When trace capturing the trace is enabled using 'tpiu_config internal'
(via the internal mode), OpenOCD can collect the trace buffers then append
it to a specified file or named pipe and propagate the trace to 'tcl_trace'
command.
This change is allowing OpenOCD to stream the captured trace over TCP.
When using this configuration OpenOCD acts like a server and multiple
clients can connect and receive the captured trace.
Example on STM32F7 running at 216MHz:
itm port 0 on
tpiu config internal :3344 uart off 216000000
Change-Id: Idea43e7e26e87b98a33da7fb9acf7ea50fe3b345
Signed-off-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
Reviewed-on: http://openocd.zylin.com/5345
Tested-by: jenkins
Reviewed-by: Karl Palsson <karlp@tweak.net.au>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
returning the created service seems useful:
as the only method to get the freshly created service is by getting the
last item in the services linked list, and this seems to be like an
intrusion to service internal mechanism.
possibly, we could get the service from a connection but this is possible
only from [new_connection|input|connection_closed]_handler_t, but this is
not always practical:
example: armv7m: add a TCP channel to stream captured trace
http://openocd.zylin.com/#/c/5345/
here we poll for trace and broadcast to all connections
outside of these xxx_handler_t functions
also, storing one of the connections in new_connection_handler_t and get
the service from it is possible, but this will make the code less readable.
Change-Id: I5fef1baecec1e054953c6faf5b99d864ecc97f02
Signed-off-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
Reviewed-on: http://openocd.zylin.com/5717
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Tested-by: jenkins
Functions and variables that are not used outside the file should
be declared as static.
Change-Id: I9f97571a528f0cb3c3c26f873577ab16fdec3cdc
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5895
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Is it possible to run OpenOCD without any target, for example to
only dump the rom-tables of an arm dap, or to perform low level
jtag operations.
But without any target created, the command 'target current'
causes OpenOCD to abruptly exit.
Handle in command 'target current' the case of no targets.
Change-Id: Ide15cb13bec84b88ccc3e7126523c04a6d70e636
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5881
Tested-by: jenkins
The default way of working is to have a single GDB attached to one
target, so OpenOCD accepts only one connection to the GDB port of
each targets and rejects any further connection.
There are some barely safe use cases in which it could get useful
having a second GDB connection to the same target.
One such use case is while using GDB as a 'non-intrusive memory
inspector', as explained in the OpenOCD documentation.
One GDB can be left running an infinite loop to dump some memory
area, or even analysing the content, while keeping a second GDB
ready for user interaction or spot memory check.
Add a target configure option to specify the maximum number of GDB
connections allowed for that target, keeping the default to 1.
Change-Id: I4985a602e61588df0b527d2f2aa5b955c93e125e
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5865
Tested-by: jenkins
Reviewed-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
To avoid code duplication, reorganize the code to replace
cti_configure() with adiv5_jim_mem_ap_spot_configure().
Reorganize 'struct arm_cti_object' and its sub-'struct arm_cti'
moving DAP and mem-AP info in a 'struct adiv5_mem_ap_spot'.
Replace cti_configure() with adiv5_jim_mem_ap_spot_configure().
Deprecate the use of '-ctibase' in favor of '-baseaddr'.
Change-Id: I43740a37c80de67c0f5e4dc79c3400b91a12e9e8
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5869
Tested-by: jenkins
This is somehow an extension of existing adiv5_jim_configure(),
but includes the 'address' in the mem_ap.
Rewrite adiv5_jim_configure() using the new helper.
Change-Id: Ia7effeeece044004d459b45126ed4961a98b8568
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5857
Tested-by: jenkins
Now that it’s possible to start profiling from either a running or a
halted state, rather than unconditionally halting after profiling
finishes, it makes more sense to restore the processor to whatever state
(running or halted) it was in before profiling started.
Change-Id: If6f6e70a1a365c1ce3b348a306c435c220b8bf12
Signed-off-by: Christopher Head <chead@zaber.com>
Reviewed-on: http://openocd.zylin.com/5237
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
The Cortex-M implementation of profiling contains a bunch of
conditionals and checks to handle both chips which have PCSR and chips
which do not. However, the net effect of the non-PCSR branches is
actually exactly the same as what target_profiling_default does. Rather
than duplicating this code, just detect the situation where PCSR isn’t
available and delegate to target_profiling_default.
Change-Id: I1be57ac77f983816ab6bf644a3cfca77b67d6f70
Signed-off-by: Christopher Head <chead@zaber.com>
Reviewed-on: http://openocd.zylin.com/5236
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
There are a handful of implementations of profiling. There is the
default implementation, which repeatedly halts and resumes the target,
sampling PC each time. There is the Cortex-M implementation, which
uses PCSR if available, otherwise falling back to halting and resuming
and sampling PC. There is the OR1K implementation, which reads NPC
repeatedly. Finally, there is the NDS32 implementation which uses some
kind of AICE commands with which I am unfamiliar.
None of these (with the possible exception of the NDS32
implementation) actually require the target to be halted when starting
profiling. The Cortex-M and OR1K actually resume the target as pretty
much their first action. The default implementation doesn’t do this,
but is written in such a way that the target just flips back and forth
between halted and running, and the code will do the right thing from
either initial state. The NDS32 implementation I don’t know about.
As such, for everything except NDS32, it is not really necessary that
the target be halted to start profiling. For the non-PCSR Cortex-M and
default implementations, there is no real harm in such a requirement,
because profiling is intrusive anyway, but there is no benefit. For
the PCSR-based Cortex-M and the OR1K, requiring that the target is
halted is annoying because it makes profiling more intrusive.
Remove the must-be-halted check from the target_profiling function.
Add it to the NDS32 implementation because I am not sure if that will
break when invoked with a running target. Do not add it to any of the
other implementations because they don’t need it.
Change-Id: I479dce999a80eccccfd3be4fa192c904f0a45709
Signed-off-by: Christopher Head <chead@zaber.com>
Reviewed-on: http://openocd.zylin.com/5235
Tested-by: jenkins
Reviewed-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
With help of actionpoint mechanism now it is possible to introduce
watchpoints support for ARC.
Change-Id: I5887335d0ba38c15c377bc1d24a1ef36e138cf65
Signed-off-by: Evgeniy Didin <didin@synopsys.com>
Reviewed-on: http://openocd.zylin.com/5867
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
The commit b68674a1da ("Upstream tons of RISC-V changes.") was
proposed well before commit 3ac010bb9f ("Fix debug prints when
loading to flash"), but the merge got in different order.
After latest merge, the master branch fails to compile.
Fix the compile error.
Change-Id: Ia3bd21d970d589343a3b9b2d58c89e0c49f30015
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5856
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Tested-by: jenkins
Reviewed-by: Jan Matyas <matyas@codasip.com>
These are all the changes from https://github.com/riscv/riscv-openocd
(approximately 91dc0c0c) made just to src/target/riscv/*. Some of the
new code is disabled because it requires some other target-independent
changes which I didn't want to include here.
Built like this, OpenOCD passes:
* All single-RV32 tests against spike.
* All single-RV64 tests against spike.
* Enough HiFive1 tests. (I suspect the failures are due to the test
suite rotting.)
* Many dual-RV32 (-rtos hwthread) against spike.
* Many dual-RV64 (-rtos hwthread) against spike.
I suspect this is an overall improvement compared to what's in mainline
right now, and it gets me a lot closer to getting all the riscv-openocd
work upstreamed.
Change-Id: Ide2f80c9397400780ff6780d78a206bc6a6e2f98
Signed-off-by: Tim Newsome <tim@sifive.com>
Reviewed-on: http://openocd.zylin.com/5821
Tested-by: jenkins
Reviewed-by: Jan Matyas <matyas@codasip.com>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-by: Karl Palsson <karlp@tweak.net.au>
The function adiv5_jim_configure() casts the void pointer
'target->private_config' to a struct adiv5_private_config pointer.
This is tricky in case of aarch64, where the private data are in a
struct aarch64_private_config that has as first element the struct
adiv5_private_config.
While the current solution is working fine, it's not clean and
requires special attention for any further code development.
Override 'target->private_config' to the correct pointer while
calling adiv5_jim_configure().
Change-Id: Ic2fc047dd1e57013943d96e6d5879a919d1eb7b3
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5847
Tested-by: jenkins
Reviewed-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
Change aarch64 to use ap-num setting if provided. Fall back to original
behavior of using first AP when ap-num is invalid.
Change-Id: I0d3624f75c86ba5fd5a322ac60856dbbb6e71eaf
Signed-off-by: Kevin Yang <kangyang@google.com>
Reviewed-on: http://openocd.zylin.com/5831
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
If a target is not examined, command vector_catch crashes while accessing
the debug_ap NULL pointer.
maskisr and reset_config commands don't require this check.
Change-Id: I949b6f6e8b983327dd98fbe403735141f8f0b5d6
Signed-off-by: Daniel Trnka <daniel.trnka@gmail.com>
Reviewed-on: http://openocd.zylin.com/5813
Tested-by: jenkins
Reviewed-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
The function free() can be called with a NULL pointer as argument,
no need to check the argument before. If the pointer is NULL, no
operation is performed by free().
Remove the occurrences of pattern:
if (ptr)
free(ptr);
In target/openrisc/jsp_server.c, an error is logged if the ptr was
already NULL. This cannot happen since the pointer was already
referenced few lines before and openocd would have been already
SIGSEGV in that case, so remove the log.
Change-Id: I290a32e6d4deab167676af4ddc83523c830ae49e
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5809
Tested-by: jenkins
The prompt pointer in the jtag serial port is never zero'd or allocated.
Completely remove it since there is not much use for it as the target
software will provide the actual prompt.
Change-Id: Id95d8ccb9f725e53b9d03386b11d91eba1cd6ef4
Signed-off-by: Stafford Horne <shorne@gmail.com>
Reviewed-on: http://openocd.zylin.com/4093
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
The usage string should contain only the command parameters.
OpenOCD will automatically prepend the command name to the usage
string while dumping the usage or help message.
Remove the repeated command name from the usage string.
Change-Id: If10a0f1c254aee302b9ca08958390b7f21cdb21b
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5824
Tested-by: jenkins
The usage string should contain only the command parameters.
OpenOCD will automatically prepend the command name to the usage
string while dumping the usage or help message.
Remove the repeated command name from the usage string.
Change-Id: I691094a6395acb0e4ea3bea2347ff38379002464
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5822
Tested-by: jenkins
Modify the format strings to properly handle uint32_t data types.
While there, fix prototype mismatch between header and C file of
the function armv7a_l1_d_cache_inval_virt().
Change-Id: I434bd241fa5c38e0c15d22cda2295097050067f5
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5818
Tested-by: jenkins
Commit 11019a824d ("adi_v5: enforce check on AP number value")
introduces the macro DP_APSEL_MAX and use it in place of hardcoded
magic numbers for the upper limit of AP selection value.
Use the macro also while defining the array of struct adiv5_ap in
struct adiv5_dap.
Change-Id: I88f53ceb710f92a48a8026a365709fbf2d9e6912
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5806
Tested-by: jenkins
While loading to flash with debug level at least 3,
OpenOCD tries to print the whole loaded bitstream.
This will be very-very-slow due to implementation of
conversion from buffer to string.
* fix condition on selected debug level in jtag/core.c
* replace slow buf_to_str function from helper/binarybuffer.c
with faster but_to_hex_str function
Change-Id: I3dc01d5846941ca80736f2ed12e3a54114d2b6dd
Signed-off-by: Samuel Obuch <sobuch@codasip.com>
Reviewed-on: http://openocd.zylin.com/5800
Tested-by: jenkins
Reviewed-by: Jan Matyas <matyas@codasip.com>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
There are failure cases of target_create() that are not checked.
Plus, in case of failure the memory allocated in not properly
released before returning error.
Check all the possible failure in target_create().
Change current_target only when target is successfully created.
Add the new target to all_targets list only when target is
successfully created.
Release all the allocated memory before quit on failure.
Use malloc() instead of calloc() for target->type, because the
struct will be fully populated with memcpy().
Change-Id: Ib6f91cbb50c28878e7c73dc070b17b8d7d4e902f
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5776
Tested-by: jenkins
If a target is not examined when the debugger tries to connect to it
then it can lead to undesired/undefined behavior.
In particular it leads to a zero pointer dereference on the aarch64.
Change-Id: I67f2b714ab8b2727fd36f3de16d7f9017b4c55fe
Signed-off-by: Mikhail Rasputin <mikhail.godlike.rasputin@yandex.ru>
Reviewed-on: http://openocd.zylin.com/5727
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
There is no deinit_target method, so few memory allocations leak
at openocd exit.
Issue identified by tracking all calls to arm_dpm_setup().
Implement the method arm11_dpm_deinit() to free all the memory
allocated in arm11_dpm_init() and call it in the new
arm11_deinit_target().
NOT TESTED on real HW.
Change-Id: Icab86e290fc2db14f70eb84c8286357aadb02a35
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5694
Tested-by: jenkins
By default GDB timeouts after 2 seconds, even if this value can be
modified with GDB command "set remotetimeout".
On OpenOCD side, the default event for GDB attach is to halt the
target and wait it to halt. But here the default timeout of the
halt command is 5 seconds!
If the target cannot be halted (e.g. it's kept in reset by another
core or the debugger doesn't have enough privileges) then GDB will
timeout while OpenOCD is still waiting and is unable to
communicate with GDB.
Decrease the halt timeout to 1 second in the default GDB attach
event handler.
Change-Id: I231c740816bb6a0d74b0bc679a368a6cbfb34824
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5687
Tested-by: jenkins
There is no method to free the register cache, allocated in
arm720t_init_target().
Issue identified by tracking all calls to arm7tdmi_init_target().
Implement the method arm720t_deinit_target() by calling directly
arm7tdmi_deinit_target().
NOT TESTED on a real arm720t target.
Tested on a arm926ejs (SPEAr320) by hacking the target type and
pretending it is a xscale:
sed -i s/arm926ejs/arm720t/ tcl/target/spear3xx.cfg
Change-Id: I53c1f46c1a355a710e8df01468b19220671569dc
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5697
Tested-by: jenkins
There is no method to free the register cache, allocated in
arm7tdmi_init_target(), so we get a memory leak.
Issue identified by tracking all calls to arm_build_reg_cache().
Implement the method arm7tdmi_deinit_target() that in turn calls
arm7tdmi_free_reg_cache().
NOT TESTED on a real arm7tdmi target.
Tested on a arm926ejs (SPEAr320) by hacking the target type and
pretending it is a arm7tdmi:
sed -i s/arm926ejs/arm7tdmi/ tcl/target/spear3xx.cfg
Change-Id: Iad465b708eb4ebb298725d7155fea76357e9045c
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5696
Tested-by: jenkins
There is no method to free the register cache, allocated in
xscale_build_reg_cache(), so we get a memory leak.
Issue identified by tracking all calls to arm_build_reg_cache().
Implement the method xscale_deinit_target() that in turn calls the
new xscale_free_reg_cache().
Fix leak of struct xscale.
NOT TESTED on a real xscale target.
Tested on a arm926ejs (SPEAr320) by hacking the target type and
pretending it is a xscale:
sed -i s/arm926ejs/xscale/ tcl/target/spear3xx.cfg
Change-Id: Ibb2104c42411b76f4bb77c2fa387d1b85a3d2d5d
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5695
Tested-by: jenkins
Similarly to the fix for arm926ejs (also base on arm9tdmi), fix
the other targets based on arm9tdmi.
The fix for arm926ejs is tested on SPEAr320 target.
This fix is proposed separately because is not tested on a correct
target device, but tested on SPEAr320 by hacking the target type
and pretending it is the correct one, e.g.:
sed -i s/arm926ejs/arm920t/ tcl/target/spear3xx.cfg
The memory leaks detected and fixed are:
- arm register cache;
- EmbeddedICE register cache;
- arm_jtag_reset_callback internal data;
- struct <target_type>_common.
Change-Id: I565f9a5bf144a9df78474434d86a64127ef0fbe5
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5699
Tested-by: jenkins
Actionpoint mechanism allows to setup HW breakpoints and watchpoints on Synopsys ARC CPUs.
This mechanism is controlled by DEBUG register and by a set of auxilary registers.
Each actionpoint is controlled by 3 aux registers: Actionpoint(AP) match mask(AP_AMM),
AP match value(AP_AMV) and AP control(AC).
Note: some fields of actionpoint_t structure will be used in further
support of watchpoints.
Change-Id: I4efb24675f247cc19d9122501c9e63c3126fcab4
Signed-off-by: Evgeniy Didin <didin@synopsys.com>
Reviewed-on: http://openocd.zylin.com/5763
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Issue identified by checkpatch script from Linux kernel v5.1 using
the command
find src/ -type f -exec ./tools/scripts/checkpatch.pl \
-q --types FUNCTION_ARGUMENTS -f {} \;
This patch also fixes an incorrect function prototype in zy1000.c.
ZY1000 minidriver implementation overrides the function
arm11_run_instr_data_to_core_noack_inner(), but the prototype is
not the same as in src/target/arm11_dbgtap.c and to avoid compile
error it was changed also the prototype of the called function
arm11_run_instr_data_to_core_noack_inner_default().
Change-Id: I476cda8cdb0e1e280795b3b43ca95c40d09e4a3d
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5630
Tested-by: jenkins
Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
The checkpatch script from Linux kernel v5.1 complains about using
space before comma, before semicolon and between function name and
open parenthesis.
Fix them!
Issue identified using the command
find src/ -type f -exec ./tools/scripts/checkpatch.pl \
-q --types SPACING -f {} \;
The patch only changes amount and position of whitespace, thus
the following commands show empty diff
git diff -w
git log -w -p
git log -w --stat
Change-Id: I1062051d7f97d59922847f5061c6d6811742d30e
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5627
Tested-by: jenkins
It is an error to prefix with "0x" the print of values in decimal.
Replace the incorrect decimal format specifier with PRIx32.
Issue identified by checkpatch script from Linux kernel v5.1 using
the command
find src/ -type f -exec ./tools/scripts/checkpatch.pl \
-q --types PRINTF_0XDECIMAL -f {} \;
Change-Id: I2eb867ef654527b2737ba573a405ec8f97c6a739
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: http://openocd.zylin.com/5624
Tested-by: jenkins
Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Before commit b3ce5a0ae5 ("target: use LOG_USER to print errors
in events") an error in an event handler was silently lost, while
now the associated message is printed out.
A "shutdown" command in a target event (e.g. in gdb-detach) causes
the event to end with error code ERROR_COMMAND_CLOSE_CONNECTION,
that triggers the error message:
shutdown command invoked
Error executing event <event-name> on target <target-name>:
The error code returned by the command "shutdown" is required to
stop the execution in a script/proc and avoid executing any
further command in the script/proc.
It is then normal to get an error code from the "shutdown" command
and it should not be printed out.
Intercept the return code of the event in case of "shutdown", then
skip scheduling other target events and return without printing
the incorrect error message.
Change-Id: Ia3085fb46beacb90a5e4bf0abf7c6e28bb9e6a9b
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reported-by: Laurent Lemele <laurent.lemele@st.com>
Reviewed-on: http://openocd.zylin.com/5710
Reviewed-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
Tested-by: jenkins
With this commit we introduce L1 and L2 cache
flush and invalidate operations which are necessary for
getting/setting actual data during memory r/w operations.
We introduce L2 cache support, which is not presented
on currently support EMSK board. But L2 is presented
on HSDK board, which soon will be introduced.
Change-Id: I2fda505a47ecb8833cc9f5ffe24f6a4e22ab6eb0
Signed-off-by: Evgeniy Didin <didin@synopsys.com>
Reviewed-on: http://openocd.zylin.com/5688
Reviewed-by: Oleksij Rempel <linux@rempel-privat.de>
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
