Re: [PATCH] pc-bios/s390x: Pack ResetInfo struct

["Jason J. Herne" <jjherne@linux.ibm.com>]

On 8/27/20 6:07 AM, Thomas Huth wrote:
> On 07/02/2020 15.02, Jason J. Herne wrote:
>> On 2/7/20 6:28 AM, Christian Borntraeger wrote:
>>> Jason,
>>>
>>> can you run objdump -Sdr on jump2ipl.o on a broken variant?
>>>
>>>
>> To keep the volume lower, I've only pasted the output that I think
>> you're interested in. If you want to see the entire thing just let me know.
>>
>> static void jump_to_IPL_2(void)
>> {
>>   1d0:    eb bf f0 58 00 24     stmg    %r11,%r15,88(%r15)
>>   1d6:    a7 fb ff 50           aghi    %r15,-176
>>   1da:    b9 04 00 bf           lgr    %r11,%r15
>>      ResetInfo *current = 0;
>>   1de:    a7 19 00 00           lghi    %r1,0
>>   1e2:    e3 10 b0 a8 00 24     stg    %r1,168(%r11)
>>
>>      void (*ipl)(void) = (void *) (uint64_t) current->ipl_continue;
>>   1e8:    e3 10 b0 a8 00 04     lg    %r1,168(%r11)
>>   1ee:    58 10 10 08           l    %r1,8(%r1)
>>   1f2:    b9 16 00 11           llgfr    %r1,%r1
>>   1f6:    e3 10 b0 a0 00 24     stg    %r1,160(%r11)
>>      *current = save;
>>   1fc:    e3 10 b0 a8 00 04     lg    %r1,168(%r11)
>>   202:    c0 20 00 00 00 00     larl    %r2,202 <jump_to_IPL_2+0x32>
>>              204: R_390_PC32DBL    .bss+0x2
>>   208:    eb 23 20 00 00 04     lmg    %r2,%r3,0(%r2)
>>   20e:    eb 23 10 00 00 24     stmg    %r2,%r3,0(%r1)
>>      ipl(); /* should not return */
>>   214:    e3 10 b0 a0 00 04     lg    %r1,160(%r11)
>>   21a:    0d e1                 basr    %r14,%r1
>> }
>>   21c:    18 00                 lr    %r0,%r0
>>   21e:    eb bf b1 08 00 04     lmg    %r11,%r15,264(%r11)
>>   224:    07 fe                 br    %r14
>>   226:    07 07                 nopr    %r7
> 
> I'm currently looking through the past s390-ccw bios patches that still
> might need attention ... was there ever a follow up on this discussion?
> Do we need to clear the registers before jumping to the OS?
> And looking at the disassembly, should we declar the ipl function
> pointer with __attribute__((noreturn)) ?
> 

Janosch has done some cleanup work that has not hit master yet. This work, if accepted, 
would fix this problem and eliminate the need for this patch. So I think we should wait to 
see how that plays out before we revisit this.

-- 
-- Jason J. Herne (jjherne@linux.ibm.com)