Hi Wolfgang,
in my example we have four FPGAs. One is memory mapped while the other three are not. They all have the same register interface which is mapped by struct ihs_fpga { u16 reflection_low; u16 versions; ... u16 mc_tx_address; u16 mc_tx_data; u16 mc_tx_cmd; ... };
To have instances of those FPGA structs, we might create an array like this: struct ihs_fpga system_fpgas[] = { (struct ihs_fpga *)CONFIG_SYS_FPGA_BASE, (struct ihs_fpga *)NULL, (struct ihs_fpga *)NULL, (struct ihs_fpga *)NULL, }; The first element ist our memory mapped FPGA with base address CONFIG_SYS_FPGA_BASE. For the other three I use NULL as base address because I don't have any better idea what else to choose.
To probe those FPGAs we might have to do something like: for (k=0; k < 4; ++k) fpga_set_reg(k, &system_fpgas[k].reflection_low, REFLECTION_TESTPATTERN);
The implementation of fpga_set_reg() might look like:
void fpga_set_reg(unsigned int fpga, u16 *reg, u16 data) { switch (fpga) { case 0: out_le16(reg, data); break; default: mclink_send(fpga - 1, (u16)reg, data); break; } }
where mclink_send() is the routine to access the non memory mapped FPGAs through the memory mapped FPGA: int mclink_send(u8 slave, u16 addr, u16 data) { ... fpga_set_reg(0, &system_fpgas[0].mc_tx_address, addr); fpga_set_reg(0, &system_fpgas[0].mc_tx_data, data); fpga_set_reg(0, &system_fpgas[0].mc_tx_cmd, (slave & 0x03) << 14); }
The cast to u16 I am talking about happens, when mclink_send() is called.
Cheers Dirk