Re: [PATCH v2 01/05] ARM: shmobile: kzm9g-reference: 0x48008000 load address

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

On Wednesday 17 December 2014 14:35:32 Geert Uytterhoeven wrote:
> On Wed, Dec 17, 2014 at 2:08 PM, Magnus Damm <magnus.damm@gmail.com> wrote:
> >> Do you boot a zImage or a uImage?
> > 
> > uImage built from zImage with appended DTB.
> > 
> > These 4 steps actually:
> > rm arch/arm/boot/zImage
> > make -j 9 ARCH=arm CROSS_COMPILE=_ dtbs zImage
> > cat arch/arm/boot/dts/sh73a0-kzm9g.dtb >> arch/arm/boot/zImage
> > make -j 9 ARCH=arm CROSS_COMPILE=_ LOADADDR=0x48008000 uImage
> 
> Thanks, that works!

And it doesn't for me :-/

Any idea why NFS root doesn't work with your wip branch and shmobile_defconfig 
(with NFSv4 and NFSv3 ACL disabled) ? The kernel stops after

[    6.630000] Sending DHCP requests ., OK
[    6.670000] IP-Config: Got DHCP answer from 192.168.1.47, my address is 
192.168.1.254
[    6.680000] IP-Config: Complete:
[    6.690000]      device=eth0, hwaddr\0:01:9b:04:04:22, 
ipaddr\x192.168.1.254, mask%5.255.255.0, gw%5.255.255.255
[    6.700000]      host\x192.168.1.254, domain=, nis-domain=(none)
[    6.700000]      bootserver\x192.168.1.47, rootserver\x192.168.1.47, 
rootpath=/home/laurent/src/iob/netboot/renesas
[    6.710000]      nameserver0\x192.168.1.47
[    6.720000] SDHI2 Vcc: disabling
[    6.720000] SDHI0 Vcc: disabling
[    6.720000] vccq_mc0: disabling
[    6.730000] ALSA device list:
[    6.730000]   #0: fsia-dai-ak4642-hifi

and then oopses after a timeout. No NFS-related packet is received by the host 
according to tcpdump.

> But the machine lost 128 MiB of RAM, compared to adjusting the "reg"
> property of the memory node :-(
> 
> >> I usually use a zImage with appended DTB on kzm9g. It fails as before:
> >> 
> >> Uncompressing Linux... done, booting the kernel.
> >> Booting Linux on physical CPU 0x0
> >> Linux version 3.18.0-kzm9g-multiplatform-11069-g3086a70162bdec42
> >> (geert@ramsan) (gcc version 4.9.0 (GCC) ) #274 SMP Wed Dec 17 13:28:08
> >> CET 2014
> >> CPU: ARMv7 Processor [412fc098] revision 8 (ARMv7), cr\x10c5387d
> >> CPU: PIPT / VIPT nonaliasing data cache, VIPT aliasing instruction cache
> >> Machine model: KZM-A9-GT
> >> bootconsole [earlycon0] enabled
> >> debug: ignoring loglevel setting.
> >> Memory policy: Data cache writealloc
> >> On node 0 totalpages: 124928
> >> free_area_init_node: node 0, pgdat c058f640, node_mem_map df428000
> >> 
> >>   Normal zone: 976 pages used for memmap
> >>   Normal zone: 0 pages reserved
> >>   Normal zone: 124928 pages, LIFO batch:31
> > 
> > Wow, earlycon. Luxury. =)
> 
> Noh, a necessity when developing ;-)
> 
> > The string above says "kzm9g-multiplatform". Did you try legacy or DT
> > reference without CCF?
> 
> Yes, they both work (with zImage and appended DTB), as before.
> 
> >> When trying with a uImage with the "new" load address i.e.
> >> 
> >> make ARCH=arm LOADADDR=0x48008000 uImage
> >> 
> >> it still fails:
> >> 
> >> ## Flattened Device Tree blob at 50000000
> >> 
> >>    Booting using the fdt blob at 0x50000000
> >>    Loading Kernel Image ... OK
> >> 
> >> OK
> >> 
> >>    Loading Device Tree to 417f8000, end 417ff7be ... OK
> 
> Aarghl, u-Boot copied the DT below 0x48000000. That's why it fails:
> >> Unable to handle kernel paging request at virtual address b97f8000
> 
> Hence no choice but to use uImage+DTB.
> 
> > You seem to be using "bootm 0x43000000 - 0x50000000".
> > In my case I simply use "bootp; bootm". I also tend to use appended
> 
> My DHCP server (OpenWRT dnsmasq) doesn't provide TFTP/NFS
> information to its clients.
> 
> > DTB and compile-in the command line to rely on the boot loader as
> > little as ever possible.
> 
> Indeed.
> 
> > One thing that bit me for quite some time was when handling the case
> > CONFIG_AUTO_ZRELADDR=n. I built the zImage using my steps above but I
> > did not pass the LOADADDR when building the zImage - so the
> > "zreladdr-y" value in Makefile.boot was still the default value which
> > ended up in the kernel image. What I mistakenly ended up doing was to
> > only adjust the uImage load address. And your bootm line above has
> > even more magic to it - so you may have 3 different versions... =)
> 
> I usually use zImage, which is better for multiplatform.
> In theory the same multiplatform zImage should boot on all machines.

-- 
Regards,

Laurent Pinchart