ARM R4F, r1p3 manual Handling TCM parity errors, Handling TCM ECC errors, TCM arbitration

Level One Memory System

Handling TCM parity errors

If a TCM interface has been built with parity error checking, you can enable this by setting the appropriate bits in the Auxiliary Control Register. See c1, Auxiliary Control Register on page 4-38.If the BTCM interface has been built with two ports, parity checking can be enabled for each port individually. You can pin-configure the processor to set the enable bits and therefore enable parity checking on reset, by tying off the PARECCENRAM input as required.

Parity bits for the data are generated on all TCM writes, regardless of whether or not the parity bits are being checked on reads. When a parity error is detected on a TCM read, a precise abort is generated. The type of the abort is shown in the appropriate Fault Status Register (FSR) as being a precise parity error. The processor cannot correct parity errors in the TCM.

When you use the parity error detection scheme, the PARLVRAM input to the processor selects between odd and even parity.

Handling TCM ECC errors

If a TCM interface has been built with either 32-bit or 64-bit ECC error checking, you can enable this by setting the appropriate bits in the Auxiliary Control Register. See c1, Auxiliary Control Register on page 4-38.On the BTCM interface, ECC checking can only be enabled for both ports or neither port. You can pin-configure the processor to set the enable bits and therefore enable ECC checking on reset, by tying off the PARECCENRAM input as required.

When a fatal error, that is, a 2-bit ECC error, is detected on a TCM read, an error is generated. Instruction and data reads generate the appropriate type of precise abort, and the AXI-slave interface returns a SLVERR response to the AXI system.

When a correctable error, that is, a 1-bit ECC error, is detected on a TCM read made by the

AXI-slave interface, the processor corrects the data inline before returning to the system.

When a correctable ECC error is detected on a TCM read made by the instruction-side or data-side, the processor normally generates the correct data and writes it back to the TCM. In the meantime, the processor retries the read to fetch the correct instruction or data. By setting the appropriate bits in the Secondary Auxiliary Control Register, you can disable this behavior. See c15, Secondary Auxiliary Control Register on page 4-41.Instead of correcting the error in the TCM, the processor generates the appropriate type of precise abort.

All ECC code generation and ECC checking must be performed on a complete data chunk, either 32-bits or 64-bits depending on the configuration. If a read access smaller than the data chunk is required, the whole chunk is read. If a write smaller than the data chunk is required, the processor must perform read-modify-write to generate the correct data and ECC code, but it only does this when ECC error checking is enabled. The data read as part of the read-modify-write sequence is checked for ECC errors, and the errors are handled in the same way as for any other TCM read. The ECC code is generated and written to the TCM for every write, regardless of whether error checking is enabled or not, but the code is only correct if the write was of a complete data chunk or if the processor performed read-modify-write to generate the complete data chunk. All data and instruction aborts generated by the ECC logic are indicated in the appropriate FSR as being a precise parity error.

8.4.4TCM arbitration

Each TCM port receives requests from the LSU, PFU, and AXI slave. In most cases, the LSU has the highest priority, followed by the PFU, with the AXI slave having lowest priority.

When a higher-priority device is accessing a TCM port, an access from a lower-priority device must stall.

ARM DDI 0363E	Copyright © 2009 ARM Limited. All rights reserved.	8-15
ID013010	Non-Confidential, Unrestricted Access