Re: [PATCH v4 1/2] m68k: Handle __generic_copy_to_user faults more carefully

[Michael Schmitz <schmitzmic@gmail.com>]

Hi Finn,

Am 09.08.2024 um 15:34 schrieb Finn Thain:
>
> On Fri, 9 Aug 2024, Michael Schmitz wrote:
>
>>>
>>> I tried out the "stress-ng --sysbadaddr -1" test, and that didn't go
>>> so well for me:
>>>
>>> # stress-ng --sysbadaddr -1
>>> stress-ng: info:  [37] defaulting to a 86400 second (1 day, 0.00 secs) run
>>> per stressor
>>> stress-ng: info:  [37] dispatching hogs: 1 sysbadaddr
>>> *** ILLEGAL INSTRUCTION ***   FORMAT=4
>>> Current process id is 39
>>> BAD KERNEL TRAP: 00000000
>>> Modules linked in:
>>> PC: [<00000000>] 0x0
>>> SR: 2004  SP: 6504e563  a2: 008ee380
>>> d0: 000000f7    d1: 00000000    d2: 00000000    d3: 00000000
>>> d4: 00a87b80    d5: bfbf3814    a0: 00000000    a1: bfbf3814
>>> Process stress-ng-sysba (pid: 39, task=4dbb2ec5)
>>> Frame format=4 eff addr=480a2004 pc=0002b154
>>> Stack from 00adff20:
>>>         00ade000 00000000 00000000 000000f7 00000000 00000004 00a87b80
>>> 00000000
>>>         00000000 00000000 00000000 008ee380 0002ab5c 00000100 00000122
>>> fffffff6
>>>         bfbf376c 0002b29e 000000f7 bfbf3814 00000000 00000000 00ade000
>>> 0002b222
>>>         00ae0800 80118988 00000000 00000005 bfbf37a0 00000005 bfbf3814
>>> 00adffcc
>>>         00023d2c 00adffcc 00000000 00000000 00000000 00000000 000000f7
>>> 00000000
>>>         80118b46 00021850 00024b00 000000f7 bfbf3814 00000000 00000000
>>> bfbf3814
>>> Call Trace: [<0002ab5c>] child_wait_callback+0x0/0x24
>>>  [<0002b29e>] sys_wait4+0x7c/0x8e
>>>  [<0002b222>] sys_wait4+0x0/0x8e
>>>  [<00023d2c>] buserr_c+0xb0/0x152
>>>  [<00021850>] buserr+0x28/0x30
>>>  [<00024b00>] system_call+0x54/0xa8
>>>
>>> But that is the same with and without these patches.
>>
>> I wonder if recent signal handling changes (e.g. commit
>> 0d4276cfbe6fd4c4a21acdee803b05a3a6192082) have rare unexpected side effects on
>> Coldfire here ... OTOH, signal handling as such works just fine, right?
>>
>
> That would be commit b845b574f86d ("m68k: Move signal frame following
> exception on 68020/030") on mainline. If it caused a regression, that
> would have first appeared in v6.4. I can't imagine how that commit could
> affect Coldfire but that's no reason not to test an older kernel.

Yes, testing older kernels is certainly the fastest way to rule out 
involvement of that commit.

On a closer look, there is no possible way in that this commit can be 
responsible for the bug unless Coldfire does use frame format B for 
access errors. I don't think that's likely?

> FWIW, my hunch is that the other stressors which call wait4() will
> probably crash too (regardless of kernel version).

Quite likely ...

Cheers,

	Michael