Commit 6541233aa7 ("Combine register lists of smp targets.")
assumes that all the targets in the SMP cluster are already
examined and unconditionally call target_get_gdb_reg_list_noread()
that will in turn return error if the target is not examined yet.
Skip targets not examined yet.
Add an additional check in case the register list cannot be built,
e.g. because no target in the SMP cluster is examined. This should
never happen, but it's better to play safe.
Change-Id: I8609815c3d5144790fb05a870cb0c931540aef8a
Fixes: 6541233aa7 ("Combine register lists of smp targets.")
Reported-by: Michele Bisogno <michele.bisogno.ct@renesas.com>
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6853
Tested-by: jenkins
Reviewed-by: Michele Bisogno <michele.bisogno.ct@renesas.com>
Reviewed-by: Tim Newsome <tim@sifive.com>
Commit 6541233aa7 ("Combine register lists of smp targets.")
unconditionally assigns the output pointers of the function
smp_reg_list_noread(), even if the function fails and returns
error.
This causes a double free from the caller, that has assigned NULL
to the pointers to simplify the error handling.
Use local variables in smp_reg_list_noread() and assign the output
pointers only on success.
Change-Id: Ic0fd2f26520566cf322f0190780e15637c01cfae
Fixes: 6541233aa7 ("Combine register lists of smp targets.")
Reported-by: Michele Bisogno <michele.bisogno.ct@renesas.com>
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6852
Tested-by: jenkins
Reviewed-by: Michele Bisogno <michele.bisogno.ct@renesas.com>
Reviewed-by: Tim Newsome <tim@sifive.com>
This adds a board file for the NXP LS1088ARDB. This only covers the
"primary" JTAG header J55, and not the PCIe header (J91). The only
oddity is that the LS1088A and CPLD are muxed by adding/removing a
jumper from J48. Unfortunately, it doesn't look like OpenOCD supports
this CPLD beyond determining the irlen, so it's not very useful. Those
who are interested in experimenting can define CWTAP to access the CPLD,
but the default is to access the CPU.
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Change-Id: Ia07436a534f86bd907aa5fe2a78a326a27855a24
Reviewed-on: https://review.openocd.org/c/openocd/+/6849
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Normally the service processor is not necessary for debugging. However,
if you are using the hard-coded RCW or your boot source is otherwise
corrupt, then the general purpose processors will never be released from
hold-off. This will cause GDB to become confused if it tries to attach,
since they will appear to be running arm32 processors. To deal with
this, we can release the CPUs manually with the BRRL register. This
register cannot be written to from the axi target, so we need to do it
from the service processor target. This involves halting the service
processor, modifying the register, and then resuming it again. We try
and determine what state the service processor was in to avoid resuming
it if it was already halted.
The reset vector for the general purpose processors is determined by the
boot logation pointer registers in the device configuration unit.
Normally these are set using pre-boot initialization commands, but if
they are not set then they default to 0. This will cause the CPU to
almost immediately hit an illegal instruction. This is fine because we
will almost certainly want to attach to the processor and load a program
anyway.
I considered adding this as an event handler for either gdb-attach or
reset-init. However, this command shouldn't be necessary most of the
time, and so I don't think we should run it automatically.
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Change-Id: I1b725292d8a11274d03af5313dc83678e10e944c
Reviewed-on: https://review.openocd.org/c/openocd/+/6850
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
The LS1088A is an octo-core aarch64 processor from NXP in the layerscape
family. The JTAG is undocumented, but I was able to figure things out
from the output of `dap info`. This is the first in-tree example of
using the hwthread rtos (as far as I know), so hopefully it can serve as
an example to other developers. There are some ETMs, but I was unable to
try them out because I got 'invalid command name "etm"' when trying to
test things out.
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Change-Id: I9b0791d27d8c41170a413a8d86431107a85feba2
Reviewed-on: https://review.openocd.org/c/openocd/+/6848
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
According to the datasheet, the minimum clock period with Vccio1 = 1.5V
(the lowest voltage supported) is 143ns, or around 6MHz. Set the default
adapter speed to 5 MHz.
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Change-Id: I21cad33fa7f1e25e81f43b5d2214d1fa4ec924de
Reviewed-on: https://review.openocd.org/c/openocd/+/6847
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
This adds some additional IDCODEs from the datasheet. It also adds
support for customizing the tap name.
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Change-Id: I7cda10b92c229b61836c12cd9ca410de358ede2e
Reviewed-on: https://review.openocd.org/c/openocd/+/6846
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Some of STM32 flash drivers read Cortex-M part number from
cortex_m->core_info.
In corner cases the core_info pointer was observed uninitialised
even if target_was_examined() returned true. See also [1]
Use the new and safe helper to get Cortex-M part number.
While on it switch also target_to_cm()/target_to_armv7m() to the safe
versions. This prevents a crash when the flash bank is misconfigured
with non-Cortex-M target.
Add missing checks for target_was_examined() to flash probes.
[1] 6545: fix crash in case cortex_m->core_info is not set
https://review.openocd.org/c/openocd/+/6545
Change-Id: If2471af74ebfe22f14442f48ae109b2e1bb5fa3b
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Fixes: f5898bd93f (flash/stm32fxx.c: do not read CPUID as this info is stored in cortex_m_common)
Reviewed-on: https://review.openocd.org/c/openocd/+/6752
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
The getter checks the magic numbers in arch_info to detect eventual
type mismatch.
Change-Id: I61134b05310a97ae9831517d0516c7b4240d35a5
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: https://review.openocd.org/c/openocd/+/6751
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
target_to_armv7m() and target_to_cm() do not match the magic number
so they are not suitable for use outside of target driver code.
Add checked versions of pointer getters. Match the magic number
to ensure the returned value points to struct of the correct type.
Change-Id: If90ef7e969ef04f0f2103e0da29dcbe8e1ac1c0d
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: https://review.openocd.org/c/openocd/+/6750
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
The third parameter of container_of() should point to the same member
as target->arch_info points to, struct arm.
It worked just because struct arm is the first member in
struct armv7m_common.
If you move arm member from the first place, OpenOCD fails heavily.
Change-Id: I0c0a5221490945563e17a0a34d99a603f1d6c2ff
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: https://review.openocd.org/c/openocd/+/6749
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
This commit corrects the erase function for stm32f2x when dealing with
sectors in bank 2, for STM32F42x/43x devices with 1MB flash.
On STM32F42x/43x with 1MB flash in dual bank configuration, the sector
numbering is not consecutive. The last sector in bank 1 is number 7, and
the first sector in bank 2 is number 12.
The sector indices used by openocd, however, _are_ consecutive (0 to 15
in this case). The arguments "first" and "last" to stm32x_erase() are of
this type, and so the logic surrounding sector numbers needed to be
corrected.
Since the two banks in dual bank mode have the same number of sectors, a
sector index in bank 2 is larger than or equal to half the total number
of sectors.
Change-Id: I15260f8a86d9002769a1ae1c40ebdf62142dae18
Signed-off-by: Simon Johansson <ampleyfly@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6810
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-by: Tarek BOCHKATI <tarek.bouchkati@gmail.com>
The host repo.or.cz is often offline, creating issues for cloning
and building OpenOCD from scratch.
Already 'jimtcl' developer has dropped repo.or.cz, triggering the
OpenOCD commit 861e75f54e ("jimtcl: switch to github").
Change also the link of the remaining submodules 'git2cl' and
'libjaylink' to their respective main repository.
Change-Id: Ib513237427635359ce36a480a8f2060e2fb12ba4
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6834
Tested-by: jenkins
Reviewed-by: zapb <dev@zapb.de>
Tested with an AZ3166 dev board (which uses the STM32F412ZGT6) running
the Azure RTOS ThreadX demonstration system.
Signed-off-by: Ben McMorran <bemcmorr@microsoft.com>
Change-Id: I44c8f7701d9f1aaa872274166321cd7d34fb1855
Reviewed-on: https://review.openocd.org/c/openocd/+/6829
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Commit 39650e2273 ("ecosboard: delete bit-rotted eCos code") has
removed eCos code but has left some empty function that was used
during non-eCos build to replace eCos mutex.
Drop the functions and the file that contain them.
Change-Id: I31bc0237ea699c11bd70921660f960ee406ffa80
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6835
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Explicitly providing a thread name in the "thread" element produces
better thread visualizations in downstream tools like IDEs.
Signed-off-by: Ben McMorran <bemcmorr@microsoft.com>
Change-Id: I102c14ddb8b87757fa474de8e3a3f6a1cfe10d98
Reviewed-on: https://review.openocd.org/c/openocd/+/6828
Tested-by: jenkins
Reviewed-by: zapb <dev@zapb.de>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
This functionality was lost in [1], which was merged as commit
615709d140 ("Upstream a whole host of RISC-V changes.").
Now it works as expected again.
Add convenience macro foreach_smp_target_direction().
Link: [1] https://github.com/riscv/riscv-openocd/pull/567
Change-Id: I1545fa6b45b8a07e27c8ff9dcdcfa2fc4f950cd1
Signed-off-by: Tim Newsome <tim@sifive.com>
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6785
Tested-by: jenkins
Use a bool flag to specify if the list should be forward or
backward iterated.
Change-Id: Ied19d049f46cdcb7f50137d459cc7c02014526bc
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6784
Tested-by: jenkins
Instead of reinventing a simply linked list, reuse the list helper
for the list of targets in a smp cluster.
Using the existing helper, that implements a double linked list,
makes trivial going through the list in reverse order.
Change-Id: Ib36ad2955f15cd2a601b0b9e36ca6d948b12d00f
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6783
Tested-by: jenkins
Continuation of the previous patch. There are two more cases
when progbuf cache in OpenOCD shall be invalidated:
- When OpenOCD resets the debug module undergoes reset (dmactive=0),
e.g. during target examination
- When the user manually performs that very same operation
(via riscv dmi_write)
Change-Id: I53f8f08250eeedcbd55ab4361d5665370b063680
Signed-off-by: Jan Matyas <matyas@codasip.com>
This commit relates to progbuf cache,
implemented in https://github.com/riscv/riscv-openocd/pull/381
Make sure the cache gets invalidated when the progbuf
contents change via other means. I've identified two
such cases where the invalidation is required:
1) When the user manually tinkers with the progbuf registers
(TCL command "riscv dmi_write")
2) When program buffer is used as a scratch memory
(scratch_write64())
Change-Id: Ie7ffb0fccda63297de894ab919d09082ea21cfae
Signed-off-by: Jan Matyas <matyas@codasip.com>
Manually applying https://review.openocd.org/c/openocd/+/6834 (which
hasn't merged yet) since repo.or.cz is down often enough that it's
affecting development.
Change-Id: Idb337c799662f3e0fa72379c59c52a61a048044e
Signed-off-by: Tim Newsome <tim@sifive.com>
Enabling a portion (0x100 - 0x107) of the user defined semihosting
operation number range (0x100 - 0x1FF) to be processed with the help of
the existing target event mechanism, to implement a general-purpose Tcl
interface for the target available on the host, via semihosting
interface.
Example usage:
- The user configures a Tcl command as a callback for one of the newly
defined events (semihosting-user-cmd-0x10X) in the configuration
file.
- The target can make a semihosting call with <opnum>, passing optional
parameters for the call.
If there is no callback registered to the user defined operation number,
nothing happens.
Example usage: Configure RTT automatically with the exact, linked
control block location from target.
Signed-off-by: Zoltán Dudás <zedudi@gmail.com>
Change-Id: I10e1784b1fecd4e630d78df81cb44bf1aa2fc247
Reviewed-on: https://review.openocd.org/c/openocd/+/6748
Tested-by: jenkins
Reviewed-by: Oleksij Rempel <linux@rempel-privat.de>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
This lets the RTOS pick the "current" target, which matters if address
translation differs between threads.
Change-Id: I5b5510ab6a06621589c902f42a91562055817dc4
Signed-off-by: Tim Newsome <tim@sifive.com>
Values defined in sys/stat.h are not guaranteed to match
the constants defined by the GDB remote protocol, which are defined in
https://sourceware.org/gdb/onlinedocs/gdb/Open-Flags.html#Open-Flags.
On my local system (Manjaro 21.2.1 x86_64), for example, O_TRUNC is
defined as 0x40, whereas GDB requires it to be 0x400,
causing all "w" file open modes to misbehave.
This patch has been tested with STM32F446.
Change-Id: Ifb2c740fd689e71d6f1a4bde1edaecd76fdca910
Signed-off-by: Pavel Kirienko <pavel.kirienko@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6804
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
This is helpful when you want to pretend to gdb that your heterogeneous
multicore system is homogeneous, because gdb cannot handle heterogeneous
systems. This won't always works, but works fine if e.g. one of the
cores has an FPU while the other does not. (Specifically, HiFive
Unleashed has 1 core with no FPU, plus 4 cores with an FPU.)
Signed-off-by: Tim Newsome <tim@sifive.com>
Change-Id: I05ff4c28646778fbc00327bc510be064bfe6c9f0
Reviewed-on: https://review.openocd.org/c/openocd/+/6362
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Some CPU wrongly indicate the bypas bit in the codeid.
It's the case of the NanoXplore NG-ULTRA chip that export a
configurable (and potentially invalid) ID for one of
its component.
Add an option to ignore it.
Signed-off-by: Adrien Grassein <adrien.grassein@gmail.com>
Change-Id: Ic59743f23bfc4d4e23da0e8535fec8ca9e87ff1a
Reviewed-on: https://review.openocd.org/c/openocd/+/6802
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-by: Oleksij Rempel <linux@rempel-privat.de>
We have the API target_event_name().
Use it to improve code readability.
Change-Id: Ic48d2227bdefe9af05aff99a871a45e0612e5254
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6790
Tested-by: jenkins
When log to file is enabled, the file is not closed by OpenOCD at
exit. This is reported by Valgrind as a memory leak that is still
reachable, as the internal buffers of 'FILE *log_output' are freed
by the automatic fclose() at exit.
Close the log file before exit.
Change-Id: Id472c0d04462035254a9b49ecb0a4037263c6f6f
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6789
Tested-by: jenkins
If the CTI is not specified OpenOCD fails target's examination
without indicating the reason.
Drop an error message about the missing CTI.
Change-Id: I344537fb21cf38785796ba938e71890e04135509
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6788
Tested-by: jenkins
With jimtcl 0.81 the syntax of the TCL command 'expr' requires the
multiple arguments to be within curly brackets.
Update the examples in the documentation to follow the new syntax.
While there, split one example to avoid it to exceed the line size
during pdf document generation.
Change-Id: I91cca419f8273415ccb0c2ce369fc6ac476e34e5
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/6809
Tested-by: jenkins
Changed flash driver to support writing to the user data page, as well as to any portion of the lockbits page above 512 bytes (the amount used for the actual page lock words). The top part of the lockbits page is used on at least the EFR32xG1 chips for the SiLabs bootloader encryption keys.
As presented to the user, the lockbits page is the same size as the other pages, but any attempt to write to its low 512 bytes is an error. To enforce this, efr32x_write is renamed to efm32x_priv_write and a wrapper function is provided in its place. If the user erases the lockbits page, the driver rewrites the cached lock words after the erase. When the driver erases the lockbits page in order to update the lock words, it first takes a copy of anything stored in the top part of the page, and re-programs it after the erase operation.
There are now multiple instances of flash_bank for each target, and the flash_bank instances must share their cached lock words to operate as intended. Therefore, when a bank is created, the global flash bank list is used to find any other banks that share the same target. Since some banks in the global list are invalid at the time free_driver_priv is called, reference counting is used to decide when to free driver_priv.
To avoid the need to find the lockbits flash_bank from another flash_bank, efm32x_priv_write and efm32x_erase_page now take an absolute address.
There didn't seem to be any reason to prohibit unprotecting individual flash pages, so that limitation is removed from efm32x_protect().
This addresses ticket #185.
Valgrind-clean, except for 2x 4kiB not freed/still reachable blocks that were allocated by libudev.
No new Clang analyzer warnings, no new sanitizer warnings.
Signed-off-by: Doug Brunner <doug.a.brunner@gmail.com>
Change-Id: Ifb22e6149939d893f386706e99b928691ec1d41b
Reviewed-on: https://review.openocd.org/c/openocd/+/6665
Tested-by: jenkins
Reviewed-by: Fredrik Hederstierna <fredrik.hederstierna@gmail.com>
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Antonio Borneo
Sean Anderson
Tomas Vanek
Simon Johansson
Erhan Kurubas
Tim Newsome
Ben McMorran
Jacek Wuwer
Jan Matyas
Erhan Kurubas
Greg Savin
Zoltán Dudás
Pavel Kirienko
Julien Massot
Hans-Erik Floryd
Adrien Grassein
Doug Brunner
Marc Schink