Skip to contentSkip to conversation

mmc: omap_hsmmc: Enable SDIO IRQ

PATCH Superseded Andreas Fenkart 3 participants 9 messages May 11, 2014 – May 17, 2014

Subsystems (3)

v11 of 14 Superseded Latest is v14 →

lore
Share & tools
lore
#0 lore raw

v11 cover mmc: omap_hsmmc: Enable SDIO IRQ

Hi Balaji, Tony, all

v12
- drop !CONFIG_OF compile break only exists when
#undef CONFIG_OF after include headers 1/7(Sebastian Reichel)
- do not emit "falling back to polling" if wake_irq not specified since MMC does not need it, and it might confuse users only emit if pinmux default/idle is not present or claiming the irq failed 2/7(Balaji)
- dropped out-of-tree patches 6/7(Balaji)
- mention "ti,am33xx-hsmmc" compatible section in bindings documentation 1/5

v11
- split !CONFIG_OF compile break into separate patch
- enable IWE/CLKEXTFREE in CON/HCTL register needed for omap4
- '&' vs '&&' in omap_hsmmc_resume, 1/5 (Andreas Müller)
- #define DLEV_DAT instead of BIT(21) 2/5 (Balaji)
- pinctrl_pm_select_default_state() removed, 4/5 (Balaji)
- drop _irqsave/_irqrestore from omap_hsmmc_wake_irq handler since it can't be preempted by same priority omap_hsmmc_irq handler 1/5(Joel Fernandes)
- replace devres_open_group by explicit devm_free calls 1/5 (Balaji)
- disable_irq_nosync wake_irq since we handle it thread safe 1/5 (Balaji)
- drop 'gpio_dat1' pinctrl states and rework documentation 5/5 (Balaji)

v10
- bug fix on multi-core, untested
- incorporated changes from Balaji
- use devres / RAII mechanism to configure wake_up / sdio irq capabilities
- drop pinctrl state 'active'
rely on driver-model states 'default', 'idle'
- add specific 'gpio_dat1' state for am335x SWAKEUP hack
- reorganized patches; +1 patch multi-core bugfix / +1 for pinctrl
- rebased 455c6fdbd21916 / cherry-picks from mmc-next

v9
- extended comment about why wake-irq is needed
- drop double '(' ')' around card_detect_irq
- drop final '.' in in subject line of patch

v8
- rebased on top of Tony Lindgren<[email protected]> changes
- improved changelog describing the earlier work
- improved wakeup irq setup
- works for am3730 es platform now
- my changes on top:
- compile tested with #undef CONFIG_OF
- disable wake_irq in handler to prevent infinite loop
- fixed typo and added comment about wake-irq

v7
- rebase on 3.14.0-rc3-49726-g77e15ec
- split omap_hsmmc_pin_init due to regression on omap-3730 platform

v6
- rebase on Linux 3.13-rc3
- reformatting debugfs

v5
- fix compile error introduced by last minute one line fix

v4:
- switch to interrupts-extended format
- drop ti,swakeup-missing flag convert to comaptible section

v3:
- removed gpio_irq from platform_data

v2:
- incorparated changes as suggested by reviewers
- simplified workaround for am335x, gpio will now only wake the module from runtime suspend, not handle the sdio irq itself

Andreas Fenkart (6):
mmc: omap_hsmmc: Enable SDIO interrupt
mmc: omap_hsmmc: enable wakeup event for sdio OMAP4
mmc: omap_hsmmc: Extend debugfs by SDIO IRQ handling, runtime state
mmc: omap_hsmmc: abort runtime suspend if pending sdio irq detected
mmc: omap_hsmmc: switch default/idle pinctrl states in runtime hooks
mmc: omap_hsmmc: Pin remux workaround to support SDIO interrupt on

    AM335x
 .../devicetree/bindings/mmc/ti-omap-hsmmc.txt      |   54 ++++
 drivers/mmc/host/omap_hsmmc.c                      |  280 ++++++++++++++++++--
 include/linux/platform_data/mmc-omap.h             |    1 +
 3 files changed, 314 insertions(+), 21 deletions(-)
Signature · 1 line
1.7.10.4
#1 lore raw

PATCH v11 1/6 mmc: omap_hsmmc: Enable SDIO interrupt

There have been various patches floating around for enabling the SDIO IRQ for hsmmc, but none of them ever got merged.

Probably the reason for not merging the SDIO interrupt patches has been the lack of wake-up path for SDIO on some omaps that has also needed remuxing the SDIO DAT1 line to a GPIO making the patches complex.

This patch adds the minimal SDIO IRQ support to hsmmc for omaps that do have the wake-up path. For those omaps, the DAT1 line need to have the wake-up enable bit set, and the wake-up interrupt is the same as for the MMC controller.

This patch has been tested on am3730 es1.2 with mwifiex connected to MMC3 with mwifiex waking to Ethernet traffic from off-idle mode. Note that for omaps that do not have the SDIO wake-up path, this patch will not work for idle modes and further patches for remuxing DAT1 to GPIO are needed.

Based on earlier patches [1][2] by David Vrabel <[email protected]>, Steve Sakoman <[email protected]>

For now, only support SDIO interrupt if we are booted with a separate wake-irq configued via device tree. This is because omaps need the wake-irq for idle states, and some omaps need special quirks. And we don't want to add new legacy mux platform init code callbacks any longer as we are moving to DT based booting anyways.

To use it, you need to specify the wake-irq using the interrupts-extended property.

[1] http://www.sakoman.com/cgi-bin/gitweb.cgi?p=linux.git;a=commitdiff_plain;h=010810d22f6f49ac03da4ba384969432e0320453
[2] http://comments.gmane.org/gmane.linux.kernel.mmc/20446

Documentation/devicetree/bindings/mmc/ti-omap-hsmmc.txt +1 -0
index ce80561..0233ba7 100644
@@ -12,6 +12,7 @@ Required properties:
  Should be "ti,omap3-hsmmc", for OMAP3 controllers
  Should be "ti,omap3-pre-es3-hsmmc" for OMAP3 controllers pre ES3.0
  Should be "ti,omap4-hsmmc", for OMAP4 controllers
+ Should be "ti,am33xx-hsmmc", for AM335x controllers
 - ti,hwmods: Must be "mmc<n>", n is controller instance starting 1
 
 Optional properties:
drivers/mmc/host/omap_hsmmc.c +189 -12

350 lines not on this page · raw message

include/linux/platform_data/mmc-omap.h +1 -0
index 2bf1b30..51e70cf 100644
@@ -28,6 +28,7 @@
  */
 #define OMAP_HSMMC_SUPPORTS_DUAL_VOLT		BIT(0)
 #define OMAP_HSMMC_BROKEN_MULTIBLOCK_READ	BIT(1)
+#define OMAP_HSMMC_SWAKEUP_MISSING		BIT(2)
 
 struct mmc_card;
 
Signature · 1 line
1.7.10.4
#2 lore raw

PATCH v11 2/6 mmc: omap_hsmmc: enable wakeup event for sdio OMAP4

In reply to Andreas Fenkart

To detect sdio irqs properly without spurious events, OMAP4 needs IWE in CON and CTPL, CLKEXTFREE in HCTL to be set

drivers/mmc/host/omap_hsmmc.c +13 -1
index edd53adc..4ee4645 100644
@@ -93,7 +93,10 @@
 #define BCE			(1 << 1)
 #define FOUR_BIT		(1 << 1)
 #define HSPE			(1 << 2)
+#define IWE			(1 << 24)
 #define DDR			(1 << 19)
+#define CLKEXTFREE		(1 << 16)
+#define CTPL			(1 << 11)
 #define DW8			(1 << 5)
 #define OD			0x1
 #define STAT_CLEAR		0xFFFFFFFF
@@ -686,6 +689,9 @@ static int omap_hsmmc_context_restore(struct omap_hsmmc_host *host)
 		capa = VS18;
 	}
 
+	if (host->mmc->caps & MMC_CAP_SDIO_IRQ)
+		hctl |= IWE;
+
 	OMAP_HSMMC_WRITE(host->base, HCTL,
 			OMAP_HSMMC_READ(host->base, HCTL) | hctl);
 
@@ -1682,19 +1688,23 @@ static void omap_hsmmc_init_card(struct mmc_host *mmc, struct mmc_card *card)
 static void omap_hsmmc_enable_sdio_irq(struct mmc_host *mmc, int enable)
 {
 	struct omap_hsmmc_host *host = mmc_priv(mmc);
-	u32 irq_mask;
+	u32 irq_mask, con;
 	unsigned long flags;
 
 	spin_lock_irqsave(&host->irq_lock, flags);
 
+	con = OMAP_HSMMC_READ(host->base, CON);
 	irq_mask = OMAP_HSMMC_READ(host->base, ISE);
 	if (enable) {
 		host->flags |= HSMMC_SDIO_IRQ_ENABLED;
 		irq_mask |= CIRQ_EN;
+		con |= CTPL | CLKEXTFREE;
 	} else {
 		host->flags &= ~HSMMC_SDIO_IRQ_ENABLED;
 		irq_mask &= ~CIRQ_EN;
+		con &= ~(CTPL | CLKEXTFREE);
 	}
+	OMAP_HSMMC_WRITE(host->base, CON, con);
 	OMAP_HSMMC_WRITE(host->base, IE, irq_mask);
 
 	/*
@@ -1744,6 +1754,8 @@ static int omap_hsmmc_configure_wake_irq(struct omap_hsmmc_host *host)
 		goto err;
 	}
 
+	OMAP_HSMMC_WRITE(host->base, HCTL,
+			 OMAP_HSMMC_READ(host->base, HCTL) | IWE);
 	return 0;
 
 err:
Signature · 1 line
1.7.10.4
#3 lore raw

PATCH v11 3/6 mmc: omap_hsmmc: Extend debugfs by SDIO IRQ handling, runtime state

In reply to Andreas Fenkart

Add SDIO IRQ entries to debugfs entry. Note that PSTATE shows current state of data lines, incl. SDIO IRQ pending

drivers/mmc/host/omap_hsmmc.c +19 -3
index 4ee4645..1a9d31f 100644
@@ -56,6 +56,7 @@
 #define OMAP_HSMMC_RSP54	0x0118
 #define OMAP_HSMMC_RSP76	0x011C
 #define OMAP_HSMMC_DATA		0x0120
+#define OMAP_HSMMC_PSTATE	0x0124
 #define OMAP_HSMMC_HCTL		0x0128
 #define OMAP_HSMMC_SYSCTL	0x012C
 #define OMAP_HSMMC_STAT		0x0130
@@ -1825,14 +1826,29 @@ static int omap_hsmmc_regs_show(struct seq_file *s, void *data)
 {
 	struct mmc_host *mmc = s->private;
 	struct omap_hsmmc_host *host = mmc_priv(mmc);
+	bool suspended;
 
-	seq_printf(s, "mmc%d:\n ctx_loss:\t%d\n\nregs:\n",
-			mmc->index, host->context_loss);
+	seq_printf(s, "mmc%d:\n", mmc->index);
+	seq_printf(s, "sdio irq mode\t%s\n",
+		   (mmc->caps & MMC_CAP_SDIO_IRQ) ? "interrupt" : "polling");
 
-	pm_runtime_get_sync(host->dev);
+	if (mmc->caps & MMC_CAP_SDIO_IRQ) {
+		seq_printf(s, "sdio irq \t%s\n",
+			   (host->flags & HSMMC_SDIO_IRQ_ENABLED) ?  "enabled"
+			   : "disabled");
+	}
+
+	suspended = host->dev->power.runtime_status != RPM_ACTIVE;
+	seq_printf(s, "runtime state\t%s\n", (suspended ?  "idle" : "active"));
 
+	seq_printf(s, "ctx_loss:\t%d\n", host->context_loss);
+
+	pm_runtime_get_sync(host->dev);
+	seq_puts(s, "\nregs:\n");
 	seq_printf(s, "CON:\t\t0x%08x\n",
 			OMAP_HSMMC_READ(host->base, CON));
+	seq_printf(s, "PSTATE:\t\t0x%08x\n",
+		   OMAP_HSMMC_READ(host->base, PSTATE));
 	seq_printf(s, "HCTL:\t\t0x%08x\n",
 			OMAP_HSMMC_READ(host->base, HCTL));
 	seq_printf(s, "SYSCTL:\t\t0x%08x\n",
Signature · 1 line
1.7.10.4
#4 lore raw

PATCH v11 4/6 mmc: omap_hsmmc: abort runtime suspend if pending sdio irq detected

In reply to Andreas Fenkart

On multicores, an sdio irq handler could be running in parallel to runtime suspend. In the worst case it could be waiting for the spinlock held by the runtime suspend. When runtime suspend is complete and the functional clock (fclk) turned off, the irq handler will continue and cause a SIGBUS on the first register access.

drivers/mmc/host/omap_hsmmc.c +21 -2
index 1a9d31f..30365f5 100644
@@ -107,6 +107,9 @@
 #define SRD			(1 << 26)
 #define SOFTRESET		(1 << 1)
 
+/* PSTATE */
+#define DLEV_DAT(x)		(1 << (20 + (x)))
+
 /* Interrupt masks for IE and ISE register */
 #define CC_EN			(1 << 0)
 #define TC_EN			(1 << 1)
@@ -2423,6 +2426,7 @@ static int omap_hsmmc_runtime_suspend(struct device *dev)
 {
 	struct omap_hsmmc_host *host;
 	unsigned long flags;
+	int ret = 0;
 
 	host = platform_get_drvdata(to_platform_device(dev));
 	omap_hsmmc_context_save(host);
@@ -2434,14 +2438,29 @@ static int omap_hsmmc_runtime_suspend(struct device *dev)
 		/* disable sdio irq handling to prevent race */
 		OMAP_HSMMC_WRITE(host->base, ISE, 0);
 		OMAP_HSMMC_WRITE(host->base, IE, 0);
-		OMAP_HSMMC_WRITE(host->base, STAT, STAT_CLEAR);
+
+		if (!(OMAP_HSMMC_READ(host->base, PSTATE) & DLEV_DAT(1))) {
+			/*
+			 * dat1 line low, pending sdio irq
+			 * race condition: possible irq handler running on
+			 * multi-core, abort
+			 */
+			dev_dbg(dev, "pending sdio irq, abort suspend\n");
+			OMAP_HSMMC_WRITE(host->base, STAT, STAT_CLEAR);
+			OMAP_HSMMC_WRITE(host->base, ISE, CIRQ_EN);
+			OMAP_HSMMC_WRITE(host->base, IE, CIRQ_EN);
+			pm_runtime_mark_last_busy(dev);
+			ret = -EBUSY;
+			goto abort;
+		}
 
 		WARN_ON(host->flags & HSMMC_WAKE_IRQ_ENABLED);
 		enable_irq(host->wake_irq);
 		host->flags |= HSMMC_WAKE_IRQ_ENABLED;
 	}
+abort:
 	spin_unlock_irqrestore(&host->irq_lock, flags);
-	return 0;
+	return ret;
 }
 
 static int omap_hsmmc_runtime_resume(struct device *dev)
Signature · 1 line
1.7.10.4
#5 lore raw

PATCH v11 5/6 mmc: omap_hsmmc: switch default/idle pinctrl states in runtime hooks

In reply to Andreas Fenkart

These are predefined states of the driver model. When not present, as if not set in the device tree, they become no-ops. Explicitly selecting the default state is not needed since the device core layer sets pin mux to "default" state before probe. This is not the simplest implementation, on AM335x at least, we could switch to idle at any point in the suspend hook, only the default state needs to be set before writing to the irq registers or an IRQ might get lost.

drivers/mmc/host/omap_hsmmc.c +10 -6
index 30365f5..3c8b183 100644
@@ -2002,7 +2002,6 @@ static int omap_hsmmc_probe(struct platform_device *pdev)
 	const struct of_device_id *match;
 	dma_cap_mask_t mask;
 	unsigned tx_req, rx_req;
-	struct pinctrl *pinctrl;
 	const struct omap_mmc_of_data *data;
 
 	match = of_match_device(of_match_ptr(omap_mmc_of_match), &pdev->dev);
@@ -2226,11 +2225,6 @@ static int omap_hsmmc_probe(struct platform_device *pdev)
 
 	omap_hsmmc_disable_irq(host);
 
-	pinctrl = devm_pinctrl_get_select_default(&pdev->dev);
-	if (IS_ERR(pinctrl))
-		dev_warn(&pdev->dev,
-			"pins are not configured from the driver\n");
-
 	/*
 	 * For now, only support SDIO interrupt if we have a separate
 	 * wake-up interrupt configured from device tree. This is because
@@ -2454,10 +2448,15 @@ static int omap_hsmmc_runtime_suspend(struct device *dev)
 			goto abort;
 		}
 
+		pinctrl_pm_select_idle_state(dev);
+
 		WARN_ON(host->flags & HSMMC_WAKE_IRQ_ENABLED);
 		enable_irq(host->wake_irq);
 		host->flags |= HSMMC_WAKE_IRQ_ENABLED;
+	} else {
+		pinctrl_pm_select_idle_state(dev);
 	}
+
 abort:
 	spin_unlock_irqrestore(&host->irq_lock, flags);
 	return ret;
@@ -2481,9 +2480,14 @@ static int omap_hsmmc_runtime_resume(struct device *dev)
 			host->flags &= ~HSMMC_WAKE_IRQ_ENABLED;
 		}
 
+		pinctrl_pm_select_default_state(host->dev);
+
+		/* irq lost, if pinmux incorrect */
 		OMAP_HSMMC_WRITE(host->base, STAT, STAT_CLEAR);
 		OMAP_HSMMC_WRITE(host->base, ISE, CIRQ_EN);
 		OMAP_HSMMC_WRITE(host->base, IE, CIRQ_EN);
+	} else {
+		pinctrl_pm_select_default_state(host->dev);
 	}
 	spin_unlock_irqrestore(&host->irq_lock, flags);
 	return 0;
Signature · 1 line
1.7.10.4
#6 lore raw

PATCH v11 6/6 mmc: omap_hsmmc: Pin remux workaround to support SDIO interrupt on AM335x

In reply to Andreas Fenkart

The am335x can't detect pending cirq in PM runtime suspend. This patch reconfigures dat1 as a GPIO before going to suspend. SDIO interrupts are detected with the GPIO, the GPIO will only wake the module from suspend, SDIO irq detection will still happen through the IP block.

Idea of remuxing the pins by Tony Lindgren. Code contributions from Tony Lindgren and Balaji T K <[email protected]>

Documentation/devicetree/bindings/mmc/ti-omap-hsmmc.txt +53 -0
index 0233ba7..76bf087 100644
@@ -57,3 +57,56 @@ Examples:
 			&edma 25>;
 		dma-names = "tx", "rx";
 	};
+
+[workaround for missing swakeup on am33xx]
+
+This SOC is missing the swakeup line, it will not detect SDIO irq
+while in suspend.
+
+                             ------
+                             | PRCM |
+                              ------
+                               ^ |
+                       swakeup | | fclk
+                               | v
+       ------                -------               -----
+      | card | -- CIRQ -->  | hsmmc | -- IRQ -->  | CPU |
+       ------                -------               -----
+
+In suspend the fclk is off and the module is disfunctional. Even register reads
+will fail. A small logic in the host will request fclk restore, when an
+external event is detected. Once the clock is restored, the host detects the
+event normally. Since am33xx doesn't have this line it never wakes from
+suspend.
+
+The workaround is to reconfigure the dat1 line as a GPIO upon suspend. To make
+this work, we need to set the named pinctrl states "default" and "idle".
+Prepare idle to remux dat1 as a gpio, and default to remux it back as sdio
+dat1. The MMC driver will then toggle between idle and default state during
+runtime.
+
+In summary:
+1. select matching 'compatible' section, see example below.
+2. specify pinctrl states "default" and "idle", "sleep" is optional.
+3. specify the gpio irq used for detecting sdio irq in suspend
+
+If configuration is incomplete, a warning message is emitted "falling back to
+polling". Also check the "sdio irq mode" in /sys/kernel/debug/mmc0/regs. Mind
+not every application needs SDIO irq, e.g. MMC cards.
+
+	mmc1: mmc@48060100 {
+		compatible = "ti,am33xx-hsmmc";
+		...
+		pinctrl-names = "default", "idle", "sleep"
+		pinctrl-0 = <&mmc1_pins>;
+		pinctrl-1 = <&mmc1_idle>;
+		pinctrl-2 = <&mmc1_sleep>;
+		...
+		interrupts-extended = <&intc 64 &gpio2 28 0>;
+	};
+
+	mmc1_idle : pinmux_cirq_pin {
+		pinctrl-single,pins = <
+		        0x0f8 0x3f      /* GPIO2_28 */
+		>;
+	};
drivers/mmc/host/omap_hsmmc.c +13 -3
index 3c8b183..d22ac3f 100644
@@ -1753,15 +1753,25 @@ static int omap_hsmmc_configure_wake_irq(struct omap_hsmmc_host *host)
 	 * and need to remux SDIO DAT1 to GPIO for wake-up from idle.
 	 */
 	if (host->pdata->controller_flags & OMAP_HSMMC_SWAKEUP_MISSING) {
-		ret = -ENODEV;
-		devm_free_irq(host->dev, host->wake_irq, host);
-		goto err;
+		if (IS_ERR(host->dev->pins->default_state)) {
+			dev_info(host->dev, "missing default pinctrl state\n");
+			ret = -EINVAL;
+			goto err_free_irq;
+		}
+
+		if (IS_ERR(host->dev->pins->idle_state)) {
+			dev_info(host->dev, "missing idle pinctrl state\n");
+			ret = -EINVAL;
+			goto err_free_irq;
+		}
 	}
 
 	OMAP_HSMMC_WRITE(host->base, HCTL,
 			 OMAP_HSMMC_READ(host->base, HCTL) | IWE);
 	return 0;
 
+err_free_irq:
+	devm_free_irq(host->dev, host->wake_irq, host);
 err:
 	dev_warn(host->dev, "no SDIO IRQ support, falling back to polling\n");
 	host->wake_irq = 0;
Signature · 1 line
1.7.10.4
#7 lore raw
In reply to Andreas Fenkart
Quoted 12 lines · Andreas Fenkart

Hi Balaji, Tony, all

v12
- drop !CONFIG_OF compile break only exists when
#undef CONFIG_OF after include headers 1/7(Sebastian Reichel)
- do not emit "falling back to polling" if wake_irq not specified since MMC does not need it, and it might confuse users only emit if pinmux default/idle is not present or claiming the irq failed 2/7(Balaji)
- dropped out-of-tree patches 6/7(Balaji)
- mention "ti,am33xx-hsmmc" compatible section in bindings documentation 1/5

Hi Andreas,

It seems like we are ready to merge this patchset - I will include it in the next PR I send to Chris.

Kind regards
Ulf Hansson

#8 lore raw
Quoted 14 lines · Ulf Hansson
Quoted 12 lines · Andreas Fenkart

Hi Balaji, Tony, all

v12
- drop !CONFIG_OF compile break only exists when
#undef CONFIG_OF after include headers 1/7(Sebastian Reichel)
- do not emit "falling back to polling" if wake_irq not specified since MMC does not need it, and it might confuse users only emit if pinmux default/idle is not present or claiming the irq failed 2/7(Balaji)
- dropped out-of-tree patches 6/7(Balaji)
- mention "ti,am33xx-hsmmc" compatible section in bindings documentation 1/5

Quoted 4 lines

Hi Andreas,

It seems like we are ready to merge this patchset - I will include it in the next PR I send to Chris.

Great! Just gave these a try again after a while, worked fine for me. Thanks Andreas for the patience with updating these patches.

Regards,

Tony