Data Transfer from FPGA-to-HPS

Can you add some signaltap probes around your master, to check that it is actually writing to the memory?

I have never developped on embedded with Linux, but basically you shoudln't need to limit the ram available to the operating system. Just allocate a block of memory (aligned with data cache lines), obtain it's real physical address (if you are using an MMU) and have your hardware write to it. Once it's done, don't forget to invalidate the data cache before accessing the data from the software.