Skip to contentSkip to conversation

Implementing a driver that listen to a GPIO

DISCUSSION Clemens Gruber 3 participants 3 messages Oct 2, 2014 – Oct 3, 2014

lore
Share & tools
lore
#0 lore raw

Hi,

I am just trying to solve a similar problem as Luca described. I have a TQMa28 board running Linux 3.17-rc7 though.
The capacitors keep the system up for about 1500 more milliseconds after the power fails. A GPIO line is set to high as soon as the main power is lost.

So I have not much time to quit an important user space process and to clean up the ext4 root filesystem before the power fails.

At the moment I pass rootflags=data=journal,commit=1 in the kernel commandline to lower the risk of data loss.
In the user space application I open important files using the O_SYNC flag.

Is there something else I should do to avoid data loss?

How would you implement the "emergency cleanup on power-failure" logic? Dedicated driver (similar to the mentioned gpio-event driver?), implement it with UIO, let a userspace application running as root select /dev/uioX and cleaning up?

Or should I create a separate partition on my eMMC just for this user space application data and use a readonly rootfs for the rest?

Thanks for your help!

Best regards,
Clemens

#1 lore raw

2014-10-02 10:01 GMT-07:00 Clemens Gruber <[email protected]>:

Quoted 15 lines

Hi,

I am just trying to solve a similar problem as Luca described. I have a TQMa28 board running Linux 3.17-rc7 though.
The capacitors keep the system up for about 1500 more milliseconds after the power fails. A GPIO line is set to high as soon as the main power is lost.

So I have not much time to quit an important user space process and to clean up the ext4 root filesystem before the power fails.

At the moment I pass rootflags=data=journal,commit=1 in the kernel commandline to lower the risk of data loss.
In the user space application I open important files using the O_SYNC flag.

Quoted 6 lines

Is there something else I should do to avoid data loss?

How would you implement the "emergency cleanup on power-failure" logic? Dedicated driver (similar to the mentioned gpio-event driver?), implement it with UIO, let a userspace application running as root select /dev/uioX and cleaning up?

how about using netlink to notify the usersapce app on gpio interrupt .

Quoted 13 lines

Or should I create a separate partition on my eMMC just for this user space application data and use a readonly rootfs for the rest?

Thanks for your help!

Best regards,
Clemens

_______________________________________________
Kernelnewbies mailing list
Kernelnewbies at kernelnewbies.org
http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies

#2 lore raw

Hi,

Quoted 16 lines · Jinqiang Zeng

2014-10-02 10:01 GMT-07:00 Clemens Gruber <[email protected]>:

Quoted 15 lines

Hi,

I am just trying to solve a similar problem as Luca described. I have a TQMa28 board running Linux 3.17-rc7 though.
The capacitors keep the system up for about 1500 more milliseconds after the power fails. A GPIO line is set to high as soon as the main power is lost.

So I have not much time to quit an important user space process and to clean up the ext4 root filesystem before the power fails.

At the moment I pass rootflags=data=journal,commit=1 in the kernel commandline to lower the risk of data loss.
In the user space application I open important files using the O_SYNC flag.

Quoted 7 lines
Quoted 6 lines

Is there something else I should do to avoid data loss?

How would you implement the "emergency cleanup on power-failure" logic? Dedicated driver (similar to the mentioned gpio-event driver?), implement it with UIO, let a userspace application running as root select /dev/uioX and cleaning up?

how about using netlink to notify the usersapce app on gpio interrupt .

thanks to all for your replies.

Anyway, I solved the issue simply using the GPIO driver (no kernel modifications needed):

- export GPIO number (es 7):

$ echo 7 > /sys/class/gpio/export

- set direction:

$ echo "in" > /sys/class/gpio/gpio7/direction

- set event that triggers interrupt:

$ echo "falling" > /sys/class/gpio/gpio7/edge

Then I create a user space application that executes a poll system call on "/sys/class/gpio/gpio7/value".

Regards

---
Luca Ellero
http://http://www.brickedbrain.com