Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

[Arnd Bergmann]

On Wednesday 07 October 2015 13:04:06 Arnd Bergmann wrote:
> On Wednesday 07 October 2015 11:45:02 Russell King - ARM Linux wrote:
> > On Wed, Oct 07, 2015 at 12:41:21PM +0200, Arnd Bergmann wrote:
> > > The virtgpu driver prints the last_seq variable using the %ld or
> > > %lu format string, which does not work correctly on all architectures
> > > and causes this compiler warning on ARM:
> > > 
> > > drivers/gpu/drm/virtio/virtgpu_fence.c: In function 'virtio_timeline_value_str':
> > > drivers/gpu/drm/virtio/virtgpu_fence.c:64:22: warning: format '%lu' expects argument of type 'long unsigned int', but argument 4 has type 'long long int' [-Wformat=]
> > >   snprintf(str, size, "%lu", atomic64_read(&fence->drv->last_seq));
> > >                       ^
> > > drivers/gpu/drm/virtio/virtgpu_debugfs.c: In function 'virtio_gpu_debugfs_irq_info':
> > > drivers/gpu/drm/virtio/virtgpu_debugfs.c:37:16: warning: format '%ld' expects argument of type 'long int', but argument 3 has type 'long long int' [-Wformat=]
> > >   seq_printf(m, "fence %ld %lld\n",
> > >                 ^
> > > 
> > > In order to avoid the warnings, this changes the format strings to %llu
> > > and adds a cast to u64, which makes it work the same way everywhere.
> > 
> > You have to wonder why atomic64_* functions do not use u64 types.
> > If they're not reliant on manipulating 64-bit quantities, then what's
> > the point of calling them atomic _64_.
> 
> I haven't checked all architectures, but I assume what happens is that
> 64-bit ones just #define atomic64_t atomic_long_t, so they don't have
> to provide three sets of functions.

scratch that, I just looked at all the architectures and found that it's
just completely arbitrary, even within one architecture you get a mix
of 'long' and 'long long', plus this gem from MIPS:

static __inline__ int atomic64_add_unless(atomic64_t *v, long a, long u)

which truncates the result to 32 bit.

	Arnd

Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

[Geert Uytterhoeven]

On Wed, Oct 7, 2015 at 1:23 PM, Arnd Bergmann <arnd@arndb.de> wrote:
> On Wednesday 07 October 2015 13:04:06 Arnd Bergmann wrote:
>> On Wednesday 07 October 2015 11:45:02 Russell King - ARM Linux wrote:
>> > On Wed, Oct 07, 2015 at 12:41:21PM +0200, Arnd Bergmann wrote:
>> > > The virtgpu driver prints the last_seq variable using the %ld or
>> > > %lu format string, which does not work correctly on all architectures
>> > > and causes this compiler warning on ARM:
>> > >
>> > > drivers/gpu/drm/virtio/virtgpu_fence.c: In function 'virtio_timeline_value_str':
>> > > drivers/gpu/drm/virtio/virtgpu_fence.c:64:22: warning: format '%lu' expects argument of type 'long unsigned int', but argument 4 has type 'long long int' [-Wformat=]
>> > >   snprintf(str, size, "%lu", atomic64_read(&fence->drv->last_seq));
>> > >                       ^
>> > > drivers/gpu/drm/virtio/virtgpu_debugfs.c: In function 'virtio_gpu_debugfs_irq_info':
>> > > drivers/gpu/drm/virtio/virtgpu_debugfs.c:37:16: warning: format '%ld' expects argument of type 'long int', but argument 3 has type 'long long int' [-Wformat=]
>> > >   seq_printf(m, "fence %ld %lld\n",
>> > >                 ^
>> > >
>> > > In order to avoid the warnings, this changes the format strings to %llu
>> > > and adds a cast to u64, which makes it work the same way everywhere.
>> >
>> > You have to wonder why atomic64_* functions do not use u64 types.
>> > If they're not reliant on manipulating 64-bit quantities, then what's
>> > the point of calling them atomic _64_.
>>
>> I haven't checked all architectures, but I assume what happens is that
>> 64-bit ones just #define atomic64_t atomic_long_t, so they don't have
>> to provide three sets of functions.
>
> scratch that, I just looked at all the architectures and found that it's
> just completely arbitrary, even within one architecture you get a mix
> of 'long' and 'long long', plus this gem from MIPS:
>
> static __inline__ int atomic64_add_unless(atomic64_t *v, long a, long u)
>
> which truncates the result to 32 bit.

Woops.

See also my unanswered question in "atomic64 on 32-bit vs 64-bit (was:
Re: Add virtio gpu driver.)", which is still valid:
https://lkml.org/lkml/2015/6/28/18

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

[Geert Uytterhoeven]

On Wed, Oct 7, 2015 at 1:23 PM, Arnd Bergmann <arnd@arndb.de> wrote:
> On Wednesday 07 October 2015 13:04:06 Arnd Bergmann wrote:
>> On Wednesday 07 October 2015 11:45:02 Russell King - ARM Linux wrote:
>> > On Wed, Oct 07, 2015 at 12:41:21PM +0200, Arnd Bergmann wrote:
>> > > The virtgpu driver prints the last_seq variable using the %ld or
>> > > %lu format string, which does not work correctly on all architectures
>> > > and causes this compiler warning on ARM:
>> > >
>> > > drivers/gpu/drm/virtio/virtgpu_fence.c: In function 'virtio_timeline_value_str':
>> > > drivers/gpu/drm/virtio/virtgpu_fence.c:64:22: warning: format '%lu' expects argument of type 'long unsigned int', but argument 4 has type 'long long int' [-Wformat=]
>> > >   snprintf(str, size, "%lu", atomic64_read(&fence->drv->last_seq));
>> > >                       ^
>> > > drivers/gpu/drm/virtio/virtgpu_debugfs.c: In function 'virtio_gpu_debugfs_irq_info':
>> > > drivers/gpu/drm/virtio/virtgpu_debugfs.c:37:16: warning: format '%ld' expects argument of type 'long int', but argument 3 has type 'long long int' [-Wformat=]
>> > >   seq_printf(m, "fence %ld %lld\n",
>> > >                 ^
>> > >
>> > > In order to avoid the warnings, this changes the format strings to %llu
>> > > and adds a cast to u64, which makes it work the same way everywhere.
>> >
>> > You have to wonder why atomic64_* functions do not use u64 types.
>> > If they're not reliant on manipulating 64-bit quantities, then what's
>> > the point of calling them atomic _64_.
>>
>> I haven't checked all architectures, but I assume what happens is that
>> 64-bit ones just #define atomic64_t atomic_long_t, so they don't have
>> to provide three sets of functions.
>
> scratch that, I just looked at all the architectures and found that it's
> just completely arbitrary, even within one architecture you get a mix
> of 'long' and 'long long', plus this gem from MIPS:
>
> static __inline__ int atomic64_add_unless(atomic64_t *v, long a, long u)
>
> which truncates the result to 32 bit.

Woops.

See also my unanswered question in "atomic64 on 32-bit vs 64-bit (was:
Re: Add virtio gpu driver.)", which is still valid:
https://lkml.org/lkml/2015/6/28/18

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

[Arnd Bergmann]

On Monday 19 October 2015 09:34:15 Geert Uytterhoeven wrote:
> On Wed, Oct 7, 2015 at 1:23 PM, Arnd Bergmann <arnd@arndb.de> wrote:
> > static __inline__ int atomic64_add_unless(atomic64_t *v, long a, long u)
> >
> > which truncates the result to 32 bit.
> 
> Woops.
> 
> See also my unanswered question in "atomic64 on 32-bit vs 64-bit (was:
> Re: Add virtio gpu driver.)", which is still valid:
> https://lkml.org/lkml/2015/6/28/18
> 

Regarding your question of

> Instead of sprinkling casts, is there any good reason why atomic64_read()
> and atomic64_t aren't "long long" everywhere, cfr. u64?


I assume the answer is that some (all?) 64-bit architectures intentionally
return 'long' here, in order for atomic_long_read() to return 'long' on
all architectures, given the definitions from
include/asm-generic/atomic-long.h

We would have to either change those, or we have to pick between
atomic_long_* or atomic64_* to have a consistent return type.

	Arnd

Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

[Geert Uytterhoeven]

On Mon, Oct 19, 2015 at 12:11 PM, Arnd Bergmann <arnd@arndb.de> wrote:
> On Monday 19 October 2015 09:34:15 Geert Uytterhoeven wrote:
>> On Wed, Oct 7, 2015 at 1:23 PM, Arnd Bergmann <arnd@arndb.de> wrote:
>> > static __inline__ int atomic64_add_unless(atomic64_t *v, long a, long u)
>> >
>> > which truncates the result to 32 bit.
>>
>> Woops.
>>
>> See also my unanswered question in "atomic64 on 32-bit vs 64-bit (was:
>> Re: Add virtio gpu driver.)", which is still valid:
>> https://lkml.org/lkml/2015/6/28/18
>>
>
> Regarding your question of
>
>> Instead of sprinkling casts, is there any good reason why atomic64_read()
>> and atomic64_t aren't "long long" everywhere, cfr. u64?
>
>
> I assume the answer is that some (all?) 64-bit architectures intentionally
> return 'long' here, in order for atomic_long_read() to return 'long' on
> all architectures, given the definitions from
> include/asm-generic/atomic-long.h
>
> We would have to either change those, or we have to pick between
> atomic_long_* or atomic64_* to have a consistent return type.

I guess the main reason is this comment in include/asm-generic/atomic-long.h,
which I hadn't noticed before:

 * Casts for parameters are avoided for existing atomic functions in order to
 * avoid issues with cast-as-lval under gcc 4.x and other limitations that the
 * macros of a platform may have.

Still, it's a pity, as printing atomic_64 is one more place where casts are
needed in callers.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

[Geert Uytterhoeven]

On Mon, Oct 19, 2015 at 12:11 PM, Arnd Bergmann <arnd@arndb.de> wrote:
> On Monday 19 October 2015 09:34:15 Geert Uytterhoeven wrote:
>> On Wed, Oct 7, 2015 at 1:23 PM, Arnd Bergmann <arnd@arndb.de> wrote:
>> > static __inline__ int atomic64_add_unless(atomic64_t *v, long a, long u)
>> >
>> > which truncates the result to 32 bit.
>>
>> Woops.
>>
>> See also my unanswered question in "atomic64 on 32-bit vs 64-bit (was:
>> Re: Add virtio gpu driver.)", which is still valid:
>> https://lkml.org/lkml/2015/6/28/18
>>
>
> Regarding your question of
>
>> Instead of sprinkling casts, is there any good reason why atomic64_read()
>> and atomic64_t aren't "long long" everywhere, cfr. u64?
>
>
> I assume the answer is that some (all?) 64-bit architectures intentionally
> return 'long' here, in order for atomic_long_read() to return 'long' on
> all architectures, given the definitions from
> include/asm-generic/atomic-long.h
>
> We would have to either change those, or we have to pick between
> atomic_long_* or atomic64_* to have a consistent return type.

I guess the main reason is this comment in include/asm-generic/atomic-long.h,
which I hadn't noticed before:

 * Casts for parameters are avoided for existing atomic functions in order to
 * avoid issues with cast-as-lval under gcc 4.x and other limitations that the
 * macros of a platform may have.

Still, it's a pity, as printing atomic_64 is one more place where casts are
needed in callers.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

[Ralf Baechle]

On Wed, Oct 07, 2015 at 01:23:07PM +0200, Arnd Bergmann wrote:

> > I haven't checked all architectures, but I assume what happens is that
> > 64-bit ones just #define atomic64_t atomic_long_t, so they don't have
> > to provide three sets of functions.
> 
> scratch that, I just looked at all the architectures and found that it's
> just completely arbitrary, even within one architecture you get a mix
> of 'long' and 'long long', plus this gem from MIPS:
> 
> static __inline__ int atomic64_add_unless(atomic64_t *v, long a, long u)
> 
> which truncates the result to 32 bit.

Eh...  The result is 0/1 so nothing is truncated.  Alpha, MIPS,
PARISC and PowerPC are using the same prototype and x86 only differs
in the use of inline instead __inline__.  And anyway, that function on
MIPS is only built for CONFIG_64BIT.

What's wrong on MIPS is the comment describing the function's return value
which was changed by f24219b4e90cf70ec4a211b17fbabc725a0ddf3c (atomic: move
atomic_add_unless to generic code) and I've queued up a patch to fix that
since a few days.  I guess that was a cut and paste error from
__atomic_add_unless which indeed does return the old value.

  Ralf

Re: [PATCH] drm/virtio: use %llu format string form atomic64_t

[Arnd Bergmann]

On Monday 19 October 2015 11:37:00 Ralf Baechle wrote:
> On Wed, Oct 07, 2015 at 01:23:07PM +0200, Arnd Bergmann wrote:
> 
> > > I haven't checked all architectures, but I assume what happens is that
> > > 64-bit ones just #define atomic64_t atomic_long_t, so they don't have
> > > to provide three sets of functions.
> > 
> > scratch that, I just looked at all the architectures and found that it's
> > just completely arbitrary, even within one architecture you get a mix
> > of 'long' and 'long long', plus this gem from MIPS:
> > 
> > static __inline__ int atomic64_add_unless(atomic64_t *v, long a, long u)
> > 
> > which truncates the result to 32 bit.
> 
> Eh...  The result is 0/1 so nothing is truncated.  Alpha, MIPS,
> PARISC and PowerPC are using the same prototype and x86 only differs
> in the use of inline instead __inline__.  And anyway, that function on
> MIPS is only built for CONFIG_64BIT.

Ah, got it. Sorry about that.

> What's wrong on MIPS is the comment describing the function's return value
> which was changed by f24219b4e90cf70ec4a211b17fbabc725a0ddf3c (atomic: move
> atomic_add_unless to generic code) and I've queued up a patch to fix that
> since a few days.  I guess that was a cut and paste error from
> __atomic_add_unless which indeed does return the old value.

Thanks!

	Arnd