ARMhf on the Kindle(s)

[knc1]

This description **was not** the major problem.
But it does describe considerations to be made with "out of tree" system libraries.

Spoiler:

Originally Posted by NiLuJe

If that helps, that's what happens when the kernel can't even run the binary (wrong arch, possibly fatally wrong CPU tuning, although that should SIGILL instead).

I have had a 4-1/2 hour (night-time) nap, and now the answer is "obvious" - - -

The runpath string is too long for ld.so and/or the kernel (untested at the moment).
But that is the only significant change as the builds progressed from "working" to "totally failing".
The runpath string was getting longer.

We all win (for reasons unknown at the time) -
You are correct: The kernel can't run it.
Our (twobob and I) plan to flatten the FHS tree into a single level.
(Everything then gets the short search path string: "$ORIGIN/../lib")

If that (relative search path) doesn't cover all use cases, then a rename of where we put things might.
Such as: "/mnt/us/esys/lib:mnt/us/esys/usr/lib"

Hmm...
#2 uses absolute pathnames -
And there are undocumented (outside of source code comments) restrictions on the use of "$ORIGIN ....".

Three level, absolute, search path with current naming: 103 characters (broken wherever tried).

Two level, absolute, search path with current naming: 63 characters (broken in some places tried).

Renamed structure, two level, absolute, search path with new naming: 36 characters.

And the winner is (should be): /mnt/us/esys/lib:mnt/us/esys/usr/lib

In: Real world use VS. Documented use
"Real World" wins this one and we just get to conform.

[knc1]

(Kterm)

Install

• Connect USB cable
  Leave in USB storage mode.
• Place downloaded tar-ball in topmost directory of USB storage.
• Safely remove and eject USB Storage.
• Kual -> Kterm
• enter: cd /mnt/us
• enter: rm -fr esys
• enter: tar -xzf base-0.3-a8_armhf_vfpv3d16.tar.gz

Usage (kterm)
While in kterm - NOTE: This is a per-session setup!
This setup will not persist between kterm sessions.

• enter: cd /mnt/us
• enter: esys/bin/busybox ash
• enter: export PATH=/mnt/us/esys/bin:/mnt/us/esys/usr/bin:$PATH

Note: You are now running the bb-1.22 shell from the bb-1.17 shell
to 'exit' you have to enter: 'exit' twice.
Note: The "exec esys/bin/busybox ash" command might work under Kterm also, it is just that I did not test it.

(ssh)

Install

• core2quad BR_Working $ scp base-0.3-a8_armhf_vfpv3d16.tar.gz kpw:/mnt/us
  base-0.3-a8_armhf_vfpv3d16.tar.gz 100% 2438KB 2.4MB/s 00:01
• core2quad BR_Working $ ssh kpw
  #################################################
  # N O T I C E * N O T I C E * N O T I C E #
  #################################################
  Rootfs is mounted read-only. Invoke mntroot rw to
  switch back to a writable rootfs.
  #################################################
• [root@kindle root]# cd /mnt/us
• [root@kindle us]# rm -fr esys
• [root@kindle us]# tar -xzf base-0.3-a8_armhf_vfpv3d16.tar.gz

Usage (ssh)
While in ssh session - NOTE: This is a per-session setup!

• [root@kindle root]# cd /mnt/us
• [root@kindle us]# exec esys/bin/busybox ash
• [root@kindle us]# export PATH=/mnt/us/esys/bin:/mnt/us/esys/usr/bin:$PATH

Note: the 'exec' replaces the running bb-1.17 shell with the bb-1.22 shell.
So you only need one "exit" command to exit.

[knc1]

Hints on running the BR image file:
http://buildroot.org/downloads/manua...yond_buildroot
https://wiki.debian.org/QemuUserEmulation
http://www.gentoo.org/proj/en/base/e...?part=1&chap=5 (for the qemu wrapper example).

[knc1]

Version 0.3 released.
Top post and its attachments updated.

A text file of "notes" can be found in the 'helpers' archive along with the post-image-building script.

[twobob]

Now works perfectly thank you


as a word to the wise:

[root@kindle root]# /mnt/us/esys/bin/busybox ash

[root@kindle root]# unset LD_LIBRARY_PATH
[root@kindle root]# /mnt/us/esys/bin/busybox ash

[knc1]

Also check your shell's environment for any alias settings of the 227 new busybox utility names.
Those might effect the bb-command-wrappers in strange ways.

(The command wrappers for bb-1.22 do not work with bb-1.17)

Just enter:
alias
Which will list all those set.

Thanks for testing this out!

[twobob]

I have no alias

the busybox and its subvariants names are subsumed thanks to the way the busybox on the kindle is built.

The soft text magic linked variants we came up with last time are a much nicer alternative and the option to eat the names can be switched off in BR at build time, which wont help until you are inside a HF shell but hey ho.

[knc1]

Originally Posted by twobob

I have no alias

the busybox and its subvariants names are subsumed thanks to the way the busybox on the kindle is built.

The soft text magic linked variants we came up with last time are a much nicer alternative and the option to eat the names can be switched off in BR at build time, which wont help until you are inside a HF shell but hey ho.

Busybox 1.22 wrapper generation does not take into consideration the "Busybox location" string in the .config (it foolishly assumes I am installing it to /bin/busybox)

Which is why my helper is writing the v-1.22 wrappers itself.
(Note: Be careful of going close to those wrappers with an editor that inserts a newline at the end of the file. There **must not** be a newline at the end of the file.)

But sure - I can try something else - paste in a code block an example of what you used before.

[twobob]

https://bitbucket.org/twobob/kual-sy...l/cp?at=master

that kind of thing I was talking about

[knc1]

Originally Posted by twobob

https://bitbucket.org/twobob/kual-sy...l/cp?at=master

that kind of thing I was talking about

That is what you have (with the addition of a second line - the wrapper parsing must have changed between versions).

After doing the set-up in an ssh session, run:
top
stretch window so you can see the entire command line that "top" wrapper expanded to.




I do intend to post the .config files for Buildroot when this gets a bit closer to being done.

There are about 1,000 applications in this version of Buildroot that I did not build.
Plus, if I reach my goal, everything in Debian/Jessie.

Historical note - it was about 2 years ago that I first started to install newer system library structure to the Kindles - have learned a bit from MR since then.

[knc1]

The following example will (behind the scenes) dynamically load libz.so (from the Amazon soft-float libraries).

Since the C-ABI of libz does not pass floating point values - this will work.
I.E: This is a general rule, the "armhf-vfpv3d16" descriptor of our build only applies to parameter passing.

Note: This version of luajit only generates ARMv5-VFP code.
(Which also runs on our ARMv7-VFP3 machine).

Ref: http://luajit.org/ext_ffi_tutorial.html
The source of this example, includes a step-by-step description.

• ssh into your Kindle
• switch to the hard-float (and much newer) shell
  

  [root@kindle root]# cd /mnt/us
  [root@kindle us]# exec esys/bin/busybox ash
  [root@kindle us]# export PATH=/mnt/us/esys/bin:/mnt/us/esys/usr/bin:$PATH

• enter: nano zip-test.lua
• cut&paste this code into the editor:
  

  local ffi = require("ffi")
  ffi.cdef[[
  unsigned long compressBound(unsigned long sourceLen);
  int compress2(uint8_t *dest, unsigned long *destLen,
  	      const uint8_t *source, unsigned long sourceLen, int level);
  int uncompress(uint8_t *dest, unsigned long *destLen,
  	       const uint8_t *source, unsigned long sourceLen);
  ]]
  local zlib = ffi.load(ffi.os == "Windows" and "zlib1" or "z")
  
  local function compress(txt)
    local n = zlib.compressBound(#txt)
    local buf = ffi.new("uint8_t[?]", n)
    local buflen = ffi.new("unsigned long[1]", n)
    local res = zlib.compress2(buf, buflen, txt, #txt, 9)
    assert(res == 0)
    return ffi.string(buf, buflen[0])
  end
  
  local function uncompress(comp, n)
    local buf = ffi.new("uint8_t[?]", n)
    local buflen = ffi.new("unsigned long[1]", n)
    local res = zlib.uncompress(buf, buflen, comp, #comp)
    assert(res == 0)
    return ffi.string(buf, buflen[0])
  end
  
  -- Simple test code.
  local txt = string.rep("abcd", 1000)
  print("Uncompressed size: ", #txt)
  local c = compress(txt)
  print("Compressed size: ", #c)
  local txt2 = uncompress(c, #txt)
  assert(txt2 == txt)

• write-out (ctrl-o) and exit (ctrl-x) the editor
• compile and run: luajit zip-test.lua
• or: time luajit zip-test.lua
• or: LD_DEBUG=libs luajit zip-test.lua
  Follow along (scroll back as required) the library loading sequence.
  Note that our ld.so.3 has no problems loading the ld.so.2 format code.

Notes:
With the recommended path setting above, any command not defined in our esys hard-float tree will be run from Amazon's soft-float tree.
I.E: All of the Amazon functions we hack with continue to work un-changed.

With the FFI C interface of LuaJit, you can define and directly access the Amazon system libraries entry points.
**ANY** of them, as long as they don't pass floating point arguments.

[hawhill]

LuaJIT's FFI is jolly good fun, indeed :-) (we rely heavily on it in Koreader now - our distributions contain a plain LuaJIT binary, the Kobo version has a armhf one, the Kindle version is arm soft-float). Hm, now I'm asking myself it LuaJIT's FFI couldn't be massaged to support *both* calling conventions...

[knc1]

Originally Posted by hawhill

LuaJIT's FFI is jolly good fun, indeed :-) (we rely heavily on it in Koreader now - our distributions contain a plain LuaJIT binary, the Kobo version has a armhf one, the Kindle version is arm soft-float). Hm, now I'm asking myself it LuaJIT's FFI couldn't be massaged to support *both* calling conventions...

It does (at least the version-2.0.3 installed here does).
Ref: http://luajit.org/changes.html see: Item #2
Ref: http://luajit.org/ext_ffi_api.html see: status = ffi.abi(param) description.

Have not (yet) tested it myself.

[knc1]

Ubuntu Core, 14.04

To quote Ubuntu:

Installation on a file system type that does not support symbolic links may lead to unpredictable results.

Yea, Right!
No comment made about installing it somewhere other than / (root) -
and on FAT32 at that.
I guess no one there ever considered that one.

Link:
http://cdimage.ubuntu.com/ubuntu-cor...e-armhf.tar.gz

Made new appointment with my head shrinker
and attached the manifest here.

[twobob]

actually a lot less than I thought it was gonna have in there. 184 aint none but meh.

Encouraging