Re: [Qemu-devel] [PATCH] fix PPC OpenHackWare for Linux 2.6

[Laurent Vivier <Laurent.Vivier@bull.net>]

Nice work,

Le mardi 06 mai 2008 à 18:00 +0200, René Rebe a écrit :
> Hi all,
> 
> after quite some debugging thru the early startup code of Linux 2.6
> I fixed some issues in the OpenHackWare "BIOS" for PPC emulation.
> 
> The diff below is against the Qemu pc-bios/ohw.diff, I can also
> provide the resulting binary if interested.
> 
> First I fixed the linker script to actually work at all (actually
> I also installed an old toolchain, just to find out that even that
> one would not link OHW, ...
> 
> Second, Linux 2.6 flatten_device_tree stuff in arch/powerpc/kernel/prom_init.c
> check for a return value of 1, as I do not have the IEEE spec on my desk
> I assumed it means success, and with the change the flattened device
> tree is no longer empty, helping a great deal continuing to boot ...
> 
> Third, when selecting a core99 machine, Linux assumes to find an NVRAM,
> and panics in some init_timer if there isn't one. Though a core99 machine
> does not yet work perfectly, further investigation to be done.
> 
> Linewise the FB CLUT is with Linux 2.6 has some issue and the FB thus
> shows false colors, but good enough to work with for now.
> 
> Oh, and last but not least with a self built prom, I hit this
> "error booting from CD-ROM bug, that Jocelyn hacked around in
> r3309. As that was just a binary only blob fix and there is no
> newer version of OHW I just commented out the "return on error"
> which allowed to boot CD images on my side.
> 
> As this is the first nightly track thru the guts of Qemu any
> comment welcome:
> 
> --- qemu-svn/pc-bios/ohw.diff	(revision 4353)
> +++ qemu-svn/pc-bios/ohw.diff	(working copy)
> @@ -1,3 +1,19 @@
> +Make it link at al.
> +
> +  - Rene Rebe <rene@exactcode.de>
> +
> +--- OpenHackWare-release-0.4/src/main.ld	2005-03-31 09:23:33.000000000 +0200
> ++++ OpenHackWare-release-0.4-hacked/src/main.ld	2008-05-06 11:23:29.000000000 +0200
> +@@ -49,7 +49,7 @@
> +         _sdata_end = . ;
> +         . = ALIGN(4) ;
> +         _ro_start = . ;
> +-        .rodata    : { *(.rodata) } > bios
> ++        .rodata    : { *(.rodata*) } > bios
> +         _ro_end = . ;
> +         . = ALIGN(4) ;
> +         _RTAS_start = .;
> +


I think you should also remove .rodata.str1.4 from .data section:

@@ -35,7 +35,7 @@
         .OpenFirmware : { *(.OpenFirmware) } > bios
         . = ALIGN(4) ;
         _data_start = . ;
-        .data     : { *(.data) *(.rodata.str1.4)    } > bios
+        .data     : { *(.data)    } > bios
         _data_end = . ;
         . = ALIGN(4) ;
         _OF_vars_start = . ;

>   diff -wruN --exclude '*~' --exclude '*.o' --exclude '*.bin' --exclude '*.out' --exclude mkdiff OpenHackWare-release-0.4.org/src/bios.h OpenHackWare-release-0.4/src/bios.h
>   --- OpenHackWare-release-0.4.org/src/bios.h	2005-04-06 23:20:22.000000000 +0200
>   +++ OpenHackWare-release-0.4/src/bios.h	2005-07-07 01:10:20.000000000 +0200
> @@ -748,24 +764,14 @@
>            {
>                /* Hack taken 'as-is' from PearPC */
>                unsigned char info[] = {
> -@@ -1596,7 +1627,9 @@
> -         OF_node_put(OF_env, brom);
> +@@ -1596,6 +1627,7 @@
>            OF_node_put(OF_env, rom);
>        }
> -+#if 0
>        /* From here, hardcoded hacks to get a Mac-like machine */
> -+    /* XXX: Core99 does not seem to like this NVRAM tree */
> ++    /* XXX: Not yet perfect, but Linux 2.6 does oops on boot on Core99 without NVRAM node */
>        /* "/nvram@fff04000" node */
>        {
>            OF_regprop_t regs;
> -@@ -1617,6 +1650,7 @@
> -         OF_prop_int_new(OF_env, chs, "nvram", OF_pack_handle(OF_env, nvr));
> -         OF_node_put(OF_env, nvr);
> -     }
> -+#endif
> -     /* "/pseudo-hid" : hid emulation as Apple does */
> -     {
> -         OF_node_t *hid;
>   @@ -1663,7 +1697,27 @@
>            }
>            OF_node_put(OF_env, hid);
> @@ -1841,3 +1847,40 @@
>        case ARCH_MAC99:
>            /* We are supposed to have 3 host bridges:
>             * - the uninorth AGP bridge at 0xF0000000
> +
> +
> +The 2.6 Linux kernel checks for 1, as I do not have the IEEE spec on
> +my desk I can only guess it should return 1 on success, not the
> +string length.
> +
> +  - Rene Rebe <rene@exactcode.de>
> +
> +--- OpenHackWare-release-0.4/src/of.c	2005-04-06 23:17:26.000000000 +0200
> ++++ OpenHackWare-release-0.4-hacked/src/of.c	2008-05-06 14:50:48.000000000 +0200
> +@@ -3841,7 +4061,7 @@
> +             OF_DPRINTF("Return property name [%s]\n", next->name);
> +             OF_sts(next_name, (void *)(next->name));
> +             OF_DUMP_STRING(OF_env, next_name);
> +-            pushd(OF_env, strlen(next->name) + 1);
> ++            pushd(OF_env, 1); /* ReneR: Linux 2.6 flatten_device_tree */
> +         }
> +     }
> + }
> +
> +
> +In qemu r3309 j_mayer did some "Quickly hack PowerPC BIOS able to boot
> +on CDROM again.", as I did not feel like disassemble to find out this
> +worked for me for now.
> +
> +  - Rene Rebe <rene@exactcode.de>
> +
> +--- OpenHackWare-release-0.4/src/bloc.c	2005-04-06 23:21:00.000000000 +0200
> ++++ OpenHackWare-release-0.4-hacked/src/bloc.c	2008-05-06 14:20:10.000000000 +0200
> +@@ -1021,7 +1038,7 @@
> +         status = ide_port_read(bd, 0x07);
> +         if (status != 0x08) {
> +             ERROR("ATAPI TEST_UNIT_READY : status %0x != 0x08\n", status);
> +-            return -1;
> ++            /*return -1;*/ /* fails to boot from cdrom? */
> +         }
> +         for (i = 0; i < 3; i++) {
> 

Regards,
Laurent
-- 
------------- Laurent.Vivier@bull.net ---------------
"The best way to predict the future is to invent it."
- Alan Kay