| Age | Commit message (Collapse) | Author |
|---|
|
Define a target_addr_t type to support 32-bit and 64-bit addresses at
the same time. Also define matching TARGET_PRI*ADDR format macros as
well as a convenient TARGET_ADDR_FMT.
In targets that are 32-bit (avr32, nds32, arm7/9/11, fm4, xmc1000)
be least invasive by leaving the formatting unchanged apart from the
type;
for generic code adopt TARGET_ADDR_FMT as unified address format.
Don't silently change gdb formatting here, leave that to later.
Add COMMAND_PARSE_ADDRESS() macro to abstract the address type.
Implement it using its own parse_target_addr() function, in the hopes
of catching pointer type mismatches better.
Add '--disable-target64' configure option to revert to previous 32-bit
target address behavior.
Change-Id: I2e91d205862ceb14f94b3e72a7e99ee0373a85d5
Signed-off-by: Dongxue Zhang <elta.era@gmail.com>
Signed-off-by: David Ung <david.ung.42@gmail.com>
[AF: Default to enabling (Paul Fertser), rename macros, simplify]
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Matthias Welwarsky <matthias.welwarsky@sysgo.com>
|
|
Avoid special rules to generate array at compile time by shipping
the generated file. Convert to Makefile build like the other
loaders.
Change-Id: I5a05edddcfaff3d395086cd3aa33120f8a7aa9dc
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-on: http://openocd.zylin.com/3864
Tested-by: jenkins
Reviewed-by: Paul Fertser <fercerpav@gmail.com>
|
|
With -Og gcc doesn't perform as many optimizations, and as a result
warns about some code that it wouldn't otherwise warn about.
These fixes all assign values to otherwise uninitialized variables.
Change-Id: I9a6ea5eadd73673891ecfec568b8b00d78b596a5
Signed-off-by: Tim Newsome <tim@sifive.com>
Reviewed-on: http://openocd.zylin.com/3779
Tested-by: jenkins
Reviewed-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-by: Freddie Chopin <freddie.chopin@gmail.com>
|
|
Also make GPL notices consistent according to:
https://www.gnu.org/licenses/gpl-howto.html
Change-Id: I84c9df40a774958a7ed91460c5d931cfab9f45ba
Signed-off-by: Marc Schink <openocd-dev@marcschink.de>
Reviewed-on: http://openocd.zylin.com/3488
Tested-by: jenkins
Reviewed-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Freddie Chopin <freddie.chopin@gmail.com>
Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
|
|
Before this change jim_target_reset() checked examined state of a target
and failed without calling .assert_reset in particular target layer
(and without comprehensible warning to user).
Cortex-M target (which refuses access to DP under active SRST):
If connection is lost then reset process fails before asserting SRST
and connection with MCU is not restored.
This resulted in:
1) A lot of Cortex-M MCUs required use of reset button or cycling power
after firmware blocked SWD access somehow (sleep, misconfigured clock etc).
If firmware blocks SWD access early during initialization, a MCU could
become completely inaccessible by SWD.
2) If OpenOCD is (re)started and a MCU is in a broken state unresponsive
to SWD, reset command does not work even if it could help to restore communication.
Hopefully this scenario is not possible under full JTAG.
jim_target_reset() in target.c now does not check examined state
and delegates this task to a particular target. All targets have been checked
and xx_assert_reset() (or xx_deassert_reset()) procedures were changed
to check examined state if needed. Targets except arm11, cortex_a and cortex_m
just fail if target is not examined although it may be possible to use
at least hw reset. Left as TODO for developers familiar with these targets.
cortex_m_assert_reset(): memory access errors are stored
instead of immediate returning them to a higher level.
Errors from less important reads/writes are ignored.
Requested reset always leads to a configured action.
arm11_assert_reset() just asserts hw reset in case of not examined target.
cortex_a_assert_reset() works as usual in case of not examined target.
Change-Id: I84fa869f4f58e2fa83b6ea75de84440d9dc3d929
Signed-off-by: Tomas Vanek <vanekt@fbl.cz>
Reviewed-on: http://openocd.zylin.com/2606
Tested-by: jenkins
Reviewed-by: Matthias Welwarsky <matthias@welwarsky.de>
Reviewed-by: Paul Fertser <fercerpav@gmail.com>
|
|
Change-Id: I1a3afbddcf950689da58e0df8850a05f558d7879
Signed-off-by: Marc Schink <openocd-dev@marcschink.de>
Reviewed-on: http://openocd.zylin.com/3222
Tested-by: jenkins
Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
|
|
Obsolete C source code semicolons were removed using the semantic patch
semicolon/semicolon.cocci, see coccinellery.org
Change-Id: I153b4995a9e028ebaf5f58c947821dc78345a777
Signed-off-by: Alexander Kurz <akurz@blala.de>
Reviewed-on: http://openocd.zylin.com/3367
Tested-by: jenkins
Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
|
|
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: I900a0787812cb24d1f74ca50eb6bb4f85375a353
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-on: http://openocd.zylin.com/2478
Tested-by: jenkins
Reviewed-by: Paul Fertser <fercerpav@gmail.com>
|
|
Change-Id: I875cfab8dec4ade72ed9c9cd7d52baaca182a1ef
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/2295
Tested-by: jenkins
|
|
Remove this underutilized feature. Despite the fact that a lot of configs
specifies a arbitrary "variant", only the xscale target actually defines
any.
In the case of xscale, the use of -variant is dubious since
1) it's used as a redundant irlen specifier,
2) it carries a comment that it doesn't really need it and
3) only two xscale configs even specify it.
If there's a future target that needs a variant set, a target specific
option could be added when needed.
Change-Id: I1ba25a946f0d80872cbd96ddcc48f92695c4ae20
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-on: http://openocd.zylin.com/2283
Tested-by: jenkins
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
Don't hardcode the type for the array, just output the array initializer
so the includer can choose the type and storage class, zero-terminate at
will and so on.
Change-Id: I6d5e0710eaaba0a218b3eb32f6569177356f4462
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-on: http://openocd.zylin.com/2176
Tested-by: jenkins
Reviewed-by: Paul Fertser <fercerpav@gmail.com>
|
|
Found by clang.
Change-Id: Ifb25dca52f8d9e8e46a35f0947a7239f26eb3757
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/2067
Tested-by: jenkins
Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-by: Jörg Wunsch <openocd@uriah.heep.sax.de>
|
|
The out only version of jtag_add_dr_scan smells like a bogus optimization
that complicates the minidriver API for questionable gain.
The function was only used by four old ARM targets. Rewrite the callers
to use the generic function and remove all implementations.
Change-Id: I13b643687ee8ed6bc9b6336e7096c34f40ea96af
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-on: http://openocd.zylin.com/1801
Tested-by: jenkins
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
This eliminates most of the warnings reported when building for
arm-none-eabi (newlib).
Hsiangkai, there're many similar warnings left in your nds32 files, I
didn't have the nerve to clean them all, probably you could pick it
up.
Change-Id: Id3bbe2ed2e3f1396290e55bea4c45068165a4810
Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Reviewed-on: http://openocd.zylin.com/1674
Tested-by: jenkins
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
Change-Id: Ia9afa83752d17f0f56bdf3bd81f5c69d108aa5e9
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/1537
Tested-by: jenkins
Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
|
|
Let the default handler issue an unsupported warning rather than using
empty handler routines that may/may not issue a unsupported warning.
Change-Id: Iafe3e45146981a4cfae39771c3ab7370ac86da48
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/1535
Tested-by: jenkins
Reviewed-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-by: Hsiangkai Wang <hsiangkai@gmail.com>
|
|
On XScale architecture, to write debug control register DCSR
and activate JTAG debug (ie. to choose Halt Mode), the
enabling can only be done while the board is held in reset
state (ie. PXAxx #RST line held low).
The current implementation writes to the register before
asserting the SRST line. Swap the order to activate the SRST
line before writing to DCSR.
Change-Id: I914b9d53d39bdeb5fe4ee5e11068cafafe0da4d2
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Reviewed-on: http://openocd.zylin.com/1458
Tested-by: jenkins
Reviewed-by: Marek Vasut <marek.vasut@gmail.com>
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
Replace hexadecimal masks for vector catching with words
representing the caught exception, such as dabt for data
abort, etc ...
This way, the new xscale command is :
- xscale vector_catch
Reads back to the user the current vector catching status
- xscale vector_catch reset dabt pabt
Sets the caught vectors to data abort and prefetch abort
for example.
This is mostly taken from Cortex-M3 openocd code.
Change-Id: I66591d5796f0e07f0f31edc8d28722e1e48aa8c5
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Reviewed-on: http://openocd.zylin.com/1456
Tested-by: jenkins
Reviewed-by: Marek Vasut <marek.vasut@gmail.com>
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
Change-Id: I429f7fd51f77b0e7c86d7a7f110ca31afd76c173
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/1426
Tested-by: jenkins
Reviewed-by: Paul Fertser <fercerpav@gmail.com>
|
|
Seems commit fc2abe63fd3cea7497da7be2955d333bd3f800b9 caused a regression
in that the arm reg cmd no longer worked. The issue was caused because we
changed the value of ARM_MODE_THREAD which was being checked in arm_init_arch_info.
Change-Id: Id571d4ab336d1b0e2b93363147af245d24b65ca5
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/1362
Tested-by: jenkins
Reviewed-by: Luca Bruno <lucab@debian.org>
Reviewed-by: Freddie Chopin <freddie.chopin@gmail.com>
|
|
Remove dummy implementations from all targets except arm7_9 and mips, which
are the only ones with real implementations. Replace with a single default
implementation simply calling target_write_memory().
Change-Id: I9228104240bc0b50661be20bc7909713ccda2164
Signed-off-by: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Reviewed-on: http://openocd.zylin.com/1213
Tested-by: jenkins
Reviewed-by: Øyvind Harboe <oyvindharboe@gmail.com>
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
Change-Id: Ia055b6d2b5f6449a38afd0539a8c66e7d7e0c059
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/430
Tested-by: jenkins
|
|
Nothing more than a name change, just to make reading
the code a bit simpler.
Change-Id: I73a16b7302b48ce07d9688162955aae71d11eb45
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/390
Tested-by: jenkins
Reviewed-by: Øyvind Harboe <oyvindharboe@gmail.com>
|
|
we should have caught them all - hopefully.
Change-Id: I35435317fccaf5ad0216244d69f76db6857bb582
Signed-off-by: Spencer Oliver <spen@spen-soft.co.uk>
Reviewed-on: http://openocd.zylin.com/381
Tested-by: jenkins
Reviewed-by: Øyvind Harboe <oyvindharboe@gmail.com>
|
|
On wrong parameters a error is signalized to the calling function.
Change-Id: I484443fdb39938e20382edc9246d5ec546a5c960
Signed-off-by: Mathias K <kesmtp@freenet.de>
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Reviewed-on: http://openocd.zylin.com/282
Tested-by: jenkins
Reviewed-by: Øyvind Harboe <oyvindharboe@gmail.com>
|
|
Change-Id: I6dee51e1fab1944085391f274a343cdb9014c7a4
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Reviewed-on: http://openocd.zylin.com/300
Tested-by: jenkins
Reviewed-by: Øyvind Harboe <oyvindharboe@gmail.com>
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
this particular edge case of the JTAG API will no
longer be supported.
the in_value buffer must be provided by the caller when
the callback needs the buffer.
Change-Id: I552c72a64af6875f4aa4fa9b923194dcf3b57b64
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Reviewed-on: http://openocd.zylin.com/265
Tested-by: jenkins
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
- fix error propagation
- add some debug output for "unused" variables
- reduce scope of variable to avoid "unread" variables.
Change-Id: I136995f2a043102e64b8fff3017502fae564013b
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Reviewed-on: http://openocd.zylin.com/175
Tested-by: jenkins
Reviewed-by: Spencer Oliver <spen@spen-soft.co.uk>
|
|
The code in xscale_receive() that tries to skip invalid reads (i.e.
reads that don't have the DBG_SR[0] 'valid' bit set) seems to be
wrong, as it only looks at the first word's valid flag rather than
each word's own valid flag. Am I reading the code correctly? If so,
the attached patch should fix it.
If this looks correct, I'll generate a proper patch and commit message.
Matt
Change-Id: I74ebe2ad7a36d340a9dd3b8487578b6ea7f3cf1e
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Reviewed-on: http://openocd.zylin.com/32
Tested-by: jenkins
Reviewed-by: Øyvind Harboe <oyvindharboe@gmail.com>
|
|
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
it would *read* instead of *write* to memory
when the MMU was disabled.
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
Do not propagate error number to user. This is for internal
programming purposes only. Error messages to the user is
reported as text via LOG_ERROR().
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
Hi everyone,
Since a call went out for patches... been sitting on this for months. For some
reason, the xscale trace buffer is automatically disabled as soon as a break
occurs and the trace data is collected. This patch was a result of the
frustration of always re-enabling it, or else hitting a breakpoint and checking
the trace data, only to discover that I forgot to re-enable it before resuming.
Don't see why it should work this way. There is no run-time penalty, AFAIK.
Along the way, I also cleaned up a little by removing the ugly practice of
recording wrap mode by setting the fill count variable to "-1", replacing it
with an enum that records the trace mode.
I've been using this for months. Comments, criticisms gratefully received.
Mike
Signed-off-by: Mike Dunn <mikedunn@newsguy.com>
|
|
These warnings are for architectures that do not
support non-aligned word access.
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
Hi everyone,
A while back I sent in a patch that adds support for watchpoint lengths greater
than four on xscale. It's been working well, until the other day, when it
caused an unexpected debug exception. Looking into this I realized there is a
case where it breaks: when the length arg is greater than the base address.
This is a consequence of the way the hardware works. Don't see a work-around,
so I added code to xscale_add_watchpoint() to check for and disallow this
combination.
Some more detail... xscale watchpoint hardware does not support a length
directly. Instead, a mask value can be specified (not to be confused with the
optional mask arg to the wp command, which xscale does not support). Any bits
set in the mask are ignored when the watchpoint hardware compares the access
address to the watchpoint address. So as long as the length is a power of two,
setting the mask to length-1 effectively specifies the length. Or so I thought,
until I realized that if the length exceeds the base address, *all* bits of the
base address are ignored by the comaparator, and the watchpoint range
effectively becomes 0 .. length.
Questions, comments, criticisms gratefully received.
Thanks,
Mike
Signed-off-by: Mike Dunn <mikedunn@newsguy.com>
|
|
Hi everyone,
Added more LOG_ERROR messsages to watchpoint and breakpoint code, given that the
infrastructure no longer interprets returned error codes. Also changed
existing LOG_INFO and LOG_WARNING to LOG_ERROR for cases where an error is
returned.
Note that the check of the target state is superflous, since the infrastruture
code currently checks this before calling target code. Is this being
reconsidered as well? Also, should we stop returning anything other than
ERROR_OK and ERROR_FAIL?
Comments gratefully received.
Thanks,
Mike
Signed-off-by: Mike Dunn <mikedunn@newsguy.com>
|
|
Hi everyone,
Version 2 of this patch. Code added to breakpoints.c was removed from previous
patch, and item 3 added, per discussion with Øyvind regarding error reporting.
Item 4 added, which I just noticed.
I tried to use a software breakpoint in thumb code on the xscale for the first
time recently, and was surprised to find that it didn't work. The result was
this patch, which does four things:
1): fix trivial cut-n-paste error that caused thumb breakpoints to not work
2): call xscale_set_breakpoint() from xscale_add_breakpoint()
3): log error on data abort in xscale_write_memory()
4): fixed incorrect error code returned by xscale_set_breakpoint() when no
breakpoint register is available; added comment
Item 2 not only makes the xscale breakpoint code consistent with other targets,
but also alerts the user immediately if an error occurs when writing the
breakpoint instruction to target memory (previously, xscale_set_breakpoint() was
not called until execution resumed). Also, calling xscale_breakpoint_set() as
part of the call chain starting with handle_bp_command() and propagating the
return status back up the chain avoids the situation where OpenOCD "thinks" the
breakpoint is set when in reality an error ocurred.
Item 3 provides a helpful message for a common reason for failure to set sw
breakpoint.
This was thoroughly tested, mindful of the fact that breakpoint management is
somewhat dicey during single-stepping.
Comments and criticisms of course gratefully received.
Mike
Signed-off-by: Mike Dunn <mikedunn@newsguy.com>
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
Hi everyone,
This simple patch fixes a problem I noticed on the xscale where incorrect values
are sometimes reported by the reg command. The problem can occur when
requesting the value of registers in the xscale-specific register cache. With a
couple of exceptions, none of the registers in the xscale register cache are
automatically retrieved on debug entry. This is probably fine, as they are
unlikely to be needed on a regular basis during a typical debug session, and
they can be retrieved when explicitly requested by name using the reg command.
The problem is that once this is done, the register remains marked as valid for
the remainder of the OpenOCD session, and the reg command will henceforth always
report the same value because it is obtained from the cache and is never again
retrieved from the debug handler on the target.
The fix is to mark all registers in the xscale register cache as invalid on
debug entry (before the two exceptions are retrieved), thus forcing retrieval
(when requested) from the target across resumptions in execution, and avoiding
the reporting of stale values.
Small addition change by Øyvind: change 'i' to unsigned to fix compiler
warning for xscale_debug_entry() fn.
Signed-off-by: Mike Dunn <mikedunn@newsguy.com>
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
error was returned instead of using assert.
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
Parameters "domain" and "ap" of function armv4_5_mmu_translate_va()
are not used.
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
|
|
Parameter "type" of function armv4_5_mmu_translate_va()
is now not used.
Remove the parameter and the "enum" listing its values.
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
|
|
Function armv4_5_mmu_translate_va() now properly signals
errors in the return value.
Remove former error handling by setting variable "type" to
value "-1".
Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com>
|
|
The return value for MMU translation was a mess, either
error or value.
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
This patch adds support for the length argument to the xscale implementation of
the wp command. Per discussion with David, the length argument specifies the
range of addresses over which a memory access should generate a debug exception.
This patch utilizes the "mask" feature of the xscale debug hardware to implement
the correct functionality of the length argument. Some limitations imposed by
the hardware are:
- The length must be a power of two, with a minumum of 4.
- Two data breakpoint registers are available, allowing for two watchpoints.
However, if the length of a watchpoint is greater than four, both registers
are used (the second for a mask value), limiting the number of watchpoints
to one.
This patch also removes a useless call to xscale_get_reg(dbcon) in
xscale_set_watchpoint() (value had already been read from the register cache,
and the same previously read value is then modified and written back).
I have been using and testing this patch for a couple days.
Questions, corrections, criticisms of course gratefully received.
|
|
This patch fixes the xscale_analyze_trace() function. This function was
defective for a trace collected in 'fill' mode (hiccups with repeated
instructions) and completely broken when buffer overflowed in 'wrap' mode. The
reason for the latter case is that the checkpoint registers were interpreted
incorrectly when two checkpoints are present in the trace (which will be true in
'wrap' mode once the buffer fills). In this case, checkpoint1 register will
contain the older entry, and checkpoint0 the newer. The original code assumed
the opposite. I eventually gave up trying to understand all the logic of the
function, and rewrote it. I think it's much cleaner and understandable now. I
have been using and testing this for a few weeks now. I'm confident it hasn't
regressed in any way.
Also added capability to handle (as best as possible) the case where an
instruction can not be read from the loaded trace image; e.g., partial image.
This was a 'TODO' comment in the original xscale_analyze_trace().
Outside of xcsale_analyze_trace(), these (related) changes were made:
- Remove pc_ok and current_pc elements from struct xscale_trace. These elements
and associated logic are useless clutter because the very first entry placed
in the trace buffer is always an indirect jump to the address at which
execution resumed. This type of trace entry includes the literal address in
the trace buffer, so the initial address of the trace is immediately
determined from the trace buffer contents and does not need to be recorded
when trace is enabled.
- Added num_checkpoints to struct xscale_trace_data, which is necessary in order
to correctly interpret the checkpoint register contents.
- In xscale_read_trace()
- Fix potential array out-of-bounds condition.
- Eliminate partial address entries when parsing trace (can occur in wrap mode).
- Count and record number of checkpoints in trace.
- Added small, inlined utility function xscale_display_instruction() to help
make the code more concise and clear.
TODO:
- Save processor state (arm or thumb) in struct xscale_trace when trace is
enabled so that trace can be analyzed correctly (currently assumes arm mode).
- Add element to struct xscale_trace that records (when trace is enabled)
whether vector table is relocated high (to 0xffff0000) or not, so that a
branch to an exception vector is traced correctly (curently assumes vectors
at 0x0).
|
|
Problem: halt at a breakpoint, enable trace buffer ('xscale trace_buffer enable
fill'), then resume. Wait for debug exception when trace buffer fills (if not
sooner due to another breakpoint, vector catch, etc). Instead, never halts.
When halted explicitly from OpenOCD and trace buffer dumped, it contains only
one entry; a branch to the address of the original breakpoint. If the above
steps are repeated, except that the breakpoint is removed before resuming, the
trace buffer fills and the debug exception is generated, as expected.
Cause: related to how a breakpoint is stepped over on resume. The breakpoint is
temporarily removed, and a hardware breakpoint is set on the next instruction
that will execute. xscale_debug_entry() is called when that breakpoint hits.
This function checks if the trace buffer is enabled, and if so reads the trace
buffer from the target and then disables the trace (unless multiple trace
buffers are specified by the user when trace is enabled). Thus you only trace
one instruction before it is disabled.
Solution: kind of a hack on top of a hack, but it's simple. Anything better
would involve some refactoring. This has been tested and trace now works as
intended, except that the very first instruction is not part of the trace when
resuming from a breakpoint.
TODO: still many issues with trace: doesn't work during single-stepping (trace
buffer is flushed each step), 'xscale analyze_trace' works only marginally for
a trace captured in 'fill' mode, and not at all for a trace captured in 'wrap'
mode.
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
|
|
This patch fixes xscale software breakpoints by cleaning the dcache and
invalidating the icache after the bkpt instruction is inserted or removed. The
icache operation is necessary in order to flush the fetch buffers, even if the
icache is disabled (see section 4.2.7 of the xscale core developer's manual).
The dcache is presumed to be enabled; no harm done if not. The dcache is also
invalidated after cleaning in order to safeguard against a future load of
invalid data, in the event that cache_clean_address points to memory that is
valid and in use.
Also corrected a confusing typo I noticed in a comment.
TODO (or not TODO...?): the xscale's 2K "mini dcache" is not cleaned. This
cache is not used unless the 'X' bit in the page table entry is set. This is a
proprietary xscale extension to the ARM architecture. If a target's OS or
executive makes use of this for memory regions holding code, the breakpoint
problem will persist. Flushing the mini dcache requires that 2K of valid
cacheable memory (mapped with 'X' bit set) be designated by the user for this
purpose. The debug handler that gets downloaded to the target will also need to
be extended.
|
