Writing driver for a net device which does not support interrupt

DISCUSSION Freeman Zhang 4 participants 6 messages Mar 23, 2015 – Mar 24, 2015

Share & tools

Freeman Zhang

author

2015-03-23 10:51 UTC #0 lore raw

Hi list,

I'm writing a net device driver for my final project in college. But the half-finished device doesn't support interrupt yet(those hardware guys...)

So I'm wondering if there is some way to poll the device for its status and events.

Google tells me 'NAPI' uses a polling mechanism, but it still needs hardware interrupt support :(

Any suggestions?

Thanks
Freeman

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
Url : http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20150323/6441e188/attachment.bin

Valdis.Kletnieks at vt.edu

2015-03-23 15:51 UTC #1 lore raw

Quoted 5 lines · Freeman Zhang

I'm writing a net device driver for my final project in college. But the half-finished device doesn't support interrupt yet(those hardware guys...)
So I'm wondering if there is some way to poll the device for its status and events.

As a realistic datapoint - if it still doesn't support interrupts, the *rest* of it is probably still so buggy that trying to write a driver isn't worth the effort.

It's doable if you are working with a group of talented and experienced engineers - but if the hardware is also being done as a final project, you're in for naught but misery.

You *could* do something like this:

      while (waiting) {
          status = read_status_bits(your_device);
          if (status & DATA_AVAIL_MASK) break;
          msleep(100);
      }

But having seen enough student hardware design projects in my life, I estimate that if it still can't signal interrupts, the status bits probably aren't valid either. (Think about it - a bog-simple implementation of interrupts would be to just feed a transition-high on the appropriate status bit to the interrupt pin, and use a read from the chip to clear the pin).

Good luck - I suspect you're going to need it....
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 848 bytes
Desc: not available
Url : http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20150323/d9f1295d/attachment.bin

Freeman Zhang

author

2015-03-24 01:24 UTC #2 lore raw

Hi Valdis,

Thanks for your warning. I feel like I believe in my group and those engineers. In fact, implementing interrupts is part of the plan, but, yes, the prototype is really buggy.... However they are trying hard to fix it, and I should not stand by just to watch, waiting for the deadline, should I?

Anyway, this is a serious project. This work need to be done. And I just figured out something that might solve the current problem and might be easily modifid when hardware interrupt is available. I think I could use soft interrupt to active NAPI polling, so that I can use existing infrastructure, as what 'snull' does. What do you think?

Thanks again!
Freeman

-------- Original Message --------

Quoted 23 lines · Freeman Zhang

Quoted 5 lines
I'm writing a net device driver for my final project in college. But the half-finished device doesn't support interrupt yet(those hardware guys...)
So I'm wondering if there is some way to poll the device for its status and events.
As a realistic datapoint - if it still doesn't support interrupts, the *rest* of it is probably still so buggy that trying to write a driver isn't worth the effort.
It's doable if you are working with a group of talented and experienced engineers - but if the hardware is also being done as a final project, you're in for naught but misery.
You *could* do something like this:
      while (waiting) {
          status = read_status_bits(your_device);
          if (status & DATA_AVAIL_MASK) break;
          msleep(100);
      }

Quoted 8 lines

But having seen enough student hardware design projects in my life, I estimate that if it still can't signal interrupts, the status bits probably aren't valid either. (Think about it - a bog-simple implementation of interrupts would be to just feed a transition-high on the appropriate status bit to the interrupt pin, and use a read from the chip to clear the pin).
Good luck - I suspect you're going to need it....

phil

2015-03-23 16:59 UTC #3 lore raw

In reply to Freeman Zhang

Quoted 5 lines · Freeman Zhang

I'm writing a net device driver for my final project in college. But the half-finished device doesn't support interrupt yet(those hardware guys...)
So I'm wondering if there is some way to poll the device for its status and events.

You should take a look at kernel timers, this is a rather old resource and was just the first hit in a google search but it should still be relevant.

http://www.ibm.com/developerworks/library/l-timers-list/

Kernel timers will sort of prepare you for interrupts i.e. a kernel timer will fire every $time_value and you can put your code which should normally run during an interrupt in the callback code.

I've used kernel timers to debug interrupts on our hardware devices, i.e. there have been times when the hardware has stopped raising interrupts and I've put timers in to cover this. For example a new firmware might work fine, but have completely broken interrupts or not fire interrupts often enough and timers can prove to the hardware guys that it's interrupts at fault ;)

hint: you might also want to look at tasklets and have your timer callback just call a tasklet.

Phil

Freeman Zhang

author

2015-03-24 01:34 UTC #4 lore raw

Hi phil,

Thanks for your reply! I googled timer, it might be a solution except that interval granularity is a little bit large. This device is for high performance network. I'm not sure whether the missmatch will cause problem.

And I just figured another potential solution, to use soft interrupt to active NAPI. I will spend some time to varify these ideas.

Thanks again!
Freeman

-------- Original Message --------

Quoted 27 lines · Freeman Zhang

Quoted 6 lines
I'm writing a net device driver for my final project in college. But the half-finished device doesn't support interrupt yet(those hardware guys...)
So I'm wondering if there is some way to poll the device for its status and events.
You should take a look at kernel timers, this is a rather old resource and was just the first hit in a google search but it should still be relevant.
http://www.ibm.com/developerworks/library/l-timers-list/
Kernel timers will sort of prepare you for interrupts i.e. a kernel timer will fire every $time_value and you can put your code which should normally run during an interrupt in the callback code.
I've used kernel timers to debug interrupts on our hardware devices, i.e. there have been times when the hardware has stopped raising interrupts and I've put timers in to cover this. For example a new firmware might work fine, but have completely broken interrupts or not fire interrupts often enough and timers can prove to the hardware guys that it's interrupts at fault ;)

Quoted 4 lines

hint: you might also want to look at tasklets and have your timer callback just call a tasklet.
Phil

Daniel Baluta

2015-03-24 16:52 UTC #5 lore raw

Quoted 5 lines · Freeman Zhang

Hi phil,
Thanks for your reply! I googled timer, it might be a solution except that interval granularity is a little bit large. This device is for high performance network. I'm not sure whether the missmatch will cause problem.

Google for high resolution timers.

Daniel.