Pin Monitor 100% CPU

[splitice]

The TIOCMIWAIT ioctl returns EINVAL on some drivers/hardware. This results in 100% CPU usage for the thread as instead of blocking the function call immediately returns.

Perhaps it would be better to just introduce a delay (sleep) in cases like this since you can't wait on pin changes on this hardware.

[jcurl]

Thanks for the bug report! I'd also like to document this in the wiki. What hardware specifically do you have that shows this behaviour?

[splitice]

It's the g_serial driver for a USB virtual serialport (/dev/ttyACM0) interfacing with a TI CC2531.

[jcurl]

Investigated the issue. The proper solution is to propogate the error to UnixNativeSerial class which will abort all pin monitoring operations. So instead of polling in case of this error, I just stop. I assume that if it doesn't work once, it will stop working. I don't expect that the ioctl() should return an error otherwise (except for EAGAIN or EWOULDBLOCK, in which case we'd try again immediately).

[splitice]

I think that it's a fair bet for EINVAL, perhaps restrict it to just that error code for now (unless a case of a different code comes up).

For our information if this functionality is (Pin Monitoring) is disabled, what effect does it have on functionality? Might be good to document too.

[jcurl]

I'll continue the loop for only EAGAIN or EWOULDBLOCK as a signal may occur at any time. All others I'll treat as fatal. When the monitoring thread stops, a user will no longer get notification of pin state changes CTS, DTR, DCD and RI.

I had an old phone that also provided a /dev/ttyACM0 interface, unfortunately, I can't test with that (the ioctl works). I've tested by changing code in such a way that the error occurs and I get the result I expect (no 100% CPU, the thread stops and the program continues as expected).

[splitice]

Could I make a suggestion?

A better failure mode for devices that don't support waiting (and those who fail at it) would be to update pin changes using a thread with a fixed wait time (i.e 10ms) IMHO.

We don't use that functionality currently, so it's not an issue for us. Just saying.

[splitice]

And I'm happy to test any changes you make on Monday on the hardware, no issue.

[jcurl]

I've made a first draft in github with the branch "bugfix/modem". There are a few other fixes present when I reviewed the code. I'll look over your suggestion and see what I can do.

[jcurl]

Hu, wondering if you had the chance to test before I merge to the main branch.

[splitice]

Sorry I haven't my hardware got feature frozen for a demos.

Unfortunately that's the only one I have built at the moment. Another one will be getting built very soon, probably early next week,

[jcurl]

Would you be able to review the code? I'd like to be ready to merge on the weekend. Thanks for your help.

[splitice]

looks fine to me

[jcurl]

Just did a review of my own after a few days. I've found a few bugs, so I'll have to clean them up first.

[jcurl]

Bugs fixed and now pushed to v2.x

[splitice]

@jcurl I tested your newest version today. The issue is back.

strace shows:

[pid  2142] ioctl(3, TIOCMIWAIT, 0x1e0) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCGICOUNT, 0xb3afedd4) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMIWAIT, 0x1e0) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCGICOUNT, 0xb3afedd4) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMIWAIT, 0x1e0) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCGICOUNT, 0xb3afedd4) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMIWAIT, 0x1e0) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCGICOUNT, 0xb3afedd4) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMIWAIT, 0x1e0) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCGICOUNT, 0xb3afedd4) = -1 EINVAL (Invalid argument)
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0
[pid  2142] ioctl(3, TIOCMGET^C, [TIOCM_DTR|TIOCM_RTS|TIOCM_CTS]) = 0

[splitice]

The problem appears to be that the pthread continues to get recreated after a failure.

[jcurl]

Just a sanity question, did you update and use the newer libnserial package? Did you compile from source or install the debian package?

[splitice]

yes using the latest libnserial compiled from git, and the latest .net side from NuGet.

I forgot to mention that that strace is above is grep'ed with ioctl.

I did some debugging and found that although the pthread is exited on failure, it gets restarted when requested again.

[pigwing]

hi, i have the same problem in raspberry pi 3 b+

[pigwing]

i found that,i am use 2.1.2 version the cpu not 100% and the 2.1.4 when i use it (serialportstream object singleton) read while 7 or 8 hours my raspberry cpu will cpu 100% all time.and that 2.1.2 will not

[jcurl]

I'm open for patches to review if you can provide a generic solution without regressions for standard hardware.

[pigwing]

i have catch crash info double free or corruption (fasttop): 0x5c011c18 ***

[pigwing]

i found that 2.1.2 always will let cpu 100%