On Mon, 2022-04-04 at 17:01 +0200, Ulf Hansson wrote: > On Tue, 29 Mar 2022 at 05:29, Axe Yang <axe.yang@xxxxxxxxxxxx> wrote: > > > > Add support for eint IRQ when MSDC is used as an SDIO host. This > > feature requires SDIO device support async IRQ function. With this > > feature, SDIO host can be awakened by SDIO card in suspend state, > > without additional pin. > > > > MSDC driver will time-share the SDIO DAT1 pin. During suspend, MSDC > > turn off clock and switch SDIO DAT1 pin to GPIO mode. And during > > resume, switch GPIO function back to DAT1 mode then turn on clock. > > > > Some device tree property should be added or modified in MSDC node > > to support SDIO eint IRQ. Pinctrls "state_eint" is mandatory. Since > > this feature depends on asynchronous interrupts, "wakeup-source", > > "keep-power-in-suspend" and "cap-sdio-irq" flags are necessary, and > > the interrupts list should be extended: > > &mmcX { > > ... > > interrupts-extended = <...>, > > <&pio xxx > > IRQ_TYPE_LEVEL_LOW>; > > ... > > pinctrl-names = "default", "state_uhs", > > "state_eint"; > > ... > > pinctrl-2 = <&mmc2_pins_eint>; > > ... > > cap-sdio-irq; > > keep-power-in-suspend; > > wakeup-source; > > ... > > }; > > > > Co-developed-by: Yong Mao <yong.mao@xxxxxxxxxxxx> > > Signed-off-by: Yong Mao <yong.mao@xxxxxxxxxxxx> > > Signed-off-by: Axe Yang <axe.yang@xxxxxxxxxxxx> > > --- > > drivers/mmc/host/mtk-sd.c | 94 > > +++++++++++++++++++++++++++++++++++---- > > 1 file changed, 86 insertions(+), 8 deletions(-) > > > > diff --git a/drivers/mmc/host/mtk-sd.c b/drivers/mmc/host/mtk-sd.c > > index 65037e1d7723..2905d7134243 100644 > > --- a/drivers/mmc/host/mtk-sd.c > > +++ b/drivers/mmc/host/mtk-sd.c > > @@ -1,6 +1,6 @@ > > // SPDX-License-Identifier: GPL-2.0-only > > /* > > - * Copyright (c) 2014-2015 MediaTek Inc. > > + * Copyright (c) 2014-2015, 2022 MediaTek Inc. > > * Author: Chaotian.Jing <chaotian.jing@xxxxxxxxxxxx> > > */ > > > > @@ -20,6 +20,7 @@ > > #include <linux/platform_device.h> > > #include <linux/pm.h> > > #include <linux/pm_runtime.h> > > +#include <linux/pm_wakeirq.h> > > #include <linux/regulator/consumer.h> > > #include <linux/slab.h> > > #include <linux/spinlock.h> > > @@ -440,8 +441,11 @@ struct msdc_host { > > struct pinctrl *pinctrl; > > struct pinctrl_state *pins_default; > > struct pinctrl_state *pins_uhs; > > + struct pinctrl_state *pins_eint; > > struct delayed_work req_timeout; > > int irq; /* host interrupt */ > > + int eint_irq; /* interrupt from sdio device for > > waking up system */ > > + int sdio_wake_irq_depth; > > struct reset_control *reset; > > > > struct clk *src_clk; /* msdc source clock */ > > @@ -465,6 +469,7 @@ struct msdc_host { > > bool hs400_tuning; /* hs400 mode online tuning */ > > bool internal_cd; /* Use internal card-detect logic > > */ > > bool cqhci; /* support eMMC hw cmdq */ > > + bool sdio_eint_ready; /* Ready to support SDIO eint > > interrupt */ > > struct msdc_save_para save_para; /* used when gate HCLK */ > > struct msdc_tune_para def_tune_para; /* default tune > > setting */ > > struct msdc_tune_para saved_tune_para; /* tune result of > > CMD21/CMD19 */ > > @@ -1527,10 +1532,12 @@ static void msdc_enable_sdio_irq(struct > > mmc_host *mmc, int enb) > > __msdc_enable_sdio_irq(host, enb); > > spin_unlock_irqrestore(&host->lock, flags); > > > > - if (enb) > > - pm_runtime_get_noresume(host->dev); > > - else > > - pm_runtime_put_noidle(host->dev); > > + if (mmc->card && !mmc_card_enable_async_irq(mmc->card)) { > > It might not be obvious, but in fact mmc->card always exists in this > case, hence there is no need to check the pointer before using it. this check is removed in v10. > > > + if (enb) > > + pm_runtime_get_noresume(host->dev); > > + else > > + pm_runtime_put_noidle(host->dev); > > + } > > } > > > > static irqreturn_t msdc_cmdq_irq(struct msdc_host *host, u32 > > intsts) > > @@ -2631,6 +2638,18 @@ static int msdc_drv_probe(struct > > platform_device *pdev) > > goto host_free; > > } > > > > + /* Support for SDIO eint irq ? */ > > + if (mmc->pm_caps & MMC_PM_WAKE_SDIO_IRQ) { > > + host->pins_eint = pinctrl_lookup_state(host- > > >pinctrl, "state_eint"); > > + if (IS_ERR(host->pins_eint)) { > > + ret = dev_err_probe(&pdev->dev, > > PTR_ERR(host->pins_eint), > > + "Cannot find pinctrl > > eint!\n"); > > + goto host_free; > > + } > > + > > + host->sdio_eint_ready = true; > > Rather than using a separate variable, couldn't we just check "if > (host->pins_eint)" to know whether wakeup settings shall be managed? This variable is removed in v10, use host->pins_eint instead. > > > + } > > + > > msdc_of_property_parse(pdev, host); > > > > host->dev = &pdev->dev; > > @@ -2722,6 +2741,21 @@ static int msdc_drv_probe(struct > > platform_device *pdev) > > if (ret) > > goto release; > > > > + if (host->sdio_eint_ready) { > > + host->eint_irq = irq_of_parse_and_map(host->dev- > > >of_node, 1); > > + ret = host->eint_irq ? > > dev_pm_set_dedicated_wake_irq(host->dev, host->eint_irq) : > > + -ENODEV; > > I think it would make better sense to call > dev_pm_set_dedicated_wake_irq() from msdc_enable_sdio_irq(). In this > way, you can still support SDIO irqs, even if there are no GPIO irq > for available for re-routing. > > In other words, if the GPIO irq is available, then that's the best > option and we should use that - otherwise we need to keep preventing > runtime-suspend by bumping the runtime PM usage count. dev_pm_set_dedicated_wake_irq() is replaced with dev_pm_set_dedicated_wake_irq_reverse() to assure msdc irq is disabled earlier than enable wake irq in rpm_suspend(), and it is moved to msdc_enable_sdio_irq() phase. > > > + > > + if (ret) { > > + dev_err(host->dev, "Failed to register > > data1 eint irq!\n"); > > + goto release; > > + } > > + > > + dev_pm_disable_wake_irq(host->dev); > > + pinctrl_select_state(host->pinctrl, host- > > >pins_uhs); > > + } > > + > > + device_init_wakeup(host->dev, true); > > I think this better belongs under the condition of whether > MMC_PM_WAKE_SDIO_IRQ and MMC_PM_KEEP_POWER caps are set. Both of them are checked before parsing eint pinstate in v10. > > > pm_runtime_set_active(host->dev); > > pm_runtime_set_autosuspend_delay(host->dev, > > MTK_MMC_AUTOSUSPEND_DELAY); > > pm_runtime_use_autosuspend(host->dev); > > @@ -2734,6 +2768,7 @@ static int msdc_drv_probe(struct > > platform_device *pdev) > > return 0; > > end: > > pm_runtime_disable(host->dev); > > + dev_pm_clear_wake_irq(host->dev); > > release: > > platform_set_drvdata(pdev, NULL); > > msdc_deinit_hw(host); > > @@ -2845,6 +2880,16 @@ static int __maybe_unused > > msdc_runtime_suspend(struct device *dev) > > struct msdc_host *host = mmc_priv(mmc); > > > > msdc_save_reg(host); > > + > > + if (host->sdio_eint_ready) { > > The core provides a helper function, sdio_irq_claimed(), that can be > used to know whether there is an SDIO irq registered. It looks like > that certainly can be useful here. Yes. We need sdio_irq_claimed() check since set dedicated wake irq is been moved to claim irq stage. > > > + disable_irq(host->irq); > > + pinctrl_select_state(host->pinctrl, host- > > >pins_eint); > > + if (host->sdio_wake_irq_depth == 0) { > > + dev_pm_enable_wake_irq(dev); > > This should be managed from the runtime PM core. See rpm_suspend(). > > Or perhaps there is a reason that doesn't work for you? > > > + host->sdio_wake_irq_depth++; > > Again, sdio_irq_claimed() is probably better to use. > > > + } > > + sdr_clr_bits(host->base + SDC_CFG, > > SDC_CFG_SDIOIDE); > > + } > > msdc_gate_clock(host); > > return 0; > > } > > @@ -2860,12 +2905,25 @@ static int __maybe_unused > > msdc_runtime_resume(struct device *dev) > > return ret; > > > > msdc_restore_reg(host); > > + > > + if (host->sdio_eint_ready) { > > + if (host->sdio_wake_irq_depth > 0) { > > + dev_pm_disable_wake_irq(dev); > > + host->sdio_wake_irq_depth--; > > + sdr_set_bits(host->base + SDC_CFG, > > SDC_CFG_SDIOIDE); > > + } else { > > + sdr_clr_bits(host->base + MSDC_INTEN, > > MSDC_INTEN_SDIOIRQ); > > + } > > + pinctrl_select_state(host->pinctrl, host- > > >pins_uhs); > > + enable_irq(host->irq); > > + } > > return 0; > > } > > > > -static int __maybe_unused msdc_suspend(struct device *dev) > > +static int __maybe_unused msdc_suspend_noirq(struct device *dev) > > { > > struct mmc_host *mmc = dev_get_drvdata(dev); > > + struct msdc_host *host = mmc_priv(mmc); > > int ret; > > > > if (mmc->caps2 & MMC_CAP2_CQE) { > > @@ -2874,16 +2932,36 @@ static int __maybe_unused > > msdc_suspend(struct device *dev) > > return ret; > > } > > > > + if (host->sdio_eint_ready) > > + enable_irq_wake(host->eint_irq); > > The PM core should manage this for us through dpm_suspend_noirq(), I > think. We just need to make sure that the runtime PM status of the > device has been set to RPM_SUSPENDED before dpm_suspend_noirq() gets > called. Have a look at dpm_suspend_noirq() -> > device_wakeup_arm_wake_irqs() -> dev_pm_arm_wake_irq(). > > In other words (and unless I am missing something), if we continue to > call pm_runtime_force_suspend() from a ->suspend() callback this > should work just fine. This part is simplified in v10. dpm/rpm system can manage wake irq well. > > > + > > return pm_runtime_force_suspend(dev); > > } > > > > -static int __maybe_unused msdc_resume(struct device *dev) > > +static int __maybe_unused msdc_resume_noirq(struct device *dev) > > { > > + struct mmc_host *mmc = dev_get_drvdata(dev); > > + struct msdc_host *host = mmc_priv(mmc); > > + > > + if (host->sdio_eint_ready) { > > + disable_irq_wake(host->eint_irq); > > + > > + /* > > + * In noirq resume stage, msdc_runtime_resume() > > + * won't be called, so disalbe wake irq here > > + * to block dedicated wake irq handler callback. > > + */ > > + if (likely(host->sdio_wake_irq_depth > 0)) { > > + dev_pm_disable_wake_irq(dev); > > + host->sdio_wake_irq_depth--; > > + } > > Similar comment as from the system suspend case. I don't think we > need > to move to the noirq phase. > > Or maybe I am missing something here? They are restored to system suspend stage in V10. But I have to use rpm_suspend()/rpm_resueme() instead of pm_runtime_force_suspend()/pm_runtime_force_resmue(). pm_runtime_force_resume() will go out directly because rpm state is RPM_SUSPENDED in system resume stage. > > > + } > > + > > return pm_runtime_force_resume(dev); > > } > > > > static const struct dev_pm_ops msdc_dev_pm_ops = { > > - SET_SYSTEM_SLEEP_PM_OPS(msdc_suspend, msdc_resume) > > + SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(msdc_suspend_noirq, > > msdc_resume_noirq) > > SET_RUNTIME_PM_OPS(msdc_runtime_suspend, > > msdc_runtime_resume, NULL) > > }; > > > > -- > > 2.25.1 > > > Regards, Axe