omapl138 large latency when dma reads uart rx

i am using edma to store incoming rx chars from uart 2 in L2 SRAM.  i am using a "bare metal" DSP application.

the uart baudrate is 460 kbaud, so 14-byte packets arrive in about 1/3 msec.  I have confirmed this using a scope.

The edma stores the incoming data correctly, but it takes several msec to do so -- i poll the dma index register CCNT periodically, so i can monitor the data transfer.

Is the latency in the uart, or the dma?  is there any way to fix this?