Greeting everyone.
I am experimenting with edma and gpmc on the beaglebone white and i have seen a behavior that i want someone to help me understand.
The setup is the following.
Pwm from waveform generator to xdma_event_intr2, Triggering edma transfer from gpmc ,to a buffer ., Triggering one gpmc read cycle .
The edma is implemented as A-sync for acnt=2 .
Gpmc is set for an 20ns read operation (cs falling to cs rising edge)
Event generation pwm and Cs0 are monitored by an osciloscope
I thought to put this setup under a frequency sweep and check for responces.
What i noticed is that initially(the first event) and for every event with the event generation pwm being lower that ~5.8 Mhz
it takes around 140 ns for the Cs0 falling edge to appear.
As the event generation pwm got at higher frequencies i started seeing that the Cs0 falling edges are begining to fall under the event rising edges.
Then as the event generation pwm got at even higher frequencies Cs0 periods started to get smaller that the 140ns i expected. They got smaller and smaller until they reached to what it seems to be the maximum of 12,5Mhz( signals started to be really messed up around 14,3Mhz).
At first i thought that it maybe because of multiple events and not one by one are going to the transfer controller but i imagine that would look like a small burst in that maximum frequency of around 12,5Mhz and it wouldn't be so proportional to the event frequency.
Is it some kind of optimization being done?
Can someone shed some light?
Thanks in advance
