98
32072H–AVR32–10/2012
AT32UC3A3
Interrupt Priority Register (IPR). The GrpReq inputs are then masked by the mask bits from the
CPU status register. Any interrupt group that has a pending interrupt of a priority level that is not
masked by the CPU status register, gets its corresponding ValReq line asserted.
Masking of the interrupt requests is done based on five interrupt mask bits of the CPU status
register, namely Interrupt Level 3 Mask (I3M) to Interrupt Level 0 Mask (I0M), and Global Inter-
rupt Mask (GM). An interrupt request is masked if either the GM or the corresponding interrupt
level mask bit is set.
The Prioritizer hardware uses the ValReq lines and the INTLEVEL field in the IPRs to select the
pending interrupt of the highest priority. If an NMI interrupt request is pending, it automatically
gets the highest priority of any pending interrupt. If several interrupt groups of the highest pend-
ing interrupt level have pending interrupts, the interrupt group with the lowest number is
selected.
The INTLEVEL and handler autovector offset (AUTOVECTOR) of the selected interrupt are
transmitted to the CPU for interrupt handling and context switching. The CPU does not need to
know which interrupt is requesting handling, but only the level and the offset of the handler
address. The IRR registers contain the interrupt request lines of the groups and can be read via
user interface registers for checking which interrupts of the group are actually active.
The delay through the INTC from the peripheral interrupt request is set until the interrupt request
to the CPU is set is three cycles of CLK_SYNC.
10.5.1
Non-Maskable Interrupts
A NMI request has priority over all other interrupt requests. NMI has a dedicated exception vec-
tor address defined by the AVR32 architecture, so AUTOVECTOR is undefined when
INTLEVEL indicates that an NMI is pending.
10.5.2
CPU Response
When the CPU receives an interrupt request it checks if any other exceptions are pending. If no
exceptions of higher priority are pending, interrupt handling is initiated. When initiating interrupt
handling, the corresponding interrupt mask bit is set automatically for this and lower levels in sta-
tus register. E.g, if an interrupt of level 3 is approved for handling, the interrupt mask bits I3M,
I2M, I1M, and I0M are set in status register. If an interrupt of level 1 is approved, the masking
bits I1M and I0M are set in status register. The handler address is calculated by logical OR of
the AUTOVECTOR to the CPU system register Exception Vector Base Address (EVBA). The
CPU will then jump to the calculated address and start executing the interrupt handler.
Setting the interrupt mask bits prevents the interrupts from the same and lower levels to be
passed through the interrupt controller. Setting of the same level mask bit prevents also multiple
requests of the same interrupt to happen.
It is the responsibility of the handler software to clear the interrupt request that caused the inter-
rupt before returning from the interrupt handler. If the conditions that caused the interrupt are not
cleared, the interrupt request remains active.
10.5.3
Clearing an Interrupt Request
Clearing of the interrupt request is done by writing to registers in the corresponding peripheral
module, which then clears the corresponding NMIREQ/IREQ signal.
The recommended way of clearing an interrupt request is a store operation to the controlling
peripheral register, followed by a dummy load operation from the same register. This causes a