Use of FIQs
===========

FIQ usage is now deprecated. When you modify the FIQ vector, you will need
to synchronise the code. However, as FIQ routines cannot call SWIs you are
unable to call OS_SynchroniseCodeAreas. For StrongARM compatibility we
recommend you use the FIQ vector as follows:

     0000001C: LDR PC,&00000020
     00000020: <address of FIQ handler>
     
and use the following code to do the synchronise manually when you write
the instruction at location &1C (no synchronise is required when you alter
the address at &20).

        .
        <alter FIQ vector>
        .
        MRC    CP15,0,R0,C0,C0           ; get processor ID
        AND    R0,R0,#&F000
        TEQ    R0,#&A000
        BNE    NotStrongARM
        MOV    R0,#&1C           
        MCR    CP15,0,R0,C7,C10,1        ; clean data cache entry for FIQ vector
        MOV    R0,#0
        MCR    CP15,0,R0,C7,C10,4        ; drain write buffer
        MCR    CP15,0,R0,C7,C5           ; flush whole instruction cache
  NotStrongARM
        .
        .
        .

If you want to write a complete FIQ handler into locations &1C to &FC, you
should clean each 32-byte data cache line containing written code thus:

  replace
        MOV     R0,#&1C
        MCR     CP15,0,R0,C7,C10,1
        
  with
        MOV     R0,#&E0              ; clean complete FIQ area
  01    MCR     CP15,0,R0,C7,C10,1   ; 32 bytes (1 cache line) at a time
        SUBS    R0,R0,#&20
        BGE     %BT01
        
This will usually be slower than the approach recommended above.

This is, of course, not a future-proof solution. We recommend that no new
products use FIQ code. If you feel you have a pressing need to use FIQs,
contact ART Developer Support for advice.