Re: + input-new-force-feedback-interface.patch added to -mm tree

Re: + input-new-force-feedback-interface.patch added to -mm tree

[Arjan van de Ven]

On Wed, 2006-05-17 at 21:46 -0700, akpm@osdl.org wrote:
> +
> +#ifdef DEBUG
> +#define debug(format, arg...) printk(KERN_DEBUG "ff-effects: " format "\n" , ## arg)
> +#else
> +#define debug(format, arg...) do {} while (0)
> +#endif

please just use the existing prdebug() thing for this, no need to invent
your own ;)

> +
> +EXPORT_SYMBOL(input_ff_allocate);
> +EXPORT_SYMBOL(input_ff_register);
> +EXPORT_SYMBOL(input_ff_upload);
> +EXPORT_SYMBOL(input_ff_erase);

should these be _GPL exports?


> +
> +#define spin_ff_cond_lock(_ff, _flags)					  \
> +	do {								  \
> +		if (!_ff->driver->playback)				  \
> +			spin_lock_irqsave(&_ff->atomiclock, _flags);	  \
> +	} while (0);
> +
> +#define spin_ff_cond_unlock(_ff, _flags)					  \
> +	do {								  \
> +		if (!_ff->driver->playback)				  \
> +			spin_unlock_irqrestore(&_ff->atomiclock, _flags); \
> +	} while (0);


hmmm conditional locking like this always makes me very nervous... what
is preventing ->playback from changing halfway a locked sequence?

> +	if (!events) {
> +		debug("no actions");
> +		del_timer(&ff->timer);

are you really sure you don't need del_timer_sync() here?



> +static void input_ff_timer(unsigned long timer_data)
> +{
> +	struct input_dev *dev = (struct input_dev *) timer_data;
> +	struct ff_device *ff = dev->ff;
> +	struct ff_effect effect;
> +	int i;
> +	unsigned long flags;
> +	int effects_pending;
> +	unsigned long effect_handled[NBITS(FF_EFFECTS_MAX)];


hmmm stack space?

> +		} else {
> +			ret = -ENOSYS;

that is almost always a wrong return value
> +	if (test_bit(FF_CONSTANT, dev->ff->flags))
> +		set_bit(FF_CONSTANT, dev->ffbit);
> +	if (test_bit(FF_SPRING, dev->ff->flags))
> +		set_bit(FF_SPRING, dev->ffbit);
> +	if (test_bit(FF_FRICTION, dev->ff->flags))
> +		set_bit(FF_FRICTION, dev->ffbit);
> +	if (test_bit(FF_DAMPER, dev->ff->flags))
> +		set_bit(FF_DAMPER, dev->ffbit);
> +	if (test_bit(FF_INERTIA, dev->ff->flags))

are you really sure you need atomic set_bit()'s here??
if so.. I think you have a race ;)

Re: + input-new-force-feedback-interface.patch added to -mm tree

[Anssi Hannula]

Arjan van de Ven wrote:
> On Wed, 2006-05-17 at 21:46 -0700, akpm@osdl.org wrote:
> 
>>+
>>+#ifdef DEBUG
>>+#define debug(format, arg...) printk(KERN_DEBUG "ff-effects: " format "\n" , ## arg)
>>+#else
>>+#define debug(format, arg...) do {} while (0)
>>+#endif
> 
> 
> please just use the existing prdebug() thing for this, no need to invent
> your own ;)

Couldn't find any info on that one, are you sure you spelled it correctly?

> 
>>+
>>+EXPORT_SYMBOL(input_ff_allocate);
>>+EXPORT_SYMBOL(input_ff_register);
>>+EXPORT_SYMBOL(input_ff_upload);
>>+EXPORT_SYMBOL(input_ff_erase);
> 
> 
> should these be _GPL exports?
> 

Well, I don't know. When should EXPORT_SYMBOLs be EXPORT_SYMBOL_GPLs?

> 
>>+
>>+#define spin_ff_cond_lock(_ff, _flags)					  \
>>+	do {								  \
>>+		if (!_ff->driver->playback)				  \
>>+			spin_lock_irqsave(&_ff->atomiclock, _flags);	  \
>>+	} while (0);
>>+
>>+#define spin_ff_cond_unlock(_ff, _flags)					  \
>>+	do {								  \
>>+		if (!_ff->driver->playback)				  \
>>+			spin_unlock_irqrestore(&_ff->atomiclock, _flags); \
>>+	} while (0);
> 
> 
> 
> hmmm conditional locking like this always makes me very nervous... what
> is preventing ->playback from changing halfway a locked sequence?

Well, it's never changed, locked or not. But maybe we can get rid of
this condlocking alltogether, see my reply for Andrew Morton.

> 
>>+	if (!events) {
>>+		debug("no actions");
>>+		del_timer(&ff->timer);
> 
> 
> are you really sure you don't need del_timer_sync() here?
> 

Yes, this function is also called from inside the timer in question and
del_timer_sync() would block.

> 
> 
>>+static void input_ff_timer(unsigned long timer_data)
>>+{
>>+	struct input_dev *dev = (struct input_dev *) timer_data;
>>+	struct ff_device *ff = dev->ff;
>>+	struct ff_effect effect;
>>+	int i;
>>+	unsigned long flags;
>>+	int effects_pending;
>>+	unsigned long effect_handled[NBITS(FF_EFFECTS_MAX)];
> 
> 
> 
> hmmm stack space?
> 

I count 76 bytes (x86), is that too much?

>>+		} else {
>>+			ret = -ENOSYS;
> 
> 
> that is almost always a wrong return value
> 

It's returned when the device is mem-capable but driver doesn't
implement set_gain() but sets FF_GAIN or when driver doesn't implement
set_autocenter() but sets FF_AUTOCENTER. But yes, if that happens, it's
a driver bug, so maybe this is not correct use for -ENOSYS. Probably
there should be BUG() too here.

>>+	if (test_bit(FF_CONSTANT, dev->ff->flags))
>>+		set_bit(FF_CONSTANT, dev->ffbit);
>>+	if (test_bit(FF_SPRING, dev->ff->flags))
>>+		set_bit(FF_SPRING, dev->ffbit);
>>+	if (test_bit(FF_FRICTION, dev->ff->flags))
>>+		set_bit(FF_FRICTION, dev->ffbit);
>>+	if (test_bit(FF_DAMPER, dev->ff->flags))
>>+		set_bit(FF_DAMPER, dev->ffbit);
>>+	if (test_bit(FF_INERTIA, dev->ff->flags))
> 
> 
> are you really sure you need atomic set_bit()'s here??
> if so.. I think you have a race ;)

Well, I'm not. Is there an alternative?


-- 
Anssi Hannula

Re: + input-new-force-feedback-interface.patch added to -mm tree

[Arjan van de Ven]

On Mon, 2006-05-22 at 19:12 +0300, Anssi Hannula wrote:
> Arjan van de Ven wrote:
> > On Wed, 2006-05-17 at 21:46 -0700, akpm@osdl.org wrote:
> > 
> >>+
> >>+#ifdef DEBUG
> >>+#define debug(format, arg...) printk(KERN_DEBUG "ff-effects: " format "\n" , ## arg)
> >>+#else
> >>+#define debug(format, arg...) do {} while (0)
> >>+#endif
> > 
> > 
> > please just use the existing prdebug() thing for this, no need to invent
> > your own ;)
> 
> Couldn't find any info on that one, are you sure you spelled it correctly?


> 
> > 
#ifdef DEBUG
#define pr_debug(fmt,arg...) \
        printk(KERN_DEBUG fmt,##arg)
#else
#define pr_debug(fmt,arg...) \
        do { } while (0)
#endif


in linux/kernel.h


> Well, I don't know. When should EXPORT_SYMBOLs be EXPORT_SYMBOL_GPLs?
> 

if it's linux-only code or otherwise very internal, _GPL is usually the
right thing

> > hmmm stack space?
> > 
> 
> I count 76 bytes (x86), is that too much?

it's sort of ok, just make sure it doesn't grow more

> > 
> > that is almost always a wrong return value
> > 
> 
> It's returned when the device is mem-capable but driver doesn't
> implement set_gain() but sets FF_GAIN or when driver doesn't implement
> set_autocenter() but sets FF_AUTOCENTER. But yes, if that happens, it's
> a driver bug, so maybe this is not correct use for -ENOSYS. Probably
> there should be BUG() too here.

-EINVAL or so would be better; -ENOSYS basically is "not implemented
system call", which is totally off topic in this context

> 
> >>+	if (test_bit(FF_CONSTANT, dev->ff->flags))
> >>+		set_bit(FF_CONSTANT, dev->ffbit);
> >>+	if (test_bit(FF_SPRING, dev->ff->flags))
> >>+		set_bit(FF_SPRING, dev->ffbit);
> >>+	if (test_bit(FF_FRICTION, dev->ff->flags))
> >>+		set_bit(FF_FRICTION, dev->ffbit);
> >>+	if (test_bit(FF_DAMPER, dev->ff->flags))
> >>+		set_bit(FF_DAMPER, dev->ffbit);
> >>+	if (test_bit(FF_INERTIA, dev->ff->flags))
> > 
> > 
> > are you really sure you need atomic set_bit()'s here??
> > if so.. I think you have a race ;)
> 
> Well, I'm not. Is there an alternative?

__set_bit() is not atomic, and thus a lot faster if you don't need
atomic behavior..