Trying to fix 3dfx.o + question about char drivers..

Trying to fix 3dfx.o + question about char drivers..

[Michael Bacarella]

So, I upgrade to 2.4.0 and it's cool, except that I can't do
anything neat with my voodoo3 anymore. I've been looking
for a solution for weeks but to no avail. 3dfx's web site
looks like it's gone and nothing on lk about it.

[ By all means, if someone has fixed this, do let me know ]

Tracing the Glide test programs shows that ioctl() is returning
-EPERM. Compiling the driver with 'make debug' shows:

[  To aid in the confusion, my machine's hostname is 'mmap' ]

[ insmod 3dfx.o ]
Feb 14 15:08:14 mmap kernel: 3dfx: Entering init_module()
Feb 14 15:08:14 mmap kernel: 3dfx: Successfully registered device 3dfx
Feb 14 15:08:14 mmap kernel: 3dfx: board vendor 4634 type 5 located at de000000/

[ ./test3Dfx ]
Feb 14 15:08:29 mmap kernel: 3dfx: Entering mmap_3dfx
Feb 14 15:08:29 mmap kernel: 3dfx: Couldn't match address 0 to a card
Feb 14 15:08:29 mmap kernel: 3dfx: Entering release_3dfx

mmap_3dfx is being called before ioctl_3dfx is ever reached. Looking
to make heads of the "Couldn't match address 0 to a card" message, I
stuck in some more debugging output:

[ VM_OFFSET is #define VM_OFFSET(vma) (vma->vm_pgoff << PAGE_SHIFT) ]

Feb 14 15:08:29 mmap kernel: 3dfx: Entering mmap_3dfx
Feb 14 15:08:29 mmap kernel: 3dfx: Entering decode_vma:
Feb 14 15:08:29 mmap kernel: 3dfx: start = c1270640, end = c5e40840
Feb 14 15:08:29 mmap kernel: 3dfx: offset = 0 (VM_OFFSET = 0)
Feb 14 15:08:29 mmap kernel: 3dfx: Leaving decode_vma
Feb 14 15:08:29 mmap kernel: 3dfx: compare de000000 or e2000000 to 0
Feb 14 15:08:29 mmap kernel: 3dfx: Couldn't match address 0 to a card
Feb 14 15:08:29 mmap kernel: 3dfx: Entering release_3dfx

Sure stumped me. By my guess, it seems that mmap_3dfx is provided by
the char driver so that a userland process can map a region of (video?)
memory on the card.

The process calls ioctl() after opening /dev/3dfx. That ioctl() triggers
an mmap() call, the driver gets addresses it's totally not expecting,
and it returns -EPERM.

Why does mmap get called first?? Am I reading this right?

-- 
Michael Bacarella <mbac@nyct.net>
Technical Staff / System Development,
New York Connect.Net, Ltd.

Re: Trying to fix 3dfx.o + question about char drivers..

[Alexander V. Lukyanov]

Michael Bacarella wrote:
> So, I upgrade to 2.4.0 and it's cool, except that I can't do
> anything neat with my voodoo3 anymore. I've been looking
> for a solution for weeks but to no avail. 3dfx's web site
> looks like it's gone and nothing on lk about it.
> 
> The process calls ioctl() after opening /dev/3dfx. That ioctl() triggers
> an mmap() call, the driver gets addresses it's totally not expecting,
> and it returns -EPERM.

struct file_operations has changed in 2.4.0. The following patch for 3dfx
module fixes the structure initialization and makes it work just fine.

--- 3dfx_driver.c	Sun Apr  9 03:44:10 2000
+++ 3dfx_driver.c	Fri Feb 23 22:42:06 2001
@@ -707,25 +707,10 @@
 #endif /* HAVE_MTRR */
 
 static struct file_operations fops_3dfx = {
-	NULL,			/* llseek (2.1) / lseek (2.0) */
-	NULL,			/* read */
-	NULL,			/* write */
-	NULL,			/* readdir */
-	NULL,			/* poll (2.1) / select (2.0) */
-	ioctl_3dfx,		/* ioctl */
-	mmap_3dfx,		/* mmap */
-	open_3dfx,		/* open */
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 1, 118)
-	NULL,			/* flush (2.1.118+) */
-#endif
-	release_3dfx,		/* release */
-	NULL,			/* fsync */
-	NULL,			/* fasync */
-	NULL,			/* check_media_change */
-	NULL,			/* revalidate */
-#ifdef KERNEL_VER_2_1
-	NULL			/* lock (2.1) */
-#endif
+	ioctl:		ioctl_3dfx,
+	mmap:		mmap_3dfx,
+	open:		open_3dfx,
+	release:	release_3dfx
 };
 
 #ifdef MODULE