[ECOS] Monitoring stack usage

[ECOS] Monitoring stack usage

[Greg Dyer]

Hello,

I am running an embedded environment using multiple threads in eCos, a
main thread and a logging thread. I've been testing stack usages vs
different stack sizes in the logging thread because I had suspected that
a stack overflow could be occurring in the logging thread causing it to
lock up. I enabled CYGFUN_KERNEL_THREADS_STACK_MEASUREMENT and started
printing the values from cyg_thread_measure_stack_usage as well as the
value from threadInfo.stack_used. 
The value from cyg_thread_measure_stack_usage and threadInfo.stack_used
are always identical. 

My testing:
Stack size of 4096 bytes. The usage got up to 3528 bytes (according to
cyg_thread_measure_stack_usage) 
Stack size of 10240 bytes. The usage got up to 3528 bytes (according to
cyg_thread_measure_stack_usage) 
Stack size of 20480 bytes. The usage got up to 5488 bytes (according to
cyg_thread_measure_stack_usage) 
I was hoping to see with the 10k stack size the usage growing above 4k.
In both the 4k and 10k stack sizes, the logging thread locked up. The
20k stack did not lock up.

I changed a few config parameters and ran the same code as the above
test (except for changing stack sizes) and found some strange results:
Stack size of 4096 bytes. The usage got up to 4096 bytes (according to
cyg_thread_measure_stack_usage) 
Stack size of 10240 bytes. The usage got up to 9820 bytes (according to
cyg_thread_measure_stack_usage) 
Stack size of 20480 bytes. The usage got up to 20208 bytes (according to
cyg_thread_measure_stack_usage) 
Stack size of 102400 bytes. The usage got up to 87350 bytes (according
to cyg_thread_measure_stack_usage) 
It seems that usage tracks fairly close to what the given stack size.
Only the 4k stack size locked up in this test.

Is there a reason that my stack usage could be growing so high in my
second test? The compiled code is identical, only some a few parameters
are changed in my running code. I was expecting to see my stack usage
possibly go above the 4k mark but not continue to grow with my stack
size. Is there a better way I can monitor stack usage to detect an
overflow so I can appropriately set a better stack size?

Thanks,
Greg

--
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

[ECOS] Re: Monitoring stack usage

[John Dallaway]

Hi Greg

On 06/03/13 16:29, Greg Dyer wrote:

> I am running an embedded environment using multiple threads in eCos, a
> main thread and a logging thread. I've been testing stack usages vs
> different stack sizes in the logging thread because I had suspected that
> a stack overflow could be occurring in the logging thread causing it to
> lock up. I enabled CYGFUN_KERNEL_THREADS_STACK_MEASUREMENT and started
> printing the values from cyg_thread_measure_stack_usage as well as the
> value from threadInfo.stack_used.

[ snip ]

> Is there a reason that my stack usage could be growing so high in my
> second test? The compiled code is identical, only some a few parameters
> are changed in my running code. I was expecting to see my stack usage
> possibly go above the 4k mark but not continue to grow with my stack
> size. Is there a better way I can monitor stack usage to detect an
> overflow so I can appropriately set a better stack size?

Of course, you are correct. Stack usage should not grow with stack size.

To help interpret what you are observing, it is important to understand
how stack usage measurement is implemented in eCos.

Enabling CYGFUN_KERNEL_THREADS_STACK_MEASUREMENT will cause a bit
pattern to be written across the entire stack during thread
initialisation. This bit pattern is then overwritten progressively as
more of the stack is used. Calls to cyg_thread_measure_stack_usage()
will calculate the usage by searching for the lowest address on the
stack that no-longer contains the bit pattern. If bad code is causing
even a single byte to be written to an unused part of the stack then
cyg_thread_measure_stack_usage() will find that byte and report
unexpectedly large stack usage.

By the way, do make sure that CYGPKG_INFRA_DEBUG and CYGDBG_USE_ASSERTS
are enabled while you track down this issue. Enabling these items will
also enable stack overflow checking (CYGFUN_KERNEL_THREADS_STACK_CHECKING).

I hope this helps...

John Dallaway
eCos maintainer
http://www.dallaway.org.uk/john

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss