ARMv7M Exception Handler

Joel Sherrill joel.sherrill at oarcorp.com
Thu Aug 27 21:06:02 UTC 2015



On 8/27/2015 3:58 PM, Daniel Gutson wrote:
> On Thu, Aug 27, 2015 at 3:53 PM, sudarshan.rajagopalan
> <sudarshan.rajagopalan at vecna.com> wrote:
>> Hey guys,
>>
>> I was working on the exception handler for the CPU hard fault. Managed to
>> register the fatal error user extension to RTEMS and call the user defined
>> function when an exception occurs. But the pointer to the stacked frame
>> didn't have the correct register values(esp. the PC register). So I looked
>> into the assembly code in /cpukit/score/cpu/arm/armv7m-exception-default.c,
>> where it decides which stack pointer was used (MSP or PSP) before the
>> exception occurred depending on the error code returned in the Link
>> Register. After carefully examining all the assembly instructions, I guess
>> theres a little bug in the code.
>>
>> The instruction "cmn r2, #3\n" in line 31 basically compares the Link
>> Register(lr) to value 0xFFFFFFFD (negative #3, because CMN negates the RHS
>> and compares with LHS) and chooses MSP or PSP in the following IT block.
>> This is pretty cool! But, it will not work if you have the floating-point
>> unit (FPU) enabled in your processor, which is enabled in mine. With FPU
>> enabled, the error code returned is either 0xFFFFFFE9 or 0xFFFFFFED, for
>> which the above assembly instruction will not work out and MSP will be
>> selected always.
>>
>> Better way to do is to check the 2nd bit of the error code to determine
>> which stack pointer was used before the exception happened - "tst lr, #4\n"
>> and change the IT block from "itt ne" to "itt eq" and the "mov" and "add"
>> within this IT block.
>>
>> Have tested this with the above changes and it works. I have sent the patch
>> "0001-Fix-exception-handler-for-supporting-FPU.patch" to the devel mailing
>> list that fixes this problem. :)
>
> Nice. Have you tested this without FPU support too?

Daniel .. when you are happy, we can push it.

Should this go on the 4.11 branch and master? If so, it needs a ticket.

>> Thanks and Regards,
>> Sudarshan
>> _______________________________________________
>> devel mailing list
>> devel at rtems.org
>> http://lists.rtems.org/mailman/listinfo/devel
>
>
>

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel.sherrill at OARcorp.com        On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
Support Available                (256) 722-9985


More information about the devel mailing list