Em Fri, 4 Sep 2020 13:45:23 +0100 Mark Brown <[email protected]> escreveu:
> On Fri, Sep 04, 2020 at 02:38:48PM +0200, Mauro Carvalho Chehab wrote: > > Em Fri, 4 Sep 2020 13:23:03 +0100 > > Mark Brown <[email protected]> escreveu: > > > > > On Fri, Sep 04, 2020 at 12:23:31PM +0200, Mauro Carvalho Chehab wrote: > > > > > > > + regulator = devm_regulator_get_optional(&pdev->dev, "hub-vdd"); > > > > + if (IS_ERR(regulator)) { > > > > + if (PTR_ERR(regulator) == -EPROBE_DEFER) { > > > > + dev_info(&pdev->dev, > > > > + "waiting for hub-vdd-supply to be > > > > probed\n"); > > > > + return PTR_ERR(regulator); > > > > + } > > > > + > > > > + /* let it fall back to regulator dummy */ > > > > + regulator = devm_regulator_get(&pdev->dev, "hub-vdd"); > > > > + if (IS_ERR(regulator)) { > > > > + dev_err(&pdev->dev, > > > > + "get hub-vdd-supply failed with error > > > > %ld\n", > > > > + PTR_ERR(regulator)); > > > > + return PTR_ERR(regulator); > > > > + } > > > > + } > > > > This seems weird - if the supply is non-optional why is the code trying > > > with devm_regulator_get_optional()? Just use normal get directly. > > > That's meant to avoid problems with EPROBE_DEFER. > > Which problems and in what way does it avoid them? > > > See, Hikey 970 need to initialize 4 drivers for the regulators: > > SPMI core, SPMI bus controller, MFD and regulator. This can take > > some time. So, a first call to *regulator_get() may return > > EPROBE_DEFER, specially if both regulator drivers and USB HUB > > are builtin. > > This is totally normal and works fine with normal regulator_get(). When I was porting the drm driver (first from OOT Kernel 4.9 to 4.19.5, and then from 4.19.5 to 5.7), I'm pretty sure I got several of those messages: "supply %s not found, using dummy regulator\n" due to EPROBE_DEFER. I remember I ended checked the implementation of the regulator code on that time. Yet, looking at the current implementation of _regulator_get(): rdev = regulator_dev_lookup(dev, id); if (IS_ERR(rdev)) { ret = PTR_ERR(rdev); I see that devm_regulator_get() will do the right thing with EPROBE_DEFER. So, I'll replace it at the driver. Thanks for the review! Thanks, Mauro
