Re: [PATCH v5] reset: add support for non-DT systems

From: Philipp Zabel
Date: Wed Feb 28 2018 - 04:17:00 EST


On Tue, 2018-02-27 at 19:07 +0100, Bartosz Golaszewski wrote:
> 2018-02-27 17:10 GMT+01:00 Philipp Zabel <p.zabel@xxxxxxxxxxxxxx>:
> > Hi Bartosz,
> >
> > thank you for the update.
> >
> > On Fri, 2018-02-23 at 12:39 +0100, Bartosz Golaszewski wrote:
> > > From: Bartosz Golaszewski <bgolaszewski@xxxxxxxxxxxx>
> > >
> > > The reset framework only supports device-tree. There are some platforms
> > > however, which need to use it even in legacy, board-file based mode.
> > >
> > > An example of such architecture is the DaVinci family of SoCs which
> > > supports both device tree and legacy boot modes and we don't want to
> > > introduce any regressions.
> > >
> > > We're currently working on converting the platform from its hand-crafted
> > > clock API to using the common clock framework. Part of the overhaul will
> > > be representing the chip's power sleep controller's reset lines using
> > > the reset framework.
> > >
> > > This changeset extends the core reset code with a new reset lookup
> > > entry structure. It contains data allowing the reset core to associate
> > > reset lines with devices by comparing the dev_id and con_id strings.
> > >
> > > It also provides a function allowing drivers to register lookup entries
> > > with the framework.
> > >
> > > The new lookup function is only called as a fallback in case the
> > > of_node field is NULL and doesn't change anything for current users.
> > >
> > > Tested with a dummy reset driver with several lookup entries.
> > >
> > > An example lookup table registration from a driver can be found below:
> > >
> > > static struct reset_control_lookup foobar_reset_lookup[] = {
> > > RESET_LOOKUP("foo.0", "foo", 15),
> > > RESET_LOOKUP("bar.0", NULL, 5),
> > > };
> > >
> > > foobar_probe()
> > > {
> > > ...
> > >
> > > reset_controller_add_lookup(&rcdev, foobar_reset_lookup,
> > > ARRAY_SIZE(foobar_reset_lookup));
> > >
> > > ...
> > > }
> > >
> > > Cc: Sekhar Nori <nsekhar@xxxxxx>
> > > Cc: Kevin Hilman <khilman@xxxxxxxxxxxx>
> > > Cc: David Lechner <david@xxxxxxxxxxxxxx>
> > > Signed-off-by: Bartosz Golaszewski <bgolaszewski@xxxxxxxxxxxx>
> > > ---
> > > v1 -> v2:
> > > - renamed the new function to __reset_control_get_from_lookup()
> > > - added a missing break; when a matching entry is found
> > > - rearranged the code in __reset_control_get() - we can no longer get to the
> > > return at the bottom, so remove it and return from
> > > __reset_control_get_from_lookup() if __of_reset_control_get() fails
> > > - return -ENOENT from reset_contol_get() if we can't find a matching entry,
> > > prevously returned -EINVAL referred to the fact that we passed a device
> > > without the of_node which is no longer an error condition
> > > - add a comment about needing a sentinel in the lookup table
> > >
> > > v2 -> v3:
> > > - added the reset id number field to the lookup struct so that we don't need
> > > to rely on the array index
> > >
> > > v3 -> v4:
> > > - separated the driver and lookup table registration logic by adding a
> > > function meant to be called by machine-specific code that adds a lookup
> > > table to the internal list
> > > - the correct reset controller is now found by first finding the lookup
> > > table associated with it, then finding the actual reset controller by
> > > the associated device
> > >
> > > v4 -> v5:
> > > - since the first user of this will be the davinci clk driver and it
> > > already registers clock lookup from within the driver code - allow
> > > drivers to register lookups with the assumption that the code can be
> > > extended to make it possible to register entries from machine code as
> > > well
> >
> > How do you imagine this may be extended? By adding an rddev_devid field
> > to the lookup, similarly to the pwm_lookup?
> > I suppose reset_controller_add_lookup could then be called with a NULL
> > rcdev to register lookups by id.
> >
>
> Yes, this is what I was thinking about more or less.

Ok. I just want to avoid having to change all users when somebody needs
that functionality, even though hopefully there won't be that many.

> > > - simplify the code - only expose a single lookup structure and a simply
> > > registration function
> > > - add the RESET_LOOKUP macro for brevity
> > >
> > > drivers/reset/core.c | 65 +++++++++++++++++++++++++++++++++++++++-
> > > include/linux/reset-controller.h | 28 +++++++++++++++++
> > > 2 files changed, 92 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/reset/core.c b/drivers/reset/core.c
> > > index da4292e9de97..75e54a05147a 100644
> > > --- a/drivers/reset/core.c
> > > +++ b/drivers/reset/core.c
> > > @@ -23,6 +23,9 @@
> > > static DEFINE_MUTEX(reset_list_mutex);
> > > static LIST_HEAD(reset_controller_list);
> > >
> > > +static DEFINE_MUTEX(reset_lookup_mutex);
> > > +static LIST_HEAD(reset_lookup_list);
> > > +
> > > /**
> > > * struct reset_control - a reset control
> > > * @rcdev: a pointer to the reset controller device
> > > @@ -148,6 +151,30 @@ int devm_reset_controller_register(struct device *dev,
> > > }
> > > EXPORT_SYMBOL_GPL(devm_reset_controller_register);
> > >
> > > +/**
> > > + * reset_controller_add_lookup - register a set of lookup entries
> > > + * @rcdev: initialized reset controller device owning the reset line
> > > + * @lookup: array of reset lookup entries
> > > + * @num_entries: number of entries in the lookup array
> > > + */
> > > +void reset_controller_add_lookup(struct reset_controller_dev *rcdev,
> > > + struct reset_control_lookup *lookup,
> > > + unsigned int num_entries)
> > > +{
> > > + struct reset_control_lookup *entry;
> > > + unsigned int i;
> > > +
> > > + mutex_lock(&reset_lookup_mutex);
> > > + for (i = 0; i < num_entries; i++) {
> > > + entry = &lookup[i];
> > > +
> > > + entry->rcdev = rcdev;
> > > + list_add_tail(&entry->list, &reset_lookup_list);
> > > + }
> > > + mutex_unlock(&reset_lookup_mutex);
> > > +}
> > > +EXPORT_SYMBOL_GPL(reset_controller_add_lookup);
> > > +
> >
> > What about reset_controller_remove_lookup?
>
> While I understand that we have those for phy or gpio, I don't really
> think they're that useful. Not many users do it anyway. Can we add it
> once it's actually needed?

Agreed.

> >
> > > static inline struct reset_control_array *
> > > rstc_to_array(struct reset_control *rstc) {
> > > return container_of(rstc, struct reset_control_array, base);
> > > @@ -493,6 +520,42 @@ struct reset_control *__of_reset_control_get(struct device_node *node,
> > > }
> > > EXPORT_SYMBOL_GPL(__of_reset_control_get);
> > >
> > > +static struct reset_control *
> > > +__reset_control_get_from_lookup(struct device *dev, const char *con_id,
> > > + bool shared, bool optional)
> > > +{
> > > + const struct reset_control_lookup *lookup;
> > > + const char *dev_id = dev_name(dev);
> > > + struct reset_control *rstc = NULL;
> > > +
> > > + if (!dev)
> > > + return ERR_PTR(-EINVAL);
> > > +
> > > + mutex_lock(&reset_lookup_mutex);
> > > +
> > > + list_for_each_entry(lookup, &reset_lookup_list, list) {
> > > + if (strcmp(lookup->dev_id, dev_id))
> >
> > This will crash if (lookup->dev_id == NULL).
> >
>
> We're stating explicitly that dev_id is required in the kernel doc but
> if you insist, I can add a check for that.

This isn't such a hot path that we couldn't afford the NULL pointer
check, but given that it is documented as required, maybe better let
reset_controller_add_lookup check for it and never add malformed lookups
to the list in the first place.

> > > + continue;
> > > +
> > > + if ((!con_id && !lookup->con_id) ||
> > > + !strcmp(con_id, lookup->con_id)) {
> >
> > This will crash if only one of con_id or lookup->con_id is NULL.
> >
> > Call strcmp only if both strings are valid.
> >
>
> Indeed, will fix it.

regards
Philipp