Embedded Processors and Peripherals

[gordwait]

V5 LX30T fpga, EDK 10.1.03,

Using xilkernel 4.00.a

I let BSB build a basic microblaze core, all default settings, and added a uart. 

I started with standalone OS, and downloaded the "TestApp_Memory" default app and ran it successfully, including the print command output to an external serial terminal via stdout. 

All I want to do next is add a "sleep(2);" command in between two prints. 

Found out the standalone OS doesn't do sleep, so I loaded xilkernel and added the hardware timer xps_timer to the soup, enabled the drivers etc. 

The "default" TestApp_Memory doesn't do anything but print two lines (since it doesn't want to test the memory the program is running from). 

The basic program unmodified does this:

   print("-- Entering main() --\r\n");

   print("-- Exiting main() --\r\n");

 This works as expected, printing each message just once. 

When I add a sleep command:

   print("-- Entering main() --\r\n");

   sleep(2);

   print("-- Exiting main() --\r\n");

The program prints the start line over and over again till I reset the cpu, with about a 2 second gap each time:

-- Entering main() --

-- Entering main() --

-- Entering main() --

-- Entering main() --

-- Entering main() --

etc. 

It's as if the sleep command  causes a CPU warm restart every time.

Any ideas? 

[bassman59]

sounds like a watchdog timer is enabled, and you are not petting the dog.

-a

[gordwait]

Nope, no watchdog present..  and I left the dog at home! :)

[gordwait]

Seems like some library issue - I tried using usleep instead to see what it would do, and the code won't even compile.

I added    "usleep(10000);"

and the EDK gives me this:

......../my_microblaze/TestApp_Memory/src/TestApp_Memory.c:71: undefined reference to `usleep'

[jagron]

The strange behavior sounds like a heap/stack overflow.  Try generating a new linker-script that has a bigger stack and heap, and re-run your program.

-JasonI

[brianhill]

I notice you mention that sleep() is called from main().  I suspect that a xilkernel thread can't be put  to sleep until the kernel has been initialized -- after

    xilkernel_main();
(which will not return) in which case you could sleep() in one of the threads that were created.  The scheduler isn't running yet otherwise.

-Brian

[aminfar1]

I have similar problem. I don't need to use any threads. I just need a sleep function to sleep the processor for some mili-seconds. However, since I needed to use sleep fucntion, I was forced to use xilkernel. The program compiled well, but when I ran it, it kept restarting. Then I found out that I have not used xilkernel_main in my program. So I added it, but the control never reached after the xilkernel_main function (that is expected). So my question is how I can use sleep function without having threads? I really need a sleep function but my program is a simple single thread program that interacts with the user.

[gordwait]

I ended up making my own gord_sleep() subroutine and bypassing all this other nonsense.

All it does is waste some cpu cycles in a for loop (loopcount supplied by the caller), so it's very approximate - all I needed..

[william.kafig]

Doesn't the sleep function wait the for the specified number of seconds? So, sleep(2) should make it sleep for 2 seconds - the exact behavior that you are observing. If you need finer granularity, try usleep()

[gordwait]

No, the original sleep was not behaving, it was causing a 2 second delay (so far so good) then rebooting the CPU so it re-enters main over and over again. It should only print the "entering main" message once.