Since kernel v5.0, there is a dedicated region of size 128MB
kept aside for BPF programs(BPF_JIT_REGION) which were
previously using module region. Also in v5.4, kernel VA space
was flipped to accomodate support for kasan shadow and 52bit VA.
Hence we are updating the VA space calculations in parser to make
it work on newer kernel.

Adding few changes on linux banner read and match for cleanup.

There was an unmonitored variable ramdump.version being used in
debug_image_v2 cauing version check failure. Adding a change to
use kernel_version.
Fix intend in debug_image_v2.

Briefly, the calculations are updated as below:

_PAGE_END(va)		= (-(UL(1) << ((va) - 1))) = 0xffffffc000000000
KASAN_SHADOW_END	= (_PAGE_END(VA_BITS_MIN)) = 0xffffffc000000000
BPF_JIT_REGION_START	= (KASAN_SHADOW_END)
BPF_JIT_REGION_SIZE	= (SZ_128M)
BPF_JIT_REGION_END	= (BPF_JIT_REGION_START + BPF_JIT_REGION_SIZE) = 0xffffffc000000000 + 0x8000000
MODULES_VADDR		= (BPF_JIT_REGION_END)
MODULES_END		= (MODULES_VADDR + MODULES_VSIZE) = 0xffffffc000000000 + 0x8000000 + 0x8000000 = 0xffffffc010000000
KIMAGE_VADDR		= (MODULES_END) = 0xffffffc010000000

Change-Id: I5a65b327e015970d68c41210fc813105fd6cb4b8
Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org>
Signed-off-by: Shivendra Pratap <spratap@codeaurora.org>

Autodump feature broke after addition of DDRCS0_2.BIN file.
Use a regex instead to (hopefully) reduce the need for future updates.

Change-Id: I04917c4e2c0a77903ab4ebd5e96050603d9aa3c3

Board definitions will be kept in a separate git repository.
They will continue to use the same format as before.

Change-Id: I00380de950a91914e3fa1873c08e16584f1f3370

Added check to make sure num_symtab is not noe
to avoid TypeError: range() where num_symtab is
expected to be integer but sometimes it returned
to be None

Change-Id: I8e7acb450e9928a1ff56bef7cba49541a0e2bb5b

With some newer versions of Trace32, cmd PER.S.F cannot be used
to setup the MMU related registers.

Use PER.S.simple instead when enabling MMU support on 32-bit
ARM platforms.

Change-Id: I692ce4a2edaf1ff575e1df247ae51ab332d74166

When parsing kernel module symbols retrieved from kallsyms
nodes, lrdp-v2 will unconditionally add address 0x0 when seeing
it, which causes task stack unwinding showing a particular
module's name for address 0x0.

Add symbol address checking when parsing info retrieved from
module specific kallsyms nodes, to avoid adding address 0x0 as
undefined symbol for a particular module.

Change-Id: I36cc8e6a3ffa3b76bf24fd881f26845983059472

Lookups should be done using a dictionary, instead of doing a
string search through the --print-kconfig parser output.

Found via:
python -m cProfile ramparse.py --autodump <path>
--print-pagetracking

Change-Id: Ia8d4e7921151f61f0c492749720ea4ed777e028a
Signed-off-by: Patrick Daly <pdaly@codeaurora.org>

Add code to retrieve module specific symbol info from per-module
kallsyms nodes when CONFIG_KALLSYMS is defined. Symbols retrieved
from kallsyms nodes will be more accurate because relocation
has already been done when the per-module kallsyms nodes are
formed.

Add the following options to control how the symbol info will
be recorded in lrdp-v2 output files:

--dump_mod_sym_tbl: dump per-module symbol info retrieved from
                    module specific symbol files.
--dump_krnl_sym_tbl: dump Linux kernel symbol info retrieved
                     from the kernel symbol file.
--dump_mod_kallsyms: dump per-module symbol info retrieved from
                     per-module kallsyms node.
--dump_glb_sym_tbl: dump everything in the global symbol lookup
                    table.

Change-Id: Ic137eaf64a5e1f7203946d323efcd7527e3f1968

pagetracking is broken because page_owner->order size is changed
to unsigned short.
Add check for unsigned short(size = 2) in read_structure_filed
if size is 2 then use the read_u16.

Change-Id: I63a790709718eda89c669b081eb59468b99cd90c

Enforce the physical offset to be picked from boards config file.
Autodump feauture takes physical offset from ddr info files which
assumes kernel is loaded at ram start address. This is no longer a
valid assumption.

Change-Id: I0d55ce2f93198a802723017e2a9f64d0877f461c
Signed-off-by: Rishabh Bhatnagar <rishabhb@codeaurora.org>

Remove symbols which are not between _head and _end from the
symbol lookup table.

Change-Id: I785d6bac029162dd012717e61f927fd64df87514

These changes add a new option --kernel-exception-level
which can be passed from command line to specify the
current exception level of kernel. This is to handle
a case where backtrace should get generated if kernel
is executing at EL2 instead of EL1.

Change-Id: If0d61119a6cc27e263d73ce9921a63cab4bdee45
Signed-off-by: Kumar Harsh <harkumar@codeaurora.org>

        In case of 32bit ramdump kimage_voffset will not be used in ramparser.
        In upstream kernel 4.14 this variable got added and set to some address.
        Due to that in case of 32 bit all calculation went worng. So forcefully setting to null if it is 32 bit dump

Change-Id: Ia341074ebfd0811543cc3a2e77d2b23ade43d520

This patch fixes the need for a hard coded value
to specify upper limit on length of the Linux Banner.

get_version() inside ramdump.py has been renamed to
get_matched_version() to reflect the true nature of
the function.

Added a variable banner_len in get_matched_version()
which is the length of linux_banner returned by
gdbmi.get_value_of_string('linux_banner').
This variable is further passed to read_cstring()
function to specify the max limit of characters to
be read. We are dynamically calculating the length
using gdb to read linux banner.

Change-Id: I571882a5999c26ccd8accd90b26d0149c839633b
Signed-off-by: Kumar Harsh <harkumar@codeaurora.org>

Add a utility function for iterating through all valid process
and thread of each process.
Use it in different parser where we are traversing through the task
list Ex : memusages.py, ion_buffer_parse.py

Change-Id: Ided26acc67e7c5dc2ba7d4c2db0935f0c9fa24f1
Signed-off-by: Ankur Bansal <ankban@codeaurora.org>

1) When lrdp-v2 is parsing RAM dumps generated on ARM-32 platforms,
   the Trace32 startup script will not contain any instructions
   to enable MMU support, which causes Linux kernel symbol loading
   failure on some newer Trace32 simulators.

   Add Trace32 startup instruction to set bit 0 of P15:0x1 if
   lrdp-v2 detects MMU support within the RAM dump.

2) Use task.symbol.loadmod instead of data.load.elf to load all
   the kernel moduels' symbols to avoid symbol loading failures.

Change-Id: Ibae79e3c80279788d7cccd28080d2cecd5432a4b

Currently LRDP-v2 will only add linux.t32 and linux.men into
the t32_startup_script.cmm script when it is parsing RAM
dumps generated on ARM 32-bit platforms.

Change the code in ramdump.py to add linux3.t32 and its
related linux.men into t32_startup_script.cmm when it is
determined that the kernel version is greater than 3.0.0.

Change-Id: Id0ddc62cc702b0bc586bb9864ceb02177d397e13

Verify that files exist before checking file size.

Change-Id: Icb55984bf33a677594726345f4c82c55c2948134

The hypervisor may populate an imem cookie with the location of
a standard dump table, instead of using the table allocated by linux.
Read the table location from this cookie instead of relying on the
linux variable 'memdump'.
On error, fall back to the old approach of reading from 'memdump'.

Change-Id: Id99bb643fb7f35de79ff7308a999cc2143d7aff0

Currently the loadble module symbol parsing code is only decoding
the symbol names without prefixing the module names. This will
result in some confusion when two module's symbol names are
identical, and things get worse espeicially when parsing RAM
dumps where two WLAN drivers are loaded.

Prefix module names to module specific symbol names when parsing
their symbol files so user can easily tell which module is
involved in a call stack.

Change-Id: I3cf5e9527be191fb5f21830b1ba9f915d734b5d2

Starting from msm-4.9, a new field named 'core_layout' will be
available to replace 'module_core' in struct module.

Add code to parse the newly introduced field for loadable module
symbol addition based on the Linux kernel version.

Change-Id: I503ba6b0d4f5928bb2fea85ab3fdd4a33150b1c0

In thread_saved_field_common_32, read_word should not be used to
retrieve the thread_info structure's address because
get_thread_info_addr is already doing the same.

Change-Id: I422c67b98d318c0084ab32fb0ad9f180da186829

        In the latest kernel version minidump_table structure got changed. Some emelments are removed.
        like smem_table, version. In those information anyhow is not being used in ramparser. So removed all those elements
Change-Id: Ibce7ab2515737a7871c1f9df70d4eddee64bdc09

Add support to parse DCC dumps with autodump option to RDP
In autodump we give only ramdumps location without --ram-file
bin locations.

Change-Id: I87aa2b337415dd93cc5b60c813bf6cd1712f2565

	Read the linux banner data from vmlinux rodata section.
	Also read the linux banner data from ramump. Do a matching of both the data.
	If success then ramparser will go ahead and proces further
Change-Id: I1b9245f1a4fcc9cdff0a3bfb7293b2385981feba

	here fp is value is coming less than page_offset.
	Due that rest of the call stack is not getting printed

Change-Id: Ie878d41f0ecc16d21f9795c699f035a7460904b4

Current Linux Ram Dump Paser scripts are only able to load kernel
symbols by default. No loadable modules' symbols will be parsed
at startup time, which may result in 'unknown symbols' in some
cases, especially when checking SLAB info on dual-wifi platforms.

With this change, loadable modules' symbols can be parsed on
script startup by the following configurations:

1) Put all symbols under a certain dir, such as:

<dump_location>/symbols/wlan.ko
<dump_location>/symbols/wlan_sdio.ko

2) Add --sym_path when invoking the scripts, such as:

python %RAM_DUMP_PARSER_PATH%\ramparse.py --sym_path
<dump_location>/symbols

With the above actions, all loadable modules' symbols will be
parsed and stored for subsequent information dumping. Symbol
loading commands will also be added into the Trace32 startup
script.

Change-Id: Idda9c9a08cfd24c19bf4021e80fee5187cd031b9

The THREAD_INFO_IN_TASK kernel feature removes the thread_info
structure from task stack. Add support for this in the parser to handle
when this option is enabled.

Change-Id: I23eda300395936618e6d69eccd6c5d610c871ba8

Abstract out whether kaslr is enabled or not to allow ddrcompare to work
regardless of whether kaslr is enabled or not.

Change-Id: I4b00ed48748ce9a5a63c0905c99200ac9bfe4482

        As DCC is not going to be enabled on secure device,
        SDI would dump required registers in standard dcc dumping format.
        So extend ramparser’s DCC parsing support for <address><value> format.
        The KMISC section would contain all registers address value pair as mentioned below.
        Parse the data and generate the “dcc_captured_data.xml” and it should be same as actual dcc dump parsed xml.

Change-Id: Ib042ebc6895f0594f5784600174087a979e8709f

Add support for generating sram.bin from DCC_SRAM.BIN
With this NEW sram will call version 2 of DCC driver
for parsing.

Change-Id: If8012b1f38ba29daa5bdac2fd15820de51ea9cab

	A minidump supports fewer features than a full ramdump,
	but does not require the user to save all 4GB+ of memory.
	Instead a small subset of DDR is saved by crashscope as an elf file.
	Currently dmesg logs, rtb logs, and the cpu context parser are supported.

	Example usage: python ramparse.py --vmlinux <vmlinux> --minidump
			--ram-elf <elf_file> --ram-file <OCIMEM.bin> --everything

Change-Id: Ib689b4811d66472ac20aa7c63893fdaed915512e

   This changes to fetch serila number of HW from dump

Change-Id: I93ee37feafd9e407fe380761c0e6469fabe40f7d

When calling read_structure_field with an undefined symbol, the
desired behavior is to return None rather than crash.

Change-Id: Iaf172a038779cf767b55c47ec37b8e90e194aa0f

Add a function for reading a signed 64 bit value from a ramdump.

Change-Id: Ic7e5a8a7cb9d5fd551ffb2c3add2499a25aff540

This prints kimage_voffset info if kaslr is enabled.

Change-Id: Ia18bba9f2df1c579721aeb44a8dd2b7143183b89

This adds remove_file function to ramdump class.

Change-Id: Ibe6196aad7ccd1c2bb381b9adab682394244f879

As of commit c4c54dd1 (kernel/cpu.c: change type of cpu_possible_bits
and friends), cpu_present_bits no longer exists. Add new method of finding
number of cpus and also change other functions to use the get_num_cpus
API instead of defining it again.

Change-Id: I6d38d505ba7674b94e8481e6679226803eba977e

Add check to avoid "object has no attribute 'kaslr_addr'" on boards
where kaslr_addr is not defined.

Change-Id: I2691b313f6a0ab75b216cd678245e97cf22ce4e7

Kaslr-offset in stored in imem by the kernel driver. This patch
enables the parser to extract the required offset before parsing.

Change-Id: Ie773a2e52989b0f2a83c19d39ffdba7f77911e39