Problem mapping GPIO regs on ppc440gx

Problem mapping GPIO regs on ppc440gx

[Travis B. Sawyer]

Greetings:

We've been using a 2.4.30 kernel (with numerous patches) on a AMCC 440gx
custom built board for quite some time now.

We're building some new hardware that forces us to go to a 2.6 kernel.
So, I've downloaded 2.6.16.16 from kernel.org and am porting everything
forward from our 2.4.30 kernel.

The problem lies in mapping the GPIO regs of the 440 to user space using
/dev/mem:

    gpio_fd = open("/dev/mem", O_RDWR | O_SYNC);

    if (0 > gpio_fd) {
        perror("mbGpioGet(): Unable to open gpio");
        return(-1);
    }

    addr = (ppc440_gpio_regs_t *)mmap(0, getpagesize() * 2,
                                      PROT_READ | PROT_WRITE,
                                      MAP_SHARED, gpio_fd,
                                      (off_t)(0x40000000));

    if (MAP_FAILED == addr) {
        perror("mbGpioGet(): map error");
        close(gpio_fd);
        return(-1);
    }

    pGpioRegs = (ppc440_gpio_regs_t *)((uint32_t)addr + 0x700);

    data = pGpioRegs->in & (MB_GPIO_PRI_N_K | MB_GPIO_SEC_PRES_K);

    munmap(0, sizeof(ppc440_gpio_regs_t));
    close(gpio_fd);

The data I get back is 0.  Always.  /SEC_PRES_K should be 1 in this case.

(gdb) print /x addr
$7 = 0x30015000
(gdb) print /x pGpioRegs
$8 = 0x30015700


Checking /proc/<PID>/map shows the mapping:
30015000-30019000 rw-p 30015000 00:00 0


On a board running the 2.4.30 kernel, I show the SAME entry in 
/proc/<PID>/map
but I do get data back.

Any idea what I'm doing wrong here?

TIA,

Travis Sawyer

Re: Problem mapping GPIO regs on ppc440gx

[Travis B. Sawyer]

Answering my own question here...

Travis B. Sawyer wrote:

>Greetings:
>
>We've been using a 2.4.30 kernel (with numerous patches) on a AMCC 440gx
>custom built board for quite some time now.
>
>We're building some new hardware that forces us to go to a 2.6 kernel.
>So, I've downloaded 2.6.16.16 from kernel.org and am porting everything
>forward from our 2.4.30 kernel.
>
>The problem lies in mapping the GPIO regs of the 440 to user space using
>/dev/mem:
>
>    gpio_fd = open("/dev/mem", O_RDWR | O_SYNC);
>
>    if (0 > gpio_fd) {
>        perror("mbGpioGet(): Unable to open gpio");
>        return(-1);
>    }
>
>    addr = (ppc440_gpio_regs_t *)mmap(0, getpagesize() * 2,
>                                      PROT_READ | PROT_WRITE,
>                                      MAP_SHARED, gpio_fd,
>                                      (off_t)(0x40000000));
>
>  
>
The above mmap call worked for 2.4.30, but for 2.6.16 I needed to change 
the offset
to be a multiple of page size, so:
(off_t)(0x40000000/getpagesize());

-travis