[USRP-users] USRP E100 FPGA - OMAP communication

Muez Berhane Reda m.reda at isispace.nl
Tue Sep 10 04:09:29 EDT 2013


On 9/9/2013 5:08 PM, Philip Balister wrote:
> On 09/09/2013 10:54 AM, Muez Berhane Reda wrote:
>> On 9/9/2013 4:35 PM, Philip Balister wrote:
>>> On 09/09/2013 09:50 AM, Muez Berhane Reda wrote:
>>>> On 9/9/2013 3:31 PM, Philip Balister wrote:
>>>>> On 09/09/2013 08:44 AM, Muez Berhane Reda wrote:
>>>>>> Thank you for the reply Philip,
>>>>>>
>>>>>> 1- The FPGA is memory mapped to GPMC chip select 4, where it writes
>>>>>> the
>>>>>> samples to and DMA of ARM is responsible for copying from GPMC to the
>>>>>> ring buffers available.
>>>>>>         So, are we assuming the DMA copy is faster than the rate
>>>>>> FPGA is
>>>>>> writing to the GPMC memory?
>>>>> The DMA is reading samples from the FPGA and writing them to RAM. The
>>>>> FPGA cannot bus master, so all the transfers are done via the OMAP DMA
>>>>> controller.
>>>> Do you mean the FPGA have its own memory or is it memory mapped to
>>>> GPMC CS?
>>> There is some FPGA memory that stores samples until the ARM copied from
>>> them into the OMAP memory (which is available to the ARM and DSP)
>> 1- Ok, So if I use DSP, instead of ARM, I have to copy from that FPGA
>> memory, not from GPMC CS 4?
>>
>> 2- How is the ARM informing FPGA it have finished DMA copying from the
>> FPGA memory? because I have to do the same after EDMA copy from DSP.
> You read from the memory at CS4. Reading from this address range gets
> the data from the FPGA.
>
> I believe we ended up copying 2048 bytes/words at a time. Each block is
> structured as a VITA-49 packet. I'd have to look at the code for the
> exact details.

In the usrp_e kernel module, I see that the size of transfer is hard-coded to 2048. It copies 2048 samples at a time.

My question is why do I stop getting GPIO interrupts to DSP(I receive only one interrupt) after disabling the GPIO pin
interrupt enable of ARM? Could it be because the DSP is not doing the signaling /communication with FPGA the same way as the ARM was doing it( via usrp_e and UHD module)?


regards,
Muez

> Philip
>
>
>>>>>> 2- I am working on offloading this task to c64x DSP on the
>>>>>> OMAP3530, but
>>>>>> I keep reading the same block of samples (using EDMA) on every GPIO
>>>>>> interrupt received by the DSP. This is when GPIO interrupt enable (IE)
>>>>>> registers of ARM and DSP are both set. When I disable the IE for
>>>>>> ARM, I
>>>>>> only get 1 data ready  GPIO interrupt to DSP at the beginning of the
>>>>>> program.
>>>>>>
>>>>> I'm not following you. There is a GPIO from the FPGA that tells the ARM
>>>>> when there is data available to read.
>>>> Yes, that GPIO pin tells the ARM when there is data available to read.
>>>> In addition to that,
>>>> The GPIO module can also send event(interrupt) to IVA2.2 Interrupt
>>>> Controller(DSP) of the OMAP. It have two IE registers for both ARM
>>>> and DSP.
>>>> The default is, IE for DSP is disabled. I enabled it and I started
>>>> getting GPIO interrupts to DSP and I
>>>> used EDMA to copy the samples from the GPMC CS4 memory address.
>>>>
>>>> But the problem is at this time both ARM (via DMA) and DSP (via EDMA)
>>>> are accessing the same
>>>> memory locations, which I think is causing the problem of reading the
>>>> same block of samples from the GPMC memory.
>>>>
>>>> As a result I disabled the ARM's IE register of the Data_Ready GPIO
>>>> pin. I
>>>> only get 1 data ready  GPIO interrupt to DSP at the beginning of the
>>>> program. I don't know why.
>>>>
>>>> I hope I make myself clear.
>>>>
>>> It sounds like you need to unload the ARM driver with rmmod so it does
>>> not conflict with your DSP based driver.
>> Yes, I am doing that after loading the FPGA image and issuing stream
>> command from the UHD program.
>>
>>> Philip
>>>
>>>
>>>>> Philip
>>>>>
>>>>>> regards,
>>>>>> Muez
>>>>>>
>>>>>> On 9/9/2013 2:29 PM, Philip Balister wrote:
>>>>>>> On 09/09/2013 06:27 AM, Muez Berhane Reda wrote:
>>>>>>>> Dear all,
>>>>>>>>
>>>>>>>> I have a question on USRP E100.
>>>>>>>>
>>>>>>>> The FPGA sends GPIO interrupts to Gumstix (OMAP) to notify that
>>>>>>>> Data is
>>>>>>>> available for it to read.
>>>>>>>> How does the FPGA know that the OMAP system have finished
>>>>>>>> accessing(reading from) the GPMC CS , so that it can write to that
>>>>>>>> memory location again. What is the signaling mechanism used to
>>>>>>>> prevent
>>>>>>>> FPGA from writing new samples while the DMA of OMAP is on the
>>>>>>>> process of
>>>>>>>> copying the old samples?
>>>>>>> There is a large buffer in the OMAP RAM that samples are copied form
>>>>>>> the
>>>>>>> FPGA to memory, so there are multiple destination buffers
>>>>>>> available for
>>>>>>> the fpga to write to. As these buffers are processed by the ARM, they
>>>>>>> are returned to the pool of buffers available for the FPGA to write
>>>>>>> samples to.
>>>>>>>
>>>>>>> Philip
>>>>>>>
>>>>>>>> regards,
>>>>>>>>
>>>> regards,
>>>> Muez
>>>>
>> thank you,
>>
>> regards,
>> Muez
>>


-- 
Muez Berhane Reda,

ISIS - Innovative Solutions In Space B.V.
Molengraaffsingel 12-14 | 2629 JD  DELFT | The Netherlands

T: +31(0)63 437 1234 |
E: m.reda at ISISpace.nl | www.ISISpace.nl





More information about the USRP-users mailing list