userspace testing an ati fb driver

userspace testing an ati fb driver

[random]

On my Powerbook, /proc/iomem lists 0xb0000000-0xb000ffff as the location
of the programming registers for my ATI Radeon R250 Lf.

This seems to be the same spot XFree86 finds in its log:
(II) RADEON(0): MMIO registers at 0xb0000000

I've been trying to mimic the XFree86 fb accel code to make a Linux 2.6 fb
port, but i've had little luck. To try to pare down the new code, I've
written a userspace program which will open "/dev/mem" and then mmap in
this file starting at '0xb0000000' going for 0x10000 bytes.  Poking at the
"registers" there seems to have no effect; reading from these "registers"
(for example, to verify the engine idling) indicates that I'm perhaps
reading from the wrong place in memory, i.e. the engine never indicates
that it's idle.  Any ideas?

-eger david

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: userspace testing an ati fb driver

[Benjamin Herrenschmidt]

[-- Attachment #1: Type: text/plain, Size: 1041 bytes --]

On Wed, 2003-12-03 at 20:44, random wrote:
> On my Powerbook, /proc/iomem lists 0xb0000000-0xb000ffff as the location
> of the programming registers for my ATI Radeon R250 Lf.
>
> This seems to be the same spot XFree86 finds in its log:
> (II) RADEON(0): MMIO registers at 0xb0000000
>
> I've been trying to mimic the XFree86 fb accel code to make a Linux 2.6 fb
> port, but i've had little luck. To try to pare down the new code, I've
> written a userspace program which will open "/dev/mem" and then mmap in
> this file starting at '0xb0000000' going for 0x10000 bytes.  Poking at the
> "registers" there seems to have no effect; reading from these "registers"
> (for example, to verify the engine idling) indicates that I'm perhaps
> reading from the wrong place in memory, i.e. the engine never indicates
> that it's idle.  Any ideas?

That should work, make sure you are using proper memory barriers for
your IO accesses though (eieio instructions)

Enclosed is an example of a little program I use to dump the radeon
registers

Ben.



[-- Attachment #2: radeondump.c --]
[-- Type: text/x-c, Size: 2339 bytes --]

#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <sys/mman.h>
#include <unistd.h>
#include <byteswap.h>

void * mmio;

//#define RADEON_RBASE	0xe3000000
//#define RADEON_RBASE	0xb0000000
#define RADEON_RBASE	0x90000000
//#define RADEON_RBASE	0x90080000

static  unsigned char regr8(volatile void *base, const unsigned long offset)
{
        unsigned char val; 
        __asm__ __volatile__(
                        "lbzx %0,%1,%2\n\t"
                        "eieio"
                        : "=r" (val)
                        : "b" (base), "r" (offset),
                        "m" (*((volatile unsigned char *)base+offset)));
        return(val);
}


static  void regw8(volatile void *base, const unsigned long offset,
			const unsigned char val)
{
	__asm__ __volatile__ ("stbx %1,%2,%3\n\t"
				"eieio"
                        : "=m" (*((volatile unsigned char *)base+offset))
                        : "r" (val), "b" (base), "r" (offset));
}


static  unsigned int regr(volatile void *base, const unsigned long offset)
{
	unsigned int val;

	__asm__ __volatile__ ("lwbrx %0,%1,%2\n\t"
				"eieio"
				: "=r" (val)
				: "b" (base), "r"(offset),
				"m" (*((volatile unsigned char *)base+offset)));
	return val;
}

static  void regw(volatile void *base, const unsigned long offset,
				const unsigned int val)
{
	__asm__ __volatile__ ("stwbrx %1,%2,%3\n\t"
				"eieio"
				: "=m" (*((volatile unsigned char *)base+offset))
				: "r"(val), "b"(base), "r"(offset));
}

static  unsigned int pllr(volatile void *base, const unsigned long index)
{
	unsigned int val;

	regw8(base, 0x8, index & 0x3f);
	return regr(base, 0xc);
}


unsigned long regs[0x800];
unsigned long pllregs[0x40];

int main() {
	int fd, i;
	
	fd = open("/dev/mem", O_RDWR);

	mmio = (void *)mmap(0, 0x80000, PROT_READ|PROT_WRITE,
			MAP_SHARED, fd, RADEON_RBASE);

	for (i=0; i<0x800; i++)
		regs[i] = regr(mmio, i*4);
	for (i=0; i<0x40; i++)
		pllregs[i] = pllr(mmio, i);

	printf("Radeon MMIO registers:\n");
	for (i=0; i<0x800; i++) {
		if (regs[i] != 0)
			printf("%04x : %08x\n", i*4, regs[i]);
	}
	printf("\nRadeon PLL registers:\n");
	for (i=0; i<0x40; i++) {
		if (pllregs[i] != 0)
			printf("  %02x : %08x\n", i, pllregs[i]);
	}
	printf("\n");
}

Re: userspace testing an ati fb driver

[David Eger]

I think this explains my problem;  output from my reg dump:

Radeon MMIO registers:
0000 : 0000c081
0004 : 00313131
0008 : 3f000000
...

output from your reg dump:

Radeon MMIO registers:
0000 : 81c00000
0004 : 31313100
0008 : 0000003f
...

thanks a bunch.

-david

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: userspace testing an ati fb driver

[Benjamin Herrenschmidt]

On Fri, 2003-12-05 at 23:07, David Eger wrote:
> I think this explains my problem;  output from my reg dump:

Are you properly endian-flipping register accesses ? My access
macro use the lwbrx/stwbrx instructions that to the endian flipping
automatically.

> Radeon MMIO registers:
> 0000 : 0000c081
> 0004 : 00313131
> 0008 : 3f000000
> ...
>
> output from your reg dump:
>
> Radeon MMIO registers:
> 0000 : 81c00000
> 0004 : 31313100
> 0008 : 0000003f
> ...

--
Benjamin Herrenschmidt <benh@kernel.crashing.org>

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/