Hello,

My hardware is creating an interrupt handled by a kernel driver.
In order to notify the application there are 2 ways (that I know about):

1. Send a signal from driver to user space
2. User space can send IOCTL answered upon interrupt.

Can you please tell if one method is better than the other ?
What method is more popular ?

Thank you,
Zvika

I suggest that you "send a signal." The signal-handler can then use IOCTL to find whatever else it needs to know – and it can do so at any convenient time as it sees fit.

The whole idea is to avoid so-called "busy waiting."

Your first-level interrupt handler should gather all of the necessary information into a place that the IOCTL can then retrieve. Then, it launches a user-space signal, unless it notices that a signal has already been launched. The userland program "only pays attention to the telephone because it is ringing."

1 members found this post helpful.

You've been asking about kernel modules/code for years at this point:
https://www.linuxquestions.org/quest...on-4175697514/

Is there some point at which you'll be able to apply things you've been told and learned?? You've been asking about/working with IOCTL for TWELVE YEARS now; and that question is nearly IDENTICAL to this one:
https://www.linuxquestions.org/quest...-space-746821/

Hi,

Strange. I downloaded long term linux-5.10.76.tar.xz
and looked for send_sig_info in drivers folder.
There are only 3 (!!!) drivers that use it.

If using signals is better than IOCTLs (for application to be notified on interrupts) why it is almost not used ?
My collegue claims that with signals, the time between the interrupt to the notification in the application is much shorter compared to using IOCTL.

Thank you,
Zvika

Still don't have any idea why you're asking. You've been working on this for twelve years...you have a 'colleague' who works with you on this. So why don't you and your 'colleague' try both approaches on your hardware, and see which works better for you??? Run some actual tests....do some actual work on your own actual hardware.

Hi,

Before changing approach to signals,
I have to be sure that this method is better than using IOCTL based on the experience of the experts here.

Thank you,
Zvika

Wrong: you need to make sure it's better by actually testing it for your own purposes on your own hardware by doing your own work (bold for emphasis). What works in one scenario may not work as well for others, so the excuse of, "I need experts else to tell me" isn't believable.

Again, you have been working on this for TWELVE YEARS now; you have repeatedly asked about kernel modules/hardware/etc., so there is *ABSOLUTELY NOTHING* stopping you from testing things. If this is for your work, you have ample time/resources to do the best job you can; there is no reason NOT to try both approaches. You and your 'colleague' can work together, can't you?? If your 'colleague' says it's better and you say different, then you can ask them to prove it.

Do actual work; run tests; benchmark things. Again, none of this is nothing you haven't been told before, MANY TIMES:
https://www.linuxquestions.org/quest...er-4175695475/
https://www.linuxquestions.org/quest...-a-4175685514/
https://www.linuxquestions.org/quest...ng-4175671805/
https://www.linuxquestions.org/quest...ct-4175669923/
https://www.linuxquestions.org/quest...-a-4175668757/

1 members found this post helpful.