problems in building evbearm7hf (Re: big-endian kernel fails to boot on Cubietruck)

[Rin Okuyama <okuyama%flex.phys.tohoku.ac.jp@localhost>]

> > Finally, I've found the real cause of the problem; my sd card is broken!
> > With a new sd card, both little- and big-endian kernels with -mtune flag
> > work properly. I'm so sorry for confusing you.
>
> No problem at all, glad you found the issue!

Thank you so much! I'm very happy to have a nice big-endian environment!

> > I prepared patches to address these problems. Could I post them to this
> > mailing list? Or would it be better to send PRs?
>
> Whatever is easier for you. For larger patches, PRs would be preferable,
> but small things can just go via the list (as long as you keep an eye
> on them and make sure they do not get lost/forgotten).

OK. Actually, these are small patches. I attached them below:

(1) build fails with STRIPFLAG=-s

With STRIPFLAG=-s in /etc/mk.conf, all userland binaries are installed
with -s flag. Since there is a trick that /usr/mdec/*.bin are built as
userland programs, installation fails with this flag. To avoid this,
ignore STRIPFLAG for /usr/mdec guys.

====
--- sys/arch/evbarm/stand/gzboot/Makefile.gzboot.orig	2015-07-14 
12:49:29.000000000 +0900
+++ sys/arch/evbarm/stand/gzboot/Makefile.gzboot	2015-07-13 
16:10:47.000000000 +0900
@@ -26,6 +26,7 @@
 LIBC=
 LIBCRTBEGIN=
 LIBCRTEND=
+STRIPFLAG=

 .PATH: ${EVBARM_STAND}/gzboot
 .PATH: ${EVBARM_STAND}/board
====

(2) byte order of sshramdisk.fs is fixed to little endian

MACHINE_ARCH should be checked as it is for ramdisk.fs.

====
--- distrib/evbarm/instkernel/sshramdisk/Makefile.orig	2015-07-13 
16:39:15.000000000 +0900
+++ distrib/evbarm/instkernel/sshramdisk/Makefile	2015-07-13 
16:40:17.000000000 +0900
@@ -13,7 +13,11 @@
 CRUNCHBIN=	ramdiskbin
 LISTS=		${.CURDIR}/list ${DISTRIBDIR}/common/list.sysinst
 MTREECONF=	${.CURDIR}/mtree.conf
+.if empty(MACHINE_ARCH:M*eb)
 IMAGEENDIAN=	le
+.else
+IMAGEENDIAN=	be
+.endif
 MAKEDEVTARGETS=	ramdisk
 IMAGEDEPENDS=	${CRUNCHBIN} \
 		dot.profile master.passwd sshd_config ssh_host_dsa_key \
====

(3) byte order of ffs partition in armv7.img can be wrong

As -B flag for makefs is missing, the byte order of ffs partition in
armv7.img become the same as host. For e.g., if you build evbearm7-eb
on amd64 box, the ffs partition become little endian. Since FFS_EI
option is disabled for Cubieboard/Cubietruck, a big-endian kernel
cannot mount the root partition. I've added -B option to the mkimage
script, and specified it explicitly for armv7.img.

====
--- distrib/utils/embedded/mkimage.orig	2015-07-13 17:54:01.000000000 +0900
+++ distrib/utils/embedded/mkimage	2015-07-13 18:01:02.000000000 +0900
@@ -84,7 +84,7 @@

 usage() {
 	cat << EOF 1>&2
-Usage: $PROG -h <host-arch> [-bdmx] [-K <kerneldir>] [-S <srcdir>] [-D 
<destdir>] [-c <custom-files-dir>] [-s <Mb size>] [<image>]
+Usage: $PROG -h <host-arch> [-bdmx] [-B <byte-order>] [-K <kerneldir>] 
[-S <srcdir>] [-D <destdir>] [-c <custom-files-dir>] [-s <Mb size>] 
[<image>]

 -b	Boot only, no sets loaded
 -r	root device kind (sd, wd, ld)
@@ -96,7 +96,7 @@
 }

 # First pass for options to get the host and src directories
-OPTS="K:D:S:bc:dh:mr:s:x"
+OPTS="B:D:K:S:bc:dh:mr:s:x"
 while getopts "$OPTS" f
 do
 	case $f in
@@ -127,11 +127,13 @@
 xsets_p=false
 minwrites=false
 rootdev=ld
+endian=

 OPTIND=1
 while getopts "$OPTS" f
 do
 	case $f in
+	B)	endian="-B $OPTARG";;
 	D)	release="$OPTARG";;
 	K)	kernel="$OPTARG";;
 	S)	;;
@@ -193,7 +195,7 @@

 if [ -z "${bootonly}" ]; then
 	echo ${bar} Populating ffs filesystem ${bar}
-	${MAKEFS} -r -N ${release}/etc -t ffs -rx \
+	${MAKEFS} -rx ${endian} -N ${release}/etc -t ffs \
 	    -O ${ffsoffset} \
 	    -o d=4096,f=2048,b=16384 -b $((${extra}))m \
 	    -F "$tmp/selected_sets" ${image} "${release}" "${mnt}"
--- etc/etc.evbarm/Makefile.inc.orig	2015-07-13 17:18:27.000000000 +0900
+++ etc/etc.evbarm/Makefile.inc	2015-07-13 17:53:35.000000000 +0900
@@ -13,6 +13,7 @@
 .endfor

 .if !empty(MACHINE_ARCH:M*eb)
+IMAGEENDIAN=	be
 # big endian boards
 KERNEL_SETS.armv4+=		IXM1200
 EVBARM_BOARDS.xscale+=		ADI_BRH
@@ -32,6 +33,7 @@
 EVBARM_BOARDS.armv7+=		HUMMINGBIRD_A31
 EVBARM_BOARDS.armv7hf+=		HUMMINGBIRD_A31
 .else
+IMAGEENDIAN=	le
 # little endian boards
 #EVBARM_BOARDS.armv4+=		ARMADILLO210
 #EVBARM_BOARDS.armv4+=		ARMADILLO9
@@ -136,7 +138,7 @@
 	TOOL_MTREE=${TOOL_MTREE} \
 	HOST_SH=${HOST_SH} \
 	${HOST_SH} ${MKIMAGE} -x -h ${.TARGET:S/smp_//} -D ${DESTDIR} \
-	    -S ${NETBSDSRCDIR} ${MKI_OPTS.${.TARGET}} \
+	    -S ${NETBSDSRCDIR} -B ${IMAGEENDIAN} ${MKI_OPTS.${.TARGET}} \
 	    ${IMAGE.dir}/${.TARGET:S/smp_//}.img.gz

 ARCHES.arm=		armv4 xscale armv5t armv6 armv7
====

(4) $kernel environment variable in uEnv.txt is expanded by /bin/sh

It is not an environment variable for /bin/sh but for u-boot.
Escape it to avoid expansion in the building process.

====
--- distrib/utils/embedded/conf/armv7.conf.orig	2015-07-13 
17:51:04.000000000 +0900
+++ distrib/utils/embedded/conf/armv7.conf	2015-07-13 17:51:21.000000000 
+0900
@@ -45,7 +45,7 @@
 populate_awin() {
 	cat >> "${mnt}/boot/uEnv.txt" << EOF
 bootargs=root=ld0a console=${console}
-uenvcmd=mmc dev 0; mmc rescan; if test \$fdtfile = 
sun7i-a20-cubieboard2.dtb; then setenv kernel netbsd-CUBIEBOARD2.ub; 
elif test \$fdtfile = sun7i-a20-cubietruck.dtb; then setenv kernel 
netbsd-CUBIETRUCK.ub; elif test \$fdtfile = sun6i-a31-hummingbird.dtb; 
then setenv kernel netbsd-HUMMINGBIRD_A31.ub; elif test \$fdtfile = 
sun7i-a20-bananapi.dtb; then setenv kernel netbsd-BPI.ub; fi; fatload 
mmc 0:1 82000000 ${kernel}; bootm 82000000
+uenvcmd=mmc dev 0; mmc rescan; if test \$fdtfile = 
sun7i-a20-cubieboard2.dtb; then setenv kernel netbsd-CUBIEBOARD2.ub; 
elif test \$fdtfile = sun7i-a20-cubietruck.dtb; then setenv kernel 
netbsd-CUBIETRUCK.ub; elif test \$fdtfile = sun6i-a31-hummingbird.dtb; 
then setenv kernel netbsd-HUMMINGBIRD_A31.ub; elif test \$fdtfile = 
sun7i-a20-bananapi.dtb; then setenv kernel netbsd-BPI.ub; fi; fatload 
mmc 0:1 82000000 \$kernel; bootm 82000000
 EOF
 }

====

Thanks,
Rin