Problems with AGP Apple UniNorth support in 2.4.19-pre7-ben0 kernel

[Bill Fink]

Hi,

I see that the changes I suggested have been incorporated into the
2.4.19-pre8-ben0 kernel.  I was able to successfully build and boot
this kernel on my dual 500 MHz G4 (UP kernel), but the DRI was no
longer functional (as reported by glxinfo).

I had this working with the 2.4.18-ben0 kernel, with modified versions
of drivers/char/drm/r128_{drv,state}.c that I got from a friend who I
believe got them from the XFree86 CVS at the time.  That r128_state.c
is now included in the 2.4.19-pre8-ben0 kernel but the new r128_drv.c
was not included.  The diff is pretty innocuous:

42c42
< #define DRIVER_DATE           "20010405"
---
> #define DRIVER_DATE           "20010917"
45,46c45,46
< #define DRIVER_MINOR          1
< #define DRIVER_PATCHLEVEL     6
---
> #define DRIVER_MINOR          2
> #define DRIVER_PATCHLEVEL     0

I figured what the heck, I didn't have anything to lose, so I went ahead
and made the change, and now the DRI is working once more!

						-Bill



On Wed, 24 Apr 2002, Bill Fink wrote:

> Hi,
>
> There seems to be problems with AGP Apple UniNorth support in the latest
> 2.4.19-pre7-ben0 kernel.  First you can't select it in make menuconfig.
> I fixed that with this simple patch:
>
> --- drivers/char/.orig/Config.in        Sat Apr 20 19:06:15 2002
> +++ drivers/char/.mod/Config.in Wed Apr 24 02:15:28 2002
> @@ -263,7 +263,7 @@
>     bool '  ALI chipset support' CONFIG_AGP_ALI
>     bool '  Serverworks LE/HE support' CONFIG_AGP_SWORKS
>     dep_bool '  HP ZX1 AGP support' CONFIG_AGP_HP_ZX1 $CONFIG_IA64
> -   dep_bool '  Apple UniNorth support' CONFIG_AGP_UNINORTH CONFIG_ALL_PPC
> +   dep_bool '  Apple UniNorth support' CONFIG_AGP_UNINORTH $CONFIG_ALL_PPC
>  fi
>
>  bool 'Direct Rendering Manager (XFree86 DRI support)' CONFIG_DRM
>
> But then I just get the following fatal error compiling agpgart_be.c:
>
> gcc -D__KERNEL__ -I/usr/src/linux-2.4-benh-19pre7/include -Wall -Wstrict-prototy
> pes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common -fomit-frame-pointer -D_
> _powerpc__ -fsigned-char -msoft-float -pipe -ffixed-r2 -Wno-uninitialized -mmult
> iple -mstring   -nostdinc -I /usr/lib/gcc-lib/ppc-yellowdog-linux/2.95.4/include
>  -DKBUILD_BASENAME=agpgart_be  -DEXPORT_SYMTAB -c agpgart_be.c
> agpgart_be.c:88: #error "Please define flush_cache."
> agpgart_be.c: In function `agp_generic_agp_enable':
> agpgart_be.c:404: warning: unused variable `cap_id'
> agpgart_be.c: In function `uninorth_insert_memory':
> agpgart_be.c:3924: warning: passing arg 1 of `flush_dcache_range' makes integer
> from pointer without a cast
> agpgart_be.c:3924: warning: passing arg 2 of `flush_dcache_range' makes integer
> from pointer without a cast
> agpgart_be.c: In function `agp_find_supported_device':
> agpgart_be.c:4622: warning: unused variable `scratch'
> agpgart_be.c:4622: warning: unused variable `cap_id'
> agpgart_be.c: At top level:
> agpgart_be.c:402: warning: `agp_generic_agp_enable' defined but not used
> agpgart_be.c:495: warning: `agp_generic_create_gatt_table' defined but not used
> agpgart_be.c:623: warning: `agp_generic_free_gatt_table' defined but not used
> agpgart_be.c:671: warning: `agp_generic_insert_memory' defined but not used
>
> The relevant code in agpgart_be.c is:
>
> static inline void flush_cache(void)
> {
> #if defined(__i386__) || defined(__x86_64__)
>         asm volatile ("wbinvd":::"memory");
> #elif defined(__alpha__) || defined(__ia64__) || defined(__sparc__)
>         /* ??? I wonder if we'll really need to flush caches, or if the
>            core logic can manage to keep the system coherent.  The ARM
>            speaks only of using `cflush' to get things in memory in
>            preparation for power failure.
>
>            If we do need to call `cflush', we'll need a target page,
>            as we can only flush one page at a time.
>
>            Ditto for IA-64. --davidm 00/08/07 */
>         mb();
> #else
> #error "Please define flush_cache."
> #endif
> }
>
> When this worked in the 2.4.18-ben0 kernel, the line with defined
> __alpha__ or __ia64__ or __sparc__ had an extra "|| defined (__powerpc__)",
> which is now missing, hence the fatal compilation error.
>
> Any comments?  Is it safe just to add back in the "|| defined (__powerpc__)"?
>
> 						-Thanks
>
> 						-Bill

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