The BMP with hardware version 4 and newer, use option bytes instead of
physical GPIO to encode the hardware version. In some older BMP there is
a chance that the user option byte is set to 255 (0x00FF pattern). This
can throw off the hardware version detection routine.
The hw rev 4 and 5 both have the version stored in the Data1 user option
byte. This frees up the hw rev strapping pins for other uses, ie swtrace
decoding using USART1 RX, and additional peripherals on the SPI bus,
like bulk flash storage and displays.
lock_flash and lock_bootprot currently support only locking the whole flash and locking the maximal leading flash chunk (32k).
An optional numerical parameter is added. It can be specified in decimal or 0x prefixed hexadecimal.
For samd21 'lock_bootprot 0' locks the first 32k of flash while 'lock_bootprot 6' locks the first 512 bytes. 'lock_bootprot 0' is equivalent to 'unlock_bootprot'.
Similarly, 'lock_flash <number>' locks the flash segments corresponding to zeros in the binary representation of the given number.
'lock_flash 0xffff' is equivalent to 'unlock_flash'.
If the optional parameter is not given both commands work as previously.
The SAMD09 CPU is used in boards such as the Adafruit Seesaw. It has a
smaller amount of memory and flash than other SAMD ports.
This was tested with an Adafruit Seesaw. These boards come with preloaded
firmware. As a test, the firmware was dumped and flash was erased. Then,
flash was verified to be all zeroes. Finally, the firmware was loaded
back in:
(gdb) p/x *(unsigned int *)0@32
$8 = {0x20000f88, 0x1db, 0x1d1, 0x1d9, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x1d9, 0x0, 0x0, 0xf5, 0x1081, 0x1d9, 0x1d9, 0x1d9, 0x1d9, 0x1d9, 0x1d9, 0x1d9, 0x0, 0x1d9, 0x1d9, 0x25e9, 0x0,
0x0, 0x1d9, 0x1d9, 0x1d9}
(gdb) dump ihex memory flash.ihex 0 8192
(gdb) mon erase_mass
Erase successful!
(gdb) p/x *(unsigned int *)0@32
$9 = {0xffffffff <repeats 32 times>}
(gdb) load flash.ihex
Loading section .sec1, size 0x2000 lma 0x0
Start address 0x00000000, load size 8192
Transfer rate: 5 KB/sec, 910 bytes/write.
(gdb) p/x *(unsigned int *)0@32
$10 = {0x20000f88, 0x1db, 0x1d1, 0x1d9, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x1d9, 0x0, 0x0, 0xf5, 0x1081, 0x1d9, 0x1d9, 0x1d9, 0x1d9, 0x1d9, 0x1d9, 0x1d9, 0x0, 0x1d9, 0x1d9, 0x25e9, 0x0,
0x0, 0x1d9, 0x1d9, 0x1d9}
(gdb)
Signed-off-by: Sean Cross <sean@xobs.io>
Various SAMD devices have different amounts of memory. Up until now, all
SAMD devices have had the same amount, and therefore this value was
hardcoded to 32k of RAM and 256k of flash.
Add a parameter to the description field and set it to default to the
previous values. Use this description field when adding memories to the
target definition.
Signed-off-by: Sean Cross <sean@xobs.io>
The variable `start_time` indicates when a CRC32 operation began. This
variable is used to benchmark the speed of the CRC32 function.
Currently, this is tied to `PC_HOSTED`. However, the actual usage is
tied to `DEBUG_WARN`. This means that the variable is undefined when
`DEBUG_WARN` is defined and we're not configured for `PC_HOSTED` mode.
Add macro guards around this variable so that it is defined when
debugging is enabled, rather than only when building on `PC_HOSTED`.
Signed-off-by: Sean Cross <sean@xobs.io>
Workaround for CMSIS-DAP/Bulk debugger orbtrace that returns NO_ACK
with high values of TRNCNT. Perhaps only STM32F767 needs write to DHCSR
with high occupancy to catch the device in a moment not sleeping.
Expect signal integrity errors when using jumper cables. Often probes switch
the SWJ GPIO with highest speed, resulting in possible reflections. Additional
ground wires may help. If there is isolation between probe and targets,
additional ground wires are a must or ground shift will wrack the transfer!
The same buffer is used for both command and response, causing part of
the command to be overwritten before it’s retried. Solve this by saving
and restoring a copy of the command before retrying.
Stlink does not like low level access and aborts with STLINK_SWD_DP_ERROR.
Either our implementation still has faults or stlink can not handle
MINDP devices with BMP/Firmware also seemm not to like low level access,
either du to some hidden error in BMP or by design
the sdid bits correspond to s32k14x as well as s32k11x but i only have
access to the k14x variants so didn't bother w/ the k11x variants.
the memory/flash sizes also can come from the sdid bits, but the
splits still need to come from a lookup.
FIXME: One remote transaction still can only do up to 64 ticks, leaving
room for speed enhancement with larger transactions.
Firmware assumes (1LL << 65) == 0LL !
Fixes bug introduced with last commit( Recover from bad AP access)
Let STM32 timers run at 100 Hz against 10 Hz before.
Programming STM32F103 failed random (#900) with 20 ms timeout requested
against the 100 ms timeout granularity provided up to now.
STM32 Firmware only ticked at 10 hertz, so the sequence "low_access",
"set timeout", "send out 8 bit command", "read 3 bit result" when
reading "wait" and timer increment tick happening during that sequence
will already hits the timeout even so only mininal time has elapsed
and not the requested timeout.
E.g. AP1 on a STM32WLE5 points to a ROM table, but access to the ROM table
via AP1 hangs forever.
- Substantial reduce timeout when wait for a response. Valid access should
succeed fast.
- Abort AP access to free DP for other accesses
- Don't throw exception, only set dp->fault
- React on higher level
This also fixes a warning about an empty body in cortexm.c
if PLATFORM_HAS_DEBUG is defined but debugging is not enabled:
if (platform_timeout_is_expired(&to))
DEBUG_WARN("Reset seem to be stuck low!\n");
If the RP2040 is using the ring oscilator (ROSC) rather than the crystal
oscilator (XOSC) then flashing program will take much longer. As the XOSC is not
enabled at boot we should not assume it will be enabled before the debugger is
connected (or indeed at all), thus should use the longer timeout during load
commands.
Increasing spinner timeout to 500 means `tc_printf` is not called during
the `_flash_range_erase`. This is important as `_flash_range_erase` is
used when loading. This fixes the issue:
https://github.com/blacksphere/blackmagic/issues/875
Also adding spinner timeout counter to reduce bus traffic.
This reverts commit 61e237ec87.
This makes a CMSIS-DAP probe with version < 1.2 work again on DP V1 devices.
Status of other combination unknown:
DP V1 V2 V2/Multidrop
Probe
< 1.2 ok ok --
>= 1.2 ? ? ?
Use one file in dfu and server
For STM32, provide 3 ways to handle
- BMP way (8 bytes)
- DFU way (12 Bytes)
- Full unique id (24 Bytes), as with STLINK(V3) Bootloader
- RP2040 show both DPs
- Multidrop test with STM32L552 and STM32H745 allows selection
with "-m 0x4500041" (H7), "-m 1" (L552) or "-m 0x01002927" (RP2040)
This adds support for the STM32WL series in stm32l4.c. These parts have
the same flash registers layout as the L4 series, but a different base.
Since there are already two sets of registers in this target file, this
adds support for register maps that can be customized for each device
ID.
This adds a TRY_CATCH around the adiv5_ap_read_id() in
adiv5_component_probe() and resets the DP when that happens.
It seems like the STM32WLE5 comes with the AP of the inactive core
enabled in a way that does not make it detectable, and the current code
times out and leaves the whole device hanging.
Catching the timeout and calling adiv5_dp_abort() seems to restore the
device to a useable state.
Tested on Seed LoRa-E5 (STM32E5JC).
This patch puts null terminating characters for the 'type',
'version', and 'serial' strings extracted from blackmagic probe
id strings on linux systems.
STM32WLE5 has the same dual core chip as STM32WL5. For the second
core, the additional AP can be see, but access to e.g. CIDR0 for that
Romtable fails.
Aborting the scan too if again the second read of CIDR0 fails makes
sense anyways!
Special iRST_SENSE handling does not apply at least for BMP 2.1c.
On V2.1c, iRST_SENSE was set as output high and xRST high shorted
iRST_SENSE to ground via Q2
FIXME: Check for older versions!
This changes the USB identification from
"Black Magic Probe(STLINK/V2) v..." to
"Black Magic Probe (STLINK/V2) v..."
Signed-off-by: Martin <Ho-Ro@users.noreply.github.com>
This adds support for the NXP S32K148. This is an automotive-grade part
that is derived from the Kinetis line, so it has a very similar
interface to other parts in the family.
Signed-off-by: Sean Cross <sean@xobs.io>
**Summary**
Modifications to fix flash support on the lpc546xx:
- fix IAP entrypoint to be `0x03000204`, not the value at that address
- add a reset and attach sequence before erasing flash sectors. there's
little documentation around this, but experimentally, erasing sector 0
fails when the ROM bootloader is mapped to it (on reset). stepping the
chip once and attaching is enough to snap the chip out of it, permitting
flash erase on sector 0.
- add a few test commands to the lpc546xx table (read_uid, erase/write
sector, etc).
- write the magic CRC value when writing to sector 0
(`lpc_flash_write_magic_vect`).
- move the lpc546xx probe to before the lpc43xx probe, to prevent
getting the lpc546xx into Lockup when reading an illegal memory location
during lpc43xx probing
Fixes#786.
I don't 100% understand the reset/load sequence of this part, but these
changes are sufficient to program and debug the part now.
I didn't do a detailed analysis of what pyocd (via st-link hardware
adapter) and segger jlink do to handle the same, but both of those
worked without modification, so there's some difference in the
sequence they're using.
**Testing**
Verified I can now successfully erase and write an executable in sector
0 (and other sectors).
Verified the new commands work correctly.
More recent versions of GNU make remove the `-` prefix from
`$(MAKEFLAGS)`, which breaks the build:
```bash
# output truncated to interesting parts
> make -C src --trace all_platforms
set -e ;\
mkdir -p artifacts/v1.7.1-91-g98b4ec5 ;\
echo "<html><body><ul>" > artifacts/index.html ;\
for i in platforms/*/Makefile.inc ; do \
...\
make w --trace --no-print-directory clean ;\
make w --trace --no-print-directory;\
...\
done ;\
echo "</ul></body></html>" >> artifacts/index.html ;\
cp artifacts/*.bin artifacts/v1.7.1-91-g98b4ec5
Building for hardware platform: f4discovery
make[1]: *** No rule to make target 'w'. Stop.
```
Per the
[documentation](https://www.gnu.org/software/make/manual/html_node/Options_002fRecursion.html),
there's no need to manually apply insert `$(MAKEFLAGS)`, it should be
automatically included when using `$(MAKE)` when recursing.
I tested on my machine (make 4.2.1) and a docker image based on
ubuntu:trusty (make 3.81), both behaved the same (directories were not
printed when recursing by default);
mean
Mike Ditto
Piotr Esden-Tempski
Fabrice Prost-Boucle
dragonmux
arpadbuermen
Uwe Bonnes
Nicolas Schodet
Paul Mulders
Frank Kunz
Sean Cross
Koen De Vleeschauwer
SG
Vegard Storheil Eriksen
Jeff Lutgen
Thiadmer Riemersma
Andy Makovec
dpc
Thiadmer Riemersma
aam335
Björn Mellström
James Turton
Fabio Baltieri
Vestrel
Stoyan Shopov
Martin
Alex Norman
Noah Pendleton
Aaron Lindsay
Noah Pendleton