Re: Boot regression in Linux 6.1 gone for Montecito in 6.3

Hi,

On 20.05.23 22:07, John Paul Adrian Glaubitz wrote:
> Hello!
>
> On Sat, 2023-05-20 at 21:51 +0200, Frank Scheiner wrote:
>> On 20.05.23 21:16, John Paul Adrian Glaubitz wrote:
>>> Hello Frank!
>>>
>>> On Sat, 2023-05-20 at 20:19 +0200, Frank Scheiner wrote:
>>>> I today noticed that a new Debian kernel is available from [1]. Thanks
>>>> to whoever built that one - most likely Adrian.
>>>
>>> The buildd built the kernel, not me ;-).
>>
>> Yeah, sure. :-) Did you also check out 6.2?
>
> No, I didn't since Debian's kernel maintainers skipped 6.2:
>
>> https://buildd.debian.org/status/logs.php?pkg=linux&arch=ia64

I see.

>>>> Earlier today I built a 6.4.0-rc2 kernel with kernel config based on
>>>> "config-6.1.0-9-mckinley" and localmodconfig matching the needed modules
>>>> for my rx2620 **and** rx2800-i2 combined.
>>>>
>>>> Unfortunately the resulting kernel "crashes" as soon as udevd starts and
>>>> that for both systems similarly. But yeah, the initramfs issues are gone
>>>> for the rx2620. If someone is interested, the error messages are on [3]
>>>> and [4] for rx2620 and rx2800-i2 respectively.
>>>>
>>>> [3]: https://pastebin.com/SAUKbG7Z
>>>>
>>>> [4]: https://pastebin.com/v1TTB2x3
>>>>
>>>> Well, rc2 might just be to early to conclude that there is a new error.
>>>> And it could also be an issue of my kernel config. I'll try to observe
>>>> how later kernels behave.
>>>
>>> Maybe you will have time to bisect it tomorrow? :-)
>>
>> Unfortunately not, I'm away tomorrow. Wouldn't it make more sense to
>> wait for the 6.4 release to be sure it's not due to something unrelated
>> to ia64 (1 run takes my rx2800-i2 40 mins and it only produces a handful
>> of modules in that time)?
>
> You should cross-compile your kernel. I can build a kernel for my rx2660
> on an AMD EPYC in 2-3 minutes.

Oh I don't have new things... :-/

My "newest" x86 MP system is a Sun X4270 w/2 x Xeon X5570s. I recently
upgraded its memory to 144 GiB so could compile the kernel(s) in a RAM
disk there, not sure how much of a boost that would be. IIRC I used it
for cross-compiling sparc64 kernels some time ago.

> And it's definitely better to catch the
> regression before the release as this way you will get the fix landed
> for 6.4 instead of 6.5.
Cheers,
Frank