[REGRESSION] Kernel booted via kexec fails to resume from hibernation

[msizanoen]

Submitted for review: 
https://lore.kernel.org/stable/20250405-fix-e820-nosave-v1-1-162633199548@qtmlabs.xyz/

On 4/4/25 12:50, msizanoen wrote:
> Here's an updated version of the patch that better handles 
> pathological e820 tables.
>
> On 4/4/25 11:56, msizanoen wrote:
>> Also, can you reproduce this issue with a target kernel (the kernel 
>> being kexec-ed) that has one of the patches attached (select the 
>> correct one according to your kernel version) applied, with either 
>> kexec_load or kexec_file_load?
>>
>> On 4/4/25 09:54, msizanoen wrote:
>>> Can you send the dmesg logs for this case (6.13 + mentioned patch 
>>> series backported as target kernel, using kexec_load)?
>>>
>>> On 4/4/25 05:00, Roberto Ricci wrote:
>>>> On 2025-04-01 19:59 +0700, msizanoen wrote:
>>>>> [snip]
>>>>> It seems like `e820__register_nosave_regions` is erroneously 
>>>>> marking some
>>>>> kernel memory as nosave in the presence of sub-page e820 regions. 
>>>>> In theory
>>>>> backporting
>>>>> https://lore.kernel.org/all/20250214090651.3331663-1-rppt@kernel.org/ 
>>>>> should
>>>>> be sufficient to avoid this but a fix for the actual root cause is
>>>>> preferred.
>>>> When using kexec_file_load, this patch series fixes the issue not only
>>>> in theory but also in practice.
>>>> But the issue with kexec_load (see
>>>> https://lore.kernel.org/all/Z-hYWc9LtBU1Yhtg@desktop0a/
>>>> ), which might be related, is not fixed.