Linux 6.7 Set To Drop Support For Itanium IA-64

[mercster]

I guess you can say is what has changed is that Linux is less of a hobby/toy project now, which means if they say they support something they really mean it and if they can't support something (i.e. they don't even have access to physical hardware to see if kernel compiles etc etc) then they will consider removing it.

Even GCC is slated to remove IA64 at some point, so the kernel doesn't really have a choice here unless they want to stick to an old GCC version.

Yeah, I understand that. Again, this is more an emotional response than anything... the world's in a different place. I remember back then, interested hackers would KILL to get some piece of rare/expensive hardware so they could port Linux. It didn't matter that it was economically justifiable. People would donate to a skilled hacker, so he could buy... oh I dunno, some weird expensive enterprise-level architecture, so that they could get to work porting Linux to it. Now, I guess hardware is much less heterogenous and all the diversity is in software.

Like I said, no judgement here. Just... sad to me. Crying over spilled milk that noone's missing, probably.

[mdedetrich]

Maybe it's just because I was using Linux pretty early, but it always feels weird when the kernel removes support for some architecture. Back in the mid-90s doing 'make config', it was always impressive to see all the architectures, devices, and protocols supported... maybe only a small number of people need that, but it was a badge of honor for the Linux kernel, that it was a sort of Swiss army knife that would suit many different purposes.

Nowadays the world is different, I understand that... and I do sympathize with the burden "dead" branches in the kernel might cause to developers and maintainers in other parts of the tree. Still... seems like there could be some kind of alternative to either keeping it in the main tree, and removing it. I dunno... maybe not. It's just an emotional reaction on my part, I reckon.

(My mind is drawn to AX.25 packet radio in the kernel, for instance. Is there a whole lot of people using it these days? Probably not... but someone out there might be. Support for all these exotic older standards is something I feel like should somehow be preserved. I don't even know if AX.25 is still in the kernel...)

(Note 2: To say "IA-64 users are all using HP-UX and OpenVMS anyway"... again seems strange to me. Back in the good ol' days, someone might have remarked "But why should their only choices be closed? Doesn't all hardware deserve a free and open alternative?" Again, I know I'm probably just being romantic.)

I guess you can say is what has changed is that Linux is less of a hobby/toy project now, which means if they say they support something they really mean it and if they can't support something (i.e. they don't even have access to physical hardware to see if kernel compiles etc etc) then they will consider removing it.

Even GCC is slated to remove IA64 at some point, so the kernel doesn't really have a choice here unless they want to stick to an old GCC version.

[jabl]

For what it's worth, I agree. I don't see the same push to remove PA or Alpha, and those are a hell of a lot older and slower than Itanium is.

PA and Alpha still have active maintainers, Itanium doesn't. Simple as that really. Maybe there's a better supply of leftover PA and Alpha HW that hobbyists can pick up on the cheap?

(Some years ago there was a bit of de-Alphaization in the core kernel going on. Alpha has the craziest least strict memory consistency model you can imagine, and many of the core kernel synchronization primitives were made with Alpha in mind as the lowest common denominator. Most of this has now been removed, making the core synchronization primitives a bit simpler, at the cost of Alpha being more tightly synchronized than it needs to.)

[mercster]

Yeah I feel ya. But now most Linux devs are corporate-powered so they can’t poopoo the cashman.

Yeah. And look, I'm not some GNU fanatic Stallman-type. It's an overall plus that big companies got behind Linux and it has corporate support and all that. It kinda has to, there's no way pure volunteers could keep it where it needs to be to be as powerful as it is in today's ecosystems. But it's different... before it was some idealistic "Let's support everything and give everyone freedom!" vibe, and now.... there are costs involved. It is just the way of things, I suppose.

That's why I said, I'm not expecting the kernel devs (to whom I freely admit I have nothing to offer) to carry this on their back but, there must be some kind of... I dunno. Just kinda let it limp along. I have a feeling they look at lots of automatic unit tests and get aggravated when something in that tree gives them a headache. I dunno. I'm guessing.

[Eirikr1848]

Maybe it's just because I was using Linux pretty early, but it always feels weird when the kernel removes support for some architecture. Back in the mid-90s doing 'make config', it was always impressive to see all the architectures, devices, and protocols supported... maybe only a small number of people need that, but it was a badge of honor for the Linux kernel, that it was a sort of Swiss army knife that would suit many different purposes.

Nowadays the world is different, I understand that... and I do sympathize with the burden "dead" branches in the kernel might cause to developers and maintainers in other parts of the tree. Still... seems like there could be some kind of alternative to either keeping it in the main tree, and removing it. I dunno... maybe not. It's just an emotional reaction on my part, I reckon.

(My mind is drawn to AX.25 packet radio in the kernel, for instance. Is there a whole lot of people using it these days? Probably not... but someone out there might be. Support for all these exotic older standards is something I feel like should somehow be preserved. I don't even know if AX.25 is still in the kernel...)

(Note 2: To say "IA-64 users are all using HP-UX and OpenVMS anyway"... again seems strange to me. Back in the good ol' days, someone might have remarked "But why should their only choices be closed? Doesn't all hardware deserve a free and open alternative?" Again, I know I'm probably just being romantic.)

Yeah I feel ya. But now most Linux devs are corporate-powered so they can’t poopoo the cashman.

IBM owns Red Hat, Red Hat does a lot of Nouveau work for example.

It’s the new corpo-open-source song and dance, my friend and ideological purists need to learn new skills to fill the void: or we need to find ways to fund legacy “cruft” (aka usable stuff for someone, somewhere, somewhen.)

——

Or move it all to Linux-Legacy and find a way to allow distros or users to enable Linux-legacy support, etc.

[Eirikr1848]

The last remaining Itanium user is shaking right now.

Naw. As the last non-corporate Itanium customer I was shaking last year when it was rumored to be removed. I bought a new HP machine in like 2021 thinking that I could slap Linux on it and use it for years to come. I… misunderstood that because I hadn’t heard of it didn’t mean it was a new cool thing… but instead was an old one on its way out.

I tried to find out how much it would cost to keep Itanium support in the kernel but no one here responded to me. Some Redditors said it would stick around for awhile.

I can’t learn how to do it myself, I couldn’t buy myself a few more years of support. So I guess. Whelp. Damnit.

[mercster]

For what it's worth, I agree. I don't see the same push to remove PA or Alpha, and those are a hell of a lot older and slower than Itanium is.

A lot of people just have a weird dislike of Itanium.

Heh... .yeah, I guess the only difference might be that Alpha and PA-RISC were around longer/had more reach. So people have more of it hanging around. There's probably the prejudice of "Haha, Itanium failed, so lame." Or maybe there's another good reason, I'm no Itanium expert.

[Dawn]

Maybe it's just because I was using Linux pretty early, but it always feels weird when the kernel removes support for some architecture. Back in the mid-90s doing 'make config', it was always impressive to see all the architectures, devices, and protocols supported... maybe only a small number of people need that, but it was a badge of honor for the Linux kernel, that it was a sort of Swiss army knife that would suit many different purposes.

Nowadays the world is different, I understand that... and I do sympathize with the burden "dead" branches in the kernel might cause to developers and maintainers in other parts of the tree. Still... seems like there could be some kind of alternative to either keeping it in the main tree, and removing it. I dunno... maybe not. It's just an emotional reaction on my part, I reckon.

(My mind is drawn to AX.25 packet radio in the kernel, for instance. Is there a whole lot of people using it these days? Probably not... but someone out there might be. Support for all these exotic older standards is something I feel like should somehow be preserved. I don't even know if AX.25 is still in the kernel...)

(Note 2: To say "IA-64 users are all using HP-UX and OpenVMS anyway"... again seems strange to me. Back in the good ol' days, someone might have remarked "But why should their only choices be closed? Doesn't all hardware deserve a free and open alternative?" Again, I know I'm probably just being romantic.)

For what it's worth, I agree. I don't see the same push to remove PA or Alpha, and those are a hell of a lot older and slower than Itanium is.

A lot of people just have a weird dislike of Itanium.

[mercster]

Maybe it's just because I was using Linux pretty early, but it always feels weird when the kernel removes support for some architecture. Back in the mid-90s doing 'make config', it was always impressive to see all the architectures, devices, and protocols supported... maybe only a small number of people need that, but it was a badge of honor for the Linux kernel, that it was a sort of Swiss army knife that would suit many different purposes.

Nowadays the world is different, I understand that... and I do sympathize with the burden "dead" branches in the kernel might cause to developers and maintainers in other parts of the tree. Still... seems like there could be some kind of alternative to either keeping it in the main tree, and removing it. I dunno... maybe not. It's just an emotional reaction on my part, I reckon.

(My mind is drawn to AX.25 packet radio in the kernel, for instance. Is there a whole lot of people using it these days? Probably not... but someone out there might be. Support for all these exotic older standards is something I feel like should somehow be preserved. I don't even know if AX.25 is still in the kernel...)

(Note 2: To say "IA-64 users are all using HP-UX and OpenVMS anyway"... again seems strange to me. Back in the good ol' days, someone might have remarked "But why should their only choices be closed? Doesn't all hardware deserve a free and open alternative?" Again, I know I'm probably just being romantic.)

[Dawn]

Windows Server wasn't an impediment to migration. Technically, neither was HP-UX as it's a Unix-like with much more in common with Unices at the time than the wholly proprietary OpenVMS. There were migration paths available to IBM AIX, SPARC/Solaris (Sun/Fujitsu and later Oracle's greed-driven mismanagement). Still, I grant the argument that people tend to stick where they are regardless until they have no choices but to move. For the low-mid tier x86-64 was "good enough", while the high end stuck with the devils they knew already. The HPC crowd would go with whoever topped the performance charts and won the contracts. The support libraries for science HPC work just expanded to accommodate which ever platforms won the lab contracts.

No real argument with most of this.

UX customers went to mission-critical x86, once the DL980 showed up to demonstrate that mission-critical x86 could be non-terrible. Big ones tended to go to AIX or z/Linux at higher rates, but a lot kept it in the house with Superdome X and its succesors. I would guess about 25% of the 2012 UX base are still on the platform, and a majority of those have ongoing migration projects.

That being said, I think you still overrate the degree to which HP cared about VMS. They were trying to get out of the VMS business long before they pulled the plug on their Itanium commitment as a whole. Itanium 9300 only supported VMS on entry level systems, and 9500 didn't support VMS at all (until HP sold VMS off and VSI added support several years later.) By 2012 Itanium was effectively UX only, with a small but profitable side gig in Nonstop.

It says a lot that UX was literally the only operating system that the Superdome2, the most profitable part of the late Itanium lineup, ever ran.