On 15.09.2015 17:17, Sjoerd Simons wrote: > On Tue, 2015-09-15 at 15:50 +0900, Krzysztof Kozlowski wrote: >> 2015-09-14 17:35 GMT+09:00 Riku Voipio <riku.voipio@xxxxxxxxxx>: >>> On 5 June 2015 at 15:45, Arnd Bergmann <arnd@xxxxxxxx> wrote: >>>> On Thursday 04 June 2015 10:47:07 Kevin Hilman wrote: >>>>> >>>>>> But I wonder why is not working, shouldn't the driver defer >>>>>> and >>>>>> be probed again once the PHY driver probe succeeds? >>>>> >>>>> Yeah, I'm not sure why that isn't working, and didn't look into >>>>> it. >>>>> >>>>> FWIW, the same problem happens when both are modules. If you >>>>> modprobe >>>>> usb3503 first, then the phy, it doesn't work. You have to load >>>>> the phy >>>>> before the usb3503. >>> >>>> The driver does not try to get a reference to the phy, and it >>>> does >>>> not return -EPROBE_DEFER in any circumstance, so I assume it just >>>> runs into an error condition on the first probe and does not >>>> try again. >>> >>>> I don't really understand why the driver registers both an >>>> i2c_driver >>>> and a platform_driver, or if that is required, but it may also >>>> complicate getting deferred probing to work here. >>> >>> Is someone looking into fixing it? >> >> Fixing what? The PHY issue? The driver not supporting deferred probe? >> >> As for module vs builtin, this is somehow orthogonal for me. >> Although modules are preferred on multi_v7 but in case of >> boot-essential drivers this should not be a requirement. I also don't >> use initrd for network boot... however my root is on MMC. Regardless >> if of that I would expect nfsroot to be working on multi_v7 kernel. > > When posting a set of multi_v7 config changes recently to improve how > we support RockChip, Thierry argued the case for building things as > modules as much as possible (even if that means needing an initramfs to > complete boot)[0].. > > You're arguing the exact oposite here, while I can see the points in > both arguments (though i'm leaning to agreeing with Thierry) it would > be nice to work out a common policy here as multi_v7 seems to be a > rather big mismatch currently. Not entirely the exact opposite. Opposite only for stuff important for booting. I agree: put into modules as much as possible. But the difference is in meaning of "possible". It is possible to network boot with network rootfs when the adapter is a module. But it is not possible to do it without initrd. :) Anyway I agree that conclusion should be made a standard (or policy) so other driver should be converted to 'm' or 'y'. > Fwiw, looking at the arm64 defconfig it currently has everything built > in. Which isn't too bad as there aren't that many arm64 boards in > mainline yet, but I bet it's going to run into similar issues at some > point. Indeed, that is the easiest option for development... until Image grows out of device partition. Best regards, Krzysztof -- To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html