x86 SIG / Dropping PAE support

[Jeff Backus]

Hi List,

I wanted to reach out to you to let you know that we have established an x86 SIG in an effort to help address x86-related issues in Fedora. To help keep track of such issues, we have two x86-related tracker bugs:
* One for General x86-related issues: https://bugzilla.redhat.com/show_bug.cgi?id=1489998
* One for x86 ExcludeArch-related issues: https://bugzilla.redhat.com/show_bug.cgi?id=179258

Please feel free to also reach out to us via the x86 mailing list([e-mail]) or via the #fedora-x86 IRC channel.

Also, I wanted to let you know that we have decided to drop PAE support. I believe the kernel team has already disabled it for F27. They pointed out that this may affect Anaconda as well. And that the Lorax templates may need to be updated?

Thanks!

jeff

[Justin Forbes]

On Sun, Sep 17, 2017 at 11:30 AM, Jeff Backus  wrote:
> Hi List,
>
> I wanted to reach out to you to let you know that we have established an
> x86 SIG in an effort to help address x86-related issues in Fedora. To help
> keep track of such issues, we have two x86-related tracker bugs:
> One for General x86-related issues: https://bugzilla.redhat.com/sh
>
> ow_bug.cgi?id=1489998
> One for x86 ExcludeArch-related issues: https://bugzilla.redhat.com/sh
>
> ow_bug.cgi?id=179258
>
> Please feel free to also reach out to us via the x86 mailing list(
> [e-mail]) or via the #fedora-x86 IRC channel.
>
> Also, I wanted to let you know that we have decided to drop PAE support. I
> believe the kernel team has already disabled it for F27. They pointed out
> that this may affect Anaconda as well. And that the Lorax templates may
> need to be updated?

We are not dropping PAE for F27, it is a bit late for that. We have
disabled for rawhide/F28

Justin

[Jeff Backus]

On Sun, Sep 17, 2017 at 12:38 PM, Justin Forbes  wrote:
> On Sun, Sep 17, 2017 at 11:30 AM, Jeff Backus  wrote:
> > Hi List,
> >
> > I wanted to reach out to you to let you know that we have established an
> > x86 SIG in an effort to help address x86-related issues in Fedora. To help
> > keep track of such issues, we have two x86-related tracker bugs:
> > One for General x86-related issues:
> >
> > https://bugzilla.redhat.com/show_bug.cgi?id=1489998
> > One for x86 ExcludeArch-related issues:
> >
> > https://bugzilla.redhat.com/show_bug.cgi?id=179258
> >
> > Please feel free to also reach out to us via the x86 mailing
> > list([e-mail]) or via the #fedora-x86 IRC channel.
> >
> > Also, I wanted to let you know that we have decided to drop PAE support. I
> > believe the kernel team has already disabled it for F27. They pointed out
> > that this may affect Anaconda as well. And that the Lorax templates may need
> > to be updated?
>
> We are not dropping PAE for F27, it is a bit late for that. We have disabled
> for rawhide/F28

Doh! Right, sorry. F28. Makes more sense. Sorry! :)

jeff

[Bruno Wolff III]

Is there a change page for this somewhere?

Note if you are using PAE kernels you just stop getting updates. Maybe there 
should be some obsoletes to make the change over more transparent?

[Justin Forbes]

On Thu, Sep 21, 2017 at 1:32 AM, Bruno Wolff III  wrote:
> Is there a change page for this somewhere?
>
> Note if you are using PAE kernels you just stop getting updates. Maybe
> there should be some obsoletes to make the change over more transparent?

Yes, measures will have to be taken for that to happen, but given that it
isn't hitting most users until the F28 cycle, they have not been put into
place yet.  In  the meantime, manually installing a single non PAE kernel
will get everything back on track since you are following rawhide.

Justin

[Bruno Wolff III]

On Thu, Sep 21, 2017 at 06:01:48 -0500,
  Justin Forbes  wrote:
> Yes, measures will have to be taken for that to happen, but given that it
> isn't hitting most users until the F28 cycle, they have not been put into
> place yet.  In  the meantime, manually installing a single non PAE kernel
> will get everything back on track since you are following rawhide.

That's what I did. I was just confused for a bit when I saw kernels weren't 
updating on my i686 machine. I eventually noticed there weren't PAE rpms 
and went looking for a discussion about it to see if it was a short term or 
long term change.

[Florian Weimer]

On 09/17/2017 06:30 PM, Jeff Backus wrote:
> Also, I wanted to let you know that we have decided to drop PAE support.

Does this disable direct NX support, too?

Thanks,
Florian

[Jeff Backus]

On Mon, Sep 18, 2017 at 8:54 AM, Florian Weimer  wrote:
> On 09/17/2017 06:30 PM, Jeff Backus wrote:
> > Also, I wanted to let you know that we have decided to drop PAE support.
>
> Does this disable direct NX support, too?
>
> Thanks,
> Florian

Hmm.... Good question. Looks like it. I'm not very familiar with the
no-execute stuff, but yes, I do believe it requires PAE.

jeff

[Florian Weimer]

On 09/19/2017 12:41 AM, Jeff Backus wrote:
> On Mon, Sep 18, 2017 at 8:54 AM, Florian Weimer  wrote:
> > On 09/17/2017 06:30 PM, Jeff Backus wrote:
> > > Also, I wanted to let you know that we have decided to drop PAE support.
> >
> > Does this disable direct NX support, too?
> >
> > Thanks,
> > Florian
>
> Hmm.... Good question. Looks like it. I'm not very familiar with the
> no-execute stuff, but yes, I do believe it requires PAE.

Isn't this a rather significant feature regression?

Florian

[Stephen John Smoogen]

On 19 September 2017 at 01:08, Florian Weimer  wrote:
> On 09/19/2017 12:41 AM, Jeff Backus wrote:
> > On Mon, Sep 18, 2017 at 8:54 AM, Florian Weimer 
> > wrote:
> > > On 09/17/2017 06:30 PM, Jeff Backus wrote:
> > > > Also, I wanted to let you know that we have decided to drop PAE support.
> > >
> > > Does this disable direct NX support, too?
> > >
> > > Thanks,
> > > Florian
> >
> > Hmm.... Good question. Looks like it. I'm not very familiar with the
> > no-execute stuff, but yes, I do believe it requires PAE.
>
> Isn't this a rather significant feature regression?

Argh. This is an oversight on my part when bringing this as an option
from the kernel team. I was not aware that the NX was dependent on
PAE. The primary hardware that people were wanting all seemed to be
below 4 GB so having PAE didn't seem to be worth the overhead. I am
not sure if the OLPC desktops even have support for either NX or PAE.
The people with P-III or no SSE2 definitely don't have PAE.

[Florian Weimer]

On 09/19/2017 08:08 PM, Stephen John Smoogen wrote:
> Argh. This is an oversight on my part when bringing this as an option
> from the kernel team. I was not aware that the NX was dependent on
> PAE. The primary hardware that people were wanting all seemed to be
> below 4 GB so having PAE didn't seem to be worth the overhead. I am
> not sure if the OLPC desktops even have support for either NX or PAE.
> The people with P-III or no SSE2 definitely don't have PAE.

PAE is *way* older than NX/XD, but as you said, it does not matter for 
the old hardware because of RAM sizes and lack of NX/XD support.

There used to be an approximation to NX/XD based on segment register 
limits, but the art of writing userspace applications to make use of 
that may have been lost (and I don't know if the corresponding kernel 
code bits still exist and are being maintained).

Thanks,
Florian

[Stephen John Smoogen]

On 19 September 2017 at 15:14, Florian Weimer  wrote:
> On 09/19/2017 08:08 PM, Stephen John Smoogen wrote:
> > Argh. This is an oversight on my part when bringing this as an option
> > from the kernel team. I was not aware that the NX was dependent on
> > PAE. The primary hardware that people were wanting all seemed to be
> > below 4 GB so having PAE didn't seem to be worth the overhead. I am
> > not sure if the OLPC desktops even have support for either NX or PAE.
> > The people with P-III or no SSE2 definitely don't have PAE.
>
> PAE is *way* older than NX/XD, but as you said, it does not matter for the
> old hardware because of RAM sizes and lack of NX/XD support.
>
> There used to be an approximation to NX/XD based on segment register limits,
> but the art of writing userspace applications to make use of that may have
> been lost (and I don't know if the corresponding kernel code bits still
> exist and are being maintained).

I thought it got removed sometime in the 4 series as broken but no one
reporting it broken for a long time.. but that could be something
else.
> Thanks,
> Florian

[Jeff Backus]

On Tue, Sep 19, 2017 at 1:08 AM, Florian Weimer  wrote:
> On 09/19/2017 12:41 AM, Jeff Backus wrote:
> > On Mon, Sep 18, 2017 at 8:54 AM, Florian Weimer 
> > wrote:
> > > On 09/17/2017 06:30 PM, Jeff Backus wrote:
> > > > Also, I wanted to let you know that we have decided to drop PAE support.
> > >
> > > Does this disable direct NX support, too?
> > >
> > > Thanks,
> > > Florian
> >
> > Hmm.... Good question. Looks like it. I'm not very familiar with the
> > no-execute stuff, but yes, I do believe it requires PAE.
>
> Isn't this a rather significant feature regression?

Only from a security standpoint? *sheepish grin* Thanks for point this
out. We'll need to revisit.

jeff

[Jeff Backus]

On Sun, Sep 17, 2017 at 12:30:23PM -0400, Jeff Backus wrote:
> Hi List,
>
> I wanted to reach out to you to let you know that we have established an x86 SIG in an effort to help address x86-related issues in Fedora. To help keep track of such issues, we have two x86-related tracker bugs:
> One for General x86-related issues: https://bugzilla.redhat.com/show_bug.cgi?id=1489998
> One for x86 ExcludeArch-related issues: https://bugzilla.redhat.com/show_bug.cgi?id=179258
>
> Please feel free to also reach out to us via the x86 mailing list([e-mail]) or via the #fedora-x86 IRC channel.
>
> Also, I wanted to let you know that we have decided to drop PAE support. I believe the kernel team has already disabled it for F27. They pointed out that this may affect Anaconda as well. And that the Lorax templates may need to be updated?

Hi all,

Just following up on this.

Ananconda Team - how should we proceed. Should I file a bug to have PAE support dropped?

Thanks!
jeff