Re: [PATCH] clk: Add clk_set_parent debugfs node

From: Sam Protsenko
Date: Thu Oct 07 2021 - 09:40:16 EST


Hi Geert,

On Tue, 5 Oct 2021 at 13:43, Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
>
> Hi Sam,
>
> On Tue, Oct 5, 2021 at 12:11 PM Sam Protsenko
> <semen.protsenko@xxxxxxxxxx> wrote:
> > On Tue, 14 Sept 2021 at 17:19, Sam Protsenko <semen.protsenko@xxxxxxxxxx> wrote:
> > > Useful for testing mux clocks. One can write the index of the parent to
> > > set into clk_set_parent node, starting from 0. Example
> > >
> > > # cat clk_possible_parrents
> > > dout_shared0_div4 dout_shared1_div4
> > > # cat clk_parent
> > > dout_shared0_div4
> > > # echo 1 > clk_set_parent
> > > # cat clk_parent
> > > dout_shared1_div4
> > >
> > > Define CLOCK_ALLOW_WRITE_DEBUGFS in drivers/clk/clk.c in order to use
> > > this feature.
> > >
> > > Signed-off-by: Sam Protsenko <semen.protsenko@xxxxxxxxxx>
> > > ---
> >
> > + Adding more folks for review
> >
> > Guys, can you please review this one?
>
> Thanks for your patch!
>

Thanks for review! :)

> > > --- a/drivers/clk/clk.c
> > > +++ b/drivers/clk/clk.c
> > > @@ -3214,6 +3214,30 @@ static int current_parent_show(struct seq_file *s, void *data)
> > > }
> > > DEFINE_SHOW_ATTRIBUTE(current_parent);
> > >
> > > +#ifdef CLOCK_ALLOW_WRITE_DEBUGFS
> > > +static int clk_set_parent_set(void *data, u64 val)
>
> u64 is overkill, num_parents is u8.
>

u64 is required by simple_attr_open() signature, because
clk_set_parent_set() is being passed there as a parameter eventually
(via DEFINE_DEBUGFS_ATTRIBUTE()). But yeah, I'll use u8 when reworking
the code for using with existing 'clk_parent' file.

> > > +{
> > > + struct clk_core *core = data, *parent;
> > > + int ret;
> > > +
> > > + if (val >= core->num_parents)
> > > + return -EINVAL;
>
> clk_core_get_parent_by_index() called below already checks this.
>

Thanks, will remove this.

> > > +
> > > + parent = clk_core_get_parent_by_index(core, val);
> > > + if (IS_ERR_OR_NULL(parent))
> > > + return PTR_ERR(parent);

Also just noticed that this block is incorrect. I should've used just
'if (!parent)' here instead. I remember Russel King was raising the
question about removing that API for good, as too many people tend to
use that incorrectly, and now I can see why.

> > > +
> > > + clk_prepare_lock();
> > > + ret = clk_core_set_parent_nolock(core, parent);
> > > + clk_prepare_unlock();
> > > +
> > > + return ret;
> > > +}
> > > +
> > > +DEFINE_DEBUGFS_ATTRIBUTE(clk_set_parent_fops, NULL, clk_set_parent_set,
> > > + "%llu\n");
> > > +#endif
> > > +
> > > static int clk_duty_cycle_show(struct seq_file *s, void *data)
> > > {
> > > struct clk_core *core = s->private;
> > > @@ -3285,9 +3309,14 @@ static void clk_debug_create_one(struct clk_core *core, struct dentry *pdentry)
> > > debugfs_create_file("clk_parent", 0444, root, core,
> > > &current_parent_fops);
> > >
> > > - if (core->num_parents > 1)
> > > + if (core->num_parents > 1) {
> > > debugfs_create_file("clk_possible_parents", 0444, root, core,
> > > &possible_parents_fops);
> > > +#ifdef CLOCK_ALLOW_WRITE_DEBUGFS
> > > + debugfs_create_file("clk_set_parent", 0200, root, core,
> > > + &clk_set_parent_fops);
> > > +#endif
>
> Why add a new file, instead of making the existing "clk_parent" file
> writable, like is done for "clk_rate"?
> Yes, "clk_parent" prints a name, while you use a parent number, but
> I guess that can be fixed? Or even both can be accepted?
>

Ok, I'll merge that feature into existing 'clk_parent' file. At the
time I implemented this I was busy with something else, and use
existing code around as an example. But it's not too hard to do this
properly, by defining 'struct file_operations' manually, like it's
done for example in dwc3_lsp_write(). Will send v2 with fixes shortly.

> > > + }
> > >
> > > if (core->ops->debug_init)
> > > core->ops->debug_init(core->hw, core->dentry);
>
> Gr{oetje,eeting}s,
>
> Geert
>
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
>
> 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