Task WDT triggered in __log2() and weird program behavior

[freemind908]

I'm looking into an issue where the system hangs and would trigger task watchdog if enabled. Backtrace reported by WDT looks like the following:

backtrace.png (136.27 KiB) Viewed 1922 times

Seems the hang happens inside a call to __log2(), which is itself a simple function:

Code: [Select all] [Expand/Collapse]

1. __uint32_t
2. __log2(num)
3.     __uint32_t num;
4. {
5.     __uint32_t i, limit;
6.  
7.     limit = 1;
8.     for (i = 0; limit < num; limit = limit << 1, i++);
9.     return (i);
10. }

GeSHi © Codebox Plus Extension

So I attached gdb to a running (hanging) instance and examined the assembly listing as running on the core:

assembly.png (50.68 KiB) Viewed 1922 times

Here is the weird part: the jump address for instruction at <__log2+17> is nowhere inside the valid range (not a start address of an instruction), right? What could be wrong?

Thanks for your help!

[freemind908]

Further digging indicates the hang is related to -1 being passed to __log2() and the way this function is implemented (left shift and compare). However, I'm still curious about the jump address not being the start address of any instruction. Is this a feature of Xtensa architecture?