On 17 June 2015 at 18:25, Fabio Estevam <festevam@xxxxxxxxx> wrote: > Hi Ulf, > > On Tue, Jun 16, 2015 at 6:31 AM, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote: > >> I had a closer look, should have done that before. >> >> I think we need to decide what "no-1-8-v" should be used for. >> Currently the DT documentation is unclear and depending on what SDHCI >> variant, the binding has different purpose. It's a mess! > > Thanks for your feedback. > >> >> sdhci-pltfm + sdhci-esdhc-imx use the "no-1-8-v" DT binding to enable >> the SDHCI_QUIRK2_NO_1_8_V. The sdhci driver uses this quirk to mask >> the bus speed modes for SD UHS cards. >> >> In this regards, consider the two below options. >> >> 1) We have DT bindings for SD UHS speed modes, which somehow makes the >> "no-1-8-v" binding redundant (or deprecated). >> >> 2) If you can model the VCCQ power supply as as a regulator, you could >> use regulator_is_supported_voltage() API to find out similar >> information used to set SDHCI_QUIRK2_NO_1_8_V. In fact, sdhci already >> does that to mask MMC_CAP2_HSX00_1_2V, unless 1.2V is supported. >> >> In a third case, sdhci-pxav3 use the "no-1-8-v" DT binding to mask >> SDHCI_CAN_VDD_180/SDHCI_CAN_VDD_330, which seems wrong, as it has >> nothing to do with the power to the card (VMMC). It's also used it to >> mask MMC_CAP_1_8V_DDR. >> >> The summary, is that I would rather see us to move away from using the >> "no-1-8-v" DT binding and use the proper SD UHS bus speed modes >> binding instead. > > Makes sense indeed. > > The challenge here is that we still need to keep the old dtb's working. We need to keep something that's been "broken" for ever to keep working. Huh, it prevents us from doing the correct solution. :-( How hard is it to update the DTBs in these cases? > > Motivation for this patch was a custom mx6sl board that works fine > with 3.14, but after upgrading the kernel to 4.0 the eMMC no longer > can mount the rootfs. This is a regression caused by commit > 312449efd16bb06, which applies 1.8V to the eMMC card that can only > operate at 3.3V. > > Same thing happens on imx6q-sabresd: we have 3.3V voltage being > supplied to the eMMC, but now we are operating it at 1.8V, even though > if we have 'no-1-8-v' property in the device tree. > > While I understand the need for improvement that you clearly > described, I would also like that the old dtb's could still work > 'as-is' with a modern kernel. > > The path I propose here was the minimal invasive method I could come > up to allow the old dtb compatibility. No, I don't want to sprinkle the core with hacks to make host driver works. Those hacks will have to reside in host drivers instead. Unless updating the DTB is out of the questions.... I suggest the following instead. 1) Let's add MMC_CAP_3V_DDR and a corresponding DT binding for it. Adopt mmc_select_hs_ddr() and mmc_select_card_type() to know about this configuration. 2) If we can assume that "no-1-8-v" for sdhci has the meaning of enabling MMC_CAP_3V_DDR, let's do that parsing in drivers/mmc/host/sdhci-pltfm.c. 3) Update the DT documentation to better describe the "no-1-8-v" binding. Let's also move it to be a sdhci specific binding and not a generic mmc binding, to limit its use. 4) Update the DTB files in the kernel for imx to use new binding for MMC_CAP_3V_DDR where applicable. Moreover convert the DTB for imx to use the proper bindings for UHS bus speed modes. Thus imx should have moved away from the "no-1-8-v" for future DTB updates. Kind regards Uffe -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
