Re: kobject_init and the zeroed-out-memory requirement

[Guenter Roeck <linux@roeck-us.net>]

On 10/05/2014 08:25 PM, Greg KH wrote:
> On Sun, Oct 05, 2014 at 07:09:50PM -0700, Guenter Roeck wrote:
>> On Sun, Oct 05, 2014 at 04:24:57PM -0700, Greg KH wrote:
>>> On Sun, Oct 05, 2014 at 06:13:05PM -0400, Jason Noakes wrote:
>>>>> No driver should be working with "raw" kobjects.
>>>>
>>>> I don't agree, but it's irrelevant.
>>>
>>> Not at all.  I'd wager that if a driver is messing around with a "raw"
>>> kobject, it is doing something seriously wrong.  Of course there are
>>> exceptions, but those are very rare, and exceptions.  A driver should be
>>> using the driver core, and the functions and objects provided there, and
>>> provided by the bus it lives on.
>>>
>>> So, have a pointer to some driver code that is calling
>>> kobject_initialize()?  I'd love to see it.
>>>
>>
>> Lots to choose from. In drivers/:
>>
>> $ git grep kobject_init | wc
>>       65     289    4880
>> $ git grep kobject_init | grep -v base | wc
>>       62     277    4694
>> $ git grep kobject_init_and_add | grep -v base | wc
>>       47     232    3698
>> $ git grep kobject_add | grep -v base | wc
>>       20     111    1486
>
> Well, you hit firmware layer, and block devices, which have to deal with
> kobjects directly at times.  But these files look "suspicious":
> 	edac/edac_device_sysfs.c
> 	edac/edac_pci_sysfs.c
> 	gpu/drm/ttm/ttm_bo.c
> 	gpu/drm/ttm/ttm_memory.c
> 	gpu/drm/ttm/ttm_page_alloc.c
> 	gpu/drm/ttm/ttm_page_alloc_dma.c
> 	infiniband/core/cm.c
> 	infiniband/core/sysfs.c
> 	infiniband/core/user_mad.c
> 	infiniband/hw/mlx4/sysfs.c
> 	infiniband/hw/qib/qib_sysfs.c
> 	infiniband/hw/usnic/usnic_ib_sysfs.c
> 	iommu/iommu.c
> 	parisc/pdc_stable.c
> 	pci/slot.c
> 	power/ab8500_fg.c
> 	power/abx500_chargalg.c
> 	scsi/iscsi_boot_sysfs.c
> 	uio/uio.c
> 	video/fbdev/omap2/dss/manager-sysfs.c
> 	video/fbdev/omap2/dss/overlay-sysfs.c
>
> I'll look into them tomorrow...
>
> But that's still a very small minority compared to the tens of thousands
> of drivers we have, so I still say it's wrong to have a driver deal with
> raw kobjects.
>

I agree. Just wanted to point out that there _are_ drivers out there
using it. I had actually looked into the power and video files,
and they looked suspicious to me as well.

Guenter