On designs using romboot (firmware embedded in ram blocks), we generally upload new firmwares with serialboot and netboot for prototyping.
Moving romboot after serialboot and netboot avoid manual interrupts of the boot sequence.
This is not strictly necessary to build libunwind (it can
be built with -DNDEBUG), but it will be handy while it is
debugged.
It can be removed afterwards.
This results in generation of .eh_frame sections. These sections
can be discarded during final linking, or included if exception
handling is desired. For exception handling to work, all sources
must be built with -fexceptions.
Note that -integrated-as is not active by default on OR1K,
so we're still shelling out to binutils to assemble.
It is not yet possible to build everything using -integrated-as.
`ranlib` is not necessary on any system we can possibly build for,
as it is superseded by `ar s` for the last ten years or so (at least).
Thus, change ar invocations to `ar crs`, also removing a `l` flag
that is ignored by binutils.
$(AS) was never used: $(assemble) invokes the C compiler instead.
In case of LLVM, this will allow us to consistently use the LLVM
internal assembler for both inline assembly in C and assembly
sources; so, avoid ever invoking binutils as explicitly.
* fixes the clearing of the rx ringbuffer on rx-overflow
* removes tx_level and tx_cts by restricting the ringbuffer
to at least one slot empty
* agnostic of the details of the tx irq: works for uarts that
generate tx interrupts on !tx-full or on tx-empty.
* only rx_produce and tx_consume need to be volatile
When firmware is small enough, it can be interesting to run code from an embedded blockram memory (faster and not impacted by memory controller activity).
It can also be a fallback option in case boot from flash failed.
To use this, define ROM_BOOT_ADDRESS and initialize the blockram with the firmware data.
Florent Kermarrec
whitequark
Sebastien Bourdeauducq
Robert Jordens
Yann Sionneau