[Trac #196] CaFe NAND short DMA is unreliable

[Zarro Boogs per Child]

#196: CaFe NAND short DMA is unreliable
-------------------------------+--------------------------------------------
 Reporter:  wmb at firmworks.com  |       Owner:  mfoster
     Type:  defect             |      Status:  new    
 Priority:  high               |   Milestone:  BTest-1
Component:  hardware           |    Keywords:         
-------------------------------+--------------------------------------------
 CaFe NAND DMA reads work reliably when the transfer length is 64 bytes or
 longer, but for shorter transfer lengths, the DMA controller gets hung.

 The external symptoms of the "hang" are that the "DMA Done Interrupt" bit
 (bit 28) in the Interrupt Register (offset 0x10) does not get set reliably
 after short transfers.

 Sometimes the first such transfer will work correctly - the DMA done bit
 will be set and the data will be transfered into memory.  On the second
 try, the data will be transferred into memory, but the DMA done bit will
 not be set.  On the third try, the data will not be transferred and the
 DMA done bit will not be set.

 The attachment shows a sequence in which numerous longer transfers work
 correctly, but the first transfer of fewer than 64 (0x40) bytes hangs (the
 hang is due to the poll for the DMA done bit never seeing that bit set).

-- 
Ticket URL: <http://dev.laptop.org/ticket/196>
One Laptop Per Child <http://laptop.org/>