Hardfault in mouse.c

[mukululul]

Hello,

after finally getting the EEPROM working which now holds the calibration data for the touchpad I get a hardfault. When the data is retrieved with ginput_lld_mouse_calibration_load(). The function returns a const char* to the previously saved calibration data.

The code now casts to the struct Calibration* and stores the pointer in pc. Then pc is dereferenced using the [] operator (same as *pc);

This causes the compiler to generate code which copies the struct completely. While executing this code I get a hardfault on a STM32F103ZET.

Exchanging the compiler generated code with memcpy all works fine.

			if (MouseConfig.fnloadcal && (pc = (Calibration *)MouseConfig.fnloadcal(instance)))
			{
				//MouseConfig.caldata = pc[0];
			        memcpy(&MouseConfig.caldata,pc,sizeof(MouseConfig.caldata));
				MouseConfig.flags |= (FLG_CAL_OK|FLG_CAL_SAVED);
				if ((MouseConfig.flags & FLG_CAL_FREE))

Anybody here who can explain this?

My compiler ist arm-none-eabi-gcc.exe (GCC) 4.7.2.

Thanks for your help.

Regards

Michael

Edited December 11, 2013 by Guest

[inmarket]

Interesting problem as the two methods should be completely equivalent.

FYI - I am currently using the same compiler with no troubles but with a different processor.

To give you some background...

• There are a number of other places in the code where structure copy is used (although not all that commonly)
  
• We generally try to avoid using memcpy() to avoid dependence on system header files and system libraries (which are not always 100% compatible)

Some things to look at:

• Stack - perhaps the stack is being overrun?
  
• Does changing the compiler optimiser flags change the code generated (ie could this be an optimiser problem)?
  
• Exactly what hard-fault is being generated as that might indicate what is going on?
  
• Perhaps it is the compiler flags for code generation are not quite right for your processor?
  
• If it is using an implicit system library call to do the structure copy, perhaps the system library has been compiled incorrectly for your processor.
  
• Is it a bus alignment issue - ie the memory returned by the load function is non-bus aligned and the structure copy is assuming bus aligned data.

If you can please investigate these things that would be appreciated as I would prefer to avoid the use of memcpy unless necessary. This problem may also affect other area's of the code if it is a systemic problem that we need to handle.

Thanks for bringing this to our attention.

[mukululul]

Hi,

thank you very much for your detailed answer.

You nailed it with your last line. It is a alignment problem.

I was using a static buffer allocated by the compiler of course it was nicely aligned.

I am storing the calibration data in that buffer plus a header infront. This header is 3 bytes which is, I agree not very smart ;-)

I will change the headersize then the return pointer is aligned again.

Thanks again.

Michael

[inmarket]

Thanks for finding it!

I will adjust the code to fix this problem. I am a little swamped with my day job so it may be a little while but it will eventually be in there.

[inmarket]

This is now fixed in the master source repository.

Thanks for bringing it to our attention.