The DMA supports an AXI bus width of 128/64 bits. In the case where the source descriptor payload ends at a non-128/64 bit aligned boundary, the DMA channel fetches the last beat as the full-128/64 bit wide bus. This is considered an over fetch. The over fetch option can be disabled, if required. In the case where an over fetch is disabled and the SRC descriptor payload ends on a non-128/64 bit boundary, the DMA fetches any remaining bytes as a single byte AXI read.
The example in This Figure uses a source descriptor size of 8190 bytes (with a start address at 0x000_0000 and end address at 0x0000_1FFD), a 128-bit wide AXI bus, and a burst length of 16, the DMA can fetch 256 bytes in a single AXI burst. Two scenarios are documented.
Scenario 1: Over Fetch is Disabled
•31 AXI read command with burst length of 16 and AXI size of 16 bytes (7936 bytes fetched)
•One AXI read command with burst length of 15 and AXI size of 16 bytes (240 bytes fetched)
•To fetch the remaining 14 bytes, the DMA channel issues 14 single-beat AXI read commands with an AXI size of 1 byte.
Scenario 2: Over Fetch is Enabled
•32 AXI burst length of 16 and AXI size of 16 bytes (8192 bytes fetched)
RECOMMENDED: If the over fetch is disabled, it could significantly impact the performance of the DMA channel. AMD recommends only disabling the over fetch when absolutely necessary.