Expose most DWT registers via Tcl; there are a few more, but
those are mostly for profiling along with the ITM. Having
this set available enables operations which aren't possible
with just the standard watchpoint operations.
The cycle counter may be interesting. Turn it on after reset
by setting the LSB of the dwt_ctrl register, and it counts
CPU clocks. You can program the comparator 0 watchpoint to
trigger on a given cycle count, rather than a data address.
Likewise, comparator 1 may be able to match data values given
address matches from one or two other comparators. (Not all
hardware supports this capability though; try it. That is
something the standard watchpoint methods should eventually
handle, for the single address case.)
Minor cleanup: remove needless functional indirection for
exposing the v7m architctural registers.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
There's no reason to read which interrupts are enabled from
the NVIC; that state isn't used. Plus, it's highly dynamic
since firmware can change it at any time; remove the support
for those state records.
Remove duplicate definition of DWT_CTRL address; shrink a line.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Fix the watchpoint error checks, and do them in add(), not later
in set() when it's mostly too late. Support the full range of
watchpoint sizes (1 to 32K bytes each), and check alignments.
Minor cleanup of DWT access: shrink lines, use "+" for address
calculations, comment a few issues. Add debug message reporting
DWT capabilities, matching the message for FBP, and some minor
code and spec review comments.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Add Doxygen for the exported ARMv7-M interfaces.
Make the non-exported stuff static. Remove functions and
data which are now observably unused.
Add comment about a small speedup that the run_algorithm()
logic could use. Shrink a few too-long lines.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
improve default target->read/write_phys_memory, produce
more sensible error messages if the mmu interface
functions have not been implemented yet vs. will
not be implemented(e.g. cortex m3).
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
The quit entry point was not being invoked. Just a source
of confusion at this point. XScale ran 100x reset upon
quit, but that code made no sense, wasn't commented
and never invoke.
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Previous patch somehow made GCC lose some of its cookies;
work around, zero-init that struct.
Clean up code from the previous patch.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
To support breakpoints, flush data cache line and invalidate
instruction cache when 4 and 2 byte words are written.
The previous code was trying to write directly to the physical
memory, which was buggy and had a number of other situations
that were not handled.
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Fixed bug: if virtual address for working memory was not specified
and MMU was enabled, then address 0 would be used.
Require working address to be specified for both MMU enabled
and disabled case.
For some completely inexplicable reason this fixes the regression
in svn 2646 for flash write in arm926ejs target. The logs showed
that MMU was disabled in the case below:
https://lists.berlios.de/pipermail/openocd-development/2009-November/011882.html
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
This change is necessary to debug AT91SAM9260 on my PC with a
FT2232H dongle.
Signed-off-by: Dimitar Dimitrov <dinuxbg@gmail.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Just use the array of names we're given, ignoring indices.
The "reserved means don't use" patch missed that change.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
I'm suspecting this code can never have worked, since the
original commit (svn #335) in early 2008.
Fix is just copy/paste from another (working) function.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Implement XSVF support for detailed state path transitions,
by collecting sequences of XSTATE transitions into paths
and then calling pathmove().
It seems that the Xilinx tools want to force state-by-state
transitions instead of relying on the standardized SVF paths.
Like maybe there are XSVF tools not implementing SVF paths,
which are all that we support using svf_statemove().
So from IRPAUSE, instead of just issuing "XSTATE DRPAUSE"
they will issue XSTATES for each intermediate state: first
IREXIT2, then IRUPDATE, DRSELECT, DRCAPTURE, DREXIT1, and
finally DRPAUSE. This works now.
Handling of paths that go *through* reset is a trifle dodgey,
but it should be safe.
Tested-by: Wookey <wookey@wookware.org>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Unneeded exports cause confusion about the module interfaces.
Make most functions static, and fix some line-too-long issues.
Delete some now-obviously-unused code.
The forward decls are just code clutter; move their references
later, after the normal declarations. (Or vice versa.)
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Unneeded exports cause confusion about the module interfaces.
Only the Feroceon code builds on this, so only routines it
reuses should be public.. Make most remaining functions
static, and fix some of the line-too-long issues.
The forward decls are just code clutter; move their references
later, after the normal declarations. Turns out we don't need
even one forward declaration in this file.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
The Hex parser uses a fixed number of sections. When the
number of sections in the file is greater than that, the
stack get corrupted and a CHECKSUM ERROR is detected
which is very confusing.
This checks the number of sections read, and increases
IMAGE_MAX_SECTIONS so it works on my file.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Only type 1 branch instruction has a condition code, not type 2.
Currently they're both tagged with ARM_B which doesn't allow for the
distinction.
Signed-off-by: Nicolas Pitre <nico@marvell.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
A Thumb BLX instruction is branching to ARM code, and therefore the
first 2 bits of the target address must be cleared.
Signed-off-by: Nicolas Pitre <nico@marvell.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
This patch includes partial support for these new JTAG adapters.
More complete support will require updates to the libftdi code,
for EEPROM access.
[dbrownell@users.sourceforge.net: fix whitespace, linelen, etc ]
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Currently, OpenOCD is always caching the PC value without the T bit.
This means that assignment to the PC register must clear that bit and set
the processor state to Thumb when it is set. And when the PC register
value is transferred to another register or stored into memory then
the T bit must be restored.
Discussion: It is arguable if OpenOCd should have preserved the original
PC value which would have greatly simplified this code. The processor
state could then be obtained simply by getting at bit 0 of the PC. This
however would require special handling elsewhere instead since the T bit
is not always relevant (like when PC is used with ALU insns or as an index
with some addressing modes). It is unclear which way would be simpler in
the end.
Signed-off-by: Nicolas Pitre <nico@marvell.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Whenever an unconditional branch with the H bits set to 0b10 is met, the
offset must be combined with the offset from the following opcode and not
ignored like it is now.
A comment in evaluate_b_bl_blx_thumb() suggests that the Thumb2 decoder
would be a simpler solution. That might be true when single-stepping of
Thumb2 code is implemented. But for now this appears to be the simplest
solution to fix Thumb1 support.
Signed-off-by: Nicolas Pitre <nico@marvell.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Calling it first with every opcodes and then testing if the opcode
was indeed a branch instruction is wasteful and rather strange.
If ever thumb_pass_branch_condition() has side effects (say, like
printing a debugging traces) then the result would be garbage for most
Thumb instructions which have no condition code.
While at it, let's make the nearby code more readable by reducing some of
the redundant brace noise and reworking the error handling construct.
Signed-off-by: Nicolas Pitre <nico@marvell.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Make the "dap info" output more comprehensible:
- Don't show CIDs unless they're incorrect (only four bits matter)
- For CoreSight parts, interpret the part type
- Interpret the part number
- Show all five PID bytes together
- Other minor cleanups
Also some whitespace fixes, and shrink a few overlong source lines.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Ignore leading '0' characters on hex strings. For example a bit
pattern consisting of 6 bits could be written as 3f, 03f or 003f and
so on.
Signed-off-by: Michael Roth <mroth@nessie.de>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
This patch adds basic autoprobing support for the JTAG scan chains
which cooperate. To use, you can invoke OpenOCD with just:
- interface spec: "-f interface/...cfg"
- possibly with "-c 'reset_config ...'" for SRST/TRST
- possibly with "-c 'jtag_khz ...'" for the JTAG clock
Then set up config files matching the reported TAPs. It doesn't
declare targets ... just TAPs. So facilities above the JTAG and
SVF/XSVF levels won't be available without a real config; this is
almost purely a way to generate diagnostics.
Autoprobe was successful with most boards I tested, except ones
incorporating C55x DSPs (which don't cooperate with this scheme
for IR length autodetection). Here's what one multi-TAP chip
reported, with the "Warn:" prefixes removed:
clock speed 500 kHz
There are no enabled taps. AUTO PROBING MIGHT NOT WORK!!
AUTO auto0.tap - use "jtag newtap auto0 tap -expected-id 0x2b900f0f ..."
AUTO auto1.tap - use "jtag newtap auto1 tap -expected-id 0x07926001 ..."
AUTO auto2.tap - use "jtag newtap auto2 tap -expected-id 0x0b73b02f ..."
AUTO auto0.tap - use "... -irlen 4"
AUTO auto1.tap - use "... -irlen 4"
AUTO auto2.tap - use "... -irlen 6"
no gdb ports allocated as no target has been specified
The patch tweaks IR setup a bit, so we can represent TAPs with
undeclared IR length.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Stop allocating three bytes per IR bit, and cope somewhat better
with IR lengths over 32 bits.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Remove needless debug handler state.
- "handler_installed" became wrong as soon as the second TRST+SRST
reset was issued ... so the handler was never reloaded after the
reset removed it from the mini-icache.
This fixes the bug where subsequent resets fail on PXA255 (if the
first one even worked, which is uncommon). Other XScale chips
would have problems too; PXA270 seems to have, IXP425 maybe not.
- "handler_running" was never tested; it's pointless.
Plus a related bugfix: invalidate OpenOCD's ARM register cache on reset.
It was no more valid than the XScale's mini-icache. (Though ... such
invalidations might be better done in "SRST asserted" callbacks.)
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Bit 5 shouldn't be used. Remove all support for modifying it.
Matches the exception vector table, of course ... more than one
bootloader uses that non-vector to help distinguish valid boot
images from random garbage in flash.
Some cosmetic cleanup, and switch to a single table mapping
between state names and symbols (vs two routines which only
share that state with difficulty).
Get rid of TAP_NUM_STATES, and some related knowledge about
how TAP numbers are assigned. Later on, this will help us
get rid of more such hardwired knowlege.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
- Use the name mappings all the other code uses:
+ name-to-state ... needed to add one special case
+ state-to-name
- Improve various diagnostics:
+ don't complain about a "valid" state when the issue
is actually that it must be "stable"
+ say which command was affected
- Misc:
+ make more private data and code be static
+ use public DIM() not private dimof()
+ shorten the affected lines
Re the mappings, this means we're more generous in inputs we
accept, since case won't matter. Also our output diagnostics
will be a smidgeon more informative, saying "RUN/IDLE" not
just "IDLE" (emphasizing that there can be side effects).
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
The wrong variable (pc instead of r0) was used. Furthermore, someone
did cover this error by stupidly silencing the compiler warning that
occurred before a dummy void reference to r0 was added to the code.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
When dumping over 100 registers (as on most ARM9 + ETM cores),
aid readability by splitting them into logical groups.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
The register names are perversely not documented as zero-indexed,
so rename them to match that convention. Also switch to lowercase
suffixes and infix numbering, matching ETB and EmbeddedICE usage.
Update docs to be a bit more accurate, especially regarding what
the "trigger" event can cause; and to split the issues into a few
more paragraphs, for clarity.
Make "configure" helptext point out that "oocd_trace" is prototype
hardware, not anything "real".
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
David Brownell
Øyvind Harboe
Krzysztof Kajstura
Dimitar Dimitrov
Spencer Oliver
Michael Roth
Franck HÉRÉSON
Nicolas Pitre
Oleg Seiljus