segger embedded studio with gcc volatile problem

RE: segger embedded studio with gcc volatile problem

Ian — Fri, 13 Oct 2017 01:50:27 GMT

I got it. Now i know how to modify my code. Thank you very much!

RE: segger embedded studio with gcc volatile problem

Nguyen Hoan Hoang — Fri, 13 Oct 2017 01:47:09 GMT

many read mean different threads but if you have the write in a different thread, volatile is required. If all are in the same thread, no volatile needed.

RE: segger embedded studio with gcc volatile problem

Ian — Fri, 13 Oct 2017 01:41:27 GMT

I am wandering how to understand "many readers", i think different threads are "many readers"， but does different functions in one thread also means "many readers"? If so,does it mean every global variable should use 'volatile' qualifier except only one function in one thread read it? I think this is unreasonable and i hope only different threads are "many readers" and different functions in one thread are not "many readers".

RE: segger embedded studio with gcc volatile problem

Nguyen Hoan Hoang — Thu, 12 Oct 2017 15:51:54 GMT

If you only have one writer and many readers, volatile is sufficient. If you have multiple writers then atomic access is required. No matter which compiler you are using. You may not get the issue now with Keil does not mean you will not get it later with optimization turn on or with newer version because this is a C language thing not compiler thing. The behaviour of GCC complies with it. That is why you get the issue.

RE: segger embedded studio with gcc volatile problem

Ian — Thu, 12 Oct 2017 15:17:48 GMT

Yes, i agree with you about this, and i understand what the effect of "volatile" is. But, in my program, there are lots of variables which used like "g_ulUartTimeoutCounter" , for example:

if ((!g_ulMainLoopCounter) && (!g_tMainLoopInterface.iap_state))
{
    //do something
}

When i use keil MDK, it works well，and when i use SES, the code in this "if" segment never be executioned, even "g_ulMainLoopCounter" used 'volatile' qualifier, because the member "iap_state" of struct "g_tMainLoopInterface" is not 'volatile', and i change the value of 'iap_state' in other function(not isr).

So, if i want use SES(or other IDE used gcc),must i check every variable in my program and determine if the variable is needed use 'volatile' qualifier(it's very difficult i think)? Is there any easy solutions?

RE: segger embedded studio with gcc volatile problem

Ian — Thu, 12 Oct 2017 14:47:45 GMT

Yes, "g_ulUartTimeoutCounter" is a global variable, in the isr, check if "g_ulUartTimeoutCounter" is not zero, decrease it, in main function, wait until "g_ulUartTimeoutCounter" decrease to zero.

RE: segger embedded studio with gcc volatile problem

Nguyen Hoan Hoang — Thu, 12 Oct 2017 13:50:43 GMT

Did you mean it worked by adding volatile in front ? That is the correct behaviour. There is no bug with GCC but might be a bug in Keil if it is compiled with full optimization. The volatile key word tells the compiler to always load data from memory. This is for this type of multi-access to guarantee that the value is loaded before comparison. Without it, optimization will preload the value in register and use locally in the loop. Hence when the interrupt updates the value it will not reflect in the main loop. So you will loop endlessly.

RE: segger embedded studio with gcc volatile problem

Aleksei Vegner — Thu, 12 Oct 2017 13:04:17 GMT

I'm a little bit confused by your description. You are decreasing "g_ulUartTimeoutCounter" variable in the isr but the loop in main has "while (g_ulMainLoopCounter)". Is it right?