In order to get meaningful const properties for the PROGRAMMER, AVRPART and
AVRMEM arguments, some code needed to be moved around, otherwise a network of
"tainted" assignments risked rendering nothing const:
- Change void (*enable)(PROGRAMMER *pgm) to void (*enable)(PROGRAMMER *pgm,
const AVRPART *p); this allows changes in the PROGRAMMER structure after
the part is known. For example, use TPI, UPDI, PDI functions in that
programmer appropriate to the part. This used to be done later in the
process, eg, in the initialize() function, which "taints" all other
programmer functions wrt const and sometimes requires other finessing with
flags etc. Much clearer with the modified enable() interface.
- Move TPI initpgm-type code from initialize() to enable() --- note that
initpgm() does not have the info at the time when it is called whether or
not TPI is required
- buspirate.c: move pgm->flag to PDATA(pgm)->flag (so legitimate
modification of the flag does not change PROGRAMMER structure)
- Move AVRPART_INIT_SMC and AVRPART_WRITE bits from the flags field in
AVRPART to jtagmkII.c's private data flags32 fiels as FLAGS32_INIT_SMC and
FLAGS32_WRITE bits
- Move the xbeeResetPin component to private data in stk500.c as this is
needed by xbee when it saddles on the stk500 code (previously, the flags
component of the part was re-dedicated to this)
- Change the way the "chained" private data are used in jtag3.c whilst
keeping the PROGRAMMER structure read-only otherwise
- In stk500v2.c move the STK600 pgm update from stk500v2_initialize() to
stk500v2_enable() so the former keeps the PROGRAMMER structure read-only
(for const assertion).
- In usbasp change the code from changing PROGRAMMER functions late to
dispatching to TPI or regular SPI protocol functions at runtime; reason
being the decision whether to use TPI protocol is done at run-time
depending on the capability of the attached programmer
Also fixes Issue #1071, the treatment of default eecr value.
This commit replaces fixed-string buffers in PROGRAMMER, AVRPART and AVRMEM
that are dealt with by the parser and grammar. Now, string assignments are
always to const char *, ie, these are read-only strings with arbitrary
length.
config_gram.y now only needs to consider one type of string assignment.
This commit also
- Replaces the simple linear-search cache_string() function with faster
hashed cache_string(). Either way, the returned value is likely to be
shared, so should never be free()'d.
- Duplicates hvupdi_support list in pgm_dup() and frees it in pgm_free()
- Adds const qualifier to some function args in avrpart.c and pgm.c
- Hardens some functions against being called with NULL pointers
- Ensures _new() and _dup() functions for parts, programmers and memory
return a suitable memory. Out of memory triggers exit in one of three
functions, cfg_malloc(), cfg_realloc() and cfg_strdup(); there is
rarely anything useful that AVRDUDE or, for that matter, any
application compiled against libavrdude can do once you run out of
memory as AVRDUDE/libavrdude rely heavily on allocation of memory.
The paged read in pickit2.c has two errors:
- It drops load extended address commands unless a paged read happens at a
64k byte boundary; this is invalid when reading files with holes
- It wrongly assumed that flash memory is byte addressed
The fix is to carry out a load extended address command, if needed, at the
beginning of each paged flash read with the correct word address. Although the
pickit2_paged_load() has independent parameters page_size, addr and n_bytes,
AVRDUDE only ever calls paged read/write functions with page_size and n_bytes
both set to mem->page_size and addr aligned with a page boundary. Therefore, it
is sufficient to set the load extended address at the beginning of each page
read.
Stefan Rueger
Marius Greuel