[PATCH 0/2] mfd: freeze delayed I/O work during suspend

[PATCH 0/2] mfd: freeze delayed I/O work during suspend

[Jiale Yao]

The SI476x polling callback and TPS65010 event worker issue I2C transfers
from delayed work. Their current workqueues remain active while devices are
being suspended, allowing pending callbacks to reach an already suspended
I2C adapter.

Move both callbacks to freezable workqueues. Preserve the TPS65010 worker's
power-efficient property by using system_freezable_power_efficient_wq.
Each patch is independent and can be applied separately.

Jiale Yao (2):
  mfd: si476x: queue polling work on freezable workqueue
  mfd: tps65010: queue I/O work on freezable workqueue

 drivers/mfd/si476x-i2c.c |  5 +++--
 drivers/mfd/tps65010.c   | 10 +++++-----
 2 files changed, 8 insertions(+), 7 deletions(-)

-- 
2.34.1

[PATCH 1/2] mfd: si476x: queue polling work on freezable workqueue

[Jiale Yao]

SI476x devices without an interrupt continuously poll their status over
I2C. The polling callback runs on the non-freezable system workqueue and
can issue a transfer after the adapter has been suspended.

Queue status polling on system_freezable_wq so it stops before device
suspend and continues after resume.

Signed-off-by: Jiale Yao <yaojiale02@163.com>
---
 drivers/mfd/si476x-i2c.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/mfd/si476x-i2c.c b/drivers/mfd/si476x-i2c.c
index 9bce720a0a08..8d952113f8a6 100644
--- a/drivers/mfd/si476x-i2c.c
+++ b/drivers/mfd/si476x-i2c.c
@@ -17,6 +17,7 @@
 #include <linux/regulator/consumer.h>
 #include <linux/i2c.h>
 #include <linux/err.h>
+#include <linux/workqueue.h>
 
 #include <linux/mfd/si476x-core.h>
 
@@ -97,8 +98,8 @@ static int si476x_core_config_pinmux(struct si476x_core *core)
 
 static inline void si476x_core_schedule_polling_work(struct si476x_core *core)
 {
-	schedule_delayed_work(&core->status_monitor,
-			      usecs_to_jiffies(SI476X_STATUS_POLL_US));
+	queue_delayed_work(system_freezable_wq, &core->status_monitor,
+			   usecs_to_jiffies(SI476X_STATUS_POLL_US));
 }
 
 /**
-- 
2.34.1

[PATCH 2/2] mfd: tps65010: queue I/O work on freezable workqueue

[Jiale Yao]

The TPS65010 event worker performs SMBus transfers and can reschedule
itself to poll the charging state. The power-efficient system workqueue
is not freezable, so pending work can access an I2C adapter after it has
been suspended.

Use the freezable power-efficient workqueue for every scheduling path.
This preserves the power-efficient behavior while freezing I/O across
system suspend.

Signed-off-by: Jiale Yao <yaojiale02@163.com>
---
 drivers/mfd/tps65010.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/drivers/mfd/tps65010.c b/drivers/mfd/tps65010.c
index 8a144ec52201..b038d67ff715 100644
--- a/drivers/mfd/tps65010.c
+++ b/drivers/mfd/tps65010.c
@@ -230,7 +230,7 @@ static int dbg_show(struct seq_file *s, void *_)
 	seq_printf(s, "mask2     %s\n", buf);
 	/* ignore ackint2 */
 
-	queue_delayed_work(system_power_efficient_wq, &tps->work,
+	queue_delayed_work(system_freezable_power_efficient_wq, &tps->work,
 			   POWER_POLL_DELAY);
 
 	/* VMAIN voltage, enable lowpower, etc */
@@ -388,7 +388,7 @@ static void tps65010_interrupt(struct tps65010 *tps)
 			&& (tps->chgstatus & (TPS_CHG_USB|TPS_CHG_AC)))
 		poll = 1;
 	if (poll)
-		queue_delayed_work(system_power_efficient_wq, &tps->work,
+		queue_delayed_work(system_freezable_power_efficient_wq, &tps->work,
 				   POWER_POLL_DELAY);
 
 	/* also potentially gpio-in rise or fall */
@@ -436,7 +436,7 @@ static irqreturn_t tps65010_irq(int irq, void *_tps)
 
 	disable_irq_nosync(irq);
 	set_bit(FLAG_IRQ_ENABLE, &tps->flags);
-	queue_delayed_work(system_power_efficient_wq, &tps->work, 0);
+	queue_delayed_work(system_freezable_power_efficient_wq, &tps->work, 0);
 	return IRQ_HANDLED;
 }
 
@@ -694,7 +694,8 @@ int tps65010_set_vbus_draw(unsigned mA)
 			&& test_and_set_bit(
 				FLAG_VBUS_CHANGED, &the_tps->flags)) {
 		/* gadget drivers call this in_irq() */
-		queue_delayed_work(system_power_efficient_wq, &the_tps->work,
+		queue_delayed_work(system_freezable_power_efficient_wq,
+				   &the_tps->work,
 				   0);
 	}
 	local_irq_restore(flags);
@@ -1052,4 +1053,3 @@ static void __exit tps_exit(void)
 	i2c_del_driver(&tps65010_driver);
 }
 module_exit(tps_exit);
-
-- 
2.34.1

Re: [PATCH 2/2] mfd: tps65010: queue I/O work on freezable workqueue

[sashiko-bot]

> The TPS65010 event worker performs SMBus transfers and can reschedule
> itself to poll the charging state. The power-efficient system workqueue
> is not freezable, so pending work can access an I2C adapter after it has
> been suspended.
> 
> Use the freezable power-efficient workqueue for every scheduling path.
> This preserves the power-efficient behavior while freezing I/O across
> system suspend.
> 
> Signed-off-by: Jiale Yao <yaojiale02@163.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261016100503.2581955-1-yaojiale02@163.com?part=2

Re: [PATCH 1/2] mfd: si476x: queue polling work on freezable workqueue

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [Medium] mfd: si476x: cross-device race on static io_errors_count
- [Low] mfd: commit message description capitalization
--

commit 42712fe1738df7fa331aa622d7a9ed0967efc78f
Author: Jiale Yao <yaojiale02@163.com>

    mfd: si476x: queue polling work on freezable workqueue

This commit changes the polling work for SI476x devices to use the freezable
system workqueue. This ensures that the continuous I2C polling stops before
device suspend and properly resumes afterward, preventing transfers after the
adapter has been suspended.

>     mfd: si476x: queue polling work on freezable workqueue

[Severity: Low]
This isn't a bug, but the MFD subsystem requires capitalizing the description
after the subsystem prefix. Should "queue" be "Queue"?

> diff --git a/drivers/mfd/si476x-i2c.c b/drivers/mfd/si476x-i2c.c
> index 9bce720a0a082..8d952113f8a6f 100644
> --- a/drivers/mfd/si476x-i2c.c
> +++ b/drivers/mfd/si476x-i2c.c

[ ... ]

> @@ -97,8 +98,8 @@ static int si476x_core_config_pinmux(struct si476x_core *core)
>  
>  static inline void si476x_core_schedule_polling_work(struct si476x_core *core)
>  {
> -	schedule_delayed_work(&core->status_monitor,
> -			      usecs_to_jiffies(SI476X_STATUS_POLL_US));
> +	queue_delayed_work(system_freezable_wq, &core->status_monitor,
> +			   usecs_to_jiffies(SI476X_STATUS_POLL_US));
>  }

[Severity: Medium]
Will this polling work cause regressions if multiple si476x devices are present?

Looking at si476x_core_i2c_xfer(), which is called during the polling loop to
check device status:

drivers/mfd/si476x-i2c.c:si476x_core_i2c_xfer() {
    ...
	static int io_errors_count;
	int err;
    ...
	if (err < 0) {
		if (io_errors_count++ > SI476X_MAX_IO_ERRORS)
			si476x_core_pronounce_dead(core);
	} else {
		io_errors_count = 0;
	}
    ...
}

Because io_errors_count is static and accessed without locking, won't
concurrent I2C transfers from different devices race to modify it?

A successful read on one device could falsely reset the error count for a
failing device, preventing it from being pronounced dead. Conversely,
concurrent errors across multiple devices could accumulate together and
falsely trigger si476x_core_pronounce_dead() prematurely.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261016100503.2581955-1-yaojiale02@163.com?part=1

Re:[PATCH 1/2] mfd: si476x: queue polling work on freezable workqueue

[jiale yao]

At 2026-10-16 18:05:02, "Jiale Yao" <yaojiale02@163.com> wrote:
>SI476x devices without an interrupt continuously poll their status over
>I2C. The polling callback runs on the non-freezable system workqueue and
>can issue a transfer after the adapter has been suspended.
>
>Queue status polling on system_freezable_wq so it stops before device
>suspend and continues after resume.
>

Fixes: 730a30ab7fba ("mfd: si476x: Add the bulk of the core driver")

>Signed-off-by: Jiale Yao <yaojiale02@163.com>
>---
> drivers/mfd/si476x-i2c.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
>diff --git a/drivers/mfd/si476x-i2c.c b/drivers/mfd/si476x-i2c.c
>index 9bce720a0a08..8d952113f8a6 100644
>--- a/drivers/mfd/si476x-i2c.c
>+++ b/drivers/mfd/si476x-i2c.c
>@@ -17,6 +17,7 @@
> #include <linux/regulator/consumer.h>
> #include <linux/i2c.h>
> #include <linux/err.h>
>+#include <linux/workqueue.h>
> 
> #include <linux/mfd/si476x-core.h>
> 
>@@ -97,8 +98,8 @@ static int si476x_core_config_pinmux(struct si476x_core *core)
> 
> static inline void si476x_core_schedule_polling_work(struct si476x_core *core)
> {
>-	schedule_delayed_work(&core->status_monitor,
>-			      usecs_to_jiffies(SI476X_STATUS_POLL_US));
>+	queue_delayed_work(system_freezable_wq, &core->status_monitor,
>+			   usecs_to_jiffies(SI476X_STATUS_POLL_US));
> }
> 
> /**
>-- 
>2.34.1

Re:[PATCH 2/2] mfd: tps65010: queue I/O work on freezable workqueue

[jiale yao]

At 2026-10-16 18:05:03, "Jiale Yao" <yaojiale02@163.com> wrote:
>The TPS65010 event worker performs SMBus transfers and can reschedule
>itself to poll the charging state. The power-efficient system workqueue
>is not freezable, so pending work can access an I2C adapter after it has
>been suspended.
>
>Use the freezable power-efficient workqueue for every scheduling path.
>This preserves the power-efficient behavior while freezing I/O across
>system suspend.
>

Fixes: 72cd799544f2 ("[PATCH] I2C: add i2c driver for TPS6501x")

>Signed-off-by: Jiale Yao <yaojiale02@163.com>
>---
> drivers/mfd/tps65010.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
>diff --git a/drivers/mfd/tps65010.c b/drivers/mfd/tps65010.c
>index 8a144ec52201..b038d67ff715 100644
>--- a/drivers/mfd/tps65010.c
>+++ b/drivers/mfd/tps65010.c
>@@ -230,7 +230,7 @@ static int dbg_show(struct seq_file *s, void *_)
> 	seq_printf(s, "mask2     %s\n", buf);
> 	/* ignore ackint2 */
> 
>-	queue_delayed_work(system_power_efficient_wq, &tps->work,
>+	queue_delayed_work(system_freezable_power_efficient_wq, &tps->work,
> 			   POWER_POLL_DELAY);
> 
> 	/* VMAIN voltage, enable lowpower, etc */
>@@ -388,7 +388,7 @@ static void tps65010_interrupt(struct tps65010 *tps)
> 			&& (tps->chgstatus & (TPS_CHG_USB|TPS_CHG_AC)))
> 		poll = 1;
> 	if (poll)
>-		queue_delayed_work(system_power_efficient_wq, &tps->work,
>+		queue_delayed_work(system_freezable_power_efficient_wq, &tps->work,
> 				   POWER_POLL_DELAY);
> 
> 	/* also potentially gpio-in rise or fall */
>@@ -436,7 +436,7 @@ static irqreturn_t tps65010_irq(int irq, void *_tps)
> 
> 	disable_irq_nosync(irq);
> 	set_bit(FLAG_IRQ_ENABLE, &tps->flags);
>-	queue_delayed_work(system_power_efficient_wq, &tps->work, 0);
>+	queue_delayed_work(system_freezable_power_efficient_wq, &tps->work, 0);
> 	return IRQ_HANDLED;
> }
> 
>@@ -694,7 +694,8 @@ int tps65010_set_vbus_draw(unsigned mA)
> 			&& test_and_set_bit(
> 				FLAG_VBUS_CHANGED, &the_tps->flags)) {
> 		/* gadget drivers call this in_irq() */
>-		queue_delayed_work(system_power_efficient_wq, &the_tps->work,
>+		queue_delayed_work(system_freezable_power_efficient_wq,
>+				   &the_tps->work,
> 				   0);
> 	}
> 	local_irq_restore(flags);
>@@ -1052,4 +1053,3 @@ static void __exit tps_exit(void)
> 	i2c_del_driver(&tps65010_driver);
> }
> module_exit(tps_exit);
>-
>-- 
>2.34.1

Re: [PATCH 1/2] mfd: si476x: queue polling work on freezable workqueue

[Lee Jones]

On Fri, 16 Oct 2026, sashiko-bot@kernel.org wrote:

> Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
> - [Medium] mfd: si476x: cross-device race on static io_errors_count

This needs an answer.

> - [Low] mfd: commit message description capitalization
> --
> 
> commit 42712fe1738df7fa331aa622d7a9ed0967efc78f
> Author: Jiale Yao <yaojiale02@163.com>
> 
>     mfd: si476x: queue polling work on freezable workqueue
> 
> This commit changes the polling work for SI476x devices to use the freezable
> system workqueue. This ensures that the continuous I2C polling stops before
> device suspend and properly resumes afterward, preventing transfers after the
> adapter has been suspended.
> 
> >     mfd: si476x: queue polling work on freezable workqueue
> 
> [Severity: Low]
> This isn't a bug, but the MFD subsystem requires capitalizing the description
> after the subsystem prefix. Should "queue" be "Queue"?
> 
> > diff --git a/drivers/mfd/si476x-i2c.c b/drivers/mfd/si476x-i2c.c
> > index 9bce720a0a082..8d952113f8a6f 100644
> > --- a/drivers/mfd/si476x-i2c.c
> > +++ b/drivers/mfd/si476x-i2c.c
> 
> [ ... ]
> 
> > @@ -97,8 +98,8 @@ static int si476x_core_config_pinmux(struct si476x_core *core)
> >  
> >  static inline void si476x_core_schedule_polling_work(struct si476x_core *core)
> >  {
> > -	schedule_delayed_work(&core->status_monitor,
> > -			      usecs_to_jiffies(SI476X_STATUS_POLL_US));
> > +	queue_delayed_work(system_freezable_wq, &core->status_monitor,
> > +			   usecs_to_jiffies(SI476X_STATUS_POLL_US));
> >  }
> 
> [Severity: Medium]
> Will this polling work cause regressions if multiple si476x devices are present?
> 
> Looking at si476x_core_i2c_xfer(), which is called during the polling loop to
> check device status:
> 
> drivers/mfd/si476x-i2c.c:si476x_core_i2c_xfer() {
>     ...
> 	static int io_errors_count;
> 	int err;
>     ...
> 	if (err < 0) {
> 		if (io_errors_count++ > SI476X_MAX_IO_ERRORS)
> 			si476x_core_pronounce_dead(core);
> 	} else {
> 		io_errors_count = 0;
> 	}
>     ...
> }
> 
> Because io_errors_count is static and accessed without locking, won't
> concurrent I2C transfers from different devices race to modify it?
> 
> A successful read on one device could falsely reset the error count for a
> failing device, preventing it from being pronounced dead. Conversely,
> concurrent errors across multiple devices could accumulate together and
> falsely trigger si476x_core_pronounce_dead() prematurely.
> 
> -- 
> Sashiko AI review · https://sashiko.dev/#/patchset/20261016100503.2581955-1-yaojiale02@163.com?part=1

-- 
Lee Jones

Re: [PATCH 2/2] mfd: tps65010: queue I/O work on freezable workqueue

[Lee Jones]

On Fri, 16 Oct 2026, Jiale Yao wrote:

> The TPS65010 event worker performs SMBus transfers and can reschedule
> itself to poll the charging state. The power-efficient system workqueue
> is not freezable, so pending work can access an I2C adapter after it has
> been suspended.
> 
> Use the freezable power-efficient workqueue for every scheduling path.
> This preserves the power-efficient behavior while freezing I/O across
> system suspend.
> 
> Signed-off-by: Jiale Yao <yaojiale02@163.com>
> ---
>  drivers/mfd/tps65010.c | 10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/mfd/tps65010.c b/drivers/mfd/tps65010.c
> index 8a144ec52201..b038d67ff715 100644
> --- a/drivers/mfd/tps65010.c
> +++ b/drivers/mfd/tps65010.c
> @@ -230,7 +230,7 @@ static int dbg_show(struct seq_file *s, void *_)
>  	seq_printf(s, "mask2     %s\n", buf);
>  	/* ignore ackint2 */
>  
> -	queue_delayed_work(system_power_efficient_wq, &tps->work,
> +	queue_delayed_work(system_freezable_power_efficient_wq, &tps->work,
>  			   POWER_POLL_DELAY);
>  
>  	/* VMAIN voltage, enable lowpower, etc */
> @@ -388,7 +388,7 @@ static void tps65010_interrupt(struct tps65010 *tps)
>  			&& (tps->chgstatus & (TPS_CHG_USB|TPS_CHG_AC)))
>  		poll = 1;
>  	if (poll)
> -		queue_delayed_work(system_power_efficient_wq, &tps->work,
> +		queue_delayed_work(system_freezable_power_efficient_wq, &tps->work,
>  				   POWER_POLL_DELAY);
>  
>  	/* also potentially gpio-in rise or fall */
> @@ -436,7 +436,7 @@ static irqreturn_t tps65010_irq(int irq, void *_tps)
>  
>  	disable_irq_nosync(irq);
>  	set_bit(FLAG_IRQ_ENABLE, &tps->flags);
> -	queue_delayed_work(system_power_efficient_wq, &tps->work, 0);
> +	queue_delayed_work(system_freezable_power_efficient_wq, &tps->work, 0);
>  	return IRQ_HANDLED;
>  }
>  
> @@ -694,7 +694,8 @@ int tps65010_set_vbus_draw(unsigned mA)
>  			&& test_and_set_bit(
>  				FLAG_VBUS_CHANGED, &the_tps->flags)) {
>  		/* gadget drivers call this in_irq() */
> -		queue_delayed_work(system_power_efficient_wq, &the_tps->work,
> +		queue_delayed_work(system_freezable_power_efficient_wq,
> +				   &the_tps->work,
>  				   0);

Use 100-chars to prevent this crazy line-wrapping.

>  	}
>  	local_irq_restore(flags);
> @@ -1052,4 +1053,3 @@ static void __exit tps_exit(void)
>  	i2c_del_driver(&tps65010_driver);
>  }
>  module_exit(tps_exit);
> -
> -- 
> 2.34.1
> 

-- 
Lee Jones