We are use a iMX6Quad CPU on one of our custom board. We are running Linux 4.19.25. We are having performance issue on the EIM bus. Our bus clock is 132MHz and the core clock is at 996MHz. The EIM bus width is 16 bits. With this configuration our access of a 16/32 data on this port is slow. A 16 bit access takes 350 nano seconds. We see this issue only while accessing through virtual memory from a user space. We have setup the EIM registers correctly as we can assess the data faster from a physical address in u-boot. But with the virtual address access it adds a ~250nanoseconds delay. Is this expected delay from the MMU for address translation? I have captured the Chip select signal on the scope. For a 32 bit access from cpu, the two 16 bit access is fast. But a two 16 bit access from the CPU is slow and is spaced 250 nanoseconds apart. I have attached the cs signal for two 16 bit cpu access and two 32 bit cpu access.
