
Hi Jens,
I took a look into the files board/mpc8560ads.c and board/tqm85xx.c and found something strange:
- In the function local_bus_init() the current CLKDIV is read from the
register LCRR as was set by Hardreset. After that, the decision is made, wether the DLL has to be enabled/disabled/overridden. Inside the if-else blocks the new CLKDIV is changed. But IMO the CLKDIV has to be set before the query.
This is the current code: clkdiv = lbc->lcrr & 0x0f; lbc_hz = sysinfo.freqSystemBus / 1000000 / clkdiv;
if (lbc_hz < 66) { lbc->lcrr = CFG_LBC_LCRR | 0x80000000; /* DLL Bypass */
} else if (lbc_hz >= 133) { lbc->lcrr = CFG_LBC_LCRR & (~0x80000000); /* DLL Enabled ... This may be the situation on other 85xx boards, too. I didn't check them all. What was the intention, DLL modification dependent on the clock set by the MPC at hardreset or dependent on the targeted frequency?
I am definitely not the specialist on these chips, but checking the 8555 and 8560 manuals, it seems that LCRR comes out of reset with 0x8000,0008. I cannot find any reference to LCRR from the POR description.
Keeping this in mind, "clkdiv" in the above code will always be 8 and so it looks to me indeed like the tests really were intended to check (CFG_LBC_LCRR & 0x0f). Can anyone more knowledgable in this area confirm this suspicion?
- The variable is named lbc_hz, but it contains a value in units of
MHz. I suggest to use the name lbc_mhz or to use Hertz values by removing the division by 1,000,000 and replacing 66 and 133 by 66666667 and 133333333. What's your opinion?
Personally I would second your reasoning. But for others to comment, it is really best to implement your proposed change and send a diff to comment on. Such a suggestion is more likely by several magnitudes to be considered ;)
- I assume, the function above is called while the MPC executes this
code from flash memory (array init_sequence[] in lib_ppc/board.c). Is there a potential that the U-Boot would hang when it changed the CLKDIV?
If you get the timings wrong, of course things could get hairy, or am I missing something?
Cheers Detlev