Re: [PATCH] mfd: omap-usb-tll: check clk_prepare return code

[Andreas Kemnade <andreas@kemnade.info>]

Am Thu, 5 Dec 2024 22:54:05 +0100
schrieb Karol P <karprzy7@gmail.com>:

> On Tue, 26 Nov 2024 at 23:06, Andreas Kemnade <andreas@kemnade.info> wrote:
> >
> > Am Tue, 19 Nov 2024 16:16:42 +0200
> > schrieb Roger Quadros <rogerq@kernel.org>:
> >  
> > > On 19/11/2024 15:56, Andreas Kemnade wrote:  
> > > > Am Tue, 19 Nov 2024 15:10:23 +0200
> > > > schrieb Roger Quadros <rogerq@kernel.org>:
> > > >  
> > > >> Hi,
> > > >>
> > > >> On 13/11/2024 23:16, Karol Przybylski wrote:  
> > > >>> clk_prepare() is called in usbtll_omap_probe to fill clk array.
> > > >>> Return code is not checked, leaving possible error condition unhandled.
> > > >>>
> > > >>> Added variable to hold return value from clk_prepare() and dev_dbg statement
> > > >>> when it's not successful.
> > > >>>
> > > >>> Found in coverity scan, CID 1594680
> > > >>>
> > > >>> Signed-off-by: Karol Przybylski <karprzy7@gmail.com>
> > > >>> ---
> > > >>>  drivers/mfd/omap-usb-tll.c | 11 +++++++----
> > > >>>  1 file changed, 7 insertions(+), 4 deletions(-)
> > > >>>
> > > >>> diff --git a/drivers/mfd/omap-usb-tll.c b/drivers/mfd/omap-usb-tll.c
> > > >>> index 0f7fdb99c809..2e9319ee1b74 100644
> > > >>> --- a/drivers/mfd/omap-usb-tll.c
> > > >>> +++ b/drivers/mfd/omap-usb-tll.c
> > > >>> @@ -202,7 +202,7 @@ static int usbtll_omap_probe(struct platform_device *pdev)
> > > >>>   struct device                           *dev =  &pdev->dev;
> > > >>>   struct usbtll_omap                      *tll;
> > > >>>   void __iomem                            *base;
> > > >>> - int                                     i, nch, ver;
> > > >>> + int                                     i, nch, ver, err;
> > > >>>
> > > >>>   dev_dbg(dev, "starting TI HSUSB TLL Controller\n");
> > > >>>
> > > >>> @@ -248,10 +248,13 @@ static int usbtll_omap_probe(struct platform_device *pdev)
> > > >>>                                   "usb_tll_hs_usb_ch%d_clk", i);
> > > >>>           tll->ch_clk[i] = clk_get(dev, clkname);
> > > >>>
> > > >>> -         if (IS_ERR(tll->ch_clk[i]))
> > > >>> +         if (IS_ERR(tll->ch_clk[i])) {
> > > >>>                   dev_dbg(dev, "can't get clock : %s\n", clkname);  
> > > >
> > > > if you want dev_err() later, then why not here?  
> > >
> > > Because clk is optional. If it is not there then we should not complain.
> > > But if it is there then it needs to be enabled successfully.
> > >  
> > I guess you mean *prepared*, the clock is enabled later (with error
> > checking). But your reasoning makes sense.
> >  
> > > >  
> > > >>> -         else
> > > >>> -                 clk_prepare(tll->ch_clk[i]);
> > > >>> +         } else {
> > > >>> +                 err = clk_prepare(tll->ch_clk[i]);
> > > >>> +                 if (err)
> > > >>> +                         dev_dbg(dev, "clock prepare error for: %s\n", clkname);  
> > > >>
> > > >> dev_err()?
> > > >>  
> > > > So why do you want a different return handling here? (I doubt there is
> > > > any clock having a real prepare() involved here)
> > > >
> > > > As said in an earlier incarnation of this patch, the real question is
> > > > whether having partial clocks available is a valid operating scenario.
> > > > If yes, then the error should be ignored. If no, then bailing out early
> > > > is a good idea.  
> > >
> > > In the DT binding, clocks is optional. So if it doesn't exist it is not
> > > an error condition.
> > >  
> > > >
> > > > clk_prepare() errors are catched by failing clk_enable() later,
> > > > ch_clk[i] is checked later, too.
> > > >  
> > > >> I think we should return the error in this case.
> > > >> (after unpreparing the prepared clocks and clk_put())
> > > >>  
> > > > and pm_runtime_put_sync(dev)  
> >
> > which can probably be done before dealing with the clocks. It is only
> > needed for the register access.  
> 
> I'm fairly new to this subsystem and I'm trying to understand the
> conclusion. In the end, we should add dev_err() here after
> clk_prepare() with appropriate handling?
> 
we must make sure pm_runtime_put/get are paired and _put is called in
any case. Looking around a bit:
I think a good solution would be along this lines:
https://lore.kernel.org/linux-omap/34ab5f0b78c2869cc43797a72d6a2f40d9b246f3.camel@siemens.com/T/#u

using devm_clk_get_prepared() things can be simplified.

Regards,
Andreas