Re: [PATCH v6 2/3] OPP: Add support for bandwidth OPP tables

From: Viresh Kumar
Date: Wed Jan 08 2020 - 23:31:19 EST


On 08-01-20, 16:58, Saravana Kannan wrote:
> On Wed, Jan 8, 2020 at 2:53 AM Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
> >
> > On 06-12-19, 16:24, Saravana Kannan wrote:
> > > Not all devices quantify their performance points in terms of frequency.
> > > Devices like interconnects quantify their performance points in terms of
> > > bandwidth. We need a way to represent these bandwidth levels in OPP. So,
> > > add support for parsing bandwidth OPPs from DT.
> > >
> > > Signed-off-by: Saravana Kannan <saravanak@xxxxxxxxxx>
> > > ---
> > > drivers/opp/core.c | 15 +++++++++--
> > > drivers/opp/of.c | 63 ++++++++++++++++++++++++++++++++--------------
> > > drivers/opp/opp.h | 5 ++++
> > > 3 files changed, 62 insertions(+), 21 deletions(-)
> > >
> > > diff --git a/drivers/opp/core.c b/drivers/opp/core.c
> > > index be7a7d332332..c79bbfac7289 100644
> > > --- a/drivers/opp/core.c
> > > +++ b/drivers/opp/core.c
> > > @@ -1282,11 +1282,21 @@ static bool _opp_supported_by_regulators(struct dev_pm_opp *opp,
> > > return true;
> > > }
> > >
> > > +int opp_compare_key(struct dev_pm_opp *opp1, struct dev_pm_opp *opp2)
> > > +{
> > > + if (opp1->rate != opp2->rate)
> > > + return opp1->rate < opp2->rate ? -1 : 1;
> > > + if (opp1->peak_bw != opp2->peak_bw)
> > > + return opp1->peak_bw < opp2->peak_bw ? -1 : 1;
> >
> > Please also add level here.
>
> I can, but I vaguely remember finding opp-levels could have
> duplicates? Am I wrong? If so I can add the opp-level comparison too.

No they can't have duplicates.

> > > + return 0;
> > > +}
> > > +
> > > static int _opp_is_duplicate(struct device *dev, struct dev_pm_opp *new_opp,
> > > struct opp_table *opp_table,
> > > struct list_head **head)
> > > {
> > > struct dev_pm_opp *opp;
> > > + int opp_cmp;
> > >
> > > /*
> > > * Insert new OPP in order of increasing frequency and discard if
> > > @@ -1297,12 +1307,13 @@ static int _opp_is_duplicate(struct device *dev, struct dev_pm_opp *new_opp,
> > > * loop.
> > > */
> > > list_for_each_entry(opp, &opp_table->opp_list, node) {
> > > - if (new_opp->rate > opp->rate) {
> > > + opp_cmp = opp_compare_key(new_opp, opp);
> > > + if (opp_cmp > 0) {
> > > *head = &opp->node;
> > > continue;
> > > }
> > >
> > > - if (new_opp->rate < opp->rate)
> > > + if (opp_cmp < 0)
> > > return 0;
> > >
> > > /* Duplicate OPPs */
> > > diff --git a/drivers/opp/of.c b/drivers/opp/of.c
> > > index 1cbb58240b80..b565da5a2b1f 100644
> > > --- a/drivers/opp/of.c
> > > +++ b/drivers/opp/of.c
> > > @@ -521,6 +521,44 @@ void dev_pm_opp_of_remove_table(struct device *dev)
> > > }
> > > EXPORT_SYMBOL_GPL(dev_pm_opp_of_remove_table);
> > >
> > > +static int _read_opp_key(struct dev_pm_opp *new_opp, struct device_node *np,
> > > + bool *rate_not_available)
> > > +{
> > > + int ret;
> > > + u64 rate;
> > > + u32 bw;
> > > +
> > > + ret = of_property_read_u64(np, "opp-hz", &rate);
> > > + if (!ret) {
> > > + /*
> > > + * Rate is defined as an unsigned long in clk API, and so
> > > + * casting explicitly to its type. Must be fixed once rate is 64
> > > + * bit guaranteed in clk API.
> > > + */
> > > + new_opp->rate = (unsigned long)rate;
> > > + goto out;
> > > + }
> > > +
> > > + ret = of_property_read_u32(np, "opp-peak-kBps", &bw);
> > > + if (!ret) {
> > > + new_opp->peak_bw = bw;
> > > +
> > > + if (!of_property_read_u32(np, "opp-avg-kBps", &bw))
> > > + new_opp->avg_bw = bw;
> >
> > Maybe
> > of_property_read_u32(np, "opp-avg-kBps", &new_opp->avg_bw);
> >
> > and same for opp-peak-kbps as well.
>
> But those are not u32. Is it always safe to directly read into it
> across all endian-ness and unsigned int sizes? I get tripped up by
> that occasionally.

It may not be safe.

> > > + }
> > > +
> > > +out:
> > > + *rate_not_available = !!ret;
> > > + /*
> > > + * If ret is 0 at this point, we have already found a key. If we
> > > + * haven't found a key yet, then ret already has an error value. In
> > > + * either case, we don't need to update ret.
> > > + */
> > > + of_property_read_u32(np, "opp-level", &new_opp->level);
> >
> > Yes, it wasn't done earlier but we should do it now. Check level as
> > well and treat it as any other key.
> >
> > I think add a preparatory patch first which does all the cleanup
> > before bandwidth thing is added.
>
> Ah come on man! You are making this too painful. It's okay to add a
> few more error checks as part of implementing a new feature. Please
> don't make me add more patches before this.

It will only make your life easier and not painful in my opinion as
the reviews are getting mixed/confused between the new things and the
old fields right now. With a separate patch introducing just the
bandwidth part, it will get reviewed in maximum 1-2 versions, else you
will keep updating the unrelated patch and I will keep reviewing it as
it is all a single patch.

It is always suggested to break patches into the smallest possible
meaningful separate things you want to achieve. You are introducing
something here and adding cleanup to that.

> > > +
> > > + return ret;
> > > +}
> > > +
> > > /**
> > > * _opp_add_static_v2() - Allocate static OPPs (As per 'v2' DT bindings)
> > > * @opp_table: OPP table
> > > @@ -558,26 +596,12 @@ static struct dev_pm_opp *_opp_add_static_v2(struct opp_table *opp_table,
> > > if (!new_opp)
> > > return ERR_PTR(-ENOMEM);
> > >
> > > - ret = of_property_read_u64(np, "opp-hz", &rate);
> > > - if (ret < 0) {
> > > - /* "opp-hz" is optional for devices like power domains. */
> > > - if (!opp_table->is_genpd) {
> > > - dev_err(dev, "%s: opp-hz not found\n", __func__);
> > > - goto free_opp;
> > > - }
> > > -
> > > - rate_not_available = true;
> > > - } else {
> > > - /*
> > > - * Rate is defined as an unsigned long in clk API, and so
> > > - * casting explicitly to its type. Must be fixed once rate is 64
> > > - * bit guaranteed in clk API.
> > > - */
> > > - new_opp->rate = (unsigned long)rate;
> > > + ret = _read_opp_key(new_opp, np, &rate_not_available);
> > > + if (ret) {
> > > + dev_err(dev, "%s: opp key field not found\n", __func__);
> > > + goto free_opp;
> > > }
> > >
> > > - of_property_read_u32(np, "opp-level", &new_opp->level);
> > > -
> > > /* Check if the OPP supports hardware's hierarchy of versions or not */
> > > if (!_opp_is_supported(dev, opp_table, np)) {
> > > dev_dbg(dev, "OPP not supported by hardware: %llu\n", rate);
> > > @@ -616,7 +640,8 @@ static struct dev_pm_opp *_opp_add_static_v2(struct opp_table *opp_table,
> > > if (of_property_read_bool(np, "opp-suspend")) {
> > > if (opp_table->suspend_opp) {
> > > /* Pick the OPP with higher rate as suspend OPP */
> > > - if (new_opp->rate > opp_table->suspend_opp->rate) {
> > > + if (opp_compare_key(new_opp,
> > > + opp_table->suspend_opp) > 1) {
> >
> > Maybe leave this place as is as we never want to compare anything else
> > but rate.
>
> We do want to support suspend bandwidth.

Yeah, I understood that from a later patch.

> So I think I should have this
> fix here so it works in general. Also, why not suspend opp-level?

Because we don't want/need to set a specific value to the
voltage-corners during suspend directly from the PM domain driver.
This should happen automatically via a call from the underlying
cpufreq or device driver.

And that's what I expected out of the interconnect thing. For example,
say during suspend you put the interconnect or PM domains to a low
bandwidth/level value, while the underlying device driver (which uses
the interconnect or domain) never requested for it. Who will be
responsible to restore the value during resume as we would be out of
sync here.

--
viresh