| Message ID | 20200714093229.28763-5-jagan@amarulasolutions.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | roc-rk3399-pc: Custom SPL | expand |
Hi Jagan, On 2020/7/14 下午5:32, Jagan Teki wrote: > Custom board_early_init_f not only deal with simple gpio > configuration but also have a possibility to access clocks > to process any clock related operations like checking reset > cause state and etc. > > So, call it once the rockchip timer initialization done instead > of calling first place of board_init_f which doesn't have any > rockchip init code before. > > This specific concern was tested with checking reset reason > via board_early_init_f, which indeed require a clk probe. board_early_init_f() is suppose to run at very beginning before the spl framework is init, so I don't agree this move. I think the led setup code definitely should go to spl_board_init(). As I said before, the setup_led should be the common feature, so I think you can have a copy of setup_led() in rk3399.c as a weak common function, and if roc-rk3399-pc is really special, then you can have its own implementation. Note that the spl_board_init in SPL is already very early stage in system init, should be enough for your case, you can do a test for the led_setup() in board_early_init_f() after this patch and in spl_board_init(), they should be very close. Thanks, - Kever > > Signed-off-by: Jagan Teki <jagan@amarulasolutions.com> > --- > Changes for v5: > - new patch > > arch/arm/mach-rockchip/spl.c | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/arch/arm/mach-rockchip/spl.c b/arch/arm/mach-rockchip/spl.c > index cddf4fd3d5..082828de66 100644 > --- a/arch/arm/mach-rockchip/spl.c > +++ b/arch/arm/mach-rockchip/spl.c > @@ -122,8 +122,6 @@ void board_init_f(ulong dummy) > debug("\nspl:debug uart enabled in %s\n", __func__); > #endif > > - board_early_init_f(); > - > ret = spl_early_init(); > if (ret) { > printf("spl_early_init() failed: %d\n", ret); > @@ -137,6 +135,9 @@ void board_init_f(ulong dummy) > /* Init ARM arch timer in arch/arm/cpu/armv7/arch_timer.c */ > timer_init(); > #endif > + > + board_early_init_f(); > + > #if !defined(CONFIG_TPL) || defined(CONFIG_SPL_RAM) > debug("\nspl:init dram\n"); > ret = dram_init();
diff --git a/arch/arm/mach-rockchip/spl.c b/arch/arm/mach-rockchip/spl.c index cddf4fd3d5..082828de66 100644 --- a/arch/arm/mach-rockchip/spl.c +++ b/arch/arm/mach-rockchip/spl.c @@ -122,8 +122,6 @@ void board_init_f(ulong dummy) debug("\nspl:debug uart enabled in %s\n", __func__); #endif - board_early_init_f(); - ret = spl_early_init(); if (ret) { printf("spl_early_init() failed: %d\n", ret); @@ -137,6 +135,9 @@ void board_init_f(ulong dummy) /* Init ARM arch timer in arch/arm/cpu/armv7/arch_timer.c */ timer_init(); #endif + + board_early_init_f(); + #if !defined(CONFIG_TPL) || defined(CONFIG_SPL_RAM) debug("\nspl:init dram\n"); ret = dram_init();
Custom board_early_init_f not only deal with simple gpio configuration but also have a possibility to access clocks to process any clock related operations like checking reset cause state and etc. So, call it once the rockchip timer initialization done instead of calling first place of board_init_f which doesn't have any rockchip init code before. This specific concern was tested with checking reset reason via board_early_init_f, which indeed require a clk probe. Signed-off-by: Jagan Teki <jagan@amarulasolutions.com> --- Changes for v5: - new patch arch/arm/mach-rockchip/spl.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)