpc/c/e708c24cd01ce80b1609d8bacc
> 
> We have agreement between the kernel and GLIBC (and the ABI). 
> 
> The issue is just coordination across communities and individuals that
> may not being paying attention to other communities dead lines.
> 
> Have you ever tried to push a string, up hill. That is open source
> development in nutshell. ;)

I know exactly what this is like.

> So it is in flight and glibc is soft/slush freeze. I would hate to
> revert this one day just to add it back to the next. Especially if those
> days straddle the hard freeze ...
> 
> So can we let this ride a day or too?

Sure. I'm not an unreasonable person.

My goal as a glibc steward is to remind IBM that our best practice is that
we *wait* until it goes into mainline before committing to glibc master.

There really isn't any reason to check this in to glibc master right now.
It could wait.

Adhemerval as a release manager is also not an unreasonable person.
I have already discussed with Tulio that he should have just waited to
commit these changes, but gotten an exception from Adhemerval to checkin
the fairly low-risk patches late in the freeze. That's exactly the purpose
of a release managers job, to grant you exceptions as we approach release,
particularly when schedules don't quite line up.

Cheers,
Carlos.

Re: [PATCH] Add hwcap2 bits for POWER9

[Michael Ellerman]

On 13 January 2016 03:39:17 GMT+11:00, Steven Munroe <munroesj@linux.vnet.ibm.com> wrote:
>On Mon, 2016-01-11 at 15:48 -0500, Carlos O'Donell wrote:
>> On 01/11/2016 02:55 PM, Tulio Magno Quites Machado Filho wrote:
>> > "Carlos O'Donell" <carlos@redhat.com> writes:
>> > 
>> >> On 01/11/2016 10:16 AM, Tulio Magno Quites Machado Filho wrote:
>> >>> Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:
>> >>>
>> >>>> On 08-01-2016 13:36, Peter Bergner wrote:
>> >>>>> On Fri, 2016-01-08 at 11:25 -0200, Tulio Magno Quites Machado
>Filho wrote:
>> >>>>>> Peter, this solves the issue you reported previously [1].
>> >>>>>>
>> >>>>>> [1] https://sourceware.org/ml/libc-alpha/2015-12/msg00522.html
>> >>>>>
>> >>>>> Agreed, thanks.  I'll also add the POWER9 support to the GCC
>side
>> >>>>> of the patch now that the glibc code is upstream.
>> >>>>
>> >>>> I do not see these bits being added in kernel side yet and GLIBC
>usual
>> >>>> only sync these kind of bits *after* they are included in kernel
>side.
>> >>>> So I would advise to either get these pieces (kernel support and
>hwcap
>> >>>> advertise) in kernel before 2.23 release, otherwise revert the
>patches.
>> >>>
>> >>> Ack.
>> >>> It has just been sent to the correspondent Linux mailing list:
>> >>>
>https://lists.ozlabs.org/pipermail/linuxppc-dev/2016-January/137763.html
>> >>
>> >> Please revert the changes from glibc until you checkin support to
>linux
>> >> kernel mainline.
>> >>
>> >> Leaving these bits in increases the risk that someone uses to
>deploy a glibc
>> >> that then may have the wrong value.
>> > 
>> > Could you clarify this statement, please?
>> > I fail to see how they could have the wrong value.
>> 
>> Until it is checked into the mainline kernel it is not canonical.
>> 
>> That's the rule. There are no other discussions to be had.
>> 
>Well is was posted to to powerpc next:
>https://git.kernel.org/powerpc/c/e708c24cd01ce80b1609d8bacc

And it is now in Linus' tree:

https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=e708c24cd01ce80b1609d8baccee40ccc3608a01

cheers

Re: [PATCH] Add hwcap2 bits for POWER9

[Carlos Eduardo Seo]

On 1/11/16 5:55 PM, Tulio Magno Quites Machado Filho wrote:
>
> However, I do agree with the concerns raised by Peter and Adhemerval: glibc
> should be in sync with the kernel by the time of the release in order to
> guarantee both bits are reserved for the exact same goal and we should have
> both AT_HWCAP and AT_PLATFORM supporting the new processor.
> With that said, I was planning to revert both commits d2de9ef7 and b1f19b8e
> if we don't get the kernel patch accepted into the powerpc tree in time for
> the release 2.23.
>

Kernel patch is in, commit f689b742f217b2ffe7925f8a6521b208ee995309

Regards,

-- 
Carlos Eduardo Seo
Software Engineer - Linux on Power Toolchain
cseo@linux.vnet.ibm.com