lkml.org 
[lkml]   [2017]   [Dec]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v1] x86/platform/intel-mid: Revert "Make 'bt_sfi_data' const"


On Thu, 28 Dec 2017, Ingo Molnar wrote:

>
> * Julia Lawall <julia.lawall@lip6.fr> wrote:
>
> >
> >
> > On Thu, 28 Dec 2017, Ingo Molnar wrote:
> >
> > >
> > > * Julia Lawall <julia.lawall@lip6.fr> wrote:
> > >
> > > > > > [...] There does seem to be a few cases where the field actually does hold an
> > > > > > integer. I guess this is not a problem?
> > > > >
> > > > > Could you point to such an example?
> > > >
> > > > drivers/thermal/intel_soc_dts_thermal.c:#define BYT_SOC_DTS_APIC_IRQ 86
> > > >
> > > > and then:
> > > >
> > > > static const struct x86_cpu_id soc_thermal_ids[] = {
> > > > { X86_VENDOR_INTEL, 6, INTEL_FAM6_ATOM_SILVERMONT1, 0,
> > > > BYT_SOC_DTS_APIC_IRQ},
> > > > {}
> > > > };
> > > >
> > > > and finally:
> > > >
> > > > soc_dts_thres_irq = (int)match_cpu->driver_data;
> > > >
> > > > Also:
> > > >
> > > > arch/x86/kernel/apic/apic.c
> > > >
> > > > #define DEADLINE_MODEL_MATCH_REV(model, rev) \
> > > > { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev
> > > > }
> > > >
> > > > DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X, 0x0b000020),
> > > > DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_CORE, 0x22),
> > > > etc. (all 2-digit numbers in the remaining case).
> > >
> > > Ok - I think in these cases the resulting long->pointer type conversion is a _lot_
> > > less dangerous than the pointer->long conversion which caused the regression.
> > >
> > > So unless the resulting code is excessively ugly, this feels like the right
> > > approach to me.
> >
> > The problem is that this case will inevitably have a cast somewhere.
>
> That's OK as long as the cast is dominantly (long)->(pointer), because that
> doesn't really risk losing the underlying type.
>
> It's the (pointer)->(pointer) and (pointer)->(long) conversions that are the most
> dangerous ones.
>
> > [...] Many of the values put into the driver_data field really are const, so
> > the type has to be const void *. When the value is extracted from the
> > structure, there will thus need to be a cast on it, and the current cast
> >
> > ddata = (struct bt_sfi_data *)id->driver_data;
> >
> > works fine, whether the original structure is const or not.
>
> So since this data structure is not size critical, I'd really suggest using two or
> three fields:
>
> ->driver_data.ptr
> ->driver_data.const_ptr
> ->driver_data.long
>
> that way the fundamental types remains.

Nothing will ensure that the data is extracted via the same field that was
used to store it. But one will get a message from the compiler if one
tries to change the properties of the stored value without changing the
field that it is stored into.

Should there be a union?

Is there some other kernel code that takes this solution?

thanks,
julia

>
> >
> > I also got a couple of complaints about non-pointer types:
> >
> > arch/x86/kernel/apic/apic.c:621:9: warning: cast from pointer to integer
> > of different size [-Wpointer-to-int-cast]
> > rev = (u32)m->driver_data;
> >
> > drivers/thermal/intel_soc_dts_thermal.c:68:22: warning: cast from pointer
> > to integer of different size [-Wpointer-to-int-cast]
> > soc_dts_thres_irq = (int)match_cpu->driver_data;
>
> These could use driver_data.long or so?
>
> Thanks,
>
> Ingo
>

\
 
 \ /
  Last update: 2017-12-28 22:23    [W:2.635 / U:0.076 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site