Clear the pipeline when no transfers are active to make sure that we do not
get residual data on the first sample for the next transfer.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
We only do have 4 channels in this design. Reducing the number of supported
channels for the dac_util_unpack core to 4 from 8 lowers the DMA alignment
requirement from 128bit to 64bit. We need this since applications only
expect a DMA alignment requirement of 64bit.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
We have enough bridge interconnect to give each DMA its own, so use them.
This makes sure that they do not interfere with each others transfers to
much. The SDRAM controller side of the FPGA2SDRAM bridges FIFO runs at a
much faster frequency then what we are able to use in the fabric. So its
better to do the arbitration on that side of the bus to make sure that we
can utilize the buses in the FPGA fabric to the maximum for each DMA core.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Always using 128bit for the input word unnecessarily increases the DMA
alignment requirements. This breaks existing software which assumes that the
DMA alignment requirement is 64bit.
So make it configurable whether we want 8 or 4 channels and while we are at
it also make the channel width configurable.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
By setting the AXI controler interface type from axi4 to axi4lite we can use
the normal toplevel file with only a simple modification to add the awprot
and arprot signals.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
We can remove the Altera toplevel wrapper if we switch the axi4 control bus
to axi4lite and add the few missing signals that are required by the Altera
interconnect to both the control and the data buses.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
This is unused and unneeded. The AXI interconnect will make sure that a
peripheral only gets requests that are meant for it, there is no need to
check the address in the peripheral itself.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Make sure that if a read and a write request arrive on the very same clock
cycle to only accept one of them. The simple solution chosen here is to only
accept the write request when this happens and delay the acceptance of the
read request until the write request is finished.
This solution is not fair since a write request will always take precedence,
which in theory allows the write bus to starve the read bus. But in practice
we should never see that many write requests that we are unable to answer
the read request.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Wait for the master to accept the response for the current transaction
before we allow a new transaction to start.
This fixes problems in case the master is not ready to accept the response
when we make it available.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
We must make sure that the response ID is the same as the request ID when we
accepted the request. Otherwise we might respond with the wrong ID and the
system will lockup.
Also set rlast to 1 instead of 0.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Otherwise we get timing errors for the reset signal that is generated in the
50MHz clock domain, but used in the VGA PLL clock domain.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Istvan Csomortani
Rejeesh Kutty
Adrian Costina
Lars-Peter Clausen
Michael Hennerich
acozma
dbogdan