Problems passing rtems test sp37 on STM32F7x

[Jay Doyle]

On 10/28/2015 04:29 PM, Joel Sherrill wrote:
>
> On October 28, 2015 2:52:34 PM EDT, Jay Doyle <jay.doyle at vecna.com> wrote:
>> All,
>>
>> We are having problems passing the RTEMS sp37 test case using our
>> STM32F7x BSP.  I did some digging around it looks like the problem
>> stems
> >from the fact that test_interrupt_body calls rtems_interrupt_disable
>> twice and the function returns with most interrupts still disabled.
>> Later in the test, a timer is used which requires the system tick
>> interrupt to fire which never happens because its interrupt is still
>> disabled.
>>
>> Am I missing something?  Presumably this test case would fail for any
>> ARMV7M core since the implementation of the arm interrupt functions are
>>
>> generic for all ARMV7M cores and not specific to our BSP.
> If this the code sequence starting around line 440, it looks like this code was reworked during SMP development and is missing a second enable. It will require two level variables as best I can tell.
>
> This needs to be a ticket because it impacts 4.11 as well. Assuming someone confirms my analysis.
>
Yeah, it works so long as you save and restore the original interrupt 
level or you remove the second invocation of rtems_interrupt_disable.  I 
suspect that the intended implementation was to just have a single 
invocation of rtems_interrupt_disable since test_interrupt_inline only 
has one invocation and it does basically the same set of tests using 
inline functions.
>> Below is an annotated copy of the step in test_interrupt_body
>>
>> // sets basepri to 0x80 and returns previous level of 0
>> level = rtems_interrupt_disable()
>>
>> // sets basepri to 0x80 and returns previous level of 0x80
>> level = rtems_interrupt_disable()
>>
>> // temporary sets basepri to 0x80 -- which does nothing because it is
>> already 0x80
>> rtems_interrupt_flash( level );
>>
>> // since level is equal to 0x80 this sets basepri to 0x80
>> rtems_interrupt_enable( level )
>> .
>> .
>> .
>> function exits with basepri still set to 0x80 instead of 0x0
>>
>> _______________________________________________
>> devel mailing list
>> devel at rtems.org
>> http://lists.rtems.org/mailman/listinfo/devel
> --joel
>