Panic in quirk_usb_early_handoff

[Ard Biesheuvel]

> On 4 Mar 2017, at 16:57, Mason <slash.tmp at free.fr> wrote:
> 
>> On 04/03/2017 09:07, Ard Biesheuvel wrote:
>>> On 4 March 2017 at 00:24, Mason wrote:
>>>> On 03/03/2017 20:02, Robin Murphy wrote:
>>>> 
>>>>> On 03/03/17 17:15, Mason wrote:
>>>>> 
>>>>> [    1.261813] Unable to handle kernel paging request at virtual address d08611e4
>>>>> [    1.269167] pgd = c0004000
>>>>> [    1.271979] [d08611e4] *pgd=8f804811, *pte=00000000, *ppte=00000000
>>>>> [    1.278394] Internal error: Oops: 7 [#1] PREEMPT SMP ARM
>>>>> [    1.283815] Modules linked in:
>>>>> [    1.286970] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.9.7-1-rc2 #157
>>>>> [    1.293614] Hardware name: Sigma Tango DT
>>>>> [    1.297726] task: cf82c9c0 task.stack: cf838000
>>>>> [    1.302364] PC is at quirk_usb_early_handoff+0x3e8/0x790
>>>>> [    1.307790] LR is at ioremap_page_range+0xf8/0x1a8
>>>>> [    1.312688] pc : [<c039fe44>]    lr : [<c02d0a10>]    psr: 000e0013
>>>>> [    1.312688] sp : cf839d78  ip : 00000000  fp : cf839e38
>>>>> [    1.324399] r10: c10248a0  r9 : 00000000  r8 : d08611e4
>>>>> [    1.329733] r7 : d084e000  r6 : 00002000  r5 : 000c0300  r4 : cfb4e800
>>>>> [    1.336377] r3 : 000131e4  r2 : 00000000  r1 : 91001e13  r0 : d084e000
>>>> 
>>>> ...and again. And always at the same PC, too.
>>> 
>>> By the way, isn't LR supposed to point to the caller of the
>>> current function? ("LR is at ioremap_page_range")
>>> 
>>> If so, why does it not appear in the back trace?
>> 
>> lr is supposed to point to the return address at function entry. After
>> that, all bets are off, really, since ARM usually pops the return
>> address from the stack straight into the pc register. So in this case,
>> it looks like it still contains the address that the most recent leaf
>> function returned to (or another function that actually restores the
>> return address into lr before branching to it). But it could easily
>> contain garbage as well.
> 
> If there is only a tiny chance that LR contains genuinely useful
> information, then what is the rationale for providing the info
> at all in the panic message?
> 
> I would argue that no info is better than info that is wrong
> most of the time.
> 

After pc, the link register is the most likely to legally point into the kernel .text section so it makes sense imo to decode the address into a function name plus offset.

Educating people about the architecture's calling convention and associated caveats is not the job of the panic handler.