tinyriscv-openocd/doc/manual/server.txt

317 lines
13 KiB
Plaintext

/** @page serverdocs OpenOCD Server APIs
OpenOCD provides support for implementing different types of servers.
Presently, the following servers have APIs that can be used.
- @subpage servergdb
- @subpage servertelnet
- @subpage serverhttp
@section serverdocsoverview Overview
What follows is a development history, and describes some of the intent
of why certain features exist within OpenOCD along with the reasoning
behind them.
This roadmap section was written May 2009 - about 9 to 12 months
after some of this work had started, it attempts to document some of
the reasons why certain features exist within OpenOCD at that time.
@section serverdocsbg Background
In early 2008, Oyvind Harboe and Duane Ellis had talked about how to
create a reasonable GUI for OpenOCD - something that is non-invasive,
simple to use and maintain, and does not tie OpenOCD to many other
packages. It would be wrong to "spider web" requirements into other
external external packages. That makes it difficult for developers to
write new code and creates a support nightmare.
In many ways, people had talked about the need for some type of
high-level interface to OpenOCD, because they only had two choices:
- the ability to script: via an external program the actions of OpenOCD.
- the ablity to write a complex internal commands: native 'commands'
inside of OpenOCD was complicated.
Fundamentally, the basic problem with both of those would be solved
with a script language:
-# <b>Internal</b>: simple, small, and self-contained.
-# <b>Cross Language</b>: script friendly front-end
-# <b>Cross Host</b>: GUI Host interface
-# <b>Cross Debugger</b>: GUI-like interface
What follows hopefully shows how the plans to solve these problems
materialized and help to explain the grand roadmap plan.
@subsection serverdocsjim Why JimTCL? The Internal Script Language
At the time, the existing "command context schema" was proving itself
insufficient. However, the problem was also considered from another
direction: should OpenOCD be first class and the script second class?
Which one rules?
In the end, OpenOCD won, the conclusion was that simpler will be better.
Let the script language be "good enough"; it would not need numerous
features. Imagine debugging an embedded Perl module while debugging
OpenOCD. Yuck. OpenOCD already has a complex enough build system, why
make it worse?
The goal was to add a simple language that would be moderately easy to
work with and be self-contained. JimTCL is a single C and single H
file, allowing OpenOCD to avoid the spider web of dependent packages.
@section serverdocstcl TCL Server Port
The TCL Server port was added in mid-2008. With embedded TCL, we can
write scripts internally to help things, or we can write "C" code that
interfaces well with TCL.
From there, the developers wanted to create an external front-end that
would be @a very usable and that that @a any language could utilize,
allowing simple front-ends to be (a) cross-platform (b) languag
agnostic, and (c) easy to develop and use.
Simple ASCII protocols are easy. For example, HTTP, FTP (control), and
SMTP are all text-based. All of these examples are widely and
well-known, and they do not require high-speed or high-volume. They
also support a high degree of interoperability with multiple systems.
They are not human-centric protocols; more correctly, they are rigid,
terse, simple ASCII protocols that are emensely parsable by a script.
Thus, the TCL server -- a 'machine' type socket interface -- was added
with the hope was it would output simple "name-value" pair type
data. At the time, simple name/value pairs seemed reasonably easier to
do at the time, though Maybe it should output JSON;
See here:
http://www.mail-archive.com/openocd-development%40lists.berlios.de/msg00248.html
The hope was that one could write a script in what ever language you want
and do things with it!
@section serverdocsgui GUI Like Interfaces
A lot has been said about various "widigit-foo-gui-library is so
wonderful". Please refer back to the domino and spider web problem of
dependencies. Sure, you may well know the WhatEver-GUI library, but
most others will not (including the next contributer to OpenOCD).
How do we solve that problem?
For example, Cygwin can be painful, Cygwin GUI packages want X11
to be present, crossing the barrier between MinGW and Cygwin is
painful, let alone getting the GUI front end to work on MacOS, and
Linux, yuck yuck yuck. Painful. very very painful.
What works easier and is less work is what is already present in every
platform? The answer: A web browser. In other words, OpenOCD could
serve out embedded web pages via "localhost" to your browser.
Long before OpenOCD had a TCL command line, Zylin AS built their ZY1000
devince with a built-in HTTP server. Later, they were willing to both
contribute and integrate most of that work into the main tree.
@subsection serverdocsother Other Options Concidered
What if a web browser is not acceptable ie: You want to write your own
front gadget in Eclipse, or KDevelop, or PerlTK, Ruby, or what ever
the latest and greatest Script De Jour is.
- Option 1: Can we transport this extra data through the GDB server
protocol? In other words, can we extend the GDB server protocol?
No, Eclipse wants to talk to GDB directly and control the GDB port.
- Option 2: SWIG front end (libopenocd): Would that work?
That's painful - unless you design your api to be very simplistic -
every language has it's own set of wack-ness, parameter marshaling is
painful.
What about "callbacks" and structures, and other mess. Imagine
debugging that system. When JimTCL was introduced Spencer Oliver had
quite a few well-put concerns (Summer 2008) about the idea of "TCL"
taking over OpenOCD. His concern is and was: how do you debug
something written in 2 different languages? A "SWIG" front-end is
unlikely to help that situation.
@subsection serverdoccombined Combined: Socket & WebServer Benifits
Seriously think about this question: What script language (or compiled
language) today cannot talk directly to a socket? Every thing in the
OpenOCD world can work a socket interface. Any host side tool can talk
to Localhost or remote host, however one might want to make it work.
A socket interface is very simple. One could write a Java application
and serve it out via the embedded web server, could it - or something
like it talk to the built in TCL server? Yes, absolutely! We are on to
something here.
@subsection serverdocplatforms Platform Permuntations
Look at some permutations where OpenOCD can run; these "just work" if
the Socket Approach is used.
- Linux/Cygwin/MinGw/MacOSx/FreeBSD development Host Locally
- OpenOCD with some dongle on that host
- Linux/Cygwin/MingW/MacOS/FreeBSD development host
- DONGLE: tcpip based ARM-Linux perhaps at91rm9200 or ep93xx.c, running openocd.
- Windows cygwin/X desktop environment.
- Linux development host (via remote X11)
- Dongle: "eb93xx.c" based linux board
@subsection serverdocfuture Development Scale Out
During 2008, Duane Ellis created some TCL scripts to display peripheral
register contents. For example, look at the sam7 TCL scripts, and the
stm32 TCL scripts. The hope was others would create more.
A good example of this is display/view the peripheral registers on
your embedded target. Lots of commercial embedded debug tools have
this, some can show the TIMER registers, the interrupt controller.
What if the chip companies behind STM32, or PIC32, AT91SAM chips -
wanted to write something that makes working with their chip better,
easier, faster, etc.
@a Question: How can we (the OpenOCD group) make that really fancy
stuff across multiple different host platforms?
Remember: OpenOCD runs on:
-# Linux via USB,
-# ARM Linux - bit-banging GPIO pins
-# MacOSX
-# FreeBSD
-# Cygwin
-# MinGW32
-# Ecos
How can we get that to work?
@subsection serverdocdebug What about Debugger Plugins?
Really GDB is nice, it works, but it is not a good embedded debug tool.
OpenOCD cannot work in a GUI when one cannot get to its command line.
Some GDB front-end developers have pedantic designs that refuse any and
all access to the GDB command line (e.g. http://www.kdbg.org/todo.php).
The TELNET interface to OpenOCD works, but the intent of that interface
is <b>human interaction</b>. It must remain available, developers depend
upon it, sometimes that is the only scheme available.
As a small group of developers, supporting all the platforms and
targets in the debugger will be difficult, as there are enough problem
with the plethora of Dongles, Chips, and different target boards.
Yes, the TCL interface might be suitable, but it has not received much
love or attention. Perhaps it will after you read and understand this.
One reason might be, this adds one more host side requirement to make
use of the feature. In other words, one could write a Python/TK
front-end, but it is only useable if you have Python/TK installed.
Maybe this can be done via Ecllipse, but not all developers use Ecplise.
Many devlopers use Emacs (possibly with GUD mode) or vim and will not
accept such an interface. The next developer reading this might be
using Insight (GDB-TK) - and somebody else - DDD..
There is no common host-side GDB front-end method.
@section serverdocschallenge Front-End Scaling
Maybe we are wrong - ie: OpenOCD + some TK tool
Remember: OpenOCD is often (maybe 99.9%) of the time used with
GDB-REMOTE. There is always some front-end package - be it command-line
GDB under DDD, Eclipse, KDevelop, Emacs, or some other package
(e.g. IAR tools can talk to GDB servers). How can the OpenOCD
developers make that fancy target display GUI visible under 5 to 10
different host-side GDB..
Sure - a <em>man on a mission</em> can make that work. The GUI might be
libopenocd + Perl/TK, or maybe an Eclipse Plug-in.
That is a development support nightmare for reasons described
above. We have enough support problems as it is with targets, dongles,
etc.
@section serverdocshttpbg HTTP Server Background
OpenOCD includes an HTTP server because most development environments
are likely contain a web browser. The web browser can talk to OpenOCD's
HTTP server and provide a high-level interfaces to the program.
Altogether, it provides a universally accessible GUI for OpenOCD.
@section serverdocshtml Simple HTML Pages
There is (or could be) a simple "Jim TCL" function to read a memory
location. If that can be tied into a TCL script that can modify the
HTTP text, then we have a simple script-based web server with a JTAG
engine under the hood.
Imagine a web page - served from a small board with two buttons:
"LED_ON" and "LED_OFF", each click - turns the LED on or OFF, a very
simplistic idea. Little boards with web servers are great examples of
this: Ethernut is a good example and Contiki (not a board, an embedded
OS) is another example.
One could create a simple: <b>Click here to display memory</b> or maybe
<b>click here to display the UART REGISTER BLOCK</b>; click again and see
each register explained in exquisit detail.
For an STM32, one could create a simple HTML page, with simple
substitution text that the simple web server use to substitute the
HTML text JIMTCL_PEEK32( 0x12345678 ) with the value read from
memory. We end up with an HTML page that could list the contents of
every peripheral register on the target platform.
That also is transportable, regardless of the OpenOCD host
platform: Linux/X86, Linux/ARM, FreeBSD, Cygwin, MingW, or MacOSX.
You could even port OpenOCD to an Google Android and use it as a
bit-bang dongle JTAG serving web pages.
@subsection serverdocshtmladv Advanced HTML Pages
Java or JavaScript could be used to talk back to the TCL port. One
could write a Java, AJAX, FLASH, or some other developer friendly
toolbox and get a real cross-platform GUI interface. Sure, the interface
is not native - but it is 100% cross-platform!
OpenOCD current uses simple HTML pages; others might be an Adobe FLASH
expert, or a Java Expert. These possibilities could allow the pages
remain cross-platform but still provide a rich user-interface
experience.
Don't forget it can also be very simple, exactly what one developer
can contribute, a set of very simple web pages.
@subsection serverdocshtmlstatus HTTP/HTML Status
As of May 2009, much of the HTML pages were contributed by Zylin AS,
hence they continue to retain some resemblance to the ZY1000 interface.
Patches would be welcome to move these parts of the system forward.
*/
/** @page servergdb OpenOCD GDB Server API
This section needs to be expanded.
*/
/** @page servertelnet OpenOCD Telnet Server API
This section needs to be expanded.
*/
/** @page serverhttp OpenOCD HTTP Server API
This section needs to be expanded.
*/