[PATCH V4] Fix pointer cast for 32 bits arch

[Peter Senna Tschudin]

On Thu, Apr 16, 2015 at 7:05 PM, Alan Cox <alan at linux.intel.com> wrote:
> On Thu, 2015-04-16 at 20:01 +0300, Dan Carpenter wrote:
>> On Thu, Apr 16, 2015 at 06:14:55PM +0200, Geert Uytterhoeven wrote:
>> > On Thu, Apr 16, 2015 at 3:39 PM, Peter Senna Tschudin
>> > <peter.senna at gmail.com> wrote:
>> > > --- a/drivers/staging/goldfish/goldfish_audio.c
>> > > +++ b/drivers/staging/goldfish/goldfish_audio.c
>> > > @@ -63,7 +63,7 @@ struct goldfish_audio {
>> > >  #define AUDIO_READ(data, addr)         (readl(data->reg_base + addr))
>> > >  #define AUDIO_WRITE(data, addr, x)     (writel(x, data->reg_base + addr))
>> > >  #define AUDIO_WRITE64(data, addr, addr2, x)    \
>> > > -       (gf_write64((u64)(x), data->reg_base + addr, data->reg_base+addr2))
>> > > +       (gf_write_ptr((void *)(x), data->reg_base + addr, data->reg_base+addr2))
>> >
>> > This one should not be converted, as all callers pass a dma_addr_t, which may
>> > be 64-bit on 32-bit systems, i.e. larger than void *.
>>
>> Ugh...  You're right.
>>
>> I've been avoiding asking this but I can't any longer.  What is
>> gf_write64() actually doing?  We are writing dma addresses, user space
>> pointers and kernel space pointers to this hardware?
>>
>> This stuff doesn't seem to make any kind of sense and I can easily
>> imagine a situation where it wrote a 64 bit pointer.  Then we partially
>> write over it with a 32 bit userspace pointer.  Then it writes somewhere
>> totally unintended.
>>
>> This thing doesn't make any sort of sense to me.
>
> Its a 64 on 64 or 32 on 32 virtual machine. Goldfish is used for Android
> emulation for all the system level phone emulation tools. On the
> emulation side it provides an interface for the emulated OS but makes no
> effort to emulate it as if it was a real hardware. If you think of it as
> a funky emulator interface all is good. If you think about it as
> "hardware" you've got the wrong model and chunks of Goldfish make less
> sense.

Is is better to leave the code as is, and ignore the compiler / sparse
warnings for i386? Or is the proposal welcome if done correctly? And
if so what would be correctly?


-- 
Peter