You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
openocd/TODO

394 lines
16 KiB

// This file is part of the Doxygen Developer Manual
/** @page tasks Pending and Open Tasks
This page lists pending and open tasks being considered or worked upon
by the OpenOCD community.
@section thelist The List
Most items are open for the taking, but please post to the mailing list
before spending much time working on anything lists here. The community
may have evolved an idea since it was added here.
Feel free to send patches to add or clarify items on this list, too.
@section thelisttcl TCL
This section provides possible things to improve with OpenOCD's TCL support.
- Fix problem with incorrect line numbers reported for a syntax
error in a reset init event.
- organize the TCL configurations:
- provide more directory structure for boards/targets?
- factor configurations into layers (encapsulation and re-use)
- Fix handling of variables between multiple command line "-c" and "-f"
parameters. Currently variables assigned through one such parameter
command/script are unset before the next one is invoked.
- Isolate all TCL command support:
- Pure C CLI implementations using --disable-builtin-tcl.
- Allow developers to build new dongles using OpenOCD's JTAG core.
- At first, provide only low-level JTAG support; target layer and
above rely heavily on scripting event mechanisms.
- Allow full TCL support? add --with-tcl=/path/to/installed/tcl
- Move TCL support out of foo.[ch] and into foo_tcl.[ch] (other ideas?)
- See src/jtag/core.c and src/jtag/tcl.c for an example.
- allow some of these TCL command modules to be dynamically loadable?
@section thelistadapter Adapter
This section list issues that need to be resolved in the Adapter layer.
@subsection thelistadapterrework Code restructuring
This section lists pending reworks to complete the restructure from the
old JTAG centric implementation to a generic Adapter layer.
This restructuring is very invasive and will prevent the merge of several
changes pending in gerrit.
- rename folder src/jtag/ to src/adapter/
- rename var "jtag" to "adapter" in src/jtag/core.c
- split content of src/adapter/ in the different protocols jtag.[ch],
swd.[ch], ...
- wrap the calls to adapter->transport_ops->api() with transport_api()
and reduce the visibility of global var "adapter"
- complete the migration of JTAG-only drivers to adapter->reset()
- try to remove JTAG_SLEEP also from JTAG mode?
- tap_set_state(TAP_RESET) is already done in src/jtag/core.c. No need
to replicate it in the drivers, apart in case the driver sets TRST
independently
- add .hla_ops to "adapter"
- HLA is a API level (.hla_ops). Transport should simply be {jtag,swd},
not {hla_jtag,hla_swd}.
@subsection thelistadapterjtagcore JTAG Core
The following tasks have been suggested for cleaning up the JTAG layer:
- use tap_set_state everywhere to allow logging TAP state transitions
- Encapsulate cmd_queue_cur_state and related variable handling.
- add slick 32 bit versions of jtag_add_xxx_scan() that avoids
buf_set_u32() calls and other evidence of poor impedance match between
API and calling code. New API should cut down # of lines in calling
code by 100's and make things clearer. Also potentially be supported
directly in minidriver API for better embedded host performance.
The following tasks have been suggested for adding new core JTAG support:
- Improve autodetection of TAPs by supporting tcl escape procedures that
can configure discovered TAPs based on IDCODE value ... they could:
- Remove guessing for irlen
- Allow non-default irmask/ircapture values
- SPI/UART emulation:
- (ab)use bit-banging JTAG interfaces to emulate SPI/UART
- allow SPI to program flash, MCUs, etc.
@subsection thelistadapterinterfaces Interface drivers
There are some known bugs to fix in Interface drivers:
- For JTAG_STATEMOVE to TAP_RESET, all drivers must ignore the current
recorded state. The tap_get_state() call won't necessarily return
the correct value, especially at server startup. Fix is easy: in
that case, always issue five clocks with TMS high.
- amt_jtagaccel.c
- arm-jtag-ew.c
- bitbang.c
- bitq.c
- gw16012.c
- jlink.c
- usbprog.c
- vsllink.c
- rlink/rlink.c
- bug: USBprog is broken with new tms sequence; it needs 7-clock cycles.
Fix promised from Peter Denison openwrt at marshadder.org
Workaround: use "tms_sequence long" @par
https://lists.berlios.de/pipermail/openocd-development/2009-July/009426.html
The following tasks have been suggested for improving OpenOCD's JTAG
interface support:
- rework USB communication to be more robust. Two possible options are:
-# use libusb-1.0.1 with libusb-compat-0.1.1 (non-blocking I/O wrapper)
-# rewrite implementation to use non-blocking I/O
- J-Link driver:
- fix to work with long scan chains, such as R.Doss's svf test.
- Autodetect USB based adapters; this should be easy on Linux. If there's
more than one, list the options; otherwise, just select that one.
The following tasks have been suggested for adding new JTAG interfaces:
- TCP driver: allow client/server for remote JTAG interface control.
This requires a client and a server. The server is built into the
normal OpenOCD and takes commands from the client and executes
them on the interface returning the result of TCP/IP. The client
is an OpenOCD which is built with a TCP/IP minidriver. The use
of a minidriver is required to capture all the jtag_add_xxx()
fn's at a high enough level and repackage these cmd's as
TCP/IP packets handled by the server.
@section thelistbs Boundary Scan Support
- add STAPL support?
- add BSDL support?
A few possible options for the above:
-# Fake a TCL equivalent?
-# Integrate an existing library?
-# Write a new C implementation a la Jim?
Once the above are completed:
- add support for programming flash using boundary scan techniques
- add integration with a modified gerber view program:
- provide means to view the PCB and select pins and traces
- allow use-cases such as the following:
- @b Stimulus
-# Double-click on a pin (or trace) with the mouse.
- @b Effects
-# The trace starts blinking, and
-# OpenOCD toggles the pin(s) 0/1.
@section thelisttargets Target Support
- Many common ARM cores could be autodetected using IDCODE
- general layer cleanup: @par
https://lists.berlios.de/pipermail/openocd-development/2009-May/006590.html
- regression: "reset halt" between 729(works) and 788(fails): @par
https://lists.berlios.de/pipermail/openocd-development/2009-July/009206.html
- registers
- add flush-value operation, call them all on resume/reset
- mcr/mrc target->type support
- missing from ARM920t, ARM966e, XScale.
It's possible that the current syntax is unable to support read-modify-write
operations(see arm966e).
- mcr/mrc - retire cp15 commands when there the mrc/mrc commands have been
tested from: arm926ejs, arm720t, cortex_a8
- ARM7/9:
- clean up "arm9tdmi vector_catch". Available for some arm7 cores? @par
https://lists.berlios.de/pipermail/openocd-development/2009-October/011488.html
https://lists.berlios.de/pipermail/openocd-development/2009-October/011506.html
- add reset option to allow programming embedded ice while srst is asserted.
Some CPUs will gate the JTAG clock when srst is asserted and in this case,
it is necessary to program embedded ice and then assert srst afterwards.
- ARM926EJS:
- reset run/halt/step is not robust; needs testing to map out problems.
- ARM11 improvements (MB?)
- add support for asserting srst to reset the core.
- Single stepping works, but should automatically
use hardware stepping if available.
- mdb can return garbage data if read byte operation fails for
a memory region(16 & 32 byte access modes may be supported). Is this
a bug in the .MX31 PDK init script? Try on i.MX31 PDK:
mdw 0xb80005f0 0x8, mdh 0xb80005f0 0x10, mdb 0xb80005f0 0x20. mdb returns
garabage.
- implement missing functionality (grep FNC_INFO_NOTIMPLEMENTED ...)
- Thumb2 single stepping: ARM1156T2 needs simulator support
- Cortex-A8 support (ML)
- add target implementation (ML)
- Cortex-M3 support
- when stepping, only write dirtied registers (be faster)
- when connecting to halted core, fetch registers (startup is quirky)
- Generic ARM run_algorithm() interface
- tagged struct wrapping ARM instructions and metadata
- not revision-specific (current: ARMv4+ARMv5 -or- ARMv6 -or- ARMv7)
- usable with at least arm_nandwrite() and generic CFI drivers
- ETM
- don't show FIFOFULL registers if they're not supported
- use comparators to get more breakpoints and watchpoints
- add "etm drivers" command
- trace driver init() via examine() paths only, not setup()/reset
- MC1322x support (JW/DE?)
- integrate and test support from JW (and DE?)
- get working with a known good interface (i.e. not today's jlink)
- AT91SAM92xx:
- improvements for unknown-board-atmel-at91sam9260.cfg (RD)
- STR9x: (ZW)
- improvements to str912.cfg to be more general purpose
- AVR: (SQ)
- independently verify implementation
- incrementally improve working prototype in trunk. (SQ)
- work out how to debug this target
- AVR debugging protocol.
- FPGA:
- Altera Nios Soft-CPU support
- Coldfire (suggested by NC)
- can we draw from the BDM project? @par
http://bdm.sourceforge.net/
or the OSBDM package @par
http://forums.freescale.com/freescale/board/message?board.id=OSBDM08&thread.id=422
@section thelistsvf SVF/XSVF
- develop SVF unit tests
- develop XSVF unit tests
@section thelistflash Flash Support
- finish documentation for the following flash drivers:
- avr
- pic32mx
- ocl
- str9xpec
- Don't expect writing all-ones to be a safe way to write without
changing bit values. Minimally it loses on flash modules with
internal ECC, where it may change the ECC.
- NOR flash_write_unlock() does that between sectors
- there may be other cases too
- Make sure all commands accept either a bank name or a bank number,
and be sure both identifiers show up in "flash banks" and "nand list".
Right now the user-friendly names are pretty much hidden...
@subsection thelistflashcfi CFI
- finish implementing bus width/chip width handling (suggested by NC)
- factor vendor-specific code into separate source files
- add new callback interface for vendor-specific code
- investigate/implement "thin wrapper" to use eCos CFI drivers (ØH)
@section thelistdebug Debugger Support
- add support for masks in watchpoints? @par
https://lists.berlios.de/pipermail/openocd-development/2009-October/011507.html
- breakpoints can get lost in some circumstances: @par
https://lists.berlios.de/pipermail/openocd-development/2009-June/008853.html
- add support for masks in watchpoints. The trick is that GDB does not
support a breakpoint mask in the remote protocol. One way to work around
this is to add a separate command "watchpoint_mask add/rem <addr> <mask>", that
is run to register a list of masks that the gdb_server knows to use with
a particular watchpoint address.
- integrate Keil AGDI interface to OpenOCD? (submitted by Dario Vecchio)
@section thelisttesting Testing Suite
This section includes several related groups of ideas:
- @ref thelistunittests
- @ref thelistsmoketests
- @ref thelisttestreports
- @ref thelisttestgenerichw
@subsection thelistunittests Unit Tests
- add testing skeleton to provide frameworks for adding tests
- implement server unit tests
- implement JTAG core unit tests
- implement JTAG interface unit tests
- implement flash unit tests
- implement target unit tests
@subsection thelistsmoketests Smoke Test Tools
-# extend 'make check' with a smoketest app
- checks for OOCD_TEST_CONFIG, etc. in environment (or config file)
- if properly set, runs the smoke test with specified parameters
- openocd -f ${OOCD_TEST_CONFIG}
- implies a modular test suite (see below)
- should be able to run some minimal tests with dummy interface:
- compare results of baseline sanity checks with expected results
-# builds a more complete test suite:
- existing testing/examples/ look like a great start
- all targets should be tested fully and for all capabilities
- we do NOT want a "lowest common denominator" test suite
- ... but can we start with one to get going?
- probably requires one test configuration file per board/target
- modularization can occur here, just like with targets/boards/chips
- coverage can increase over time, building up bundles of tests
-# add new 'smoketest' Makefile target:
- calls 'make check' (and the smoketest app)
- gather inputs and output into a report file
@subsection thelisttestreports Test Feedback Tools
These ideas were first introduced here: @par
https://lists.berlios.de/pipermail/openocd-development/2009-May/006358.html
- provide report submission scripts for e-mail and web forms
- add new Makefile targets to post the report:
- 'checkreportsend' -- send to list via e-mail (via sendmail)
- 'checkreportpost' -- send web form (via curl or other script)
@subsection thelisttestgenerichw Generic Hardware Tester
- implement VHDL to use for FPGA-based JTAG TAP testing device
- develop test suite that utilizes this testing device
@section thelistautotools Autotools Build System
- make entire configure process require less user consideration:
- automatically detect the features that are available, unless
options were specifically provided to configure
- provide a report of the drivers that will be build at the end of
running configure, so the users can verify which drivers will be
built during 'make' (and their options) .
- eliminate sources of confusion in @c bootstrap script:
-# Make @c bootstrap call 'configure --enable-maintainer-mode \<opts\>'?
-# Add @c buildstrap script to assist with bootstrap and configure steps.
- automatically build tool-chains required for cross-compiling
- produce mingw32, arm-elf, others using in-tree scripts
- build all required target code from sources
- make JTAG and USB debug output a run-time configuration option
@section thelistarchitecture Architectural Tasks
The following architectural tasks need to be accomplished and should be
fairly easy to complete:
- use dynamic allocations for working memory. Scan & fix code
for excessive stack allocations. take linux/scripts/checkstack.pl and
see what the worst offenders are. Dynamic stack allocations are found
at the bottom of the list below. Example, on amd64:
$ objdump -d | checkstack.pl | head -10
0x004311e3 image_open [openocd]: 13464
0x00431301 image_open [openocd]: 13464
0x004237a4 target_array2mem [openocd]: 4376
0x0042382b target_array2mem [openocd]: 4376
0x00423e74 target_mem2array [openocd]: 4360
0x00423ef9 target_mem2array [openocd]: 4360
0x00404aed handle_svf_command [openocd]: 2248
0x00404b7e handle_svf_command [openocd]: 2248
0x00413581 handle_flash_fill_command [openocd]: 2200
0x004135fa handle_flash_fill_command [openocd]: 2200
- clean-up code to match style guides
- factor code to eliminate duplicated functionality
- rewrite code that uses casts to access 16-bit and larger types
from unaligned memory addresses
- libopenocd support: @par
https://lists.berlios.de/pipermail/openocd-development/2009-May/006405.html
- review and clean up interface/target/flash APIs
The following strategic tasks will require ambition, knowledge, and time
to complete:
- overhaul use of types to improve 32/64-bit portability
- types for both host and target word sizes?
- can we use GDB's CORE_TYPE support?
- Allow N:M:P mapping of servers, targets, and interfaces
- loadable module support for interface/target/flash drivers and commands
- support both static and dynamic modules.
- should probably use libltdl for dynamic library handing.
@section thelistadmin Documentation Tasks
- Develop milestone and release guidelines, processes, and scripts.
- Develop "style" guidelines (and scripts) for maintainers:
- reviewing patches
- committing to git
- Review Users' Guide for documentation errors or omissions
- "capture" and "ocd_find" commands
- Update Developer's Manual (doxygen output)
- Add documentation describing the architecture of each module
- Provide more Technical Primers to bootstrap contributor knowledge
*/
/** @file
This file contains the @ref thelist page.
*/