Linksys NSLU2 (evbarm, armeb) userspace regression origin found

[Eddy Petrișor <eddy.petrisor+netbsd.org%gmail.com@localhost>]

Hello,

(The first paragraph contains some background information for people
that might start directly with this mail.)

After my previously reported struggles with booting NetBSD on one of
my Linksys NSLU2[s], finding a different mode the 6.99.17 kernel was
failing[D], the success I had with booting NetBSD 6.1 from netbsd_6_1
branch[b] and confirming the NetBSD 6.1 kernel can work with NetBSD
userspace, then hacking together a dedicated (but strongly dependent
on the exact same version of the RedBoot bootloader on the slug)
second stage bootloader to load automatically a netbsd-nfs.bin image
over TFTP based on apex[a] and even changing the default commands
executed in RedBoot (after getting and checking my JTAG dongle)[j], I
finally was in a good position use the netbsd_6_1 kernel to bisect
through the NetBSD history to find out when the regression in
userspace NetBSD 7 was introduced.


I have managed to bisect through the history of the NetBSD 7 and found
out that the offending commit leading to the second type of failure[D]
(manifesting through endless repetition of the error message 'Process
(pid 1) got sig 11' when trying to run init) was 1.15 of
external/gpl3/gcc/dist/gcc/config.gcc:

===============
Revision 1.15 / (download) - annotate - [selected], Thu Jan 31
06:53:21 2013 UTC (2 years, 3 months ago) by matt
Branch: MAIN
Changes since 1.14: +7 -2 lines
Diff to previous 1.14 (colored)

Cleanup arm--netbsdelf-*eabi*
===============

This was the only modified file at the same time with this commit
message and the commit before that was done approximately one minute
before it at 06:52:16 (in case somebody wants to replicate).

The unified diff is quoted below in the mail. The side-by side changes
are visible online at:

http://cvsweb.netbsd.org/bsdweb.cgi/src/external/gpl3/gcc/dist/gcc/config.gcc.diff?r1=1.14&r2=1.15&f=h



To make sure I haven't made any mistake:
1) I have made sure I had a clean tree then using build.sh I did
'build', 'distribution' and 'sets' did the necessary preparations
(automated through a script, so there couldn't be any errors there)
and tried to start netbsd_6_1 kernel with the userspace generated from
the code right after the change; it failed to boot manifesting as
described in [D].
2) I did the same with the code before the change and confrimed this can boot
3) I have confirmed that reverting the change in question fixes the issue



Notes:
a) I suspect reverting only the first chunks to contain
TARGET_ENDIAN_DEFAULT=MASK_BIG_END would revert the problem, but I am
unsure, I haven't tested yet, I haven't anlaysed the cause, but I
don't have the necessary NetBSD+GCC knowledge to confirm this; help is
greatly appreciated
b) Not sure if this is change refers to the problem reported in PR
#48337 (reported on 23rd of Oct 2013 against 6.99.24), but I am
willing to do the necessary testing to confirm this.
c) I tried reverting the change in a more recent code, but the code
base changed too much and I lack the necessary knowledge to adapt the
revert to check; I am willing and can try proposed fixes, if some kind
netbsd person has the skills and will to provide such a fix



Help is greatly appreciated!




===================================================================
RCS file: /ftp/cvs/cvsroot/src/external/gpl3/gcc/dist/gcc/config.gcc,v
rcsdiff: /ftp/cvs/cvsroot/src/external/gpl3/gcc/dist/gcc/config.gcc,v:
warning: Unknown phrases like `commitid ...;' are present.
retrieving revision 1.14
retrieving revision 1.15
diff -u -p -r1.14 -r1.15
--- src/external/gpl3/gcc/dist/gcc/config.gcc 2012/12/20 07:12:26 1.14
+++ src/external/gpl3/gcc/dist/gcc/config.gcc 2013/01/31 06:53:21 1.15
@@ -724,11 +724,11 @@ arm*-*-freebsd*)
 arm*-*-netbsdelf*)
  tm_file="dbxelf.h elfos.h netbsd.h netbsd-elf.h arm/elf.h arm/aout.h
arm/arm.h"
  case ${target} in
- armeb*) tm_defines="${tm_defines} TARGET_ENDIAN_DEFAULT=MASK_BIG_END" ;;
+ armeb*) tm_defines="${tm_defines} TARGET_BIG_ENDIAN_DEFAULT=1" ;;
  esac
  tmake_file="${tmake_file} arm/t-arm"
  case ${target} in
- arm*-*-netbsdelf-*eabi)
+ arm*-*-netbsdelf-*eabi*)
     tm_file="$tm_file arm/bpabi.h arm/netbsd-elf.h arm/netbsd-eabi.h"
     tmake_file="$tmake_file arm/t-arm-elf arm/t-bpabi"
     tmake_file="$tmake_file arm/t-netbsd-eabi"
@@ -744,6 +744,11 @@ arm*-*-netbsdelf*)
     tmake_file="$tmake_file arm/t-netbsd"
     ;;
  esac
+ case ${target} in
+ arm*-*-netbsdelf-*eabihf*)
+    tm_defines="${tm_defines} TARGET_DEFAULT_FLOAT_ABI=ARM_FLOAT_ABI_HARD"
+    ;;
+ esac
  ;;
 arm*-*-netbsd*)
  tm_file="arm/aout.h arm/arm.h netbsd.h netbsd-aout.h arm/netbsd.h"





[i] http://mail-index.netbsd.org/port-arm/2015/02/19/msg002863.html
[D] http://mail-index.netbsd.org/port-arm/2015/03/04/msg002893.html
[b] http://mail-index.netbsd.org/port-arm/2015/03/24/msg002963.html
[a] http://mail-index.netbsd.org/port-arm/2015/04/26/msg003143.html
[j] http://mail-index.netbsd.org/port-arm/2015/05/04/msg003157.html