If bsp_early_malloc() is called early during boot room will be
allocated after BSS END. If the function is called after boot
is will call malloc() instead. The returned memory is not freeable
and always 8-byte aligned.
If the bsp_early_malloc() isn't called the function is not
dragged in and the workspace will be unmodified in size.
Signed-off-by: Daniel Hellstrom <daniel@gaisler.com>
Fixed a bug where the vector number is used to clean and unmask
the IRQ at the IRQ controller, the irq number must be used.
Signed-off-by: Daniel Hellstrom <daniel@gaisler.com>
Adds to confdefs a way to specify rtems_resource_unlimited for classic and
posix objects using a new macro CONFIGURE_OBJECTS_UNLIMITED.
Use CONFIGURE_OBJECTS_ALLOCATION_SIZE to declare the allocation size for
extending the set of objects at runtime. Updates the unlimited sample
to demonstrate how to use the new macros. Also adds new documentation in
the C User's Manual regarding configuring with unlimited objects.
When data cache snooping is not present the cache needs
flushing, the SPARC LEON CPUs does not have to ability
to flush individual cache rows and flushing all cache is
expensive. Instead the LDA instruction is used to force
cache miss on individual loads during the IP-align copy
operation required anyway.
GRETH GBIT non-snooping systems are still unsupported,
since it use zero-copy (can deal with unaligned DMA).
Let the bsp.h select if the GRETH driver is supported.
Currently only the LEON2/LEON3 platforms BSPs builds the
driver.
Signed-off-by: Daniel Hellstrom <daniel@gaisler.com>
Moved the intelligence whether the driver is supported or not
completely to the BSP, now that bsp.h is included (even though it
is a driver... hmm). The ERC32 was never supported, so HAS_SMC91111
is not added to erc32/include/bsp.h.
Signed-off-by: Daniel Hellstrom <daniel@gaisler.com>
The low level routines can be used in different occasions, it will be
required when accessing PCI.
Note the difference between byteorder.h (inlined functions) and access.S
where the functions will be declared in the library archive librtemscpu.a.
Function names starting with _ are in library and can be referenced by
function pointers.
Signed-off-by: Daniel Hellstrom <daniel@gaisler.com>
The compaction process needs erased blocks. It is only possible to
erase an entire segment. Thus in order to make a progress we always
need enough erased blocks to empty a used or available segment which can
be erased in turn. A (possibly the worst case) lower bound of erased
blocks is the block count of the largest segment. The number of
unavailable blocks specified by the configuration will be used to
determine the erase blocks starvation situation. The number of
unavailable blocks must be greater than or equal to the number of blocks
in the largest segment.
In case rtems_bdbuf_read() returns an error status, the block device
buffer pointer will be set to NULL. In RFS the chain node of the block
device buffer will be used for RFS purposes. We must not do this after
an erroneous read.
The functions
o rtems_bdbuf_get(),
o rtems_bdbuf_read(),
o rtems_bdbuf_syncdev(), and
o rtems_bdbuf_purge_dev(),
use now the disk device instead of the device identifier. This makes
bdbuf independent of rtems_disk_obtain() and rtems_disk_release(). It
is the responsiblity of the file system to obtain the disk device. This
also reduces the overhead to get a buffer.
The key for the AVL tree uses now the disk device instead of the device
identifier. The pointer is interpreted as an unsigned integer. This
reduces the memory overhead and makes the comparison operation a bit
faster.
Removed function rtems_bdbuf_purge_major(). This function was too
destructive and could have unpredictable side effects.
