Type Iii) Correcting Faults Via Rte - Motorola CPU32 Reference Manual

6.3.2.6 (Type III) Correcting Faults via RTE

The preferred method of MOVEM bus fault recovery is to correct the cause of the fault
and then execute an RTE instruction without altering the stack contents.
The RTE recognizes that MOVEM was in progress when a fault occurred, restores the
appropriate machine state, refetches the instruction, repeats the faulted transfer, and
continues the instruction.
MOVEM is the only instruction continued upon return from an exception handler. Al-
though the instruction is refetched, the effective address is not recalculated, and the
mask is rescanned the same number of times as before the fault — modifying the code
prior to RTE can cause unexpected results.
6.3.2.7 (Type IV) Correcting Faults via Software
BERR exceptions can occur during exception processing while the processor is fetch-
ing an exception vector or while it is stacking. The same stack frame and SSW are
used in both cases, but each has a distinct fault address. The stacked faulted excep-
tion format/vector word identifies the type of faulted exception and the contents of the
remainder of the frame. A fault address corresponding to the vector specified in the
stacked format/vector word indicates that the processor could not obtain the address
of the exception handler.
A BERR exception handler should execute RTE after correcting a fault. RTE restores
the internal machine state, fetches the address of the original exception handler, rec-
reates the original exception stack frame, and resumes execution at the exception
handler address.
If the fault is intractable, the exception handler should rewrite the faulted exception
stack frame at SP + $14 + $06 and then jump directly to the original exception handler.
The stack frame can be generated from the information in the BERR frame: the pre-
exception status register (SP + $0C), the format/vector word (SP + $0E), and, if the
frame being written is a six-word frame, the program counter of the instruction causing
the exception (SP + $10). The return program counter value is available at SP + $02.
A stacked fault address equal to the current stack pointer may indicate that, although
the first exception received a BERR while stacking, the BERR exception stacking was
successfully completed. This is an extremely improbable occurrence, but the CPU32
supports recovery from it. Once the exception handler determines that the fault has
been corrected, recovery can proceed as described previously. If the fault cannot be
corrected, move the supervisor stack to another area of memory, copy all valid stack
frames to the new stack, create a faulted exception frame on top of the stack, and re-
sume execution at the exception handler address.
6.4 CPU32 Stack Frames
The CPU32 generates three different stack frames — the normal four- and six-word
frames, and the twelve-word BERR stack frame.
CPU32
REFERENCE MANUAL
EXCEPTION PROCESSING MOTOROLA
6-21