alevt fixed for amd64 - does it still work on i386?

[Edgar Toernig]

Oliver Lehmann wrote:
>
> It looks like I fixed alevt for amd64. But before I commit them, I want
> to be shure that they won't break alevt support on i386. Can someone with
> an i386 put the 3 attached files in his/her /usr/ports/misc/alevt/files
> directory, do a make install and tell me if alevt is still working?

> -    seq = *(u32 *)&rawbuf[n - 4];
> +    seq = *(u64 *)&rawbuf[n - 8];

Without testing I can tell you that it won't work on 32bit systems
or on 64bit Linux.  IMHO the right place to fix that would be the
kernel driver - the frame counter is supposed to be a u32 not a u64
or a long.

[UPDATE] I just browsed the FreeBSD sources.  The relevant sections:

src/sys/dev/bktr/bktr_reg.h:
|
|    u_long      vbi_sequence_number;    /* sequence number for VBI */

src/sys/dev/bktr/bktr_core.c:
|
|    unsigned int *seq_dest;
|...
|    /* Write the VBI sequence number to the end of the vbi data */
|    /* This is used by the AleVT teletext program */
|    seq_dest = (unsigned int *)((unsigned char *)bktr->vbibuffer
|            + bktr->vbiinsert
|            + (VBI_DATA_SIZE - sizeof(bktr->vbi_sequence_number)));
|    *seq_dest = bktr->vbi_sequence_number;

As one can see, the code is broken for 64bit.  It stores an int
but uses an offset for a long.

It seems, someone (cognet?) already changed a couple of places to fix
64bit issues (changed a lot of u_long to uint32_t) and just forgot
this one.  Afaics, the patch is easy:

src/sys/dev/bktr/bktr_reg.h:
-    u_long      vbi_sequence_number;    /* sequence number for VBI */
+    int         vbi_sequence_number;    /* sequence number for VBI */

Ciao, ET.


PS: The right change on AleVT to work around the driver bug would be:

vbi.c:
-    seq = *(u32 *)&rawbuf[n - 4];
+    seq = *(u32 *)&rawbuf[n - 8];

but of course that breaks all other platforms.