Skip to contentSkip to conversation

pwm: lpss: Misc. cleanups / improvements

PATCH Superseded Hans de Goede 2 participants 7 messages Oct 30, 2020 – Nov 9, 2020

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover pwm: lpss: Misc. cleanups / improvements

Hi All,

Now that the pwm-lpss / pwm-crc / i915 atomic PWM conversion has landed in 5.10-rc1 here is a small follow up series with some misc. improvements.

Regards,

Hans

#1 lore raw

PATCH 1/3 pwm: lpss: Log error from pwm_lpss_is_updating() if the update bit is still set

pwm_lpss_is_updating() does a sanity check which should never fail. If the check does actually fail that is worth logging an error, especially since this means that we will skip making the requested changes to the PWM settings.

---
 drivers/pwm/pwm-lpss.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
drivers/pwm/pwm-lpss.c +6 -1
index 3444c56b4bed..939de93c157b 100644
@@ -76,7 +76,12 @@ static int pwm_lpss_wait_for_update(struct pwm_device *pwm)
 
 static inline int pwm_lpss_is_updating(struct pwm_device *pwm)
 {
-	return (pwm_lpss_read(pwm) & PWM_SW_UPDATE) ? -EBUSY : 0;
+	if (pwm_lpss_read(pwm) & PWM_SW_UPDATE) {
+		dev_err(pwm->chip->dev, "PWM_SW_UPDATE is still set, skipping update\n");
+		return -EBUSY;
+	}
+
+	return 0;
 }
 
 static void pwm_lpss_prepare(struct pwm_lpss_chip *lpwm, struct pwm_device *pwm,
Signature · 1 line
2.28.0
#2 lore raw

PATCH 2/3 pwm: lpss: Use DPM_FLAG_NO_DIRECT_COMPLETE instead of declaring a prepare handler

In reply to Hans de Goede

ACPI LPSS devices use direct-complete style suspend/resume handling by default. We set the DPM_FLAG_SMART_PREPARE and define a prepare handler to disable this on Cherry Trail devices.

Clean this up a bit by setting the DPM_FLAG_NO_DIRECT_COMPLETE flag for Cherry Trail devices, instead of defining a prepare handler.

While at it also improve the comment explaining why this is necessary.

---
 drivers/pwm/pwm-lpss-platform.c | 35 ++++++++++++++-------------------
 1 file changed, 15 insertions(+), 20 deletions(-)
drivers/pwm/pwm-lpss-platform.c +15 -20
index c6502cf7a7af..ac33861edb48 100644
@@ -58,7 +58,21 @@ static int pwm_lpss_probe_platform(struct platform_device *pdev)
 
 	platform_set_drvdata(pdev, lpwm);
 
-	dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_SMART_PREPARE);
+	/*
+	 * On Cherry Trail devices the GFX0._PS0 AML checks if the controller
+	 * is on and if it is not on it turns it on and restores what it
+	 * believes is the correct state to the PWM controller.
+	 * Because of this we must disallow direct-complete, which keeps the
+	 * controller (runtime)suspended on resume, to avoid 2 issues:
+	 * 1. The controller getting turned on without the linux-pm code
+	 *    knowing about this. On devices where the controller is unused
+	 *    this causes it to stay on during the next suspend causing high
+	 *    battery drain (because S0i3 is not reached)
+	 * 2. The state restoring code unexpectedly messing with the controller
+	 */
+	if (info->other_devices_aml_touches_pwm_regs)
+		dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
+
 	pm_runtime_set_active(&pdev->dev);
 	pm_runtime_enable(&pdev->dev);
 
@@ -73,24 +87,6 @@ static int pwm_lpss_remove_platform(struct platform_device *pdev)
 	return pwm_lpss_remove(lpwm);
 }
 
-static int pwm_lpss_prepare(struct device *dev)
-{
-	struct pwm_lpss_chip *lpwm = dev_get_drvdata(dev);
-
-	/*
-	 * If other device's AML code touches the PWM regs on suspend/resume
-	 * force runtime-resume the PWM controller to allow this.
-	 */
-	if (lpwm->info->other_devices_aml_touches_pwm_regs)
-		return 0; /* Force runtime-resume */
-
-	return 1; /* If runtime-suspended leave as is */
-}
-
-static const struct dev_pm_ops pwm_lpss_platform_pm_ops = {
-	.prepare = pwm_lpss_prepare,
-};
-
 static const struct acpi_device_id pwm_lpss_acpi_match[] = {
 	{ "80860F09", (unsigned long)&pwm_lpss_byt_info },
 	{ "80862288", (unsigned long)&pwm_lpss_bsw_info },
@@ -104,7 +100,6 @@ static struct platform_driver pwm_lpss_driver_platform = {
 	.driver = {
 		.name = "pwm-lpss",
 		.acpi_match_table = pwm_lpss_acpi_match,
-		.pm = &pwm_lpss_platform_pm_ops,
 	},
 	.probe = pwm_lpss_probe_platform,
 	.remove = pwm_lpss_remove_platform,
Signature · 1 line
2.28.0
#3 lore raw

PATCH 3/3 pwm: lpss: Set DPM_FLAG_SMART_SUSPEND on Cherry Trail devices

In reply to Hans de Goede

As the comment above the code setting the DPM_FLAG_NO_DIRECT_COMPLETE flag explains:

/*
* On Cherry Trail devices the GFX0._PS0 AML checks if the controller
* is on and if it is not on it turns it on and restores what it
* believes is the correct state to the PWM controller.
* Because of this we must disallow direct-complete, which keeps the
* controller (runtime)suspended, on resume to avoid 2 issues:
* 1. The controller getting turned on without the linux-pm code
* knowing about this. On devices where the controller is unused
* this causes it to stay on during the next suspend causing high
* battery drain (because S0i3 is not reached)
* 2. The state restoring code unexpectedly messing with the controller */

The pm-core must not skip resume to avoid the GFX0._PS0 AML code messing with the PWM controller behind our back. But leaving the controller runtime-suspended (skipping runtime-resume + normal-suspend) during suspend is fine. Set the DPM_FLAG_SMART_SUSPEND flag to allow this.

---
 drivers/pwm/pwm-lpss-platform.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/pwm/pwm-lpss-platform.c +2 -1
index ac33861edb48..38d17e2e2b4c 100644
@@ -71,7 +71,8 @@ static int pwm_lpss_probe_platform(struct platform_device *pdev)
 	 * 2. The state restoring code unexpectedly messing with the controller
 	 */
 	if (info->other_devices_aml_touches_pwm_regs)
-		dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
+		dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE|
+						    DPM_FLAG_SMART_SUSPEND);
 
 	pm_runtime_set_active(&pdev->dev);
 	pm_runtime_enable(&pdev->dev);
Signature · 1 line
2.28.0
#4 lore raw

Re: [PATCH 3/3] pwm: lpss: Set DPM_FLAG_SMART_SUSPEND on Cherry Trail devices

Quoted 17 lines · Hans de Goede

As the comment above the code setting the DPM_FLAG_NO_DIRECT_COMPLETE flag explains:

/*
* On Cherry Trail devices the GFX0._PS0 AML checks if the controller
* is on and if it is not on it turns it on and restores what it
* believes is the correct state to the PWM controller.
* Because of this we must disallow direct-complete, which keeps the
* controller (runtime)suspended, on resume to avoid 2 issues:
* 1. The controller getting turned on without the linux-pm code
* knowing about this. On devices where the controller is unused
* this causes it to stay on during the next suspend causing high
* battery drain (because S0i3 is not reached)
* 2. The state restoring code unexpectedly messing with the controller */

Quoted 6 lines

The pm-core must not skip resume to avoid the GFX0._PS0 AML code messing with the PWM controller behind our back. But

leaving the controller
runtime-suspended (skipping runtime-resume + normal-suspend) during suspend is fine.

This paragraph is good to have in the comment of the code I think.

Quoted 24 lines

Set the DPM_FLAG_SMART_SUSPEND flag to allow this.

---
 drivers/pwm/pwm-lpss-platform.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/pwm/pwm-lpss-platform.c +2 -2
index ac33861edb48..38d17e2e2b4c 100644
@@ -71,7 +71,8 @@ static int pwm_lpss_probe_platform(struct platform_device *pdev)
         * 2. The state restoring code unexpectedly messing with the controller
         */
        if (info->other_devices_aml_touches_pwm_regs)
-               dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
+               dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE|
+                                                   DPM_FLAG_SMART_SUSPEND);
 
        pm_runtime_set_active(&pdev->dev);
        pm_runtime_enable(&pdev->dev);
--

2.28.0

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#5 lore raw

Re: [PATCH 3/3] pwm: lpss: Set DPM_FLAG_SMART_SUSPEND on Cherry Trail devices

Hi,

Quoted 18 lines · Andy Shevchenko
Quoted 17 lines · Hans de Goede

As the comment above the code setting the DPM_FLAG_NO_DIRECT_COMPLETE flag explains:

/*
* On Cherry Trail devices the GFX0._PS0 AML checks if the controller
* is on and if it is not on it turns it on and restores what it
* believes is the correct state to the PWM controller.
* Because of this we must disallow direct-complete, which keeps the
* controller (runtime)suspended, on resume to avoid 2 issues:
* 1. The controller getting turned on without the linux-pm code
* knowing about this. On devices where the controller is unused
* this causes it to stay on during the next suspend causing high
* battery drain (because S0i3 is not reached)
* 2. The state restoring code unexpectedly messing with the controller */

Quoted 8 lines
Quoted 6 lines

The pm-core must not skip resume to avoid the GFX0._PS0 AML code messing with the PWM controller behind our back. But

leaving the controller
runtime-suspended (skipping runtime-resume + normal-suspend) during suspend is fine.

This paragraph is good to have in the comment of the code I think.

Agreed, I've added this as comment for v2 of this patch set and I will send out a v2 of the entire series with your reviewed-by added.

Thanks & Regards,

Hans

Quoted 27 lines
Quoted 25 lines

Set the DPM_FLAG_SMART_SUSPEND flag to allow this.

---
 drivers/pwm/pwm-lpss-platform.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/pwm/pwm-lpss-platform.c +2 -2
index ac33861edb48..38d17e2e2b4c 100644
@@ -71,7 +71,8 @@ static int pwm_lpss_probe_platform(struct platform_device *pdev)
         * 2. The state restoring code unexpectedly messing with the controller
         */
        if (info->other_devices_aml_touches_pwm_regs)
-               dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
+               dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE|
+                                                   DPM_FLAG_SMART_SUSPEND);
 
        pm_runtime_set_active(&pdev->dev);
        pm_runtime_enable(&pdev->dev);
--

2.28.0

Reviewed-by on the v1 series
#6 lore raw
In reply to Hans de Goede
Quoted 5 lines · Hans de Goede

Hi All,

Now that the pwm-lpss / pwm-crc / i915 atomic PWM conversion has landed in 5.10-rc1 here is a small follow up series with some misc. improvements.

One nit against 3/3, in any case looks very good to me,

Signature · 2 lines
With Best Regards,
Andy Shevchenko