Re: [PATCH v4 03/12] PCI: brcmstb: Use common error handling code in brcm_pcie_probe()

[Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>]

On Fri, Jul 26, 2024 at 02:34:54PM -0400, Jim Quinlan wrote:
> On Fri, Jul 26, 2024 at 1:04 AM Manivannan Sadhasivam
> <manivannan.sadhasivam@linaro.org> wrote:
> >
> > On Thu, Jul 25, 2024 at 03:45:59PM -0400, Jim Quinlan wrote:
> > > On Thu, Jul 25, 2024 at 12:31 AM Manivannan Sadhasivam
> > > <manivannan.sadhasivam@linaro.org> wrote:
> > > >
> > > > On Tue, Jul 16, 2024 at 05:31:18PM -0400, Jim Quinlan wrote:
> > > > > o Move the clk_prepare_enable() below the resource allocations.
> > > > > o Add a jump target (clk_out) so that a bit of exception handling can be
> > > > >   better reused at the end of this function implementation.
> > > > >
> > > > > Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com>
> > > > > Reviewed-by: Stanimir Varbanov <svarbanov@suse.de>
> > > > > Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
> > > > > ---
> > > > >  drivers/pci/controller/pcie-brcmstb.c | 29 +++++++++++++++------------
> > > > >  1 file changed, 16 insertions(+), 13 deletions(-)
> > > > >
> > > > > diff --git a/drivers/pci/controller/pcie-brcmstb.c b/drivers/pci/controller/pcie-brcmstb.c
> > > > > index c08683febdd4..c257434edc08 100644
> > > > > --- a/drivers/pci/controller/pcie-brcmstb.c
> > > > > +++ b/drivers/pci/controller/pcie-brcmstb.c
> > > > > @@ -1613,31 +1613,30 @@ static int brcm_pcie_probe(struct platform_device *pdev)
> > > > >
> > > > >       pcie->ssc = of_property_read_bool(np, "brcm,enable-ssc");
> > > > >
> > > > > -     ret = clk_prepare_enable(pcie->clk);
> > > > > -     if (ret) {
> > > > > -             dev_err(&pdev->dev, "could not enable clock\n");
> > > > > -             return ret;
> > > > > -     }
> > > > >       pcie->rescal = devm_reset_control_get_optional_shared(&pdev->dev, "rescal");
> > > > > -     if (IS_ERR(pcie->rescal)) {
> > > > > -             clk_disable_unprepare(pcie->clk);
> > > > > +     if (IS_ERR(pcie->rescal))
> > > > >               return PTR_ERR(pcie->rescal);
> > > > > -     }
> > > > > +
> > > > >       pcie->perst_reset = devm_reset_control_get_optional_exclusive(&pdev->dev, "perst");
> > > > > -     if (IS_ERR(pcie->perst_reset)) {
> > > > > -             clk_disable_unprepare(pcie->clk);
> > > > > +     if (IS_ERR(pcie->perst_reset))
> > > > >               return PTR_ERR(pcie->perst_reset);
> > > > > +
> > > > > +     ret = clk_prepare_enable(pcie->clk);
> > > > > +     if (ret) {
> > > > > +             dev_err(&pdev->dev, "could not enable clock\n");
> > > > > +             return ret;
> > > > >       }
> > > > >
> > > > >       ret = reset_control_reset(pcie->rescal);
> > > > > -     if (ret)
> > > > > +     if (ret) {
> > > > >               dev_err(&pdev->dev, "failed to deassert 'rescal'\n");
> > > > > +             goto clk_out;
> > > >
> > > > Please use a descriptive name for the err labels. Here this err path disables
> > > > and unprepares the clk, so use 'clk_disable_unprepare'.
> > > ack
> > > >
> > > > > +     }
> > > > >
> > > > >       ret = brcm_phy_start(pcie);
> > > > >       if (ret) {
> > > > >               reset_control_rearm(pcie->rescal);
> > > > > -             clk_disable_unprepare(pcie->clk);
> > > > > -             return ret;
> > > > > +             goto clk_out;
> > > > >       }
> > > > >
> > > > >       ret = brcm_pcie_setup(pcie);
> > > > > @@ -1676,6 +1675,10 @@ static int brcm_pcie_probe(struct platform_device *pdev)
> > > > >
> > > > >       return 0;
> > > > >
> > > > > +clk_out:
> > > > > +     clk_disable_unprepare(pcie->clk);
> > > > > +     return ret;
> > > > > +
> > > >
> > > > This is leaking the resources. Move this new label below 'fail'.
> > > What resources is it leaking?  At "clk_out" the return value will be negative
> > > and only managed resources have been allocated at that juncture.
> > >
> >
> > Right, but what about the err path below this one? If that path is taken, then
> > clks won't be released, right?
> No, that is the same situation.  The clock is originally allocated
> with "devm_clk_get_optional()", i.e. it is a managed resource.
>  If the probe fails, and it does in both of these error paths,
> Linux deallocates the newly formed device structure and all of its resources.
> Perhaps I am missing something?
> 

No, I missed the fact that __brcm_pcie_remove() is freeing all resources. But
grouping all release functions in a single helper and using it in multiple err
paths even when the err path need not release everything the helper is
releasing, warrants trouble.

- Mani

-- 
மணிவண்ணன் சதாசிவம்