[U-Boot] [PATCH v3][Net] Convert CS8900 Ethernet driver to CONFIG_NET_MULTI API

[Wolfgang Denk]

Dear Ben,

in message <4A9069E9.2030000 at gmail.com> you wrote:
>
> > Tested on mx31ads: works like a charm.
> >   
> Great news!  Thanks!

Wait... we've not reached the goal yet, I'm afraid...

> > Will try to debug the trab issues later tonight (if I manage to find a
> > matching JTAG connector for this board ).

This turns out to be a nasty one.

The good news is: the hang on trab is not caused by your patch.

Fact is that  it  is  instead  caused  by  using  a  somewhat  recent
(gcc-4.2.2  based)  tool  chain.  This compiler does some ugly things
when attempting to access 32 bit wide registers which  cause  one  of
the  trab-specific drivers to hang in the initialization. If I use an
older tool chain (say, gcc-4.0.0 based),  your  patch  works  like  a
charm, too.

The bad news ist: I don't know how to fix this.


In the first attempt I noticed that the (old) trab code does not use
any I/O accessors at all. My guesses that a "volatile" might be
missing didn't help.

So I changed the code in question to use readl() / writel() - and
this did not help either.

Then I noticed that readl() / writel() on ARM are mapped to
__arch_getl() / __arch_putl() which in turn are just plain stupid
volatile pointer accesses.

So I replaced these #defines by the following code:

static inline uint32_t __arch_getl (uint32_t V_ATTRIB * const addr)
{
        uint32_t val;
        val = *addr;
        return val;
}

static inline void __arch_putl (uint32_t val, uint32_t V_ATTRIB * const addr)
{
        *addr = val;
}


With this code, I see the following behaviour:

1) If I use "#define V_ATTRIB volatile" which I think is needed and
   technically correct, the trab-specific driver code starts working
   again; however, the CS8900 Ethernet driver does not work. It
   gives:

	TRAB # run load
	Using CS8900-0 device
	TFTP from server 192.168.1.1; our IP address is 192.168.20.2
	Filename '/tftpboot/trab/u-boot.bin'.
	Load address: 0xc100000
	Loading: T T T T T T T T T T 
	Retry count exceeded; starting again
	Using CS8900-0 device
	TFTP from server 192.168.1.1; our IP address is 192.168.20.2
	Filename '/tftpboot/trab/u-boot.bin'.
	Load address: 0xc100000
	Loading: T T T T T T T T T T 
	Retry count exceeded; starting again
	Using CS8900-0 device
	TFTP from server 192.168.1.1; our IP address is 192.168.20.2
	Filename '/tftpboot/trab/u-boot.bin'.
	Load address: 0xc100000
	Loading: T T T T T T T T T T 
	Retry count exceeded; starting again
	...

   All this in quick sequence, without the usual longish delays /
   timeouts.

2) If I use "#define V_ATTRIB" instead (i. e. if I do _not_ use
   "volatile"), then the trab driver code hangs again, but the CS8900
   Ethernet driver works fine.


It seems the "volatile" causes the  compiler  to  generate  different
code  - with the volatile it seems that 4 x 8 bit reads are performed
instead of 1 x 32 bit read. Unfortunaltely I wasn't  able  to  reduce
this to a trival test case yet.


Is there anybody out there with good C compiler / assembler knowhow
on ARM - and with an idea how to fix this issue?

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
The thing is, as you progress in the Craft,  you'll  learn  there  is
another rule... When you break rules, break 'em good and hard.
                                    - Terry Pratchett, _Wyrd Sisters_