[NUCLEO_xxx] Fix issue with us_ticker_read() when -O3 -Otime options are used. #166
Conversation
Fix issue when the hal is compiled with -O3 -Otime
|
What was the nature of the issue that this commit is fixing? When I see code changes like this, I am pretty suspicious that there is just a bug in the code which is being covered up. One thing I notice is that the static SlaveCounter variable is referenced in both an IRQ handler and us_ticker_read() but yet doesn't have a volatile qualifier. |
|
The issue is that the wait timing is not correct when the -O3/-Otime is used. I have already made a test with volatile on the SlaveCounter variable but it didn't solve the problem. I didn't have the time yet to investigate in detail inside the assembly code. This is why I used the pragma -O0/-OSpace as a quick workaround. |
|
Hi, @bcostm : can you be more specific how incorrect is it? What times do you get with other optimization levels? I would add volatile qualifier as Adam suggested. Regards, |
|
Declaring SlaveCounter as volatile is a better solution. |
- Semaphore returns 0 on timeout, and negative was incorrect used for errors - Correctly checked error code on tcp_connect thanks to @LiyouZhou fixes #284, fixes #285, fixes #166
- Semaphore returns 0 on timeout, and negative was incorrect used for errors - Correctly checked error code on tcp_connect thanks to @LiyouZhou fixes ARMmbed#284, fixes ARMmbed#285, fixes ARMmbed#166
* Add macros to help define offsets in config * Use build system provided macros for flash size and addr
Updating mbed-os to mbed-os-5.9.3
The workaround is not needed anymore (v32 is breaking this reindexing as they created new command to do that anyway). Co-authored-by: Martin Kojtal <[email protected]>
No description provided.