Add support to locate TMC trace buffer in memory when the
address width is greater than 32 bits.

Change-Id: I7d916a6f9cfe0312764702b67d410778eaa55a04

Add support to the ramdump parser to parse debug image v2
format. Current support is for parsing cpu context and QDSS
dumps.

Change-Id: Ic7437e3b999845d607366bcf727f00078f81915a

We don't support python3 at all and python2.6 can only be used if the
user installs an extra package (OrderedDict [1]). Ask for python2
explicitly in the shebang line to resolve the python3 issues and print
some instructions to the user regarding python2.6 when that is used.

If the user insists on using python2.6, provide a command line
switch (--force-26) to skip the version check. To reduce clutter, don't
document this switch in the --help text but inform the user about its
existence when we error out due to the python2.6 check.

In order to even print an error message in python3 we have to first
ensure that our code is valid in python3, otherwise the interpreter
fails to even start executing our code (so we can't print any
messages). Use 2to3 to patch up some print statements, etc. Also fix
inconsistent whitespace, which is a syntax error on python3.

[1] https://pypi.python.org/pypi/ordereddict

Change-Id: Ie2c0a200e60ec90bf6cf49789f2cc75f181fa94b

ListWalker.__init__ was recently changed to remove some unneeded
arguments. The vmalloc parser was also updated at roughly the same time
with a new usage of the ListWalker class and was using the old
constructor. Update it.

Change-Id: Ia47c22511aab7207b65813d25c9f97d65c24e2db

It's not always useful to pass extra data to the list walker function.
For example, if the list walker function is a class method, state can be
passed around by simply modifying class instance variables. Update
`ListWalker.walk' to accept extra arguments optionally, not require
them.

Change-Id: I433f1e34db5c143c838f0eb2ed7f58203712ccb3

The parser now requires Python 2.7 so that 2.7-specific features can be
leveraged. Update the README to reflect this.

Change-Id: I7e1ba9f99f077c7f48ed7b44ab24f56f0f18c3ea

[29d3f2ed: "lrdp-v2: refactor Iommu domain parsing code"] introduced
an extra argument for rb_tree `walker' functions, but didn't update the
runqueue rb_tree user, so the ramdump parser currently fails to print
the runqueue state. Fix this.

For certain dumps, before this change we would see the following
backtrace:

    Traceback (most recent call last):
      File "ramparse.py", line 266, in <module>
        p.cls(dump).parse()
      File "parsers/runqueue.py", line 196, in parse
        self.print_cfs_state(cfs_rq_addr)
      File "parsers/runqueue.py", line 101, in print_cfs_state
        rb_walker.walk(tasks_timeline_addr, self.cfs_node_func)
      File "rb_tree.py", line 54, in walk
        self._walk(node, func, [], extra)
      File "rb_tree.py", line 39, in _walk
        self._walk(left_node, func, seen, extra)
      File "rb_tree.py", line 39, in _walk
        self._walk(left_node, func, seen, extra)
      File "rb_tree.py", line 41, in _walk
        func(node, extra)
    TypeError: cfs_node_func() takes exactly 2 arguments (3 given)

Change-Id: Ic19ec42d6cfbe87b6e0a0e99c6adf7c41e5b35e9

Because pages are constantly being allocated and freed, the pagetypeinfo
may not be in a consistent state when dumps are collected. Check
to make sure there are no cycles in the list for more than just the
next item.

Change-Id: Ie986bbb78f39923dbe11ef0446912518bc6c7fc5

Sizes for 64-bit are different. Update rtb accordingly.

Change-Id: Ie7c4014cc22b7c071017bb3d76519542ba58082d

It's bothersome at a fundamental level that the first line of the
stack trace is printed out on the same line as the 'Stack:' tag.
Fix it.

Change-Id: I895ad0969c4afb5b081ad10dca180a2bb1178291

The number of interrupts is an int which is different from the word
size. Switch to it.

Change-Id: I06b5186ac0121854870606c78488c26f22d79f2c

The way the vmalloc list is walked has changed to use a different
structure. The old structure is now freed at init time. This may
not be noticed if init memory isn't freed. Switch to a method that
does not rely on not freeing initmem.

Change-Id: I9f9b095ac4740ff13c1f54e39c641f286941a63f

Many of the lists are very large and may overflow if traversed
recursively. Switch to an iterative method instead.

Change-Id: I8bf476eefae7e223a16fdd50bb9b10d7c24cb247

Reading 64 bits of data is needed at times. Add a dedicated function

Change-Id: Ibef3777a574fb4ea87af51ab8230af7d4402c4f9

Currently, if the user doesn't specify a vmlinux image or an
autodumpdir (like when running ramparse.py with absolutely no arguments)
they get a big ugly python stacktrace and no helpful error message. Fix
this.

Change-Id: I2d5aaaae5c8afd651ef48cfbd8f974227c2032fe

The IOMMU parser currently only supports short-descriptor format page
tables but hardware supports long-descriptor format page tables. Add
support for the long-descriptor format. Most of the existing LPAE MMU
code can be re-used for this, although it must be refactored a bit.

Change-Id: I464d7edeb8818c0be9de1e1a8224bc9b478eac37

Register objects are `equal' if they have the same fields and the values
of all of those fields are equal. Overload Register.__eq__ to reflect
this.

Change-Id: I680e869470b20d6dc88d4bd78da9f002980aea4f

The Register class supports single-bit field definitions. Add a doctest
for example purposes and regression testing.

Change-Id: Ib04762aca8cb00256e16c29ac2ab71d47d0c06da

The Register class currently can't handle setting the `value'
attribute. Fix this. Also add a `zero' convenience method for zeroing
out a register.

After adding the following doctest:

    >>> abc.value = 0
    >>> abc.value
    0

but before fixing __setattr__, I get:

    **********************************************************************
    File "register.py", line 42, in __main__.Register
    Failed example:
        abc.value = 0
    Exception raised:
        Traceback (most recent call last):
          File "/usr/lib/python2.7/doctest.py", line 1289, in __run
            compileflags, 1) in test.globs
          File "<doctest __main__.Register[10]>", line 1, in <module>
            abc.value = 0
          File "register.py", line 106, in __setattr__
            raise AttributeError
        AttributeError
    **********************************************************************
    File "register.py", line 43, in __main__.Register
    Failed example:
        abc.value
    Expected:
        0
    Got:
        49
    **********************************************************************
    1 items had failures:
       2 of  20 in __main__.Register
    ***Test Failed*** 2 failures.

After the fix in __setattr__, all tests pass successfully.

Change-Id: I6d39261b139aae0def58f38a4291baad651b10e3

It can be useful to extract the dmesg from contexts other than the
--dmesg parser. Refactor the code into a top-level "library" module and
hook the --dmesg parser into that.

Change-Id: I935c05a081b1096d67343f5ce8617febb674b2e0

It's useful to be able to dump regions of memory a la `xxd'. Add some
machinery to parser_util and RamDump to do this.

Example:

    $ ramparse.py -a . -o parsed --shell
    >>> dump.hexdump(0, 100, virtual=False)
    00000000: fffe ffde 0200 0000 ffff b7bf adf5 a5ff  ................
    00000010: e7bf edfe c6fe edf6 fefa fffe befa fefe  ................
    00000020: bfbe ffef aebd a5ef ffff 7f7e fefe 5afb  ...........~..Z.
    00000030: 1b00 2000 3300 4300 1c00 2100 3400 4800  .. .3.C...!.4.H.
    00000040: ffff ff7f fafe ffdb effd a5af adf5 a5bf  ................
    00000050: 4444 4444 4444 4444 4444 4444 4444 4444  DDDDDDDDDDDDDDDD
    00000060: 5fff ffff                                _...
    >>> dump.hexdump(0xc0000000, 100, virtual=True)
    c0000000: fffe ffde 0200 0000 ffff b7bf adf5 a5ff  ................
    c0000010: e7bf edfe c6fe edf6 fefa fffe befa fefe  ................
    c0000020: bfbe ffef aebd a5ef ffff 7f7e fefe 5afb  ...........~..Z.
    c0000030: 1b00 2000 3300 4300 1c00 2100 3400 4800  .. .3.C...!.4.H.
    c0000040: ffff ff7f fafe ffdb effd a5af adf5 a5bf  ................
    c0000050: 4444 4444 4444 4444 4444 4444 4444 4444  DDDDDDDDDDDDDDDD
    c0000060: 5fff ffff                                _...
    >>> dump.hexdump(dump.addr_lookup('linux_banner'), 144, virtual=True)
    c0b0006a: 4c69 6e75 7820 7665 7273 696f 6e20 332e  Linux version 3.
    c0b0007a: 3130 2e32 382d 6765 3232 3362 6632 3830  10.28-ge223bf280
    c0b0008a: 662d 6469 7274 7920 286d 6974 6368 656c  f-dirty (mitchel
    c0b0009a: 6840 6d69 7463 6865 6c68 2d6c 696e 7578  h@mitchelh-linux
    c0b000aa: 2920 2867 6363 2076 6572 7369 6f6e 2034  ) (gcc version 4
    c0b000ba: 2e37 2028 4743 4329 2029 2023 3720 534d  .7 (GCC) ) #7 SM
    c0b000ca: 5020 5052 4545 4d50 5420 5475 6520 4170  P PREEMPT Tue Ap
    c0b000da: 7220 3820 3134 3a32 303a 3132 2050 4454  r 8 14:20:12 PDT
    c0b000ea: 2032 3031 340a 007c 2f2d 5c00 0000 0000   2014..|/-\.....

Change-Id: Iecaf80f72845f052085a60eadedd2bb24743224c

The LPAE MMU code can be re-used for other modules that need page table
parsing (like the LPAE IOMMU page tables). Refactor the code a bit so
that it can be easily re-used later.

Change-Id: I0deb80a69f2fca83c3ddcc78867fa2e5f5d18bb2

The Iommu domain parsing code is common across different Iommu page
table types (normal vs LPAE). Pull this code out so that it can be
easily re-used later when LPAE page table support is added.

Change-Id: Ifc39c12428a9c090f8197d997ddd293e0ae1f0b1

register.py is almost ready to handle doctest testing. Fix up one typo
and add the necessary call in if __name__ == "__main__".

Change-Id: I0a79060bb3bb58e807e7699ec0c21ef0eddb062d

If a Register is created with a None value, it currently blows up. Fix
this.

Change-Id: Ic652620e367c64e0a8444ef4b3e5ea7d77edcd82

do_fl_sl_level_lookup is a class method, so it needs to be called on a
class object. Add some missing `self' class objects.

Change-Id: I8b9932ec2bd2eb7c2296355afff4783612171c7d

MMU.virt_to_phys currently lacks a docstring. They're useful, so add
one.

Change-Id: I95b0a7a8f97912f178157eb4b68c453585e547bd

Some common operations include getting the parent structure of some
embedded field as well as getting "sibling" fields within the parent
structure. Add some convenience methods to the gdbmi module (as well as
wrappers in the RamDump class) for this.

Change-Id: Ic65d7e60b930af8c73384a30b6b0dd84fb55f09c

Move to the new sizes module to avoid repetition.

Change-Id: I5b18837bb47c0077976e84bcc3af59a30ce42614

Add a convenience module with definitions for common sizes, conversions
to strings, functions for working with orders, etc.

Change-Id: I922d6daa0c021a0d189b53aac76ed211e8864923

out_dir is not used anymore. Remove it.

Change-Id: I082e0642a352e91b232ce02fe40e20a200de9fc8

Currently we error out and ask the user to create the output directory
if it doesn't exist. As a convenience, just create the directory for
them.

Change-Id: I0e9caa8294d644621bd166a06e2a70c8ca5715fc

The stack values may be bogus for a variety of reasons. Bounds
check these values and error out.

Change-Id: I646d7f6e8d2f0e03f0b36bc30d7b1bbaa604fc98

There have been requests to use CorexA7 instead of CortexA53 for
dumps. Switch.

Change-Id: I5926e361bfc012edf26f043adb30b4254b96bc8a

It's not uncommon for the vmlinux file to live in the same directory as
the actual RAM dumps. If the users doesn't explicitly specify a vmlinux
image to use, look for it in the autodump dir. If we still can't find it
there, error out as usual.

Change-Id: I702c5584477e26912e0857dbd03b02491bd33c6c

When local_settings.py is missing variables we just error out when
trying to access those attributes. Improve the error message by
mentioning local_settings.py and pointing the user to README.txt.

Change-Id: Iaa2a1ab96ca889d75972bbfa1d4e127a8bf5bec1

Add support to RamDump.virt_to_phys for automatically looking up
addresses of symbols when a string is passed in. This will simplify code
that uses the RamDump.read_* routines.

Old way:

    mydata_addr = ramdump.addr_lookup('data_var')
    mydata2_addr = ramdump.addr_lookup('data2_var')

    mydata = ramdump.read_word(mydata_addr)
    mydata2 = ramdump.read_word(mydata2_addr)

New way:

    mydata = ramdump.read_word('data_var')
    mydata2 = ramdump.read_word('data2_var')

Of course, the "Old way" will still work.

Change-Id: I0c006f5135b48f8385d2107d9b9d2fe767a266b0

During development, it's quite convenient to load up an interactive
python interpreter with a RamDump instance loaded. Add some command line
switches to launch an interactive shell with the RamDump instance that
gets created as a result of whatever other command line options are
passed.

If IPython is installed, an IPython interpreter will be launched,
otherwise a classic Python interpreter will be launched. The classic
interface can also be forced with --classic-shell.

Change-Id: Id59acb97a830055212de9db3eaf05f18358f757f

When calculating pfn_to_page, the pfn needs to be offset before
indexing into the array. Account for this.

Change-Id: If207447a91537ab7670058fede65473d3ea93281

When checking for rodata differences, we scan the ELF headers of
the vmlinux. Newer vmlinux files seem to have extra ELF headers
with addresses outside the valid mapped range. If we encounter
one of those headers and read an invalid address, skip it.

Change-Id: I2a8af784df7216bf9d890be42eec8599fbaa7f72