(Apologies - resending because I wasn't subscribed earlier)
Hi list,
I'm contacting you on behalf of the x86 SIG. Today FESCo approved our request to continue to support Fedora on x86 hardware, provided that we do our part to keep things running.
I encourage you to reach out to us when things do come up. You can find us at x86@lists.fedoraproject.org or on #fedora-x86. We will likewise try to be proactive in tracking down and triaging x86-related issues as well as helping test and debug things.
One caveat that FESCo attached to our request is that if you, the kernel team are cleared to treat i686 as any other secondary arch when you run into build issues. That is, you are allowed to ExcludeArch i686 until these issues are resolved. We ask that you block FE-ExcludeArch-x86 so that we can track these issues.
Additionally, FESCo would like us to establish a minimum level of hardware supported. We are working on this list and will follow up with you once we have it completed. In the interim, we did want to address a couple of concerns that were passed along by Stephen Smoogen:
* We have decided to drop support for PAE, so please feel free to disable it on the next build * We have decided to continue to support pre-SSE2 hardware for the time being
Please don't hesitate to contact us with any questions/comments/concerns.
Thanks!
jeff
Additionally, FESCo would like us to establish a minimum level of hardware supported.
We are
working on this list and will follow up with you once we have it completed. In
the interim, we did
want to address a couple of concerns that were passed along by
Stephen Smoogen:
- We have decided to drop support for PAE, so please feel
free to disable it on the next build
- We have decided to continue to support pre-SSE2
hardware for the time being
I'm very glad to hear this. Also makes me wonder if, since we're supporting non-SSE2 hardware, maybe we should retarget for i586, for the maximum compatibility?
x86_32 is, at this point, useful to people who want a modern OS on ancient hardware, and the more ancient of hardware we're able to support, the more use this could potentially have. This would allow supporting a number of VIA, AMD, and even old Intel CPUs.
Since we're going to deal with the quagmire of packages with sse2 as inline assembly already, and since many of these packages are the same ones that have problems with i586 in my experience (for example ffmpeg, or it used to, and mesa), why not aim for maximum compatibility? There is a performance hit for switching to i586 (which just drops CMOV/conditional move instructions), but the same is true for supporting SSE2, and I think at this point people are far more concerned with compatibility than a 2% or so performance penalty.
I might be wrong. Just putting it out there.
On 09/09/2017 07:46 AM, Daniel Hopson wrote:
Since we're going to deal with the quagmire of packages with sse2 as inline assembly already, and since many of these packages are the same ones that have problems with i586 in my experience (for example ffmpeg, or it used to, and mesa),
Just to mention it:
We also seem to be having problems with packages using -maes.
I am not 100% sure yet, but I recently have been hit by fc26 updates (nspr*), which appear to work on Atoms (w/ aes) but broke on original Intel i686 (w/o aes).
why not aim for maximum compatibility?
Well, for me, it would be sufficient if Fedora >= 26 would support the i686 like Fedora < 26 did. I am not much interested in i586s or VIA chipsets.
Ralf
On Sat, Sep 9, 2017 at 3:57 AM, Ralf Corsepius rc040203@freenet.de wrote:
Just to mention it:
We also seem to be having problems with packages using -maes.
I am not 100% sure yet, but I recently have been hit by fc26 updates (nspr*), which appear to work on Atoms (w/ aes) but broke on original Intel i686 (w/o aes).
That's a good point, thanks for bringing it up Ralf. We need to include it in our target hardware discussion.
why not aim for maximum compatibility?
Well, for me, it would be sufficient if Fedora >= 26 would support the i686 like Fedora < 26 did. I am not much interested in i586s or VIA chipsets.
Yes, I think that is a good goal at a minimum. If we can cover i586/VIA with a minimum effort, I'm interested in trying. I consider it a "stretch goal", so to speak.
jeff
On 9 September 2017 at 01:46, Daniel Hopson subsentient@universe2.us wrote:
Additionally, FESCo would like us to establish a minimum level of hardware supported. We are working on this list and will follow up with you once we have it completed. In the interim, we did want to address a couple of concerns that were passed along by Stephen Smoogen:
- We have decided to drop support for PAE, so please feel free to disable
it on the next build
- We have decided to continue to support pre-SSE2 hardware for the time
being
I'm very glad to hear this. Also makes me wonder if, since we're supporting non-SSE2 hardware, maybe we should retarget for i586, for the maximum compatibility?
x86_32 is, at this point, useful to people who want a modern OS on ancient hardware, and the more ancient of hardware we're able to support, the more use this could potentially have. This would allow supporting a number of VIA, AMD, and even old Intel CPUs.
The first question that has to be asked when doing any of this is how does this fit in with the Fedora mindset of pushing boundaries and moving forward. These architectures have not been supported by Fedora in over a decade and a lot of software inside of Fedora would probably not work in such a hardware setting.
The second question is how much work are you going to put into making i586 work? The people who got other architectures in place have gone into building the os on their own for a while to get most of the kinds out of the way and to see what was broken before proposing it to be included.
If you have good answers to both of them and can show that i586 can be built then I expect it would be possible. Otherwise it would be better to look at the distros which do have i586 support and concentrate there.
https://archives.gentoo.org/gentoo-project/message/fdcf90f02738ead4fd7722ea5...
Since we're going to deal with the quagmire of packages with sse2 as inline assembly already, and since many of these packages are the same ones that have problems with i586 in my experience (for example ffmpeg, or it used to, and mesa), why not aim for maximum compatibility? There is a performance hit for switching to i586 (which just drops CMOV/conditional move instructions), but the same is true for supporting SSE2, and I think at this point people are far more concerned with compatibility than a 2% or so performance penalty.
I might be wrong. Just putting it out there. _______________________________________________ X86 mailing list -- x86@lists.fedoraproject.org To unsubscribe send an email to x86-leave@lists.fedoraproject.org
On Sat, Sep 9, 2017 at 3:17 PM, Stephen John Smoogen smooge@gmail.com wrote:
On 9 September 2017 at 01:46, Daniel Hopson subsentient@universe2.us wrote:
x86_32 is, at this point, useful to people who want a modern OS on ancient hardware, and the more ancient of hardware we're able to support, the more use this could potentially have. This would allow supporting a number of VIA, AMD, and even old Intel CPUs.
The first question that has to be asked when doing any of this is how does this fit in with the Fedora mindset of pushing boundaries and moving forward. These architectures have not been supported by Fedora in over a decade and a lot of software inside of Fedora would probably not work in such a hardware setting.
The second question is how much work are you going to put into making i586 work? The people who got other architectures in place have gone into building the os on their own for a while to get most of the kinds out of the way and to see what was broken before proposing it to be included.
If you have good answers to both of them and can show that i586 can be built then I expect it would be possible. Otherwise it would be better to look at the distros which do have i586 support and concentrate there.
https://archives.gentoo.org/gentoo-project/message/fdcf90f02738ead4fd7722ea5...
Good point. I think we should take this in 'baby steps'. First, we stabilize i686 and then we can look at expanding, if there is sufficient interest. If we take on too much, we risk getting the boot.
The question as to 'how does this help move Fedora forward', is something we need to keep in mind. While I don't think there is anything about x86_32 that explicitly moves Fedora forward, I believe that supporting as much hardware as we can enables other innovations that will push Fedora forward. Particularly since these older platforms are still sufficient for developing most types of applications. Plus, you'll notice a memory leak on a machine with 512MB of RAM much quicker than one with 16GB. ;-)
The point you raise is important to keep in mind because if we hold the rest of the community back the community won't hesitate to move on.
jeff
On Sat, Sep 9, 2017 at 1:46 AM, Daniel Hopson subsentient@universe2.us wrote:
Additionally, FESCo would like us to establish a minimum level of hardware supported. We are working on this list and will follow up with you once we have it completed. In the interim, we did want to address a couple of concerns that were passed along by Stephen Smoogen:
- We have decided to drop support for PAE, so please feel free to disable
it on the next build
- We have decided to continue to support pre-SSE2 hardware for the time
being
I'm very glad to hear this. Also makes me wonder if, since we're supporting non-SSE2 hardware, maybe we should retarget for i586, for the maximum compatibility?
x86_32 is, at this point, useful to people who want a modern OS on ancient hardware, and the more ancient of hardware we're able to support, the more use this could potentially have. This would allow supporting a number of VIA, AMD, and even old Intel CPUs.
Since we're going to deal with the quagmire of packages with sse2 as inline assembly already, and since many of these packages are the same ones that have problems with i586 in my experience (for example ffmpeg, or it used to, and mesa), why not aim for maximum compatibility? There is a performance hit for switching to i586 (which just drops CMOV/conditional move instructions), but the same is true for supporting SSE2, and I think at this point people are far more concerned with compatibility than a 2% or so performance penalty.
Agreed. I think there is interest in doing so. It mostly depends on availability of hardware and interested warm bodies. We should discuss it in a "target hardware" thread, which I'll start once I get caught up. :)
On Fri, Sep 8, 2017 at 9:41 PM, Jeff Backus jeff.backus@gmail.com wrote:
(Apologies - resending because I wasn't subscribed earlier)
Hi list,
I'm contacting you on behalf of the x86 SIG. Today FESCo approved our request to continue to support Fedora on x86 hardware, provided that we do our part to keep things running.
I encourage you to reach out to us when things do come up. You can find us at x86@lists.fedoraproject.org or on #fedora-x86. We will likewise try to be proactive in tracking down and triaging x86-related issues as well as helping test and debug things.
One caveat that FESCo attached to our request is that if you, the kernel team are cleared to treat i686 as any other secondary arch when you run into build issues. That is, you are allowed to ExcludeArch i686 until these issues are resolved. We ask that you block FE-ExcludeArch-x86 so that we can track these issues.
Happy to do so. There wasn't much need on the current issue because there
are other problems keeping us from getting a successful build just yet. Yay, merge window. I really appreciate the SIG stepping forward and finding the solution for i686 though in a timely manner.
Additionally, FESCo would like us to establish a minimum level of hardware
supported. We are working on this list and will follow up with you once we have it completed. In the interim, we did want to address a couple of concerns that were passed along by Stephen Smoogen:
- We have decided to drop support for PAE, so please feel free to disable
it on the next build
- We have decided to continue to support pre-SSE2 hardware for the time
being
Done, rawhide going forward will drop PAE. Current stable Fedora releases will continue to build PAE as it wouldn't be wise to drop support on an existing release.
Thanks, Justin
On Mon, Sep 11, 2017 at 1:22 PM, Justin Forbes jforbes@redhat.com wrote:
On Fri, Sep 8, 2017 at 9:41 PM, Jeff Backus jeff.backus@gmail.com wrote:
(Apologies - resending because I wasn't subscribed earlier)
Hi list,
I'm contacting you on behalf of the x86 SIG. Today FESCo approved our request to continue to support Fedora on x86 hardware, provided that we do our part to keep things running.
I encourage you to reach out to us when things do come up. You can find us at x86@lists.fedoraproject.org or on #fedora-x86. We will likewise try to be proactive in tracking down and triaging x86-related issues as well as helping test and debug things.
One caveat that FESCo attached to our request is that if you, the kernel team are cleared to treat i686 as any other secondary arch when you run into build issues. That is, you are allowed to ExcludeArch i686 until these issues are resolved. We ask that you block FE-ExcludeArch-x86 so that we can track these issues.
Happy to do so. There wasn't much need on the current issue because there
are other problems keeping us from getting a successful build just yet. Yay, merge window. I really appreciate the SIG stepping forward and finding the solution for i686 though in a timely manner.
Additionally, FESCo would like us to establish a minimum level of hardware
supported. We are working on this list and will follow up with you once we have it completed. In the interim, we did want to address a couple of concerns that were passed along by Stephen Smoogen:
- We have decided to drop support for PAE, so please feel free to disable
it on the next build
- We have decided to continue to support pre-SSE2 hardware for the time
being
Done, rawhide going forward will drop PAE. Current stable Fedora releases will continue to build PAE as it wouldn't be wise to drop support on an existing release.
This might require changes in Anaconda. IIRC, it will look at the cpuflags during install and look to install the PAE kernel if that flag is present. Unless the standard i686 kernel does a fake Provides, it might cause install issues if it's missing. At least worth verifying.
josh
On Mon, Sep 11, 2017 at 6:43 PM, Josh Boyer jwboyer@fedoraproject.org wrote:
On Mon, Sep 11, 2017 at 1:22 PM, Justin Forbes jforbes@redhat.com wrote:
On Fri, Sep 8, 2017 at 9:41 PM, Jeff Backus jeff.backus@gmail.com wrote:
(Apologies - resending because I wasn't subscribed earlier)
Hi list,
I'm contacting you on behalf of the x86 SIG. Today FESCo approved our request to continue to support Fedora on x86 hardware, provided that we do our part to keep things running.
I encourage you to reach out to us when things do come up. You can find us at x86@lists.fedoraproject.org or on #fedora-x86. We will likewise try to be proactive in tracking down and triaging x86-related issues as well as helping test and debug things.
One caveat that FESCo attached to our request is that if you, the kernel team are cleared to treat i686 as any other secondary arch when you run into build issues. That is, you are allowed to ExcludeArch i686 until these issues are resolved. We ask that you block FE-ExcludeArch-x86 so that we can track these issues.
Happy to do so. There wasn't much need on the current issue because there
are other problems keeping us from getting a successful build just yet. Yay, merge window. I really appreciate the SIG stepping forward and finding the solution for i686 though in a timely manner.
Additionally, FESCo would like us to establish a minimum level of hardware
supported. We are working on this list and will follow up with you once we have it completed. In the interim, we did want to address a couple of concerns that were passed along by Stephen Smoogen:
- We have decided to drop support for PAE, so please feel free to disable
it on the next build
- We have decided to continue to support pre-SSE2 hardware for the time
being
Done, rawhide going forward will drop PAE. Current stable Fedora releases will continue to build PAE as it wouldn't be wise to drop support on an existing release.
This might require changes in Anaconda. IIRC, it will look at the cpuflags during install and look to install the PAE kernel if that flag is present. Unless the standard i686 kernel does a fake Provides, it might cause install issues if it's missing. At least worth verifying.
I'm pretty sure it will, it'll also need changes in the lorax templates.
Peter
On Mon, Sep 11, 2017 at 06:48:05PM +0100, Peter Robinson wrote:
On Mon, Sep 11, 2017 at 6:43 PM, Josh Boyer jwboyer@fedoraproject.org wrote:
On Mon, Sep 11, 2017 at 1:22 PM, Justin Forbes jforbes@redhat.com wrote:
Done, rawhide going forward will drop PAE. Current stable Fedora releases will continue to build PAE as it wouldn't be wise to drop support on an existing release.
This might require changes in Anaconda. IIRC, it will look at the cpuflags during install and look to install the PAE kernel if that flag is present. Unless the standard i686 kernel does a fake Provides, it might cause install issues if it's missing. At least worth verifying.
I'm pretty sure it will, it'll also need changes in the lorax templates.
Good point. Forgive my ignorance, but should I contact someone in addition to the Anaconda team to see that that is addressed?
Thanks!
jeff
On Mon, Sep 11, 2017 at 01:43:59PM -0400, Josh Boyer wrote:
On Mon, Sep 11, 2017 at 1:22 PM, Justin Forbes jforbes@redhat.com wrote:
Done, rawhide going forward will drop PAE. Current stable Fedora releases will continue to build PAE as it wouldn't be wise to drop support on an existing release.
This might require changes in Anaconda. IIRC, it will look at the cpuflags during install and look to install the PAE kernel if that flag is present. Unless the standard i686 kernel does a fake Provides, it might cause install issues if it's missing. At least worth verifying.
Good point. I'll reach out to the Anaconda team shortly. Sorry for taking so long to respond. Life got in the way. :)
jeff
On Mon, Sep 11, 2017 at 12:22:28PM -0500, Justin Forbes wrote:
On Fri, Sep 8, 2017 at 9:41 PM, Jeff Backus jeff.backus@gmail.com wrote:
(Apologies - resending because I wasn't subscribed earlier)
Hi list,
I'm contacting you on behalf of the x86 SIG. Today FESCo approved our request to continue to support Fedora on x86 hardware, provided that we do our part to keep things running.
I encourage you to reach out to us when things do come up. You can find us at x86@lists.fedoraproject.org or on #fedora-x86. We will likewise try to be proactive in tracking down and triaging x86-related issues as well as helping test and debug things.
One caveat that FESCo attached to our request is that if you, the kernel team are cleared to treat i686 as any other secondary arch when you run into build issues. That is, you are allowed to ExcludeArch i686 until these issues are resolved. We ask that you block FE-ExcludeArch-x86 so that we can track these issues.
Happy to do so. There wasn't much need on the current issue because there are other problems keeping us from getting a successful build just yet. Yay, merge window. I really appreciate the SIG stepping forward and finding the solution for i686 though in a timely manner.
Its our pleasure. All part of being part of a community! :)
Additionally, FESCo would like us to establish a minimum level of hardware supported. We are working on this list and will follow up with you once we have it completed. In the interim, we did want to address a couple of concerns that were passed along by Stephen Smoogen:
- We have decided to drop support for PAE, so please feel free to disable
it on the next build
- We have decided to continue to support pre-SSE2 hardware for the time
being
Done, rawhide going forward will drop PAE. Current stable Fedora releases will continue to build PAE as it wouldn't be wise to drop support on an existing release.
Yes, waiting until the next release makes sense. Thanks!
jeff
-
Daniel Hopson -
Jeff Backus -
Josh Boyer -
Justin Forbes -
Peter Robinson -
Ralf Corsepius -
Stephen John Smoogen