> I'm not a fan of ia64 as an architecture, but it's a bit sad to remove
> it entirely. It's not like it's been a huge maintenance burden in
> general.

Maybe you don't see others pain?  I added Al Viro ... perhaps
he'll replay some of his thoughts from trying to make signals
and other stuff work correctly on ia64.

-Tony

On Wed, Feb 15, 2023 at 11:43 AM Luck, Tony <tony.luck@intel.com> wrote:
>
> Maybe you don't see others pain?  I added Al Viro ... perhaps
> he'll replay some of his thoughts from trying to make signals
> and other stuff work correctly on ia64.

Well, as long as it's ia64-specific, I'll just go "hey, it was Al's
choice to look at that code".

IOW, I'm more worried about "ia64 makes it a pain to make _generic_ changes".

IOW, doing something like this:

    git log -p --no-merges --since=1.year arch/ia64/

to see what kind of pain ia64 parts of patches have caused, about a
third of them are that "look, somebody cared about ia64 explicitly".

And then the rest are trivial fixups for generic changes that aren't
any different from any other architecture. The only half-way
complicated one is the SET_FS removal, and I don't think it was any
worse than most other architectures.

IOW, it doesn't look like ia64 causes any huge issues _per_se_. I
suspect alpha continues to be more of a pain.

That said, it's entirely possible I've missed some particular painpoint.

But when it's actively known to be broken and nobody has time or
interest to look at it, at that point the "it doesn't look any more
painful than other architectures" becomes kind of moot.

                  Linus

Hi Linus,

On Wed, Feb 15, 2023 at 12:08:28PM -0800, Linus Torvalds wrote:
> On Wed, Feb 15, 2023 at 11:43 AM Luck, Tony <tony.luck@intel.com> wrote:
> >
> > Maybe you don't see others pain?  I added Al Viro ... perhaps
> > he'll replay some of his thoughts from trying to make signals
> > and other stuff work correctly on ia64.
> 
> Well, as long as it's ia64-specific, I'll just go "hey, it was Al's
> choice to look at that code".
> 
> IOW, I'm more worried about "ia64 makes it a pain to make _generic_ changes".
> 
> IOW, doing something like this:
> 
>     git log -p --no-merges --since=1.year arch/ia64/
> 
> to see what kind of pain ia64 parts of patches have caused, about a
> third of them are that "look, somebody cared about ia64 explicitly".
 
I remember that when I was doing cleanups of mm initialization, ia64
required special care several times.

> That said, it's entirely possible I've missed some particular painpoint.

The largest painpoint IMO is absence of any ability to test ia64 except
sending patches to Adrian in a hope he has time to give them a whirl.
 
>                   Linus

-- 
Sincerely yours,
Mike.

On Wed, 2023-02-15 at 12:08 -0800, Linus Torvalds wrote:
> But when it's actively known to be broken and nobody has time or
> interest to look at it, at that point the "it doesn't look any more
> painful than other architectures" becomes kind of moot.

Let me look after it in the weekend and let's see whether we can unbreak
it. I don't think there is really a big issue. The last time we had a
similar issue was the regression introduced by 974b9b2c68f3 which got fixed
with the simple fix in bd05220c7be3.

It's probably similarly trivial to fix the current regression.

Adrian

-- 
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

On 2023-02-15 18:13, John Paul Adrian Glaubitz wrote:
> On Wed, 2023-02-15 at 12:08 -0800, Linus Torvalds wrote:
>> But when it's actively known to be broken and nobody has time or
>> interest to look at it, at that point the "it doesn't look any more
>> painful than other architectures" becomes kind of moot.
> 
> Let me look after it in the weekend and let's see whether we can 
> unbreak
> it. I don't think there is really a big issue. The last time we had a
> similar issue was the regression introduced by 974b9b2c68f3 which got 
> fixed
> with the simple fix in bd05220c7be3.
> 
> It's probably similarly trivial to fix the current regression.
> 
> Adrian

Just for reference, this specific bug does not seem to be universal, but 
possibly only applies to a specific configuration.  I have observed no 
problems with 6.1 on my rx 2800 i2 and just booted 6.2 with no issues.  
Please feel free to try out my kernel config here:  
https://dpaste.com/43CACUUG8.txt

A possible guess is initramfs-related, as according to your logs on the 
Debian ML the hang happens shortly after initramfs unpacking, and I do 
not use an initramfs.