Paul Gortmaker wrote:
I was updating a sbc8560 from u-boot v1.2.0 to git-current, and found that I'd loose the kernel serial console when the 8250 driver took over from udbg0 when using u-boot 1.3.x (booting via tftp'ing the dtb and the uImage separately)
I eventually tracked it down to mpc85xx/fdt.c stomping on the contents of the UART clockrate in the dtb. Since the sbc8560 is sort of different -- in that it has external UARTs (instead of SOC/CPM), I'm guessing few other boards see this problem, because bi_busfreq is really their correct UART clk.
I fixed it with this simple patch, but I was wondering whether it would be better to alternatively just check if the dtb value is zero, and
Not defined: I would have serious heartburn if the .dts/dtb value were defined to be zero. Define or define not, there should be no zero (with apologies to Yoda).
then only insert a value; otherwise leave it alone...
Paul.
Changing the fixup last parameter (create) from "1" to "0" will prevent overriding an existing value. Note that this will *not* override a silly value (e.g. 0), so, if you do this, the dtb must either not have the clock-frequency defined or have a correct value in that property. This would have the good effect of allowing the board specific clock to be defined in the dtb where it belongs, but still support the older style approach of compiling it in.
This seems like a good improvement (your patch, but changing the "1" to "0") to me, but I'm not a 85xx expert.
diff --git a/cpu/mpc85xx/fdt.c b/cpu/mpc85xx/fdt.c index 92952e6..bd43073 100644 --- a/cpu/mpc85xx/fdt.c +++ b/cpu/mpc85xx/fdt.c @@ -224,7 +224,7 @@ void ft_cpu_setup(void *blob, bd_t *bd)
#ifdef CFG_NS16550 do_fixup_by_compat_u32(blob, "ns16550",
"clock-frequency", bd->bi_busfreq, 1);
"clock-frequency", CFG_NS16550_CLK, 1);#endif
#ifdef CONFIG_CPM2
Best regards, gvb