Re: Regression in IPMI on 4.15.6

[Laura Abbott <labbott@redhat.com>]

On 03/02/2018 05:46 AM, Corey Minyard wrote:
> On 02/28/2018 01:07 PM, Corey Minyard wrote:
>> On 02/28/2018 08:17 AM, Corey Minyard wrote:
>>> On 02/28/2018 07:53 AM, Corey Minyard wrote:
>>>> On 02/27/2018 05:55 PM, Laura Abbott wrote:
>>>>> Hi,
>>>>>
>>>>> Fedora got a bug report of a crash in IPMI on 4.15.6
>>>>> https://bugzilla.redhat.com/show_bug.cgi?id=1549316
>>>>> Unfortunately, it's only a screenshot but it's fairly
>>>>> clear. It looks like a panic in the error handling path
>>>>> in platform_device_unregister. Any ideas?
>>>>>
>>>>
>>>
>>> You may also run into another issue.  You can pull the
>>> individual patch at
>>>
>>> https://github.com/cminyard/linux-ipmi.git c8a1972e77dbe321ce5ce0247056e727234cbaec
>>
>> Actually, it needed a few more tweaks.  Can you do change
>> 426fa6179dae677134dfb37b21d057819418515b
>> instead?  It's "ipmi: Fix some error cleanup issues"
>>
>> I can send you patches, if you like.  If you could test and get back
>> to me, that would be great.
> 
> Laura, have you had a chance to test this?  I'd like to get it in soon,
> if possible.
> 
> Thanks,
> 
> -corey
> 

I think "ipmi: Re-use existing macros for built-in properties" is broken:

In file included from ./include/linux/acpi.h:28:0,
                  from ./include/linux/ipmi.h:21,
                  from drivers/char/ipmi/ipmi_dmi.c:7:
drivers/char/ipmi/ipmi_dmi.c: In function ‘dmi_add_platform_ipmi’:
./include/linux/property.h:236:1: error: expected expression before ‘{’ token
  {       \
  ^
./include/linux/property.h:244:2: note: in expansion of macro ‘PROPERTY_ENTRY_INTEGER’
   PROPERTY_ENTRY_INTEGER(_name_, u8, _val_)
   ^~~~~~~~~~~~~~~~~~~~~~
drivers/char/ipmi/ipmi_dmi.c:79:15: note: in expansion of macro ‘PROPERTY_ENTRY_U8’
    p[pidx++] = PROPERTY_ENTRY_U8("ipmi-type", si_type);
                ^~~~~~~~~~~~~~~~~
./include/linux/property.h:236:1: error: expected expression before ‘{’ token
  {       \
  ^
./include/linux/property.h:244:2: note: in expansion of macro ‘PROPERTY_ENTRY_INTEGER’
   PROPERTY_ENTRY_INTEGER(_name_, u8, _val_)
   ^~~~~~~~~~~~~~~~~~~~~~
drivers/char/ipmi/ipmi_dmi.c:81:14: note: in expansion of macro ‘PROPERTY_ENTRY_U8’
   p[pidx++] = PROPERTY_ENTRY_U8("slave-addr", slave_addr);
               ^~~~~~~~~~~~~~~~~
./include/linux/property.h:236:1: error: expected expression before ‘{’ token
  {       \
  ^
./include/linux/property.h:244:2: note: in expansion of macro ‘PROPERTY_ENTRY_INTEGER’
   PROPERTY_ENTRY_INTEGER(_name_, u8, _val_)
   ^~~~~~~~~~~~~~~~~~~~~~
drivers/char/ipmi/ipmi_dmi.c:82:14: note: in expansion of macro ‘PROPERTY_ENTRY_U8’
   p[pidx++] = PROPERTY_ENTRY_U8("addr-source", SI_SMBIOS);
               ^~~~~~~~~~~~~~~~~
./include/linux/property.h:236:1: error: expected expression before ‘{’ token
  {       \
  ^
./include/linux/property.h:246:2: note: in expansion of macro ‘PROPERTY_ENTRY_INTEGER’
   PROPERTY_ENTRY_INTEGER(_name_, u16, _val_)
   ^~~~~~~~~~~~~~~~~~~~~~
drivers/char/ipmi/ipmi_dmi.c:107:15: note: in expansion of macro ‘PROPERTY_ENTRY_U16’
    p[pidx++] = PROPERTY_ENTRY_U16("i2c-addr", base_addr);
                ^~~~~~~~~~~~~~~~~~

I don't think that macro is actually a replacement?

Thanks,
Laura

>>
>> BTW, the IPMI setup in your system is incorrect.  SMBIOS says it's at a
>> memory address, but it's at an I/O address.  And the address given
>> doesn't appear to be a valid address, the value read doesn't appear
>> to be a valid value.
>>
>> -corey
>>
>>>
>>> for that fix.
>>>
>>> -corey
>>>
>>>> Yeah, this is fixed by 174134ac7602 "ipmi_si: Fix error
>>>> handling of platform device" in mainstream.
>>>>
>>>> I guess I need to request a backport of this.
>>>>
>>>> Thanks for reporting.
>>>>
>>>> -corey
>>>>
>>>>
>>>>> Thanks,
>>>>> Laura
>>>>
>>>>
>>>
>>
>