The XDR library has a problem on architectures with short enums like the
default ARM EABI. Short enums means that the size of the enum type is
variable and the smallest integer type to hold all enum values will be
selected. For many enums this is char. The XDR library uses int32_t
for enum_t. There are several evil casts from an enum type to enum_t
which leads to invalid memory accesses on short enum architectures. A
workaround is to add appropriate dummy enum values.
The file size was wrong in the no space left on device condition. This
resulted in turn in a read of an invalid block which lead to an EIO
error status.
Read-ahead requests were previously executed in the context of the
reading task. This blocks the reading task until the complete read
with read-ahead transfer is finished. A read-ahead task is introduced
to off-load the read-ahead transfer. This allows the reading task to
work with the requested block more quickly. The read-ahead is triggered
after two misses of ascending consecutive blocks or a read hit of a
block read by the most-recent read-ahead transfer. The read-ahead
feature is configurable and can be disabled.
PowerPC cores with the SPE (Signal Processing Extension) have 64-bit
general-purpose registers. The SPE context switch code has been merged
with the standard context switch code. The context switch may use cache
operations to increase the performance. It will be ensured that the
context is 32-byte aligned (PPC_DEFAULT_CACHE_LINE_SIZE). This
increases the overall memory size of the context area in the thread
control block slightly. The general-purpose registers GPR2 and GPR13
are no longer part of the context. The BSP must initialize these
registers during startup (usually initialized by the __eabi() function).
The new BSP option BSP_USE_DATA_CACHE_BLOCK_TOUCH can be used to enable
the dcbt instruction in the context switch.
The new BSP option BSP_USE_SYNC_IN_CONTEXT_SWITCH can be used to enable
sync and isync instructions in the context switch. This should be not
necessary in most cases.
The chroot() is an implicit chdir("/"). Set the current directory to
"/" after session close. This helps to avoid references to mounted file
systems in dynamic media.
+ libbsp/sparc/shared/include/pci.h was largely a copy of
an older version of the cpukit pci.h. Removed much of the
contents and included <rtems/pci.h>.
+ sparc/*/pci*.c - Move to <rtems/pci.h> required updating
to use uint32_t for dword accesses.
+ Rename PCI_MULTI_FUNCTION to PCI_HEADER_TYPE_MULTI_FUNCTION
+ Define PCI_HEADER_TYPE_MULTI_FUNCTION in cpukit pci.h and remove
PCI_MULTI_FUNCTION definitions in C files.
+ Move PCI_INVALID_VENDORDEVICEID definitions from various C files
to cpukit pci.h
All constants that should have been in cpukit pci.h were moved there.
One naming style was corrected. This impacted if_wm.c.
All other contents were the same as powerpc/shared/pci.h so BSP
specific pci.h could be eliminated.
The header file <rtems/malloc.h> provides now also the malloc_walk()
prototype.
The malloc_walk() prototype reflects now the _Protected_heap_Walk() API.
The return status helps to print only in case of an error.
The pathconf_limits_and_options field of
rtems_filesystem_mount_table_entry_t is now a const pointer to reduce
the read-write memory demands of file system instances.
The scope of the file system operations is the file system instance.
The scope of the file system node handlers is the file location. The
benefit of moving the operations to the mount table entry is a size
reduction of the file location (rtems_filesystem_location_info_t). The
code size is slightly increased due to additional load instructions.
Restructure rtems_filesystem_mount_table_entry_t to improve cache
efficiency.
It is now the responsibility of the read() and write() handler to update
the offset field of the IO descriptor (rtems_libio_t). This change
makes it possible to protect the IO descriptor from concurrent access by
per file locks.
These modifications were required by hand after running the script.
In some cases, the file names did not match patterns. In others,
the format of the file did not match any common patterns.
Script does what is expected and tries to do it as
smartly as possible.
+ remove occurrences of two blank comment lines
next to each other after Id string line removed.
+ remove entire comment blocks which only exited to
contain CVS Ids
+ If the processing left a blank line at the top of
a file, it was removed.
The fat_file_datasync() read every cluster of the file into the cache
and then synchronized it step-by-step. For unmodified buffers this is a
non-operation. For modified buffers this will wake-up the swapout task
which performs then a single buffer write operation. This is usually
quite inefficient. Firstly we do single buffer writes, secondly we
may perform a lot of unnecessary read operations (for huge files this is
really bad), and thirdly this leads likely to cache evictions.
The synchronization procedure is replaced by a simple
rtems_bdbuf_sync_dev(). This has the side-effect that also buffers not
related to the file are synchronized, but since the modified list is
normally short this should be acceptable.
