i.MX28: Add delay after CPU bypass is cleared

This solves issues when larger amount of DRAM is used, like 256MB.
Behave the same in case of CPU bypass as we do in case of EMI
bypass, but wait 15 ms. We need to wait until the clock domain
stabilizes.

This issue seemed to have been caused by not waiting after frobbing
with the CPU bypass, it was unrelated to memory, but had a direct
impact, causing trouble. This was yet another X-File of the
imx-bootlets, sigh. The conclusion is, trying a semi-random delay
(there is delay after the EMI bypass change), the issue is fixed.

Another possible explanation is that we do not do the "simple memory
test" FSL does in their imx-bootlets (1000 R/W cycles to/from piece of
the memory, while also outputing something on the serial port). This
might have caused the similar delay in the imx-bootlets and therefore
they didn't need to add this explicitly.

For now, this seems good fix enough, but to me, whole that memory
init code in imx-bootlets is completely flunked and it'd need deeper
investigation.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Wolfgang Denk <wd@denx.de>
Cc: Detlev Zundel <dzu@denx.de>
Cc: Stefano Babic <sbabic@denx.de>
Cc: Fabio Estevam <festevam@gmail.com>
Acked-by: Stefano Babic <sbabic@denx.de>
Acked-by: Detlev Zundel <dzu@denx.de>

arch/arm/cpu/arm926ejs/mx28/spl_mem_init.c

@@ -149,6 +149,8 @@ void mx28_mem_setup_cpu_and_hbus(void)
 	/* Disable CPU bypass */
 	writel(CLKCTRL_CLKSEQ_BYPASS_CPU,
 		&clkctrl_regs->hw_clkctrl_clkseq_clr);
+
+	early_delay(15000);
 }
 
 void mx28_mem_setup_vdda(void)