Paul Fertser [Fri, 13 Mar 2015 14:55:51 +0000 (17:55 +0300)]
target/arm_adi_v5, cortex_m: retry ahbap_debugport_init few times in case of an error
Some targets need arbitrary amount of time (usually not too long)
after reset (both sysresetreq and srst) to do initialisation, and
SWD/JTAG is not available during that. According to PSoC4 docs, the
debugger should try connecting until it succeeds.
Also ahbap_debugport_init might be necessary to perform after using
hardware srst too, so add it there (except for the targets that
support srst_nogate since they are very unlikely to need it).
Change-Id: I3598d5ff7b8e0bf3a5566a57dec4b0b2b243d297 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2601 Tested-by: jenkins
Paul Fertser [Sat, 14 Mar 2015 05:15:12 +0000 (08:15 +0300)]
target/cortex_a: emit a clear error message when dbgbase can't be detected
In some cases (the most obvious are TI's SoCs) ROM table lacks entries
for the cores, so OpenOCD has no way to determine what debug base to
use. Due to an error fixed in ec9ccaa28849 it wasn't handled properly,
and OpenOCD would continue to try using dbgbase = 0, which happened to
work for e.g. AM437x.
This patch adds a clear indication to the user that to access such a
target, dbgbase must be set manually in the config.
Reported by Felipe Balbi on IRC.
Change-Id: Id8533e708f44b76550eb8b659564f5f45717c298 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2603 Tested-by: jenkins
Paul Fertser [Sat, 4 Apr 2015 21:16:59 +0000 (00:16 +0300)]
flash/nor/lpc2000: free allocated working area when target_write fails
In some circumstances (e.g. inappropriate jtag clock)
target_write_memory in lpc2000_iap_working_area_init might fail. The
allocated working area should be freed inside
lpc2000_iap_working_area_init in this error case.
This was leading to a weird segfault due to stack corruption later
when reset was executed.
Reported by quitte (Jonas Meyer).
Change-Id: Ia2ed42a9970a4d771727fd516a6eea88e9b859e2 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2696 Tested-by: jenkins
Paul Fertser [Mon, 6 Apr 2015 13:04:05 +0000 (16:04 +0300)]
jtag/adi_v5_jtag: fix infinite recursion in jtagdp_transaction_endcheck()
Calling ahbap_debugport_init() is wrong here because the actions
performed by it might lead to jtagdp_transaction_endcheck errors thus
leading to infinite recursion.
The removed code is not needed now because target polling should lead
to reexamination automatically, and both cortex_a and cortex_m call
ahbap_debugport_init() as part of their target examine handler.
This was reported as a real life issue on IRC by Weaselweb with
Cortex-A target. Quitte reports similar results in some circumstances
(adapter_khz too high) with LPC17xx.
Change-Id: I7148022f76a1272b5262d251f2e807ffb1543547 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2697 Tested-by: jenkins
Paul Fertser [Sat, 4 Apr 2015 17:50:37 +0000 (20:50 +0300)]
flash/nor/lpc2000: properly free working area used in get_lpc2000_part_id()
The IAP working area needs to be freed here, just like in all the
other driver functions since an automatic local variable is used to
store a pointer to it.
This was reported by quitte (Jonas Meyer) on IRC as a strange totally
unrelated segfault after doing certain operations (leading to target
reset) from GDB. He has provided me with remote access to the specific
machine and configuration that exposed the issue, and after some
debugging it became apparent that a auto local variable (holding the
gdb connection pointer) gets overwritten somehow. Placing an
appropriate breakpoint just before the event and using a watchpoint
made the cause apparent: reset lead to freeing of all working areas,
and there was one holding a pointer to a variable that was auto local
in get_lpc2000_part_id().
Change-Id: I7e634d890135ca0f3b4b311e09e8385a03982bd6 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2695 Tested-by: jenkins
Paul Fertser [Fri, 3 Apr 2015 04:55:50 +0000 (07:55 +0300)]
jtag/startup.tcl: fix regression with autoselecting JTAG
This regression was introduced with d90b86d8. "transport select" doesn't
throw an error anymore and autoselects the first available transport on
its own.
Reported by moyix on IRC.
Change-Id: I3f303c0372e915931cca4b28af450694abc1a63e Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2693 Tested-by: jenkins
Richard Braun [Wed, 11 Mar 2015 13:04:15 +0000 (14:04 +0100)]
ChibiOS: fix crash on auto detection
The detection framework assumes rtos->symbols is dynamically allocated,
an assumption that the ChibiOS variant breaks by providing a raw statically
allocated symbol list.
Change-Id: I379bcc2af99006912608ddd3f646ff7085606f47 Signed-off-by: Richard Braun <rbraun@sceen.net>
Reviewed-on: http://openocd.zylin.com/2597 Tested-by: jenkins Reviewed-by: Stian Skjelstad <stian@nixia.no> Reviewed-by: Tomas Vanek <vanekt@fbl.cz> Reviewed-by: Paul Fertser <fercerpav@gmail.com>
This patch might influence openocd Tcl commands behaviour in subtle
ways, please give it a nice testing.
The idea is that if an OpenOCD Tcl command returns an error, an
exception is raised, and then the return code is propogated all the
way up (or to the "catch" if present). This allows to detect
"shutdown" which is not actually an error but has to raise an
exception to stop execution of the commands that follow it in the
script.
openocd_thread special-cases shutdown because it should then terminate
OpenOCD with a success error code, unless shutdown was called with an
optional "error" argument which means terminate with a non-zero exit
code.
Paul Fertser [Thu, 26 Feb 2015 17:18:38 +0000 (20:18 +0300)]
jtag/tcl: fix incorrect memcpy in jim_newtap_expected_id
Found by clang static checker.
On the very first call of jim_newtap_expected_id() pTap->expected_ids
and expected_len are null, and there's nothing to copy. This patch
changes this cryptic code to use realloc() instead.
Nemui Trinomius [Mon, 16 Mar 2015 15:00:40 +0000 (00:00 +0900)]
lpc2000: Removed cmd51_can_xxxxb variables.
Current flash driver can fail to write across the sector boundary.
This patch fixes "thisrun_bytes" set to "cmd51_dst_boundary" value instead of "cmd51_can_xxxxb"
Referred to SevenW's post and fix,thanks.
http://sourceforge.net/p/openocd/mailman/openocd-devel/thread/fa32e7d6654df22470dc5f4a3dbc984a%40familiebenschop.nl/#msg33594152
Tested on below listed chips.
LPC810,LPC811,LPC812,LPC824,LPC1115,LPC1343,LPC1347,LPC1227,
LPC1769,LPC1788,LPC54102,LPC4088 and LPC2388.
Ed Beroset [Tue, 3 Mar 2015 20:01:38 +0000 (15:01 -0500)]
startup: Fixed measure_clk to return kHz instead of MHz.
The original code had iterated 10,000,000 times and taken the elapsed
time divided by 10,000, to yield kHz which is mathematically correct
only if we were measuring time in seconds, but we are measuring time in
milliseconds, so the correct divisor is actually 10,000,000. Previous
code would report 0.500 for actual measured speed of 500 kHz.
Change-Id: Iba4c4961fe3973e7ccfa6dfa11d606a966ceb50c Signed-off-by: Ed Beroset <beroset@ieee.org>
Reviewed-on: http://openocd.zylin.com/2573 Tested-by: jenkins Reviewed-by: Jens Bauer <jens@gpio.dk> Reviewed-by: Paul Fertser <fercerpav@gmail.com>
Cortex-A: Don't flush the data/unified cache if MMU is off
When the SCTLR has C set but M unset (i.e. Caching on, but MMU off) the cache
if effectively off. So only flush the cache if MMU is on, otherwise stale
entries might be committed to memory.
Change-Id: Iaff8b6f25b7a41ba838b91d45684c98f99fc0b27 Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-on: http://openocd.zylin.com/2429 Tested-by: jenkins Reviewed-by: Christopher Head <chead@zaber.com> Reviewed-by: Paul Fertser <fercerpav@gmail.com> Reviewed-by: Vladimir Svoboda <ze.vlad@gmail.com>
Paul Fertser [Mon, 2 Mar 2015 14:52:47 +0000 (17:52 +0300)]
tcl/interface/parport: default to sane value on non-windows systems
When using ppdev driver 0 is the most appropriate default value as it
corresponds to /dev/parport0. Raw port address is suitable only for
direct access (I think that's parport-giveio on windows).
Reported by danitool on IRC.
Change-Id: I983c22251de6601b433ad31aaf660fb664cee7e9 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2572 Reviewed-by: Andreas Färber <afaerber@suse.de> Tested-by: jenkins Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Paul Fertser [Mon, 9 Mar 2015 15:55:03 +0000 (18:55 +0300)]
jtag/core: honour SRST timings in SWD mode
Since SWD is currently abusing the JTAG command queue for reset and
sleep handling (and all other operations are performed in a different
way), sleep needs to be forced explicitly to ensure correct timings.
Change-Id: I5b0da6cbb7d0560154e4077b261aa6828cefc892 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2591 Tested-by: jenkins Reviewed-by: Tomas Vanek <vanekt@fbl.cz> Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Felipe Balbi [Wed, 18 Mar 2015 21:01:00 +0000 (16:01 -0500)]
tcl: target: am437x: pass coreid
This commit is only for the sake of completeness as
default coreid is zero. In any case, coreids 1-4 are
used for the different PRU cores inside the SoC.
Change-Id: I775f2f444b1a908ffaf7bdbc43bcc966f19668c4 Signed-off-by: Felipe Balbi <balbi@ti.com>
Reviewed-on: http://openocd.zylin.com/2621 Tested-by: jenkins Reviewed-by: Tom Rini <trini@konsulko.com> Reviewed-by: Paul Fertser <fercerpav@gmail.com>
Felipe Balbi [Wed, 18 Mar 2015 20:50:53 +0000 (15:50 -0500)]
target: am437x: use more descriptive names
Use more descriptive names for JRC and DAPs
so they more closely match documentation.
For example there's no Cortex A9 DAP, that's
the DebugSS DAP where Cortex A9 target sits. In
that same DAP we have have ETM, STM and both
dual-PRU subsystems.
Change-Id: I0e66ebb6299763f96606fae3e4c62e5785c804f2 Signed-off-by: Felipe Balbi <balbi@ti.com>
Reviewed-on: http://openocd.zylin.com/2620 Tested-by: jenkins Reviewed-by: Tom Rini <trini@konsulko.com> Reviewed-by: Paul Fertser <fercerpav@gmail.com>
Felipe Balbi [Mon, 16 Mar 2015 04:15:15 +0000 (23:15 -0500)]
tcl: am437x: disable watchdog on reset-end
sometimes, watchdog might be left running and
it could expire in the middle of a debug session,
to prevent that, just make sure to disable watchdog
on reset-end if current state is 'halted'.
Change-Id: Ib4f2a2321cba17cd8c56ca3ae63114a563a6de90 Tested-by: Tom Rini <trini@konsulko.com> Signed-off-by: Felipe Balbi <balbi@ti.com>
Reviewed-on: http://openocd.zylin.com/2615 Tested-by: jenkins Reviewed-by: Tom Rini <trini@konsulko.com> Reviewed-by: Paul Fertser <fercerpav@gmail.com>
Felipe Balbi [Fri, 13 Mar 2015 22:10:42 +0000 (17:10 -0500)]
tcl: target: am437x.cfg: pass correct dbgbase
Since commit ec9ccaa28849 (arm_adi_v5: make dap_lookup_cs_component()
traverse subtables and handle multicore) AM437x devices can't be used
with OpenOCD anymore. The reason is that dbgbase used to be set to zero
before that commit and that just happens to work with AM437x devices.
A more robust solution is to pass correct dbgbase when creating the
target, which this commit does.
Signed-off-by: Felipe Balbi <balbi@ti.com>
Change-Id: Iaf2617804324de8094b25137943e08b84f14c75f
Reviewed-on: http://openocd.zylin.com/2602 Tested-by: jenkins Reviewed-by: Paul Fertser <fercerpav@gmail.com> Reviewed-by: Tom Rini <trini@konsulko.com>
DmitryShpak [Wed, 25 Mar 2015 11:31:16 +0000 (14:31 +0300)]
target/target.c: fixed rp check bug in asynchronous flash write algorithm.
Bug in read pointer check within flash write algorithm made incorrect check
if block size is more than 4 bytes (bug was detected with 16 bytes block size).
Angus Gratton [Tue, 24 Feb 2015 21:19:15 +0000 (08:19 +1100)]
transport: make 'transport select' auto-select the first available transport if not set
This should allow most of the existing configurations for older
versions to remain compatible without forcing the user to change his
or her config to explicitly select transport.
Also in some circumstances can remove the need to chain a "-c transport
select X" when building custom configs on the command line, which seems
like a common new user pitfall.
Paul Fertser [Wed, 11 Mar 2015 08:33:55 +0000 (11:33 +0300)]
target/adi_v5_swd, cortex_m: properly handle more cases requiring reconnect
This brings SWD reconnection procedure in line with the ARM
documentation and changes cortex_m reset procedure to make use of it.
The motivation behind this patch is to make SAM4L "reset" and "reset
halt" properly without SRST. The complication here is that EDBG issues
an additional read of DP_RDBUFF automatically right after writing
SYSRESETREQ, that leads to a FAULT which needs to be dealt with
properly. With this patch the very first ahbap_debugport_init DAP
access will make SWD layer properly reinitialise the link before
continuing.
Runtime tested with mbed CMIS-DAP + KL25 only.
Change-Id: Ic506f9db30931dfa60860036b83f73b897975909 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2596 Tested-by: jenkins Reviewed-by: Andrey Yurovsky <yurovsky@gmail.com> Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Paul Fertser [Wed, 11 Feb 2015 08:08:40 +0000 (11:08 +0300)]
server, target, cortex_m: add deinit_target to the API to free resources
This should facilitate dynamic target creation and removal.
Currently it helps with getting 0 bytes lost report from Valgrind on
exit (after talking to a nucleo board). However, 1,223,886 bytes in
5,268 blocks are still reachable which means the app holds pointers to
that data on exit. The majority comes from the jtag command queue,
there're also many blocks from TCL command registration.
Paul Fertser [Mon, 9 Feb 2015 13:43:58 +0000 (16:43 +0300)]
target/target: call event handlers around examine when polling resumes
The target might be using Tcl examine-start and examine-end handlers,
they need to be called when the target gets reexamined after polling
succeeds again.
Paul Fertser [Wed, 11 Feb 2015 07:51:17 +0000 (10:51 +0300)]
target: fix timer callbacks processing
Warning, behaviour change: before this patch if a timer callback
returned an error, the other handlers in the list were not called.
This patch fixes two different issues with the way timer callbacks are
called:
1. The function is not designed to be reentrant but a nested call is
possible via: target_handle timer event -> poll -> target events
before/after reexaminantion -> script_command_run ->
target_call_timer_callbacks_now . This patch makes function a no-op
when called recursively;
2. The current code can deal with the case when calling a handler
leads to its removal but not when it leads to removal of the next
callback in the list. This patch defers actual removal to consolidate
it with the calling loop.
These bugs were exposed by Valgrind.
Change-Id: Ia628a744634f5d2911eb329747e826cb9772e789 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2541 Tested-by: jenkins Reviewed-by: Stian Skjelstad <stian@nixia.no> Reviewed-by: Andreas Färber <afaerber@suse.de> Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Paul Fertser [Thu, 13 Mar 2014 09:27:45 +0000 (13:27 +0400)]
arm11: initialise DPM and register cache before reading DSCR for the first time
When target was already halted during the initial examination,
arm11_check_init() was trying to read, store and interpret DSCR
contents before the DPM structure is initialised. This caused
a segfault like described on
http://sourceforge.net/apps/trac/openocd/ticket/65 .
This is a totally untested attempt to fix this issue.
Change-Id: I2fff115679a3f0023e7a88c749ccb5f045d6cf01 Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/2043 Tested-by: jenkins Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Paul Fertser [Sun, 1 Apr 2012 13:18:02 +0000 (15:18 +0200)]
armv7m: add FPU registers support
This patch adds the fpv4-sp-d16 registers to the armv7m register set.
The work is inspired by Mathias K but takes a different approach:
instead of having both double and single presicion registers in the
cache this patch works only with the doubles and counts on GDB to
split the data in halves whenever needed.
Tested with HLA only (on an STM32F334 disco board).
Currently this patch makes all ARMv7-M targets report an FPU-enabled
target description to GDB. It shouldn't harm if the user is not trying
to access non-existing FPU. However, the plan is to make this depend
on actual FPU presence later.
Change-Id: Ifcc72c80ef745230c42e4dc3995f792753fc4e7a Signed-off-by: Mathias K <kesmtp@freenet.de>
[fercerpav@gmail.com: rework to fit target description framework] Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/514 Tested-by: jenkins Reviewed-by: Peter Stuge <peter@stuge.se> Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Change-Id: I1beccff885b5b37747edd0b2e9fb2297ce466a00 Signed-off-by: pierre Kuo <vichy.kuo@gmail.com>
Reviewed-on: http://openocd.zylin.com/2548 Tested-by: jenkins Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Marian Cingel [Fri, 3 Oct 2014 12:15:16 +0000 (14:15 +0200)]
rtos: Freescale MQX rtos support
ARMv7E-M (CortexM4) architecture
- fix position offset of r2,r3 registers on exception stack
- switch 'calloc' arguments
- remove prototypes of internal function and typedefs
- add NULL check for alloc functions
- remove last line of license "Franklin Street, Fifth Floor"
because of 'checkpatch' validation
- environment: jlink + twrk60n512
Change-Id: I70840ded15b17dd945ca190ce31e2775078da2d9 Signed-off-by: Marian Cingel <cingel.marian@gmail.com>
Reviewed-on: http://openocd.zylin.com/2353 Tested-by: jenkins Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
Arne Wichmann [Thu, 25 Sep 2014 05:21:24 +0000 (07:21 +0200)]
target & board: AT91SAM7A2 and Olimex SAM7-LA2
Initial Support for AT91SAM7A2 on Olimex SAM7-LA2 board.
The board seems not to be able to reset into halted mode, as srst is
connected to NRESET of the cpu (configured srst_pulls_trst).
JTAG RCLK is connected to CLK.
Tested with interface/ftdi/olimex-arm-usb-ocd-h.cfg.
Change-Id: I2bdd67e3683e45f1119c5850bad294aa107891d8 Signed-off-by: Arne Wichmann <arne.wichmann@gmail.com>
Reviewed-on: http://openocd.zylin.com/2318 Tested-by: jenkins Reviewed-by: Paul Fertser <fercerpav@gmail.com>
Tomas Vanek [Wed, 25 Feb 2015 21:21:58 +0000 (22:21 +0100)]
psoc4 flash driver: cleaned printf PRI... formats
Failed build on Mac OS X 10.10.2 was reported in OpenOCD-devel.
Cleaning types and printf formats. uint32_t prefered for flash/sector sizes.
2 minor changes in comments.
Removed redundant bracket.
Change-Id: Ia06b77af59c2c0ffd10869a4b263a760ca8b0a7a Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: http://openocd.zylin.com/2558 Tested-by: jenkins Reviewed-by: Paul Fertser <fercerpav@gmail.com> Reviewed-by: Xiaofan <xiaofanc@gmail.com>
Theodore A. Roth [Thu, 29 Jan 2015 00:23:16 +0000 (17:23 -0700)]
nrf51: Update known devices table.
Added new entries to the nrf51_known_devices_table array. New entries
are documented in the "nRF51 Series Compatability Matrix V1.0" found on
the Nordic Semi web site. Reordered entries to match the order found in
the document.
Also added an entry for an undocumented hwid discovered while flashing
the PCA10031 and PCA10028 dev boards.
Change-Id: Icca7da103d437dc28e651f27ab937fe953b9aac9 Signed-off-by: Theodore A. Roth <troth@openavr.org>
Reviewed-on: http://openocd.zylin.com/2514 Tested-by: jenkins Reviewed-by: Paul Fertser <fercerpav@gmail.com>
Remove long-deprecated "target count" and "target number" commands.
Given that the manual states that these two subcommands are
deprecated and were scheduled to be removed back in 2010,
remove them and the corresponding documentation from the
manual.
Change-Id: Iaac633349d7fcb8b7f964109c7d26dd0cc5fc233 Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Reviewed-on: http://openocd.zylin.com/1860 Tested-by: jenkins Reviewed-by: Paul Fertser <fercerpav@gmail.com>
Based on the initial work on bcm2835gpio.c by Paul Fertser with many
additions. Modifications to the GPIO handling was minimal in this
patch. A more big modification is required before cleanup the
interface between bitbang and sysfsgpio.
Change-Id: I54bf2a2aa2ca059368b0e0e105dff6084b73d624 Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>
Reviewed-on: http://openocd.zylin.com/2438 Tested-by: jenkins Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk> Reviewed-by: Paul Fertser <fercerpav@gmail.com>
This is based on the initial work by Paul Fertser with addition of the
switch sequences and new ACK handling. In case of WAIT response, the
sticky bits are cleared and the last operation is repeated. The ACK
handling is based on the interpretation of the 8 February 2006 ARM
Debug Interface v5 Architecture Specification
Change-Id: Id50855b1ffff310177ccf9883dc9eb0d1b4458c8 Signed-off-by: Jean-Christian de Rivaz <jc@eclis.ch>
Reviewed-on: http://openocd.zylin.com/2437 Tested-by: jenkins Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk> Reviewed-by: Paul Fertser <fercerpav@gmail.com> Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Matej Kupljen [Wed, 11 Feb 2015 08:28:41 +0000 (09:28 +0100)]
gdb_server: ignore stray + in ACK mode
I couldn't make OpenOCD to work with GDB. I was always getting this in GDB:
(gdb) target remote localhost:3333
Remote debugging using localhost:3333
Ignoring packet error, continuing...
Ignoring packet error, continuing...
Ignoring packet error, continuing...
Ignoring packet error, continuing...
Malformed response to offset query, timeout
(gdb)
While debugging gdb remote protocol, I have seen that gdb responds with:
w ++$?#3f
And those two '+' seems to confuse the OpenOCD parser, if it sees another
'+' sign it emits the DEBUG output and sets the noack_mode to 2. The
problem is that we weren't even IN noack mode, this was set to 0 and then
it explicitly sets it to 2 and thus turning the noack mode on.
Change-Id: If267c9226e57fa83121ded09cf69829f8f0b4b93 Signed-off-by: Matej Kupljen <matej.kupljen@gmail.com>
Reviewed-on: http://openocd.zylin.com/2545 Tested-by: jenkins Reviewed-by: Tomas Vanek <vanekt@fbl.cz> Reviewed-by: Paul Fertser <fercerpav@gmail.com>
Cortex A: fix extra memory read and non-word sizes
Without this patch, to perform a memory read, OpenOCD first issues an
LDC instruction into DBGITR in Stall mode (thus executing the
instruction), then switches to Fast mode and reads from DBGDTRTX once
for each word to transfer.
At the very end of the transfer, the final Fast mode read of DBGDTRTX
has, as always, the side effect of re-issuing the LDC instruction. This
causes two problems:
(1) If the word immediately beyond the end of the requested region is
inaccessible, this spurious LDC will cause a fault. On a fast CPU, the
LDC will finish executing by the time the poll of DSCR takes place,
failing the entire memory read. On a slow CPU, the LDC might finish
executing later, leaving an unexpected and confusing sticky fault lying
around for the next operation to see.
(2) If the LDC succeeds, it will leave the loaded word in DBGDTRTX, thus
setting DBGDSCR.TXFULL=1. The cortex_a_read_apb_ab_memory routine
completes without consuming that last word, thus confusing the next
routine that tries to use DBGDTRTX (this may not have any visible effect
on some implementations, because writing to DBGDTRTXint when TXFULL=1 is
defined as Unpredictable, but I believe it caused a visible problem for
me).
With this patch, the bulk mem_ap_sel_read_buf_noincr is modified to omit
the last word of the block. The second-to-last read of DBGDTRTX by that
function will cause the issue of the LDC for the last word. After
switching back to Normal mode and waiting for that instruction to
finish, do a final read of DBGDTRTX to extract the last word into the
buffer, leaving TXFULL=0.
Without this patch, memory accesses are always expanded such that they
are aligned to the access size. With this patch, accesses are issued
exactly as ordered by the caller. The caller is expected to handle
fragments at the beginning and end of the transfer if the address is
unaligned and an unaligned access is not desired.
Without this patch, the DFAR and DFSR registers, which report the
location and status of data faults, are ignored while performing memory
accesses, which could cause problems debugging an OS page fault handler.
With this patch, DFAR and DFSR are preserved across memory accesses, and
DFSR is decoded in the event of a synchronous fault to provide the
caller with more information about the reason for failure.
Thanks to Boris Brezillon for the original patch whose ideas led to the
non-word access mechanism implemented here and to various code reviewers
for their comments.
Change-Id: I11ae7104fbe69a522efadefc705c9a217a7eef41 Signed-off-by: Christopher Head <chead@zaber.com>
Reviewed-on: http://openocd.zylin.com/2381 Tested-by: jenkins Reviewed-by: Olivier Schonken <olivier.schonken@gmail.com> Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
projects / openocd / log