Re: [PATCH 5/5] s390x: Enable and document boot device fallback on panic

[Thomas Huth <thuth@redhat.com>]

On 17/06/2024 01.44, Jared Rossi wrote:
> 
> 
> On 6/7/24 1:57 AM, Thomas Huth wrote:
>> On 05/06/2024 16.48, Jared Rossi wrote:
>>>
>>>>> diff --git a/pc-bios/s390-ccw/s390-ccw.h b/pc-bios/s390-ccw/s390-ccw.h
>>>>> index c977a52b50..de3d1f0d5a 100644
>>>>> --- a/pc-bios/s390-ccw/s390-ccw.h
>>>>> +++ b/pc-bios/s390-ccw/s390-ccw.h
>>>>> @@ -43,6 +43,7 @@ typedef unsigned long long u64;
>>>>>   #include "iplb.h"
>>>>>     /* start.s */
>>>>> +extern char _start[];
>>>>>   void disabled_wait(void) __attribute__ ((__noreturn__));
>>>>>   void consume_sclp_int(void);
>>>>>   void consume_io_int(void);
>>>>> @@ -88,6 +89,11 @@ __attribute__ ((__noreturn__))
>>>>>   static inline void panic(const char *string)
>>>>>   {
>>>>>       sclp_print(string);
>>>>> +    if (load_next_iplb()) {
>>>>> +        sclp_print("\nTrying next boot device...");
>>>>> +        jump_to_IPL_code((long)_start);
>>>>> +    }
>>>>> +
>>>>>       disabled_wait();
>>>>>   }
>>>>
>>>> Honestly, I am unsure whether this is a really cool idea or a very ugly 
>>>> hack ... but I think I tend towards the latter, sorry. Jumping back to 
>>>> the startup code might cause various problem, e.g. pre-initialized 
>>>> variables don't get their values reset, causing different behavior when 
>>>> the s390-ccw bios runs a function a second time this way. Thus this 
>>>> sounds very fragile. Could we please try to get things cleaned up 
>>>> correctly, so that functions return with error codes instead of 
>>>> panicking when we can continue with another boot device? Even if its 
>>>> more work right now, I think this will be much more maintainable in the 
>>>> future.
>>>>
>>>>  Thomas
>>>>
>>>
>>> Thanks Thomas, I appreciate your insight.  Your hesitation is perfectly 
>>> understandable as well.  My initial design was like you suggest, where 
>>> the functions return instead of panic, but the issue I ran into is that 
>>> netboot uses a separate image, which we jump in to at the start of IPL 
>>> from a network device (see zipl_load() in pc-bios/s390-ccw/bootmap.c). I 
>>> wasn't able to come up with a simple way to return to the main BIOS code 
>>> if a netboot fails other than by jumping back.  So, it seems to me that 
>>> netboot kind of throws a monkeywrench into the basic idea of reworking 
>>> the panics into returns.
>>>
>>> I'm open to suggestions on a better way to recover from a failed netboot, 
>>> and it's certainly possible I've overlooked something, but as far as I 
>>> can tell a jump is necessary in that particular case at least. Netboot 
>>> could perhaps be handled as a special case where the jump back is 
>>> permitted whereas other device types return, but I don't think that 
>>> actually solves the main issue.
>>>
>>> What are your thoughts on this?
>>
>> Yes, I agree that jumping is currently required to get back from the 
>> netboot code. So if you could rework your patches in a way that limits the 
>> jumping to a failed netboot, that would be acceptable, I think.
>>
>> Apart from that: We originally decided to put the netboot code into a 
>> separate binary since the required roms/SLOF module might not always have 
>> been checked out (it needed to be done manually), so we were not able to 
>> compile it in all cases. But nowadays, this is handled in a much nicer 
>> way, the submodule is automatically checked out once you compile the 
>> s390x-softmmu target and have a s390x compiler available, so I wonder 
>> whether we should maybe do the next step and integrate the netboot code 
>> into the main s390-ccw.img now? Anybody got an opinion on this?
>>
>>  Thomas
>>
> 
> Hi Thomas,
> 
> I would generally defer the decision about integrating the netboot code to 
> someone with more insight than me, but for what it's worth, I am of the 
> opinion that if we want to rework all of panics into returns, then it would 
> make the most sense to also do the integration now so that we can avoid 
> using jump altogether.  Unless I'm missing something simple, I don't think 
> the panic/return conversion will be trivial, and actually I think it will be 
> quite invasive since there are dozens of calls to panic and assert that will 
> need to be changed.   It doesn't seem worthwhile to do all of these 
> conversions in order to avoid using jump, but then still being exposed to 
> possible problems caused by jumping due to netboot requiring it anyway.

Agreed, we should either do it right and merge the two binaries, or it does 
not make too much sense to only partly convert the code.

I can look into merging the two binaries, but it might also take some time. 
So for the time being, I'm fine if we include the panic-jumping hack for 
now, we can still then clean it up later.

  Thomas