Re: [PATCH] Bluetooth: hci_core: Use ERR_PTR instead of NULL

[Khalid Masum]

On Sun, Jul 17, 2022 at 10:17 PM Pavel Skripkin <paskripkin@xxxxxxxxx> wrote:
>
> Hi Khalid,
>
> Khalid Masum <khalid.masum.92@xxxxxxxxx> says:
> > Failure of kzalloc to allocate memory is not reported. Return Error
> > pointer to ENOMEM if memory allocation fails. This will increase
> > readability and will make the function easier to use in future.
> >
> > Signed-off-by: Khalid Masum <khalid.masum.92@xxxxxxxxx>
> > ---
>
> [snip]
>
> > index a0f99baafd35..ea50767e02bf 100644
> > --- a/net/bluetooth/hci_core.c
> > +++ b/net/bluetooth/hci_core.c
> > @@ -2419,7 +2419,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
> >
> >       hdev = kzalloc(alloc_size, GFP_KERNEL);
> >       if (!hdev)
> > -             return NULL;
> > +             return ERR_PTR(-ENOMEM);
> >
>
> This will break all callers of hci_alloc_dev(). All callers expect NULL
> in case of an error, so you will leave them with wrong pointer.

You are right. All callers of hci_alloc_dev() need to be able to handle
the error pointer. I shall send a V2 with all the callers of hci_alloc_dev
handling the ERR_PTR.

> Also, allocation functionS return an error only in case of ENOMEM, so
> initial code is fine, IMO
>

I think it makes the memory allocation error handling look to be a bit
different from what we usually do while allocating memory which is,
returning an error or an error pointer. Here we are returning a NULL
without any context, making it a bit unreadable. So I think returning
an error pointer is better. If I am not mistaken, this also complies with
the return convention:
https://www.kernel.org/doc/htmldocs/kernel-hacking/convention-returns.html

>
> Thanks,
> --Pavel Skripkin


Thanks,
  -- Khalid Masum