Re: [RFC PATCH] iio: ak8975: Make sure chipset is always initialized

[Pandruvada, Srinivas]

On Sat, 2015-01-24 at 00:38 +0100, Hartmut Knaack wrote:
> Pandruvada, Srinivas schrieb am 19.01.2015 um 17:56:
> > On Mon, 2015-01-19 at 18:49 +0200, Daniel Baluta wrote: 
> >> On Mon, Jan 19, 2015 at 6:44 PM, Pandruvada, Srinivas
> >> <srinivas.pandruvada@xxxxxxxxx> wrote:
> >>> On Mon, 2015-01-19 at 16:40 +0200, Daniel Baluta wrote:
> >>>> Hello,
> >>>>
> >>>> On Sat, Dec 20, 2014 at 11:29 PM, Pandruvada, Srinivas
> >>>> <srinivas.pandruvada@xxxxxxxxx> wrote:
> >>>>> +Mika
> >>>>>
> >>>>> On Sat, 2014-12-20 at 13:26 -0800, Srinivas Pandruvada wrote:
> >>>>>> On Sat, 2014-12-20 at 00:25 +0200, Daniel Baluta wrote:
> >>>>>>> On Sat, Dec 20, 2014 at 12:16 AM, Hartmut Knaack <knaack.h@xxxxxx> wrote:
> >>>>>>>> Daniel Baluta schrieb am 18.12.2014 um 18:16:
> >>>>>>>>> When using ACPI, if acpi_match_device fails then chipset enum will be
> >>>>>>>>> uninitialized and &ak_def_array[chipset] will point to some bad address.
> >>>>>>>>>
> >>>>>> I am missing something. You are enumerated over i2c device, which was
> >>>>>> created from ACPI PNP resource. There is a valid handle or and the
> >>>>>> device has an ACPI companion at the least. If this failing, I have to
> >>>>>> check the code for acpi i2c.
> >>>>>> Can you check why this check failed? We may have bug in i2c handling.
> >>>>
> >>>> You are right about this. Under normal circumstances, if probe is called
> >>>> then acpi_match_device will not fail. I even tried to remove the
> >>>> device after probe
> >>>> but before acpi_match_device, anyhow acpi_match_device was still successful :)
> >>>>
> >>>> This is more a matter of code correctness.
> >>>>
> >>>> In ak8975_match_acpi_device we have:
> >>>>
> >>>> Â       const struct acpi_device_id *id;
> >>>>
> >>>> Â       id = acpi_match_device(dev->driver->acpi_match_table, dev);
> >>>> Â       if (!id)
> >>>> Â       Â       return NULL;
> >>>> Â       *chipset = (int)id->driver_data;
> >>>>
> >>>> Compiler complains on the fact that chipset might be uninitialized
> >>>> if this returns NULL, and we shouldn't ignore this warning even this case
> >>>> will never happen.
> >>>>
> >>> Will this fix?
> >>> data->chipset = AK8975;
> >>> before
> >>> ak8975_match_acpi_device(&client->dev, &data->chipset);
> >>>
> 
> This would fix the compiler warning, but doesn't seem the right solution for
> this issue. Quoting the description of acpi_match_device:
> "Return a pointer to the first matching ID on success or %NULL on failure."
> So, even if it is very unlikely to for it to fail - if it does fail, the
> error should be handled as quick as possible. I would favor Daniels solution
> to check for a valid assignment of name.
> 
This should never fail as the device is enumerated by this. So it
doesn't matter as long as you silent compiler warning.
> >>
> >> Yes, this is done in the original patch:
> >>
> >> +       *chipset = AK_MAX_TYPE;
> > Since data memory is not zero alloced, other member of data are anyway
> > initialized, so adding this also may be better. 
> 
> If there did not occur an error condition, it will be assigned a value
> before being checked for valid ranges. And if there is an error, probe
> should be aborted, anyway. So initializing *chipset doesn't seem to add
> any benefit IMHO.
> 
> >>
> >> .. and fixes the warning.
> >>
> >> Daniel.
> > 
> > NïïïïïrïïyïïïbïXïïÇvï^ï)Þ{.nï+ïïïï{ïï*"ïï^nïrïïïzïïïhïïïï&ïïïGïïïhï(ïéïÝj"ïïïmïïïïïzïÞïïïfïïïhïïï~ïmml==
> > 
>