Change-Id: I1453f4f3dc0add529da20577e38b8b82d7d00366
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
Reviewed-on: http://openocd.zylin.com/18
Reviewed-by: Alex Austin <alex.austin@spectrumdsi.com>
Tested-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Let the target algorithm be running in the background and buffer data
continuously through a FIFO. This reduces or removes the effect of latency
because only a very small number of queue executions needs to be done per
buffer fill. Previously, the many repeated target state changes, register
accesses (really inefficient) and algorithm uploads caused the flash
programming to be latency bound in many cases. Now it should scale better
with increased throughput.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Access the different flash banks' registers using a bank specific register
base and a register specific offset. This is equivalent but feels more
natural.
Some accesses were discovered that maybe should not be hard coded to bank0
registers. Add a note about that.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Split armv7m_run_algorithm into two pieces and use them to reimplement it.
The arch_info parameter is used to keep context between the two calls, so
both calls must refer to the same armv7m_algorithm struct. Ugly but works
for a proof-of-concept.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
On supported targets, this may be used to start a long running algorithm in
the background so the target may be interacted with during execution and
later wait for its completion.
The most obvious use case is a double buffered flash algorithm that can
upload the next block of data while the algorithm is flashing the current.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Target events are added to get better gdb support. The run
algorithm functionality are implemented to support feature
fast flash write functionality. The new r/w buffer api is now
used to support the special memory address handling. The output
of the md command was fixed.
This is a proof of concept to get access to the debug port of a
secured kinetis cpu. On full flash erase the cpu is automatically
secured and the debug port is not accessible.
To get this to work the srst line is needed and the necessary
configuration should be added to the configuration file.
The driver sends ascii encoded bitbang commands over unix sockets or TCP to
another process. This driver is useful for debugging software running on
processors which are being simulated.
added an attempt to use the non-reseting halting sequence. if it fails, then the full sequence will be attempted. this makes things a bit faster most of the time.
changed the location of a function, avoiding a forward def
added an alternative way to enter debug mode, which does not require restarting the chip.
this will not always work, but in general it will (failure 0.3%), and failure is not a dramatic issue, simply have to use the full sequence.
the user can only access "halt", which uses the full sequence, so the user should not have any problems.
restarting the chip requires reconfiguring the flash module. the doc is very poor, so i'd rather have the two methods, and live with the 0.3%.
sometimes the master tap will be enabled, since tap switching is required during halt/lock/unlocking procedures.
now irscan handles this, avoiding unnecessary warnings and preventing errors.
got new info regarding setting the chip to debug mode, and locking/unlocking flash memory.
the newer implementation is a bit slower, but always works.
the previous implementation would randomly (as once every 25k-70k times) get the chip into a state where the freescale tool would be necessary. this is fixed now.
added functions to play around with the jtag state machine. they are not the happiest, but are necessary to be able to execute the halting/locking/unlocking sequences.
Conflicts:
src/target/dsp5680xx.c
the user can execute halt, but no enter_debug_mode. modified the error handling to suite this.
the new implementation of unlocking will use enter_debug_mode, and should not get the same errors as the user would, because not being able to enter debug mode is actually success when checking for locked flash.
crc check was always performed on newly flashed data, now it is optional
flash mem can be locked by writing a specific word to a specific address in flash.
to verify flash, target must be halted, and this will (when the new halt sequence is implemented) require reseting the chip. if the target is reset after writing the lock words, then it will lock, hence the CRC will fail because it is not possible to read stuff from the target.
also added a function that resets the jtag state machine.
this is not used yet, but will be soon.
it is implemented to allow strict control over JTAG state machine, necessary to implement to halt and unlocking sequences.
Posted by telekatz@gmx.de in the bettyhacks forum for openocd
0.4.0 and integrated into 0.5.0 by Gunnar Henne.
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
Due to build warnings introduced in newer versions of ftd2xx we
use strings to report errors rather than result codes. This also
gives us the same behaviour as libftdi.
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
Scripts sourced via 'script' should evaluate in the global
scope to make it easy to set and reference global variables.
Signed-off-by: Steve Bennett <steveb@workware.net.au>
i had started my code from dsp5683xx, i renamed a bunch of stuff to names i consider to be better.
i believe no one is using this code, so nobody should be affected. (it's not too late to do this change)
After correcting the reply size counter, it should be safe to rely on it
for the number of bytes expected in the USB read, instead of reading the
endpoint maximum. This doesn't make things go any faster but it's nicer and
removes the local buffer.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Add a helper function for running the queue if it would overflow otherwise.
Use it to simplify the queue fill level checks and optimize in a few cases
that would previously run the queue prematurely.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
After the reply_index handling is fixed, there's no need to special case
the out scan.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
dtc_queue.reply_index was wrongly being increased during out scans, causing
the queue to be sent before the out buffer was full. This patch increases
raw upload speed by 50% or so.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Indentation was inconsistent and some lines not indented at all. Quickfix
using Eclipse's auto-indentation.
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Function mips_m4k_write_memory() does endianess byte swap,
but this procedure break one byte access (temporary array
overwrites content in buffer).
As a fix, this endianess swap and buffer affecting
is preformed only on hword and word accesses (not on byte access).
Pprogram that loads another program into memory is actually writing the
D- side cache.
The instructions it has loaded can't be executed until they reach the
I-cache.
After the instructions have been written, the loader should arrange to
write back any containing D-cache line and invalidate any locations
already in the I-cache.
For the MIPS Architecture Release2 cores, we can use synci command
that does this job.
For Release1 we must use "cache" instruction.
This patch adds MIPS32 CP0 coprocessor R/W routines,
as well as adequate commands to use these routines via
telnet interface.
Now is becomes possible to affect CP0 internal registers
and configure CPU directly from OpenOCD.
For all architectures we use distinct common magic number,
and this should be a uint32_t type.
Otherwise, comparison with macros will yield compilation
warning.
if tap enable/disable failed then a warning was written to the log, but JIM_OK was returned. if using openocd via a TCP interface to the TCL port, there is no way to catch that the command failed (it didn't enable the tap, so it failed)
now it return an error if it fails.
There are some older atmel nor chips which have negated logic for
TopBottom detection. This patch adds a special handling for the old
chips. This is the same mechanism as implemented in linux kernel.
Signed-off-by: Andreas Bießmann <andreas.devel@gmail.com>
before doing anything with the flash module (FM) the clock divider must be set.
if erase_check was the first thing done with the FM after reset then an error would be generated because the clk divider was not set.
now erase_check sets the clk divider.
reorganized code to get rid of compiler warnings
the warning were related to allignment, i do not get these warning on my build system (i've tried setting the compiler flag but it doesn't work, still working on why) so i cannot detect them (yet.)
This corrects two issues found with openocd.
d7f71e7fe9 removed some code that was
being used.
The above then caused even more code to get removed by commit 1cfb2287a6.
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
Issue warning when the old cmd is used and redirect to new supported one.
These deprecated cmds will be removed at some point.
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
Use consistent names for the stm32 family flash drivers, eg.
stm32x -> stm32f1x
stm32f2xxx -> stm32f2x
this makes it easier to add support for newer stm32 families.
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
When building official releases from tarball, git commit info is not
available in the building environment. Thus, automake should not try to
append the git commit to the version string.
Signed-off-by: Luca Bruno <lucab@debian.org>
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
The default is -Werror, so warnings become errors
Signed-off-by: Steve Bennett <steveb@workware.net.au>
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
This patch extends the cortex_m3 maskisr command by a new option 'auto'.
The 'auto' option handles interrupts during stepping in a way they are
processed but don't disturb the program flow during debugging.
Before one had to choose to either enable or disable interrupts. The former
steps into interrupt handlers when they trigger. This disturbs the flow during
debugging, making it hard to follow some piece of code when interrupts occur
often.
When interrupts are disabled, the flow isn't disturbed but code relying on
interrupt handlers to be processed will stop working. For example a delay
function counting the number of timer interrupts will never complete, RTOS
task switching will not occur and output I/O queues of interrupt driven
I/O will stall or overflow.
Using the 'maskisr' command also typically requires gdb hooks to be supplied
by the user to switch interrupts off during the step and to enable them again
afterward.
The new 'auto' option of the 'maskisr' command solves the above problems. When
set, the step command allows pending interrupt handlers to be executed before
the step, then the step is taken with interrupts disabled and finally interrupts
are enabled again. This way interrupt processing stays in the background without
disturbing the flow of debugging. No gdb hooks are required. The 'auto'
option is the default, since it's believed that handling interrupts in this
way is suitable for most users.
The principle used for interrupt handling could probably be used for other
targets too.
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
In order to compare data read from the target with some marcros or data
defined on the host, we must transform this read data from target
endianess to host endianess.
target_read_memory() gets bytes from target to the host, but keeps them in _target_
endianess. This is OK if we just want to temporary keep this data on the
host, like keeping breakpoint->orig_instr. But if we want to use this
data for any ispections and comparisons on the host side, we must
transform it to _host_ endianess, by using target_buffer_get_u32()
function.
Currently this transformation is missing, and check current_instr ==
MIPS32_SDBBP will never pass if target and host endianess differ,
because current_instr will be kept in _target_ endianess and
MIPS32_SDBBP will be kept in _host_ endianess,
The patch fix this issue by using target_buffer_get_u32() to transform current_instr to
_host_ endianess before comparison.
Use "git revert <commit>" to revert this commit, then build and
repair and post patch to the mailing list.
Warnings generated with:
nios2-elf-gcc (GCC) 3.4.6 (Altera Nios II 9.1 b222)
openocd/src/target/dsp5680xx.c: In function 'eonce_rx_upper_data':
openocd/src/target/dsp5680xx.c:252: warning: cast increases required
alignment of target type
openocd/src/target/dsp5680xx.c: In function 'eonce_rx_lower_data':
openocd/src/target/dsp5680xx.c:268: warning: cast increases required
alignment of target type
openocd/src/target/dsp5680xx.c: In function 'eonce_pc_store':
openocd/src/target/dsp5680xx.c:508: warning: dereferencing type-punned
pointer will break strict-aliasing rules
openocd/src/target/dsp5680xx.c: In function 'dsp5680xx_read':
openocd/src/target/dsp5680xx.c:736: warning: cast increases required
alignment of target type
openocd/src/target/dsp5680xx.c:737: warning: cast increases required
alignment of target type
openocd/src/target/dsp5680xx.c: In function 'dsp5680xx_write_8':
openocd/src/target/dsp5680xx.c:823: warning: cast increases required
alignment of target type
openocd/src/target/dsp5680xx.c: In function 'dsp5680xx_write':
openocd/src/target/dsp5680xx.c:938: warning: cast increases required
alignment of target type
openocd/src/target/dsp5680xx.c:941: warning: cast increases required
alignment of target type
openocd/src/target/dsp5680xx.c: In function 'dsp5680xx_f_wr':
openocd/src/target/dsp5680xx.c:1355: warning: cast increases required
alignment of target type
This patch extends the cortex_m3 maskisr command by a new option 'auto'.
The 'auto' option handles interrupts during stepping in a way they are
processed but don't disturb the program flow during debugging.
Before one had to choose to either enable or disable interrupts. The former
steps into interrupt handlers when they trigger. This disturbs the flow during
debugging, making it hard to follow some piece of code when interrupts occur
often.
When interrupts are disabled, the flow isn't disturbed but code relying on
interrupt handlers to be processed will stop working. For example a delay
function counting the number of timer interrupts will never complete, RTOS
task switching will not occur and output I/O queues of interrupt driven
I/O will stall or overflow.
Using the 'maskisr' command also typically requires gdb hooks to be supplied
by the user to switch interrupts off during the step and to enable them again
afterward.
The new 'auto' option of the 'maskisr' command solves the above problems. When
set, the step command allows pending interrupt handlers to be executed before
the step, then the step is taken with interrupts disabled and finally interrupts
are enabled again. This way interrupt processing stays in the background without
disturbing the flow of debugging. No gdb hooks are required. The 'auto'
option is the default, since it's believed that handling interrupts in this
way is suitable for most users.
The principle used for interrupt handling could probably be used for other
targets too.
Signed-off-by: Spencer Oliver <ntfreak@users.sourceforge.net>
A new variable "nobase_dist_pkglib_DATA" is introduced to install
the OpenULINK firmware image to $PREFIX/lib/openocd/OpenULINK/ulink_firmware.hex
Also, the variable "EXTRA_DIST" is used to include the OpenULINK firmware source
in the OpenOCD source distribution.
So far image_load command tries to load ELF binaries to address
discovered by reading p_paddr member of a Program header of an ELF
segment.
However, ELF specifications says for p_paddr : ...Because System V
ignores physical addressing for application programs, this member has
unspecified contents for executable files and shared objects.
ARM ELF specifiaction goes even further, demanding that this member
be set to zero, using the p_vaddr as a segment load address.
To avoid the cases to wrong addr where p_paddr is zero,
we are now using p_vaddr to as a load destination in case that *all*
p_paddr == 0. Basically, this patch re-implements the approach present in
BDF's elf.c, which is used by GDB also (so that we can be consistent).
Spencer Oliver
Spencer Oliver
Michel Jaouen
Andreas Fritiofson
Simon Barner
Clément Burin des Roziers
Steve Bennett
Mathias K
Luca Bruno
Richard Uhler
Martin Schmölzer
Rodrigo L. Rosa
Øyvind Harboe
simonqian.openocd
Heythem Bouhaja
Evan Hunter
Jonathan Dumaresq
Jie Zhang
Evan
Andreas Bießmann
olivier Schonken
Gunnar Henne
SimonQian
Stefan Mahr
Drasko DRASKOVIC
Jie Zhang
Ronny Strutz
Peter Horn
Tomek CEDRO
Freddie Chopin
Damjan Marion
Alan Bowman