Re: [PATCH 1/1] staging/comedi: remove unnecessary check around pci_dev_put

[Ian Abbott <abbotti@mev.co.uk>]

On 25/06/15 12:21, Dan Carpenter wrote:
> On Thu, Jun 25, 2015 at 04:20:26PM +0530, Maninder Singh wrote:
>> pci_dev_put cehcks for NULL pointer itself,
>> reported by coccinelle
>>
>
> This patch is correct but the motivation is wrong.
>
> The check in pci_dev_put() is like a sanity check.  There are many
> functions which have a sanity check and many which do not, it is
> impossible for a human to remember the complete list of each.  When we
> remove explicit checks for NULL and instead rely on the sanity checks
> it sometimes makes the code more subtle and difficult to read.
>
> In this case, "pcidev" can never be NULL so the check is misleading and
> makes the code more complicated.  Removing it is a good thing.  Also
> the attach function does not have a NULL check so when we remove this
> check we make the code more consistent.

Actually, it is possible for pcidev to be NULL here (pci9118_detach() in 
drivers/staging/comedi/drivers/adl_pci9118.c).  This driver supports 
both the auto-attach (via PCI driver probe) and "legacy" attach (via 
COMEDI_DEVCONFIG ioctl).  For the auto-attach case, pcidev will never be 
NULL.  For the "legacy" attach case, pcidev can be NULL if the call to 
pci9118_find_pci() from pci9118_attach() failed to find a matching PCI 
device and so returned an error before the call to comedi_set_hw_dev(). 
  The comedi core then calls pci9118_detach() which sets pcidev to the 
return value from comedi_to_pci_dev() which will be NULL in this case.

> But in other cases, if "pcidev" could be NULL then we should keep the
> check so that the code is easier to read.

So we should keep the check in this case too.

-- 
-=( Ian Abbott @ MEV Ltd.    E-mail: <abbotti@mev.co.uk> )=-
-=(                          Web: http://www.mev.co.uk/  )=-