The definition for malloc() says that it should return a pointer
to the allocated memory which is suitably aligned for any built-in
type. This requirement was lost in commit 0c15627cc1 ("lib: Remove
sys_mem_pool implementation") where the entire memory pool used to
have an explicit alignment of 16.
Fix this by allocating memory with sys_heap_aligned_alloc() using
__alignof__(z_max_align_t) which will automatically get the needed
alignment on each platform.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
The commit adds initializations of fs_file_t variables in preparation
for fs_open function change that will require fs_file_t object, passed
to the function, to be initialized before first usage.
Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>
An assignment from one multi-word union field to another was not safe
from corruption. Copy the value out to a local value before storing it
to the preferred union field.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
This allows applications that may not use minimal libc avoid the cost
of a second printf-like formatting infrastructure by using printfcb()
instead of printf() for output. It also helps make sure that the
formatting support (e.g. floats) is consistent between user-directed
output and the logging infrastructure.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
Calculate crc32 4 bits at a time. The return value of the calculation is
identical to the previous 1 bit at a time implementation.
Results in a speed up of a factor 3 at the cost of using 64 bytes of
flash for a crc table.
Calculating crc32 of 128kB of flash on a 120MHz Kinetis MKE16F512
Cortex-M4 takes 99ms using the 1 bit at a time implementation, and 30ms
using the 4 bits at a time implementation.
The crc32 routine is used by subsys/canbus/canopen/canopen_program.c to
calculate crc of flash images.
Signed-off-by: Klaus H. Sorensen <khso@vestas.com>
This option allows forcing big heap mode. Useful on for getting 8-byte
aligned blocks on 32-bit machines.
Signed-off-by: Martin Åberg <martin.aberg@gaisler.com>
reallocarray() is defined in terms of realloc(). From OpenBSD manual
pages:
"Designed for safe allocation of arrays, the reallocarray()
function is similar to realloc() except it operates on nmemb
members of size size and checks for integer overflow in the
calculation nmemb * size."
The return value of sys_heap_realloc() is not compatible with that
of realloc().
Signed-off-by: Martin Åberg <martin.aberg@gaisler.com>
Previously, newlib claimed all free physical memory in the
system.
Now, the kernel manages this, allowing for memory to be
used via k_mem_map() calls.
Establish an upper bound to how much newlib will try to
claim on system startup, instead of trying to take all
of it, allowing other parts of the system to also map
anonymous memory.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
We now draw heap memory from an anonymous memory mapping
instead of a hard-coded region past the kernel image,
which is no longer mapped by default.
Some readability cleanups were made to a particuarly
horrible set of nested ifdefs. A few types were adjusted.
sbrk()'s count argument is an intptr_t, not an int.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
directly convert ticks to nsecs in the clock_* posix
functions which will provide the best resolution the
system allows
Signed-off-by: Nicholas Lowell <nlowell@lexmark.com>
The strategy used in z_heap_aligned_alloc() was to allocate an extra
align-sized memory block for storing a pointer to the memory heap.
This is wasteful in terms of memory usage when alignment is larger
than a pointer width. A loop is needed to find the initial memory
start when freeing it which isn't optimal either.
Instead, let's have sys_heap_aligned_alloc() rewind a pointer after
it is aligned to make just enough room for storing our heap reference.
This way the heap reference is always located immediately before the
aligned memory and any unused memory is returned to the heap.
The rewind and alignment values may coincide in which case only
the alignment is necessary anyway.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Previously, newlib claimed all free physical memory in the
system.
Now, the kernel manages this, allowing for memory to be
used via k_mem_map() calls.
Establish an upper bound to how much newlib will try to
claim on system startup, instead of trying to take all
of it, allowing other parts of the system to also map
anonymous memory.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
We now draw heap memory from an anonymous memory mapping
instead of a hard-coded region past the kernel image,
which is no longer mapped by default.
Some readability cleanups were made to a particuarly
horrible set of nested ifdefs. A few types were adjusted.
sbrk()'s count argument is an intptr_t, not an int.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
Macros like INT64_C(x) convert x to a constant integral expression,
i.e. one that can be used in preprocessor code. Implement wrappers
that use the GNUC intrinsics to perform the translation.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
Provide data structures to capture a timestamp in two different
clocks, monitor the drift between those clocks, and using a base
instant with estimated drift convert between the clocks.
This provides the core technology to convert between system uptime and
an external continuous time scale like TAI (UTC without applying leap
seconds).
Signed-off-by: Peter A. Bigot <pab@pabigot.com>
Let's do it upfront only once for each entry point and dispense
with overflow checks later to keep the code simple.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Fixes: #28650
Linking with newlib now defines the following linker flags as:
```
${CMAKE_C_LINKER_WRAPPER_FLAG}${CMAKE_LINK_LIBRARY_FLAG}c
${CMAKE_C_LINKER_WRAPPER_FLAG}${CMAKE_LINK_LIBRARY_FLAG}gcc
c
```
This is needed because when linking with newlib on aarch64, then libgcc
has a link dependency to libc (strchr), but libc also has dependencies
to libgcc.
CMake is capable of handling circular link dependencies for CMake
defined static libraries, which can be further controlled using
`LINK_INTERFACE_MULTIPLICITY`.
However, libc and libgcc are not regular CMake libraries, and is seen as
linker flags by CMake, and thus symbol de-duplications will be
performed.
CMake link options cannot be used, as that will place those libs first
on the linker invocation. -Wl,--start-group is problematic as the
placement of -lc and -lgcc is not guaranteed in case later libraries are
also using -lc / -libbgcc as interface linker flags.
Thus, we resort to use
`${CMAKE_C_LINKER_WRAPPER_FLAG}${CMAKE_LINK_LIBRARY_FLAG}`
as this ensures the uniqueness and thus avoids symbol de-duplication
which means libc will be followed by libgcc, which is finally followed
by libc again.
It would have been possible to use `-lc` directly, but there is a risk
that an externally library is also adding `-lc` and thus de-duplication
and re-arrangement of this flag happens. This risk is in theory also
existing with this fix, but the long nature of this link flag with using
`${CMAKE_C_LINKER_WRAPPER_FLAG}` would likely indicate a similar fix and
thus those libraries will stay in order.
Signed-off-by: Torsten Rasmussen <Torsten.Rasmussen@nordicsemi.no>
The 'fputs' has flaw in the implementation. It almost always
returns 'EOF' even if completed successfully.
This happens because we compare 'fwrite' return value which is
"number of members successfully written" (which is 1 in current
implementation) to the total string size:
----------------------------->8-----------------------
int fputs(const char *_MLIBC_RESTRICT string,
FILE *_MLIBC_RESTRICT stream)
{
int len = strlen(string);
int ret;
ret = fwrite(string, len, 1, stream);
return len == ret ? 0 : EOF;
}
----------------------------->8-----------------------
In result 'fputs' return 'EOF' in case of string length bigger
than 1.
There are several fixes possible, and one of the fixes is to
swap number of items (1) with size (string length) when we
are calling 'fwrite'. The only difference will be that
'fwrite' will return actual numbers of bytes written which
can be compared with the string length.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
First, the maximum heap size must fit in 31 bits worth of chunks
because the internal 32-bit field holding the size is shared with
the `used` bit.
Then the mention of a 256-byte block in the doc is no longer
relevant. That pertained to the previous allocator implementation.
And ditto for the HEAP_MEM_POOL_MIN_SIZE kconfig option.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
This adds a somewhat special purpose IPC mechanism. It's intended for
applications which have a "work queue" like architecture of discrete
callback items, but which need the ability to schedule those items
independently in separate threads across multiple CPUs. So P4 Work
items:
1. Can run at any Zephyr scheduler priority and with any deadline
(this feature assumes EDF scheduling is enabled)
2. Can be submitted at any time and from any context, including being
resubmitted from within their own handler.
3. Will preempt any lower priority work as soon as they are runnable,
according to the standard rules of Zephyr priority scheduling.
4. Run from a pool of worker threads that can be allocated efficiently
(i.e. you need as many as the number of CPUs plus the number of
preempted in-progress items, but no more).
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The l length modifier can apply to the c format specifier; in that
case the expected value is of type wint_t. Minimal libc doesn't
define wint_t, and it is complex to do so correctly (must add
<wchar.h>, and use a lot of conditional tricks).
wint_t can differ from wchar_t in rank when wchar_t undergoes default
integral promotion, which it does on xtensa (wchar_t is unsigned
short). So we can use wchar_t as an approximation, except in va_arg
where we need to use a wider type: int covers this case.
Note that we still don't format wide characters, but we do want to
consume the correct amount of data for a default-promoted extended
character.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
Whether char is signed or unsigned is toolchain and target specific.
Rather than assume it's signed (which is true for x86, but not for
ARM), do the right thing based on whether the minimum representable
value is less than zero.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
It may not be clear that the length modifiers reference native C types
with specific ranks. Document the core type for each modifier.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
This function was designed to support the logging infrastructure's
need to copy values from va_list structures. It did not meet that
need, since some values need to be changed based on additional data
that is only available when the complete format specification is
examined. Remove the function as unnecessary.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
Providing a literal width or precision that exceeds the non-negative
range of int does not appear to be rejected by the standard, but it
does produce a build diagnostic so we can't test it. Switch to an
equivalent form that doesn't affect line coverage.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
Just like commit 0ae04f01b6 ("lib/os/heap: make some checks more
assertive") we shouldn't validate the externally provided align
argument only when CONFIG_SYS_HEAP_VALIDATE is set.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
If the new size amounts to the same number of chunks then:
- If right-chunk is used then we needlessly allocate new memory and
copy data over.
- If right-chunk is free then we attempt to split it with a zero size
which corrupts the prev/next list.
Make sure this case is properly handled and add a test for it.
While at it, let's simplify the code somewhat as well.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
If a precision flag is included for s formatting that bounds the
maximum output length, so we need to use strnlen rather than strlen to
get the amount of data to emit. With that flag we can't expect there
to be a terminating NUL following the text to print.
Also fix handling of an empty precision, which should behave as if a
precision of zero was provided.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
While documenting the float conversion code, I found there was room
for some optimization. In doing so I added test cases to cover edge
cases e.g. making sure proper rounding is applied and that no loss
of precision was introduced. Compiled code should be smaller and
faster.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Use the core k_heap API pervasively within our tree instead of the
z_mem_pool wrapper that provided compatibility with the older mempool
implementation.
Almost all of this is straightforward swapping of one alloc/free call
for another. In a few cases where code was holding onto an old-style
"mem_block" a local compatibility struct with a single field has been
swapped in to keep the invasiveness of the changes down.
Note that not all the relevant changes in this patch have in-tree test
coverage, though I validated that it all builds.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Add an optimized realloc() implementation that can successfully expand
allocations in place if there exists enough free memory after the
supplied block.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Mark all k_mem_pool APIs deprecated for future code. Remaining
internal usage now uses equivalent "z_mem_pool" symbols instead.
Fixes#24358
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The biggest required padding is equal to `align - chunk_header_bytes`
and not `align - 1` given that the header already contributes to the
padding.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Most of kernel files where declaring os module without providing
log level. Because of that default log level was used instead of
CONFIG_KERNEL_LOG_LEVEL.
Signed-off-by: Krzysztof Chruscinski <krzysztof.chruscinski@nordicsemi.no>
LLVM building for qemu_x86 appears to have an optimization bug where a
union that is assigned to hold values read from va_args() is inferred
to be a constant value, so is placed in ROM with an all-zero content.
Prevent this by packing the conversion state and the value union into
a single container structure that's stack allocated.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
When no message was get from kscan_msgq queue, the prev state
(x/y point) was processed as a new state. During process its
coordinates could be inverted or modified, depending on the
LVGL_POINTER_KSCAN_SWAP_XY, LVGL_POINTER_KSCAN_INVERT_X,
LVGL_POINTER_KSCAN_INVERT_Y or display orientation configuration.
In these cases, it could cause wrong input data.
Signed-off-by: Robin-Charles Guihéneuf <robin-charles@hotmail.fr>
Add support for abs with additional integer types.
This is needed to make LLVM quiet and stop warning about abs being used
with int64_t and such.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
So far data that went to stderr was simply dropped in case of minimal
libc. In case of newlib stderr was treated same like stdout
(e.g. fprintf(stderr, ...) was equivalent to fprintf(stdout, ...).
Extend filter on stream pointer to allow both stdout and stderr to pass
data to stdout hook (which is Zephyr console backend in most cases).
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
Although flags with pointers are not defined behavior, there is a
desire to have them work, so add a test and fix the complete
implementation so it passes.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
