> Date: Sat, 10 Mar 2012 22:59:24 +0100
> From: Tobias Ulmer <tobi...@tmux.org>
>
> (Ivan: would be nice if you could test it and tell us about any
> success/failure)
>
> socppc saves registers r3 (fwfdtsave) and r29 (fwargsave) in locore.S.
> initppc() checks for the absence of a flat device tree structure, in which
> case it uses the older bd_t aka bootinfo.
>
> It then checks whether the FDT checksum is valid and eventually uses that.
>
> If fwfdtsave (r3) contains rubbish and u-boot does not supply a valid
> FDT, it will go into neither branch, doesn't set the amount of memory
> and silently panics in pmap.c.
>
> This diff changes the logic to first try the FDT and
> unconditionally fall back to bootinfo. If that doesn't work, we're toast
> anyway.
>
> I suspect r3 was zero by pure chance and the code works on a different
> model. The one I have here has a 8 pin serial connector, unlike what's
> described in the install document.
>
> U-boot's README.standalone only mentions r29 being set and I can't find
> any reference where it sets r3 in do_go().
>
> Can a kind soul please tell me how to cross-compile a RAMDISK kernel? I
> keep running into errors no matter which way I try. It would be nice to
> install it, now that it boots :-)
>
> Index: socppc/machdep.c
> ===================================================================
> RCS file: /home/vcs/cvs/openbsd/src/sys/arch/socppc/socppc/machdep.c,v
> retrieving revision 1.33
> diff -u -p -r1.33 machdep.c
> --- socppc/machdep.c 29 Aug 2011 20:21:44 -0000 1.33
> +++ socppc/machdep.c 10 Mar 2012 20:52:06 -0000
> @@ -177,21 +177,6 @@ initppc(u_int startkernel, u_int endkern
> /* Make a copy of the args! */
> strlcpy(bootpathbuf, args ? args : "wd0a", sizeof bootpathbuf);
>
> - if (fwfdtsave == NULL) {
> - /*
> - * We were loaded by an old U-Boot that didn't provide
> - * a flattened device tree. It should have provided a
> - * valid bootinfo structure which we'll use to build
> - * such a device tree ourselves.
> - *
> - * XXX We don't build a flattened device tree yet.
> - */
> - memcpy(&bootinfo, *fwargsave, sizeof bootinfo);
> -
> - extern uint8_t dt_blob_start[];
> - fdt_init(&dt_blob_start);
> - }
> -
> if (fwfdtsave && fwfdtsave->fh_magic == FDT_MAGIC) {
> /*
> * Save the FDT firmware blob passed by the bootloader
> @@ -231,6 +216,19 @@ initppc(u_int startkernel, u_int endkern
> if (fdt_node_property(node, "mac-address", &addr))
> memcpy(bootinfo.bi_enetaddr, addr, 6);
> }
> + } else {
> + /*
> + * We were loaded by an old U-Boot that didn't provide
> + * a flattened device tree. It should have provided a
> + * valid bootinfo structure which we'll use to build
> + * such a device tree ourselves.
> + *
> + * XXX We don't build a flattened device tree yet.
> + */
> + memcpy(&bootinfo, *fwargsave, sizeof bootinfo);
> +
> + extern uint8_t dt_blob_start[];
> + fdt_init(&dt_blob_start);
> }
>
> proc0.p_cpu = &cpu_info[0];
Thanks. Verified that this doesn't break the rb600. Dereferencing a
garbage pointer is still a bit risky I suppose, but this is better
than what's currently in the tree. So I have committed this.
New snapshots should start appearing soon. Not sure if the first one
will have this diff, but the one after that surely will.
Cheers,
Mark
