Re: [PATCH] RDMA: remove null check after call container_of()

[Jason Gunthorpe]

On Tue, May 17, 2022 at 07:54:38PM +0200, Christophe JAILLET wrote:
> Le 17/05/2022 à 14:16, Jason Gunthorpe a écrit :
> > On Tue, May 17, 2022 at 09:33:28AM +0800, Haowen Bai wrote:
> > > container_of() will never return NULL, so remove useless code.
> > 
> > It is confusing here, but it can be null.
> 
> Hi,
> 
> out of curiosity, can you elaborate how it can be NULL?

It is guarented/required that that container_of is a 0 offset. The
normal usage of the ib_alloc_device macro:

#define ib_alloc_device(drv_struct, member)                                    \
	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
				      BUILD_BUG_ON_ZERO(offsetof(              \
					      struct drv_struct, member))),    \
		     struct drv_struct, member)

Enforces this property with a BUILD_BUG_ON

So, if the input pointer to container_of is reliably NULL or ERR_PTR
then the output pointer will be the same.

The rvt code here open codes the call because it is a mid-layer and
the sizeof() calculation above is not correct for it.

Jason