Re: [PATCH4/4] [POWERPC] Fix cpm_uart driver

[Jochen Friedrich <jochen@scram.de>]

Hi Scott,
> Yikes.  Please don't change cpm_uart_cpm1.h, as it's correct for 
> arch/powerpc, and there are numerous other places that assume cpmp is 
> virtual (including in the very same function that assigns it a 
> physical address).

I'm still not convinced cpm_uart_cpm1.h is correct:

pinfo->rx_bd_base and pinfo->tx_bd_base are both initialized with an 
address obtained by cpm_dpram_addr(). In both ppc and powerpc, this is 
an address relative to dpram_vbase in commproc.c

In cpm_uart_core.c, the operation "pinfo->rx_bd_base - DPRAM_BASE" is 
used to calculate the DPRAM offset. So DPRAM_BASE must be relative to 
dpram_vbase in commproc.c as well. However, cpm_uart_cpm1.h uses cpmp in 
commproc.c to initialize DPRAM_BASE.

On ARCH=ppc, cpmp is a physical address with 1:1 virtual mapping ("well 
known address"). On ARC=powerpc, this is an address obtained by 
ioremap(), however it's a different ioremap() call than dpram_vbase is 
obtained from, so noone can guarantee
cpmp is always the same as dpram_vbase even on ARCH=powerpc.

To me, it looks like setting DPRAM_BASE to cpm_dpram_addr(0) is the fix 
as it makes the DPRAM offset a defined result.

Thanks,
Jochen