| Message ID | 20180521223503.152131-1-evgreen@chromium.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, May 22, 2018 at 1:35 AM, Evan Green <evgreen@chromium.org> wrote: Some style concerns. > int mmc_gpio_get_cd(struct mmc_host *host) > { > + int can_sleep; > struct mmc_gpio *ctx = host->slot.handler_priv; I would rather name it int cansleep; and put after ctx definition. > > if (!ctx || !ctx->cd_gpio) > return -ENOSYS; > > - if (ctx->override_cd_active_level) > - return !gpiod_get_raw_value_cansleep(ctx->cd_gpio) ^ > - !!(host->caps2 & MMC_CAP2_CD_ACTIVE_HIGH); > + can_sleep = gpiod_cansleep(ctx->cd_gpio); > + if (ctx->override_cd_active_level) { > + int value = can_sleep ? > + gpiod_get_raw_value_cansleep(ctx->cd_gpio) : > + gpiod_get_raw_value(ctx->cd_gpio); > + return !value ^ !!(host->caps2 & MMC_CAP2_CD_ACTIVE_HIGH); > + } > > - return gpiod_get_value_cansleep(ctx->cd_gpio); > + return can_sleep ? gpiod_get_value_cansleep(ctx->cd_gpio) : > + gpiod_get_value(ctx->cd_gpio); Perhaps same style as you use above woule be better, i.e. retrun cansleep ? true_path: false_path; > } > EXPORT_SYMBOL(mmc_gpio_get_cd);
On 22 May 2018 at 00:35, Evan Green <evgreen@chromium.org> wrote: > This change uses the appropriate _cansleep or non-sleeping API for > reading GPIO card detect state. This allows users with GPIOs that > never sleep to avoid a warning when certain quirks are present. > > The sdhci controller has an SDHCI_QUIRK_NO_CARD_NO_RESET, which > indicates that a controller will not reset properly if no card is > inserted. With this quirk enabled, mmc_get_cd_gpio is called in > several places with a spinlock held and interrupts disabled. > gpiod_get_raw_value_cansleep is not happy with this situation, > and throws out a warning. > > For boards that a) use controllers that have this quirk, and b) wire > card detect up to a GPIO that doesn't sleep, this is a spurious warning. > This change silences that warning, at the cost of pushing this problem > down to users that have sleeping GPIOs and controllers with this quirk. > > Signed-off-by: Evan Green <evgreen@chromium.org> > --- > This is my initial solution to the warning I was trying to face down > in my previous RFC [1]. Upsides of this solution is it manages to > avoid the warning in places where the warning doesn't apply, and is > low-risk. Other approaches I considered: > > 1. Changing sdhci_get_cd to reach through mmc and get the GPIO value > directly (without sleeping) rather than calling mmc_gpio_get_cd, > which uses the cansleep functions. I didn't love this because a) it > seemed ugly to be reaching into mmc_host like that and b) I'd have to > duplicate the logic in mmc_gpio_get_cd, which seemed brittle. > > 2. Using mmc_gpio_set_cd_isr to record when the card detect pin > changes, and then in sdhci_do_reset we wouldn't actually need to > get CD state, just look at the cached value. This seemed risky to me > since it would affect all sdhci controllers, and I don't know that > everybody using gpio-cd can do interrupts on their pins. > > 3. Adding mmc_gpio_get_cd_nosleep(). That one's still doable if > preferred, but I'd have to give some thought to when sdhci wires > up to which one. First, I thought I preferred this option, as it becomes clear of what goes on. However I then realize, that it may not be worth it, because in the end I guess the caller (sdhci), will not be able to deal with error codes. For example, what would it do if it receives a -ENOTSUPP from mmc_gpio_get_cd_nosleep()? > > I'm up for other suggestions. This one most just seemed like the > best first stab of minimally addressing the warning without rocking > the boat otherwise. > > [1] https://patchwork.kernel.org/patch/10374633/ > --- > drivers/mmc/core/slot-gpio.c | 14 ++++++++++---- > 1 file changed, 10 insertions(+), 4 deletions(-) > > diff --git a/drivers/mmc/core/slot-gpio.c b/drivers/mmc/core/slot-gpio.c > index 31f7dbb15668..0f497b69f4b8 100644 > --- a/drivers/mmc/core/slot-gpio.c > +++ b/drivers/mmc/core/slot-gpio.c > @@ -75,16 +75,22 @@ EXPORT_SYMBOL(mmc_gpio_get_ro); > > int mmc_gpio_get_cd(struct mmc_host *host) > { > + int can_sleep; > struct mmc_gpio *ctx = host->slot.handler_priv; > > if (!ctx || !ctx->cd_gpio) > return -ENOSYS; > > - if (ctx->override_cd_active_level) > - return !gpiod_get_raw_value_cansleep(ctx->cd_gpio) ^ > - !!(host->caps2 & MMC_CAP2_CD_ACTIVE_HIGH); > + can_sleep = gpiod_cansleep(ctx->cd_gpio); > + if (ctx->override_cd_active_level) { > + int value = can_sleep ? > + gpiod_get_raw_value_cansleep(ctx->cd_gpio) : > + gpiod_get_raw_value(ctx->cd_gpio); > + return !value ^ !!(host->caps2 & MMC_CAP2_CD_ACTIVE_HIGH); > + } > > - return gpiod_get_value_cansleep(ctx->cd_gpio); > + return can_sleep ? gpiod_get_value_cansleep(ctx->cd_gpio) : > + gpiod_get_value(ctx->cd_gpio); > } > EXPORT_SYMBOL(mmc_gpio_get_cd); > > -- > 2.13.5 > Kind regards Uffe -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, May 22, 2018 at 5:08 AM Ulf Hansson <ulf.hansson@linaro.org> wrote: > First, I thought I preferred this option, as it becomes clear of what > goes on. However I then realize, that it may not be worth it, because > in the end I guess the caller (sdhci), will not be able to deal with > error codes. For example, what would it do if it receives a -ENOTSUPP > from mmc_gpio_get_cd_nosleep()? Uffe, Yeah, that would end up looking more like my original RFC patch, where we just kind of assume the card is there (and therefore ignore the quirk) if we're not in a preemptible state. Sounds like I should make another spin this way, at least for consideration. Andy, Thanks for the review. I will incorporate your feedback into the next spin. -Evan -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/mmc/core/slot-gpio.c b/drivers/mmc/core/slot-gpio.c index 31f7dbb15668..0f497b69f4b8 100644 --- a/drivers/mmc/core/slot-gpio.c +++ b/drivers/mmc/core/slot-gpio.c @@ -75,16 +75,22 @@ EXPORT_SYMBOL(mmc_gpio_get_ro); int mmc_gpio_get_cd(struct mmc_host *host) { + int can_sleep; struct mmc_gpio *ctx = host->slot.handler_priv; if (!ctx || !ctx->cd_gpio) return -ENOSYS; - if (ctx->override_cd_active_level) - return !gpiod_get_raw_value_cansleep(ctx->cd_gpio) ^ - !!(host->caps2 & MMC_CAP2_CD_ACTIVE_HIGH); + can_sleep = gpiod_cansleep(ctx->cd_gpio); + if (ctx->override_cd_active_level) { + int value = can_sleep ? + gpiod_get_raw_value_cansleep(ctx->cd_gpio) : + gpiod_get_raw_value(ctx->cd_gpio); + return !value ^ !!(host->caps2 & MMC_CAP2_CD_ACTIVE_HIGH); + } - return gpiod_get_value_cansleep(ctx->cd_gpio); + return can_sleep ? gpiod_get_value_cansleep(ctx->cd_gpio) : + gpiod_get_value(ctx->cd_gpio); } EXPORT_SYMBOL(mmc_gpio_get_cd);
This change uses the appropriate _cansleep or non-sleeping API for reading GPIO card detect state. This allows users with GPIOs that never sleep to avoid a warning when certain quirks are present. The sdhci controller has an SDHCI_QUIRK_NO_CARD_NO_RESET, which indicates that a controller will not reset properly if no card is inserted. With this quirk enabled, mmc_get_cd_gpio is called in several places with a spinlock held and interrupts disabled. gpiod_get_raw_value_cansleep is not happy with this situation, and throws out a warning. For boards that a) use controllers that have this quirk, and b) wire card detect up to a GPIO that doesn't sleep, this is a spurious warning. This change silences that warning, at the cost of pushing this problem down to users that have sleeping GPIOs and controllers with this quirk. Signed-off-by: Evan Green <evgreen@chromium.org> --- This is my initial solution to the warning I was trying to face down in my previous RFC [1]. Upsides of this solution is it manages to avoid the warning in places where the warning doesn't apply, and is low-risk. Other approaches I considered: 1. Changing sdhci_get_cd to reach through mmc and get the GPIO value directly (without sleeping) rather than calling mmc_gpio_get_cd, which uses the cansleep functions. I didn't love this because a) it seemed ugly to be reaching into mmc_host like that and b) I'd have to duplicate the logic in mmc_gpio_get_cd, which seemed brittle. 2. Using mmc_gpio_set_cd_isr to record when the card detect pin changes, and then in sdhci_do_reset we wouldn't actually need to get CD state, just look at the cached value. This seemed risky to me since it would affect all sdhci controllers, and I don't know that everybody using gpio-cd can do interrupts on their pins. 3. Adding mmc_gpio_get_cd_nosleep(). That one's still doable if preferred, but I'd have to give some thought to when sdhci wires up to which one. I'm up for other suggestions. This one most just seemed like the best first stab of minimally addressing the warning without rocking the boat otherwise. [1] https://patchwork.kernel.org/patch/10374633/ --- drivers/mmc/core/slot-gpio.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-)