The correct way to overrride the malloc family of functions in newlib-nano is
to provide the *_r (reentrant) variants. Newlib implements the "normal"
functions on top of these (see the newlib source code). Also, internally it calls
the *_r functions when allocating buffers.
If only the "normal" non-reentrant functions are provided this will mean that
some of the code will still use the vanilla newlib allocator. Furthermore, if
one uses the whole heap as a pool for TLSF then the system may in the best case
crash as there is no enough memory for its internall allocations or in the worst
case function eratically (this depends on how the heap reserved, there is an
upcomming series of commits in that direction).
This commit splits the handling between newlib and native. It also prepares the
ground for future work on the pool initialization.
Right now I could only test this in ARM and native and I cannot ensure it will
work on other platforms. Replacing the system's memory allocator is not something
that can be taken lightly and will inevitably require diving into the depths of
the libc. Therefore I would say that using TLSF as a system wide allocator is ATM
supported officially only on those plaftorms.
Testing:
Aside from reading the newlib sources, you can see the issue in a live system
using the debugger.
Compile any example (with or without tlsf-malloc), grab a debugger and place
a breakpoint in sbrk and _sbrk_r. Doing a backtrace will reveal it gets called
by _malloc_r.
A (void*) function was declared as (void**) because one of the void pointers
was hidden behind a typedef. Because of the way a void* works, this has no
consequences, but it is confusing.
Use test_utils_interactive_sync for synchronizing some case treat
the output before `reset` as the start of the test,
which fails for some boards/configurations.
Add an implementation that waits for 's' to print 'START' and return.
If 'r' is given is prints 'READY' to allow querying for state.
The help and answered string have to be different to not match the other.
Using puts/getchar was smaller than using `stdio_read/stdio_write` on the
example I tested with `esp32`.
This remove executing buildtest `for` loop in docker.
When building completely in docker, 'buildtest' would hide issues when
the host toolchain would be used when doing `make all` directly.
It has the consequence that it now starts a container for each
compilation which is slower.
The previous behavior can be reproduced by using
BUILD_IN_DOCKER=1 make buildtest-indocker
A side effect is also that now `BUILDTEST_MAKE_REDIRECT` would work when
doing `buildtest` with docker.
Add a 'buildtest-indocker' that forces executing 'buildtest' for loop
completely inside the container.
It prevents starting one container per compilation wich is slower but
it could hide errors where the host toolchain would be used
It is currently equivalent to `buildtest` but will change when the
`buidtest` handling will be move outside of `BUILD_IN_DOCKER`.
Display an error when executed without BUILD_IN_DOCKER=1.
Memory management function like `malloc`, `calloc`, `realloc` and `free` must not be preempted when they operate on allocator structures. To avoid such a preemption, wrappers around these functions are used which simply disable all interrupts for the time of their execution.
When running tests using 'nordic_softdevice_ble', the 'softdevice.hex'
file must also be taken into account for the test hashing and be
uploaded to the separated murdock testing boards.
When listed as dependency from `test-input-hash` the file must have a
target, which he has not. The file is implicitly created when compiling
`ELFFILE` so declare it as order only dependency.
In practice `BASELIBS` could be enough or even `pkg-build-softdevice`
but I do not want to be depend that much on internals there.
if xtimer_set spins the callback is executed in the thread context.
comment to explain irq_disable
and when this line could be removed
(when xtimer stops executing the callback funtion from thread context)
