Previously, only the first 32 GPIO were supported on the BCM2835.
Performance was cited as being the primary justification for not
supporting all 54 pins, notably:
1. There is overhead for calculating the memory offset for the pin
2. GPIO values cannot be written in bulk if pins span memory offsets
Now, all 54 GPIO pins are supported by the driver.
Since pins may use different offsets, multiple pins cannot be toggled
with one memory store. Multiple stores now need to occur when one
sufficed before.
To offset some of the performance overhead for the additional stores,
memory addresses, masks, and shift bits are calculated once and cached
into struct. Calculating these once reduces the number of instructions
a function needs to run in order to manipulate a given GPIO.
The following functions have been updated to leverage the new struct
as they represent some of the hottest paths:
bcm2835_swdio_drive
bcm2835_swdio_read
bcm2835gpio_swd_write_fast
bcm2835gpio_read
bcm2835gpio_write
For `bcm2835gpio_swd_write_fast`, performance should be roughly the same
as the number of memory stores hasn't changed.
For `bcm2835_write`, there is a slight performance degradation since
TMS/TDI/TCK are set separately which incurs an additional memory store.
Instruction counts across the above functions are reduced by ~10-40%.
Macros to access registers have been reworked into inline functions to
support access to all pins and to avoid checkpatch headaches.
The `initial_gpio_state.output_level` member has been retyped to bool to
better align with the expected values.
Support for adjusting pads for the expanded pin range has been left out
as support for manipulating these settings should be moved out of this
driver and into its own utility.
Change-Id: I18853d1a2c86776658630326c71a6bf236fcc6da
Signed-off-by: Vincent Fazio <vfazio@xes-inc.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/7732
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Tested-by: jenkins
The cmsis-dap driver is not added to the list of drivers if none of the
USB backends is available.
Add cmsis-dap driver also if TCP backend is available.
Change-Id: I877fac528e7102af74ee54dfcca77c5aded6a7ce
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/9162
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Use the macro 'LOG_LEVEL_IS()' to test 'debug_level'.
While there, use the macro 'LOG_LVL_*' in place of the numeric
value.
Skip all riscv code, as it is going to be updated soon from the
external fork.
Change-Id: Icad7e879e040d3b9cf1cc004c433f28725017493
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9070
Tested-by: jenkins
The cmsis-dap core driver depends on libusb-related code which breaks
the build when libusb is not available.
Remove libusb dependency of the core driver to fix the build issue. For
now, use an own timeout #define with the value of LIBUSB_TIMEOUT_MS but
timeout handling should be better moved to the backends. However, this
should be addressed in a dedicated patch.
Change-Id: Ic5da392f8ab26b47466be199432432cdc08712ab
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/9161
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-by: <niklaus.leuenb@gmail.com>
Tested-by: jenkins
Allow to specify the parallel port by its device file. Deprecate port
number support but keep it for backward compatibility.
This is one necessary step to remove direct I/O support for the parallel
port driver.
While at it, consistently return ERROR_JTAG_INIT_FAILED in case of a
failure in parport_init().
Change-Id: Ie68087f05ece4b32ccab9d9bdfbf7e1a779e9031
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/9152
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Tested-by: jenkins
We deprecate direct I/O support in favor of ppdev for the following
reasons:
- Linux supports ppdev since ~2.4 (released ~24 years ago) and it is
enabled by default on major distros (Ubuntu, Fedora). So it is
effectively ubiquitous
- FreeBSD provides no direct I/O support, so ppdev (ppi) is the only
viable option
- Direct I/O requires root/elevated privileges which is inadvisable
- Removing direct I/O reduces build and driver complexity and yields
a smaller, easier-to-maintain codebase
- Supporting only ppdev allows us to simplify the codebase by using
device files (e.g., /dev/parport0) instead of numeric identifiers
Windows is the only rationale to keep direct I/O, but the user base
appears minimal to nonexistent and no active contributors can test the
Windows driver.
Change-Id: Ia6d5ed6e8c5faa2a9b4919ca97c5cf9033372a64
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/9151
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Tested-by: jenkins
This finalizes the work that has begun to emulate v2 API
for older libgpiod versions.
It also add the required autotools/pkg-config stuff to
detect the available libgpiod version.
Change-Id: I2c3a60ce607ed9601b01d22d5d9b8af953944941
Signed-off-by: Michael Heimpold <mhei@heimpold.de>
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: https://review.openocd.org/c/openocd/+/8226
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Tested-by: jenkins
Work in progress. Target is to replace v1 gpiod_line_request()
Define and use dummy functions from v2 for v1:
- gpiod_line_config_new(),
- gpiod_line_config_free(),
Not yet used:
- gpiod_line_config_add_line_settings(),
- gpiod_chip_request_lines().
Change-Id: I9e2d3013845c5d12942f5e07c9721fcd151d6840
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/8207
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Tested-by: jenkins
Reorganize the code of helper_get_line() for label err_out.
Define and use dummy functions from v2 for v1:
- gpiod_request_config_new(),
- gpiod_request_config_free(),
- gpiod_request_config_set_consumer().
Change-Id: Ide7cd8459941a5a863c425a2b5abf1fb4a075874
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/8202
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Preparing for libgpiod v2, replace v1 specific function
gpiod_chip_open_by_number() with the common v1 & v2 function
gpiod_chip_open().
Change-Id: I5f0f9bef739f4097fb74f78b5b6e2fe823c1ab87
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/8201
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
For some hosts build fails after the recent cmsis_dap_tcp addition
(see commit fcff4b712c ("jtag/drivers/cmsis_dap: add new backend
cmsis_dap_tcp") or https://review.openocd.org/c/openocd/+/8973)
- Header 'hidapi.h' may not be available and should not be needed here.
- Global pointer variable is not guaranteed to be treated as a constant
expression even with const modifier. Use global array instead, to avoid
'error: initializer element is not constant', as address of a global
array is a constant expression.
Change-Id: I0c72ff52340f546a5f635663a8fde28c99176d1b
Signed-off-by: Samuel Obuch <samuel.obuch@espressif.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9115
Tested-by: jenkins
Reviewed-by: Brian Kuschak <bkuschak@gmail.com>
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
The coding style requires the 'case' to be at the same indentation
level of its 'switch' statement.
Checkpatch is not able to detect numbers in exponential format,
like 6E-5, and complains about missing space around the operator
minus.
To complete the alignment of switch and case statements in this
file, use a separate commit with the special checkpatch ignore tag
in the commit message.
Align the code accordingly.
No changes are reported by
git log -p -w --ignore-blank-lines --patience
Checkpatch-ignore: SPACING
Change-Id: Ibe70c4a4d3f0bb44c03007103910b778944bb90e
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9056
Tested-by: jenkins
Reviewed-by: zapb <dev@zapb.de>
The coding style requires the 'case' to be at the same indentation
level of its 'switch' statement.
Align the code accordingly.
Skip the driver angie that is going to be updated soon with a new
commit and skip the folder OpenULINK that contains a firmware for
the dongle.
No changes are reported by
git log -p -w --ignore-blank-lines --patience
Change-Id: I960a89a45bc9956f98676a019f0d1a652601155f
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9055
Tested-by: jenkins
Reviewed-by: zapb <dev@zapb.de>
To prepare for aligning switch and case statements, fix in advance
some checkpatch error due to existing code:
- remove 'else' after return and break;
- use '__func__' in place of hardcoded function name;
- remove useless parenthesis;
- don't end line with an open parenthesis.
Change-Id: I6a9905e5a30c90456de562e727dd2dfe2fda10c4
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9054
Reviewed-by: zapb <dev@zapb.de>
Tested-by: jenkins
The coding style requires the 'case' to be at the same indentation
level of its 'switch' statement.
Align the code accordingly.
No changes are reported by
git log -p -w --ignore-blank-lines --patience
Change-Id: Iaf368b0bdd7c797b0e4cfb91e838696d706fdcce
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9053
Tested-by: jenkins
Reviewed-by: zapb <dev@zapb.de>
To prepare for aligning switch and case statements, fix in advance
some checkpatch error due to existing code:
- use '__func__' in place of hardcoded function name.
Change-Id: Ib90811c7fffa15702fb710345c5ca3c7331d5ad6
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9052
Reviewed-by: zapb <dev@zapb.de>
Tested-by: jenkins
Make the driver more flexible and define adapter-specific configurations
in Tcl instead of C using the adapter GPIO subsystem.
The rework also includes coding style fixes and improvements of the
documentation. All modifications are done such that backwards
compatibility is ensured.
Tested with Olimex ARM-JTAG cable [1] and APM32F103 target device on
Linux and FreeBSD. The driver works on Linux using direct I/O and PPDEV.
On FreeBSD, only PPDEV works. The build with direct I/O already failed
before the patch. This problem will be fixed in a subsequent patch.
The patch is not tested on Windows because there is no documentation
for it.
[1] https://www.olimex.com/Products/ARM/JTAG/ARM-JTAG/
Change-Id: Ib671d52a919eaf2959cf6365f2c8004257ae074c
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/8943
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Tested-by: jenkins
Add a user-specific signal 'user0' which can be used for custom
functionality.
For now, only a single user-specific signal is supported. Additional
signals 'userX' or a more generic handling can be added in the future.
Also, the signal state can only be configured during the adapter
initialization and termination. Commands to change the signal state at
run-time may be added in the future.
Change-Id: I3f31242f6a693e11565542c3bd4521a245b4ff95
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/9088
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Tested-by: jenkins
In addition to a signal's 'init state', add support to configure its
'exit state'. This option determines the state and configuration of a
pin when the adapter terminates its operation.
Change-Id: I0577ed9e3bc79715290b4e6a21e3f64c66f2ff9a
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/9087
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Create a new backend for cmsis_dap driver that allows CMSIS-DAP protocol
to run over TCP/IP instead of USB.
An example implementation of the firmware for an SWD programmer that
uses this cmsis_dap_tcp protocol can be found at the link below.
https://github.com/bkuschak/cmsis_dap_tcp_esp32
Using this cmsis_dap_tcp backend with the firmware above on an ESP32-C6
programmer and STM32F401RE target shows the following performance:
- loading 96KB image to RAM: 80 KB/sec
- dumping 96KB image from RAM: 72 KB/sec
- flashing 512KB image completes in about 13.5 seconds (including erase,
program, and verify).
Change-Id: I6e3e45016bd16ef2259561b1046788f5536b0687
Signed-off-by: Brian Kuschak <bkuschak@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/8973
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Tested-by: jenkins
Drop some useless initialization in the driver.
Replace a constant value returned through a variable with the
value itself.
Change-Id: I7b7c0b30c6d36e9763ff78bf826742792546fa7f
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9095
Tested-by: jenkins
Reviewed-by: Matthias Jentsch <info@easydevkits.com>
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
The newly merged driver ch347 triggers two new scan-build warnings
about dead assignment to variables.
Fix them.
Change-Id: Ided14272c3573be1498584e68ac4653cde029f31
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Fixes: 7d0e125896 ("jtag/drivers: Add support for CH347-based JTAG adapters")
Reviewed-on: https://review.openocd.org/c/openocd/+/9094
Tested-by: jenkins
Reviewed-by: Matthias Jentsch <info@easydevkits.com>
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
CH347 chip firmware probably does not service USB poll/requests
from the host during processing of SWD packet. If the processing
takes longer than approx 8 ms, USB host disconnects the adapter
from the bus as if it were dead.
To avoid disconnect, compute approximate processing time
of SWD transactions and flush the queue if the total time
gets over the limit (use 7 ms to keep some safety margin).
Allow ch347_swd_send_idle() to split the requested delay
to more partial sequences. It prevents USB disconnect at beginning
of target examine, when memaccess 255 clk is used.
Also memaccess > 255 clk is now possible.
Set the maximal divisor to fit the longest ADI sequence
into 7 ms processing time.
Tested on a linux x64 PC and RPi 5.
Change-Id: Ibdcff4de52e3eb4d86ed83af81a1c64f1f9b5d24
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: https://review.openocd.org/c/openocd/+/8746
Tested-by: jenkins
Reviewed-by: ZhiYuanNJ <871238103@qq.com>
Remove annoying and not too informative error message `ack != SWD_ACK_OK`.
Fix copy&paste error in message `SWD Read data parity mismatch` logged
in case of a write returning bad ack.
Raise log level of read/write descriptive message to DEBUG when reg
read/writes returns bad ack.
Change-Id: Ic3433ae8bd02472756adf269658bfba0ba34dc26
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: https://review.openocd.org/c/openocd/+/8747
Reviewed-by: ZhiYuanNJ <871238103@qq.com>
Tested-by: jenkins
SWD read_reg() and write_reg() methods are declared with
void return. Save the error code to ch347_swd_context.queued_retval
(will be returned later by SWD run method) instead of ignoring it.
Change-Id: Ib95a1bc3398712ac2f8520c79d281633d75f0335
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: https://review.openocd.org/c/openocd/+/8745
Reviewed-by: ZhiYuanNJ <871238103@qq.com>
Tested-by: jenkins
ch347_swd_run_queue() fulfilled ADI requirement for idle cycles
after the last transaction just if there was room in the buffer.
A lazy programmer avoided this way the danger of recurrent call.
Introduce ch347_swd_run_queue_inner() without sending idle cycles.
They are useless when flushing the queue to get room for the next
transaction.
ch347_swd_run_queue() now can make room for idle cycles
in the queue without recursion.
While on it remove two useless debug logs showing ap_delay_clk
value and prevent the overflow of ap_delay_clk forcing it to be <= 255.
Change-Id: Ia7b7f0d373ff463e2f0742bdd068c3833c57f340
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: https://review.openocd.org/c/openocd/+/8744
Reviewed-by: ZhiYuanNJ <871238103@qq.com>
Tested-by: jenkins
SWD has completely different speed setting than JTAG.
There is no official chip documentation. The trial and error
method on SWD init USB packet discovered a byte parameter used
as a divisor from 1 MHz maximal clock.
Prevent setting too low clocks as it may trigger USB disconnects.
CH347T/F versions 5.44/1.1 add 5 MHz SWD speed at divisor/index 0.
CH347F related parts co-authored by ZhiYuanNJ <871238103@qq.com>
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Change-Id: I5a990a43dcd7a6d73c71b795283a9fbdff489bf4
Reviewed-on: https://review.openocd.org/c/openocd/+/8743
Reviewed-by: ZhiYuanNJ <871238103@qq.com>
Tested-by: jenkins
This adds general support for JTAG adapters based on the WCH CH347T and
CH347F chips.
The CH347T must be configured in mode 3 to provide both UART and JTAG
interfaces. The CH347F exposes UART and JTAG directly. Similar to FTDI
FT2232-based devices, the CH347T/F can be used for simultaneous UART and
JTAG communication.
The ch347 driver code is based on:
https://github.com/WCHSoftGroup/ch347
Bug fixes and adjustments were made to align the code with OpenOCD style
and ensure compatibility. The driver was integrated into the OpenOCD
build system.
The USB packet format was identified using public GitHub sources and
documented here:
https://www.easydevkits.com/wch-ch347-jtag-interface/
Change-Id: I5fca9dd015111e4410fea029611fdeedbb228fdb
Signed-off-by: EasyDevKits <info@easydevkits.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/7937
Tested-by: jenkins
Reviewed-by: ZhiYuanNJ <871238103@qq.com>
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Update jtag driver code to reflect these changes and properly drive
Angie probe.
The rationale behind this is to increase the probe performances,
especially in use cases when large files shall be loaded on a target.
The USB transfer performances are now close to those obtained with a
standard FTDI probe.
Change-Id: I3b31d75a3f66c2d07fed8c7423f765acc30925f8
Signed-off-by: Adrien Charruel <acharruel@nanoxplore.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/8711
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Correct the issue of usb descriptor does not appear
near the PID and VID, Cypress USB controller cannot
handle direct pointers to memory CODE area, so we
copy the data in the external RAM area and point
to it.
Change-Id: I3221627dc8576f6341b444acd9c554fd5cc47918
Signed-off-by: Ahmed BOUDJELIDA <aboudjelida@nanoxplore.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/8736
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Add a check if i2c SDA pin state is HIGH.
if its HIGH, the dev-board is ON, we receive
number of Ack.
in its LOW, the board is OFF, we send this
information back to driver.
Change-Id: Ia40d3910675cc10e0208d8bc0060a19c12b1409d
Signed-off-by: Ahmed BOUDJELIDA <aboudjelida@nanoxplore.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/8716
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
This new code implement two FIFOs for handling TX and RX
JTAG data transfers, its simply receives data and send it
OUT to target chip in respect of JTAG protocol timing
constraints.
The IN FIFO receives data from target chip and send it
back to openocd.
Change-Id: I17c1231e7f4b0a6b510359fe147b609922e0809e
Signed-off-by: Ahmed BOUDJELIDA <aboudjelida@nanoxplore.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/8715
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Display only the quirk state without additional text. This makes
processing via Tcl easier. Also, do not echo the selected quirk state.
Change-Id: I2e8de2742ffc10c7995a30a2a2d8a383b0cfaa69
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/9068
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Simplify syntax error handling and make the documentation and code
consistent.
Change-Id: Ib8ee5adff2071964fc6d8e153f3eb82dd20054f3
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/9067
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Format strings are often split to allow using the conversion
specifiers macros from <inttypes.h>.
When the format string ends with one of such macros, there is no
need to add an empty string "" after the macro.
In current code we have 203 cases of empty string present, against
1159 cases of string ending with the macro.
Uniform the style across OpenOCD by removing the empty string.
Don't modify the files 'angie.c' and 'max32xxx.c' as they are
already changed by other independent commits.
Change-Id: I23f1120101ce1da67c6578635fc6507a58c803e9
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
Reviewed-on: https://review.openocd.org/c/openocd/+/9065
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Print a debug message rather than an info message because this
information is not of importance for normal users.
While at it, fix the 'EMBEDDED_FUNCTION_NAME' checkpatch issue.
Change-Id: I9d458a6dd6818fdedc488e39d2daa6d72a21b743
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/8994
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
Print a debug message rather than an info message because this
information is not of importance for normal users.
Change-Id: Ie91565df455ffc0bfe976d1782dd4318bfd2d30b
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/8986
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
