> >> Shouldn't PWM_DEVICES select HAVE_PWM?
> >
> >
> > No not required, the entire concept is to remove HAVE_PWM and use
> PWM_CORE.
>
> Well in patch 4 you say that PWM_CORE is currently limited to ARM.
> Furthermore you
> change the pwm-backlight and pwm-led Kconfig entries to depend on
> HAVE_PWM ||
> PWM_CORE. Adding a select HAVE_PWM here would make those changes
> unnecessary.
HAVE_PWM is retained just because the mips pwm driver is not aligned with the pwm core driver.
On mips pwm driver aligning to the pwm core driver HAVE_PWM will be replaced by PWM_CORE.
> HAVE_PWM should be set, when pwm_* functions are available. When your
> pwm-core driver
> is selected they are available.
On applying this patch set pwm_* function will be exported in pwm_core driver and in mips pwm driver.
Since mips pwm driver is not aligned with the pwm core, HAVE_PWM is retained and removed in places where pwm drivers register to pwm core driver.
> > pwm_device will be passed to each and every pwm driver that are
> registered as client with pwm core.
> > The list consists of the registered pwm drivers and is to be handled
> by pwm core.
> > Why should each and every pwm driver get to know about the entire pwm
> driver list?
> Declare the list field to be private, by saying that it should only be
> touched by the
> core. Right now you allocate a rather small additional structure for
> each registered
> device. This could be easily be avoided by embedding the list field
> into the
> pwm_device struct.
The one that is being allocated in register is the pwm_device and this has to. Because each pwm driver will have their own data related to ops, pwm_id.
Also note that there exists an element "data" that points to the pwm device specific information. Hence this allocation is required.
> >>> + }
> >>> + pwm->pwm_dev = pwm_dev;
> >>> + list_add_tail(&pwm->list, &di->list);
> >>> + up_write(&pwm_list_lock);
> >>> +
> >> I guess you only need to lock the list when accessing the list and
> >> adding the new
> >> pwm_dev.
> >
> > Oops, thanks for pointing out, will implement this in the v2 patch.
Coming back to this, I guess the locking has to be done while traversing the list also, as my present pointer in the list my get over written by the time I add an element to list. Please let me know if I am wrong.
> >>> +struct pwm_ops {
> >>> + int (*pwm_config)(struct pwm_device *pwm, int duty_ns, int
> >> period_ns);
> >>> + int (*pwm_enable)(struct pwm_device *pwm);
> >>> + int (*pwm_disable)(struct pwm_device *pwm);
> >>> + char *name;
> >>> +};
> >>> +
> >> Shouldn't name be part of the pwm_device? That would allow the ops
> to
> >> be shared
> >> between different devices.
> > Good catch, the reason being that 2 or more devices can share the
> same ops and get registered to pwm core.
> > But the catch lies while identifying the pwm device while the clients
> are requesting for.
> > The pwm backlight will request the pwm driver by name. This is
> parameter that distinguishes among different pwm devices irrespective
> of same ops or not.
> Yes. And thats why it should go into the pwm_device struct itself.
>
> If an additional ops struct is allocated for each device anyway we
> would be better of
> embedding it directly into the device struct instead of just holding a
> pointer to it.
Yes ops structure will be allocated. But how can we get access to the ops structure of another driver?
And moreover two pwm driver sharing same ops ideally means a single pwm module. If not everything atleast the pwm registers of two different modules changes. So this scenario can never occur.
> >>> #endif /* __LINUX_PWM_H */
> >> It might be also a good idea to add a device class for pwm devices.
> > Sure, but can you please explain with an example the use case.
> >
> Well, for one it helps to keep data structured.
> And there would be functions to traverse all devices of a class, so you
> could get rid
> of your "di" list.
Sorry, I didn't get you can you please elaborate more?
Thanks and Regards,
Arun R Murthy
-------------
