Re: [REGRESSION] Power9 pseries boot hang in v7.2-rc6, narrowed to PowerPC GENERIC_ENTRY conversion window

[Venkat Rao Bagalkote <venkat88@linux.ibm.com>]

Hi Shrikanth,


On 04/08/26 8:23 pm, Shrikanth Hegde wrote:
> Hi Venkat.
>
> On 8/4/26 2:04 PM, Venkat Rao Bagalkote wrote:
>> Hi,
>>
>> I'm seeing a boot regression on a Power9 pseries LPAR. The system 
>> hangs very early during boot after:
>>
>> Quiescing Open Firmware ...
>> Booting Linux via __start() ...
>>
>
> That is strange. not sure where to look at without any logs.
> Is it a specific config that fails or this specific LPAR that fails?
>
>> No further kernel output is produced.
>>
>> Known good:8cd9520d35a6 ("Linux 7.1")
>> Known bad: 075b74841bd0 ("Linux 7.2-rc6")
>>
>
> Is the kernel built correct at 075b74841bd0 without any errors?

The known bad revision 075b74841bd0 ("Linux 7.2-rc6") builds 
successfully with my configuration and the issue is observed at boot 
time only.

>
>> Initial bisect:
>>
>> A full bisect from v7.1 to v7.2-rc6 converged to:
>>
>> c5e8e224eb7a
>> MIPS: RB532: attach the software node to its target GPIO controller
>>
>>
>> However this commit only modifies:
>>
>> arch/mips/rb532/devices.c
>> arch/mips/rb532/gpio.c
>>
>> and appears unrelated to a Power9/pseries boot hang.
>>
>> I then narrowed the investigation to the PowerPC pull.
>>
>> The following commits were tested and boot successfully:
>>
>> baa590018fbf powerpc: rename arch_irq_disabled_regs
>> dec63ea6faff powerpc: Prepare to build with generic entry/exit framework
>> 09a9d3a8499d powerpc: introduce arch_enter_from_user_mode
>> 893082ac769b powerpc: Prepare for IRQ entry exit
>>
>> The following commit reproduces the hang:
>>
>> 334f3f6d7a16 powerpc/entry: Disable interrupts before irqentry_exit
>>
>>
>>
>> Between the last known good and first known bad points:
>>
>>
>> 893082ac769b powerpc: Prepare for IRQ entry exit
>> ...
>> bee25f97ad24 powerpc: Enable GENERIC_ENTRY feature
>> ...
>> 334f3f6d7a16 powerpc/entry: Disable interrupts before irqentry_exit
>>
>>
>> The only substantial architectural/runtime change in this window 
>> appears to be:
>>
>> bee25f97ad24 powerpc: Enable GENERIC_ENTRY feature
>>
>>
>>
>> I attempted to test:
>>
>> bee25f97ad24 powerpc: Enable GENERIC_ENTRY feature
>>
>> directly, however that commit does not build with my configuration 
>> due to:
>>
>> arch/powerpc/kernel/ptrace/ptrace.c
>>
>> error: implicit declaration of function 'trace_sys_enter'
>> error: implicit declaration of function 'trace_sys_exit'
>>
>> Therefore I could not directly classify bee25f97ad24 as good or bad.
>>
>>
>
> Can you try one commit after this and see if it builds?
> i.e  "powerpc: Remove unused functions"
>
I tested:

6ed60999d33d ("powerpc: Remove unused functions")

The kernel builds successfully with my configuration, but the system 
still hangs during boot on the Power9 pseries LPAR after:

   Quiescing Open Firmware ...
   Booting Linux via __start() ...

with no further output.

So currently I have:

   893082ac769b ("powerpc: Prepare for IRQ entry exit")  GOOD
   bee25f97ad24 ("powerpc: Enable GENERIC_ENTRY feature") UNTESTABLE
   6ed60999d33d ("powerpc: Remove unused functions")      BAD

Given that 6ed60999d33d builds and reproduces the hang, the regression 
appears to be present immediately after the introduction of GENERIC_ENTRY.


Regards,

Venkat.

>> Please let me know if there are specific debug patches or boot 
>> parameters you'd like me to test.
>>
>>
>> Attached is the config file I used.
>>
>>
>> Regards,
>>
>> Venkat.
>
>