Hello Tom,
Am Dienstag, den 12.02.2013, 12:24 -0700 schrieb Tom Warren:
Lucas,
tegra_mmc_init should not be called from every individual board file, but from the common nvidia tegra board file. Only the pinmux should stay in the individual board code, same thing as was done to all the other functions like NAND and USB.
True. I was originally just adapting the current config-file driven MMC to DT step-by-step, but you're right - it should be called just once for all boards in the common board file. I'll change it in V3.
I've looked into this some more, and it appears that I can't just add a call to tegra_mmc_init() from board_init() in boards/nvidia/common/board.c.
board_init() is where the other periphs do their pin_mux and xxx_init() calls (USB, SPI, etc.). board_init() is called early in board_init_r(), before mmc_initialize() is called. mmc_initialize() is needed before tegra_mmc_init() can use the mmc_device struct, etc. So tegra_mmc_init() needs to be called after mmc_initialize(), and right now that's in each board's board_mmc_init(). In board_mmc_init(), each board sets up any power rails needed for SD-card or eMMC access, sets up it's pin muxes for MMC, and then calls tegra_mmc_init() to parse the DT file and populate the mmc structs.
I could move the pin_mux_mmc() function calls from each board file into nvidia/common/board.c's board_init(), but it wouldn't really change much. So I'll leave it as it is for now, with pin_mux_mmc() and tegra_mmc_init() being called from each board's 'board' file (seaboard.c, colibri_t20_iris.c, etc).
Let me know if you see another way to move Tegra MMC init to a common board file that doesn't break the MMC driver flow.
I didn't look up the flow myself, as I don't have time for that right now, but I think I've got a pretty good picture from your description.
I think we should really try to make the Tegra MMC init flow as similar as possible to the other peripherals, so I suggest doing the following: 1. Provide a pin_mux_mmc() (possibly with a weak define as done with other peripherals). Boards should do pinmux and rail enabling within this function. 2. Move board_mmc_init() into nvidia/common/board.c, this function should call into the board specific pinmux function and then call tegra_mmc_init().
So even while we don't get the complete same flow as for other peripherals as the board_mmc_init() still has to be a freestanding function, instead of those things being folded into board_init(), we at least gain a clear distinction between the board specific parts and Tegra common code.
Regards, Lucas