On Tue, Oct 21, 2014 at 01:02:28AM +0100, Doug Anderson wrote: > Mark, > > On Mon, Oct 20, 2014 at 3:41 PM, Mark Rutland <mark.rutland@xxxxxxx> wrote: > > On Mon, Oct 20, 2014 at 08:26:55PM +0100, Doug Anderson wrote: > >> Mark, > >> > >> On Mon, Oct 20, 2014 at 11:41 AM, Mark Rutland <mark.rutland@xxxxxxx> wrote: > >> > On Mon, Oct 20, 2014 at 04:31:18PM +0100, dinguyen@xxxxxxxxxxxxxxxxxxxxx wrote: > >> >> From: Dinh Nguyen <dinguyen@xxxxxxxxxxxxxxxxxxxxx> > >> >> > >> >> Without this patch, the booting the SOCFPGA platform would hang at the > >> >> SDMMC driver loading. There were 2 patches that caused this to happen: > >> >> > >> >> - Patch 9795a846e10 "mmc: dw_mmc: remove dw_mci_of_cd_gpio/wp_gpio()" removed > >> >> looking for "cd-gpios", since mmc_of_parse was getting called. > >> >> - Patch 3cf890fc42b "mmc: dw_mmc: Pass back errors from mmc_of_parse()" would > >> >> hang the system at the SDMMC driver loading. > >> > > >> > Regardless of which patches caused the issue, the existing DTB should > >> > continue to function. This is a kernel bug, not a DTB bug. > >> > >> Right. The kernel bug is that there is no "dtb fixup" stage of the > >> kernel to fix up old dtbs with this dtb bug. > >> > >> Said another way: > >> > >> 1. The old dtb was (possibly) not specifying the cd-gpio properly. > >> > >> 2. The kernel had a bug where it was ignoring that error. Things may > >> have been working because of some other side effect (maybe polling was > >> working). > >> > >> 3. If we fix the kernel bug, what should we do? The only sensible > >> thing (if we need to support old DTB with no changes) is to add a DTB > >> fixup stage. > >> > >> ...or did someone add that stage and I missed it? > > > > Unfortunately, we have no generic DTB fixup stage currently. > > Right. ...and that's the bug. > > I think we may need to modify the general inclination to respond to > dts change requests with "the DTS can't have a bug in it". DTS files > can indeed have bugs in it. In this case the dts file was claiming > that the card detect GPIO was a GPIO on a controller that the same dts > claimed was "disabled". If that's not a bug in the DTS I'm not sure > what it is. While it's unusual, it's not necessarily a bug in general-- the link might be true (i.e. that particular GPIO might be attached to the CD line), but for some reason the GPIO controller is unusable on a particular board. Perhaps we need to distinguish not ready yet from will never be ready -- at least for provider nodes with status = "disabled" that should be obvious. That said, this was not an intentional property of this DTB. > There are all sorts of broken ways that we could work around this in > the driver. We could pretend that EPROBE_DEFER really meant "I'm all > good". We could add a special case for this particular board in > dw_mmc (do we check the overall device tree compatible string?). We > could do all sorts of hacks. None of them are right. The "right" > behavior if we really care about maintaining compatbility with old DTB > files is to add a fixup stage for this particular broken board. While a DTB fixup stage is something which we are likely to need at some point, it comes with its own (rather large) cost. I disagree that DTB fixup is the one and only way of handling this kind of issue. > Given that no such fixup stage exists, if someone really wants old DTB > files to work then we should add one. Perhaps. In this case I guess this comes down to whether socfpga users are happy to update their DTBs. Thanks, Mark. -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
