A number of the ROM functions were pointing to wrappers that saved and
restored the $a0 register, but the functions they called either did
not use $a0 or saved and restored $a0 as needed anyway.
Memory layout is now split into two linker scripts for OTA vs
non-OTA (different starting offsets), remaining functionality in
common.ld.
As discussed in #64, progress towards #38
mbedTLS version 2.1.0 (current stable)
Has some known issues/hacks:
* Entropy source not hooked in at all
* Linker script has a messy hack in it to store some (not all) data in
irom
Prerequisite for calling Cache_Read_Disable() for OTA updates.
Massive thanks due to @foogod and their xtobjdis tool, creating the call
graph of where the wDev_ProcessFiq touched (including in newlib, etc.)
would have otherwise been very painful:
https://bitbucket.org/foogod/xtobjdis
At the moment these are almost bit-for-bit copies of the vector
behaviour in the RTOS SDK, but hopefully that will start to change.
Progress towards #2
IRAM/IROM sections.
esp-open-rtos compiled code: .text goes into irom by default, symbols
explicitly marked with IRAM attribute go into iram.
SDK code: .text goes into iram by default, symbols marked with
ICACHE_FLASH_ATTR go into irom.
libgcc functions also go into iram by default.
Closes#7.
It'd be nice to change the linker script so .text is always linked to
IROM instead of needing to objcopy each compiled file. The sticking
point for this is libgcc & libhal, which have symbols in .text that need
to be loaded to IRAM.
* This fixes the problem of axTLS symbols hmac_md5/hmac_sha1 having same
name as symbols in libwpa (which have incompatible signatures)
* Also allows for easier identification and piece-by-piece removal of
binary functions.
* Some libc symbols are not renamed, list is in lib/symbols_norename.txt
