[PATCH] omap: Fix DEBUG_LL uart to access phys addr when MMU isn't enable

[Eric Miao]

On Tue, Aug 3, 2010 at 2:57 PM, Tony Lindgren <tony at atomide.com> wrote:
> * Eric Miao <eric.y.miao at gmail.com> [100803 07:29]:
>> On Tue, Aug 3, 2010 at 11:44 AM, Jason Wang <jason77.wang at gmail.com> wrote:
>> > Now we use a memory address to store the debug port info, So we need
>> > to read/write this address when we choose DEBUG_LL. When MMU isn't
>> > enable(I.E. the begining part of init stage of the linux kernel boot),
>> > we need to access physical address instead of virtual address,
>> > otherwise the kernel will crash.
>> >
>> > Signed-off-by: Jason Wang <jason77.wang at gmail.com>
>> > ---
>> >  arch/arm/mach-omap1/include/mach/debug-macro.S |   10 +++++--
>> >  arch/arm/mach-omap2/include/mach/debug-macro.S |   29 +++++++++++++++++------
>> >  2 files changed, 28 insertions(+), 11 deletions(-)
>> >
>> > diff --git a/arch/arm/mach-omap1/include/mach/debug-macro.S b/arch/arm/mach-omap1/include/mach/debug-macro.S
>> > index e8a8cf3..671408e 100644
>> > --- a/arch/arm/mach-omap1/include/mach/debug-macro.S
>> > +++ b/arch/arm/mach-omap1/include/mach/debug-macro.S
>> > @@ -33,7 +33,7 @@ omap_uart_virt:       .word   0x0
>> >                /* Use omap_uart_phys/virt if already configured */
>> >  9:             mrc     p15, 0, \rx, c1, c0
>> >                tst     \rx, #1                 @ MMU enabled?
>> > -               ldreq   \rx, =omap_uart_phys    @ physical base address
>> > +               ldreq   \rx, =__virt_to_phys(omap_uart_phys)    @ physical base address
>>
>> Along with introduction of run-time phys_offset, I'm afraid this
>> macro won't work in assembly any longer.
>
> /me scratches head wondering how this has been working so far on the
> systems tested.. Looks like we've been lucky and the omap_uart_virt
> and phys have been initialized twice to different memory addresses,
> first with MMU off then on.
>

Yeah, you lucky guys. Or possibly because it's not even used
when MMU is off in the early stage?

> For the run-time phys_offset, can we use the same logic here and
> set it based on the PC?
>

That's exactly what the run-time phys_offset is calculated.

>> And having a variable version of debugging uart base, in my POV,
>> may not be generic enough to handle the fundamental issue. Tho
>> I'm not sure if it will feasible to have a generic debug function in
>> 'struct machine_desc' and make that structure available as early
>> as possible.
>
> That might cover the most typical debug cases, but is not available
> right from the start.
>
> What if we initialize the low level debug functions in some static
> memory location already in uncompress.h? That would help sorting
> out the debug-macro.S issue from compiling in multile ARM archs
> point of view.
>

We also need to take care of Jeremy's recent patchset to remove
.io_pg_offst and .phys_io, which modifies all the addruart macros.

Considering:

1) the debug UART might not be 8250 compatible (mostly are, but
some like ICEDCC)

2) and the settings are different between boards (different register
space, etc)

So it is actually board specific code, which could be encoded into
'struct machine_desc', yet not sure how to be generic enough.