On Tue, Feb 23, 2021 at 10:17:59PM +0100, Krzysztof Wilczyński wrote:
> Hi Fabio,
>
> Thank you for sending the patch over!
>
> [...]
> > This fixes the following Coverity error:
> >
> > CID 1472841: Error handling issues (CHECKED_RETURN)
> > Calling "phy_power_on" without checking return value (as is done elsewhere 40 out of 50 times).
> > phy_power_on(ep->phy);
> > phy_init(ep->phy);
>
> This is good, however, you would need to wrap long lines, and that would
> make the message from Coverity harder to read, etc. Thus, it might be
> better to use the "Addresses-Coverity-ID" which is becoming a de-facto
> standard for referencing Coverity defects. Check the following for some
> examples:
>
> git log drivers/pci | grep 'Addresses-Coverity-ID:'
>
> [...]
> > + ret = phy_power_on(ep->phy);
> > + if (ret < 0)
> > + return ret;
>
> I wonder if you would also have to call phy_exit() here, even though
> eventually exynos_pcie_probe() would call it once the error propagates
> all the way up the call stack.
>
> Additionally, exynos_pcie_resume_noirq() does not do any error checking
> after calling exynos_pcie_host_init() and does not call phy_exit()
> either, and I am not sure if it should, though.
>
> See some comments below.
>
> > +
> > phy_init(ep->phy);
> [...]
>
> A small nit here. You can check for any non-zero return value, as
> anything would indicate an error here.
>
> I also have a suggestion. Would you also be interested in addressing
> two Coverity defects that were detected in exynos_pcie_host_init()?
>
> These would be the one you addressed here (CID 1472841) in this patch
> and the other would be:
>
> CID 1471267 (#1 of 1): Unchecked return value (CHECKED_RETURN)
>
> Which is about checking return value from phy_init() that is called
> immediately after phy_power_on() in exynos_pcie_host_init().
>
> The error propagates from exynos_pcie_host_init() as follows:
>
> struct exynos_pcie_host_ops{}
> .host_init = exynos_pcie_host_init
>
> exynos_pcie_probe() <-- phy_exit() called here if exynos_add_pcie_port() fails.
> exynos_add_pcie_port()
> dw_pcie_host_init()
> exynos_pcie_host_init() <-- phy_power_on() and phy_init() called here.
> dw_pcie_host_init()
> struct pcie_port{}
> struct dw_pcie_host_ops{}
> .host_init <-- exynos_pcie_host_init() called via struct exynos_pcie_host_ops{}.
>
> struct exynos_pcie_pm_ops{}
> .suspend_noirq = exynos_pcie_suspend_noirq
> .resume_noirq = exynos_pcie_resume_noirq
>
> exynos_pcie_resume_noirq()
> exynos_pcie_host_init() <-- called here, but without any error checking.
>
> Thus, we could handle propagating error from both the phy_power_on() and
> phy_init() in the same time, perhaps even in a single patch, or a small
> series.
>
> Also, since there is no error checking and/or handling that might be
> returned from exynos_pcie_host_init() in the exynos_pcie_resume_noirq()
> callback, then perhaps adding some error messages to be printed should
> something bad happens regarding power management. But this would
> becompletely optional as there there is also no error checking and
> handling in exynos_pcie_suspend_noirq() either.
Fabio, what's the plan with this patch ?
Lorenzo
