On Wed, Nov 13, 2024 at 03:47:16PM +0100, Caleb Connolly wrote: > > > On 13/11/2024 15:24, Tom Rini wrote: > > On Wed, Nov 13, 2024 at 06:30:08AM +0100, Caleb Connolly wrote: > > > >> It may be the case that MMC support is enabled even though the board > >> we're booting on doesn't have any MMC devices. Move the print over to > >> the print_mmc_devices() function where we can only print it if we > >> actually have MMC devices. > >> > >> Signed-off-by: Caleb Connolly <caleb.conno...@linaro.org> > > > > I'm not sure I like this. What we do / don't find on startup is part of > > the not-exactly-API. It's true that if we don't print an MMC line at > > all, and we should have MMC, the user (and any scripts that parse > > console output) but now we're also increasing the code size a little bit > > too. I can be convinced this is a good idea, but I'm not there yet. > > Hmm, fair enough. I'll offer some more context, maybe there's a smarter > approach here I'm not seeing. > > The main place this shows up is on Qualcomm boards. Since all Qualcomm > armv8 targets are supported with qcom_defconfig (just by adjusting which > DTB is used), we can't know at build time whether the board has MMC. > > I guess my thinking behind this patch comes from a bigger picture desire > to get UFS and MMC more aligned. The number of devices with UFS is > definitely going up, and I would argue that U-Boot's inconsistent > treatment of these two storage classes (obviously a result of their > relative age and support in the codebase) is really unintuitive and > weird for users (nevermind that the "scsi" command is used for UFS > devices, cute though it is). > > I'm really wary to open this whole can of worms, since I guess it would > require some larger efforts and collaboration to fix. But maybe this > patch (or one like it) would be better suited in the context of some > larger effort to unify storage backends?
I get it now, thanks. And yeah, this is part of our inconsistency in printing information. I'd rather leave things alone here (assuming the MMC: printout is reasonable in the case of no devices, similar to how the Net: printout is reasonable in that case) and wait for a longer term / high level solution wherein for example as we go down the device tree we give some consistent level of information for everything (and more or less info depending on verbosity configured perhaps). -- Tom
signature.asc
Description: PGP signature
