
62
PRELIMINARY
82454KX/GX (PB)
A
3.4
Data Integrity and Error Handling
Several data integrity features are included in the PB. These include ECC on the host data bus (450GX only),
parity on the host address, parity on the CPU Request/Response signals, and parity on the PCI bus. Error
logging (setting a status bit) and reporting (generating an error signal) are controlled by the PCICMD Register
(04–05h), PCISTS Register (06–07h), ERRCMD Register (70h), ERRSTS Register (71h), EXERRCMD
Register (C0–C3h), and EXERRSTS Register (C4–C7h).
3.4.1
HOST BUS ERRORS
The PB detects errors on the host bus by checking the ECC provided with data (450GX only) and the parity
provided with control signals. In turn, the PB will generate ECC with data (450GX only) and parity with control
signals so that bus errors can be detected by receiving clients.
Request Parity (RP#) is the parity signal for ADS# and REQ[4:0]# and is computed as even parity. AP[1:0]# are
the parity signals for A[35:3]# and are computed as even parity (AP1# is for A[35:24]# and AP0# is for A[23:3]#.
RSP# is the parity signal for RS[2:0]# and is computed as even parity. In addition, certain host bus protocol
violations are detected by the PB.
The PB (Compatibility PB in a 82454GX dual PB system) is responsible for responding to any unclaimed trans-
actions on the host bus. The PB uses a watchdog timer to monitor host response phases. The timer is started
at the end of a response phase if the In-Order Queue is not empty. If the timer expires before the next host
response phase, a host bus time-out has occurred. The time-out window for such an event is programmable to
1.5 or 30 milliseconds via the PBC Register (4Ch). This allows for several host to PCI transactions, which may
be blocking the progress of the In-Order Queue, to undergo multiple retries. When a host bus time-out occurs,
the PB (Compatibility PB in an 450GX dual PB system) claims the transaction by returning all 1’s to a read
transaction or “pretending” to accept data for a write transaction. This event is logged in the EXERRSTS
Register and can generate a hard fail or SERR#, if enabled in the EXERRCMD Register.
AERR#. If AERR# observation is enabled, then AERR# to NMI should be enabled in the EXERRCMD register
(C0-C3h). This allows software to accept an NMI to log or recover from the event.
BINIT#. A BINIT# on the Host bus creates a PCIRST# and resets the 450KX/GX PCIset host bus state
machines. This allows for logging or recovery from catastrophic bus errors.
3.4.2
PCI BUS ERRORS
The PB always detects address parity errors when it is not the PCI master, even if it is not the selected target.
The PB detects data parity errors if it is either the master or the target of a transaction, and optionally reports
them to the system. Address parity errors are reported using the SERR# signal. Data parity errors are reported
using the PERR# signal.
On the 450GX, ECC error checking is used for the data bus. The ECC check bits are provided by the
DEP[7:0]# signals.