Daniel F. Dickinson wrote at Wed Aug 1 14:17:41 PDT 2018:
> Is this reported to do nasty things *without* debugging?
Daniel,
I compiled for WNDR3800 and your commits totally broke the switch/interface
detection/initialisation. That was with the default Openwrt kernel options.
No adjustments for debugging.
Like I wrote in my detailed bug message attached to your original PR and
referenced in jow's commit message:
https://github.com/openwrt/openwrt/pull/1217#issuecomment-409708087
* reverting both your commits fix things in WNDR3800
* reverting only the OOPS prevention leads into OOPS (no surprise), as your
first commit adds the dev_info message code likely generating the OOPS.
Apparently your first commit tries to give so a detailed error message that
all needed info is not available and that leads to OOPS. Then your second
commit tries to prevent the OOPS (which you also noticed??), but you placed
all your OOPS detection code before the actual tasks of the function, so that
the actual tasks are not run for some devices.
Additionally, you adopted a new return code ENODEV that may cause unexpected
things in the calling function. Earlier this function has returned 0 or
EINVAL, but you added ENODEV as a return code. I have not checked there, so
just an idea...
Things might work if the OOPS prevention code was inside the (!ret) condition
block where the values are actually needed, and it would not invent new
return codes. Simply check that the needed data is available for the detailed
dev_info message that you are creating. If all data is not available, write a
simpler dev_info message, but do not adjust return value of the function itself.
_______________________________________________
openwrt-devel mailing list
[email protected]
https://lists.openwrt.org/mailman/listinfo/openwrt-devel
