Seungwon, On Wed, Oct 23, 2013 at 12:25 PM, Seungwon Jeon <tgih.jun@xxxxxxxxxxx> wrote: >> >> + if (card->type == MMC_TYPE_SDIO || >> >> + card->type == MMC_TYPE_SD_COMBO) { > && card->quirks & MMC_QUIRK_BROKEN_CLK_GATING > How about considering MMC_QUIRK_BROKEN_CLK_GATING? > Some sdio device can work with gating clock. > For this, mmc_fixup_device() should be called prior to init_card() in core(sdio.c). > I guess you found that. By SDIO devices, are you referring to actual SDIO cards or some implementations of dw_mmc? As far as I understand in the CLKENA description in the generic documentation from Synopsys it say that for SDIO cards you must not stop the clock if interrupts must be detected. To me, that means that all dw_mmc implementations require this change if they support SDIO interrupts (hence checking for MMC_CAP_SDIO_IRQ). I guess I did make the assumption in this change that all (reasonable) SDIO cards would be using SDIO interrupts if they are available. If we could find out ahead of time if a given SDIO driver was planning to use interrupts we could do better. The old solution did better in this way and we could probably make it work (and still fix the read-modify-write race) if we thought this was really important. The code gets a little more twisted to try to avoid holding the IRQ-safe spinlock while updating the clock, but I could attempt it... NOTE: We've recently found that there are still occasions when we lose SDIO interrupts with dw_mmc and our Marvell WiFi driver, especially when those interrupts happen back-to-back. That problem appears unrelated to this one (it's what Bing was investigating when he found the original race). Anyone that has any thoughts, please let me know... -Doug -- 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