BSP Developer's Guide

VxWorks 5.5
BSP Developer’s Guide
The combination of copyback cache without snooping is particularly dangerous,
although the risk is reduced if all user buffers are positioned so that they do not
share cache lines with any other buffer. The user can insure that the front end of
any user buffer is aligned on a cache boundary by setting the variable
memDefaultAlignment to the cache line size. This results in all malloc( ) requests
being aligned on the start of a cache line.
The user can protect the rear end of any buffer by increasing the size of the memory
request by one cache line. This guarantees that no other buffer shares a cache line
with this buffer. Having done all this, memory buffers for DMA operations are
relatively safe from the effect of memory coherency in a copyback situation.
Many new processors implement write pipes that can buffer write operations
when the bus controller is busy. This requires device drivers to make use of the
CACHE_PIPE_FLUSH macros for generic drivers. A CACHE_PIPE_FLUSH operation
should be inserted between any I/O write operation and a I/O read operation. If
a routine begins with an I/O read then you should assume that an I/O write
operation precedes it.
4.2.3 MMU Support
VxWorks supports several different MMUs through a virtual memory library.
Because virtual memory is inappropriate for some real-time applications,
VxWorks can be configured to not include virtual memory support.
For more information on VxWorks virtual memory support, see the following:
■
the reference entries for vmBaseLib and vmLib
■
■
VxWorks Programmer’s Guide: Virtual Memory Interface
VxWorks Programmer’s Guide: Configuration
■
4.5 Enabling the Virtual Memory Library, p.83
4.2.4 Floating-Point Support
Floating point is supported as a tasking extension to avoid costly context switch
times for tasks that do not use floating-point operations. Tasks can be spawned
with a special floating-point option, and context switch callouts provide the
mechanism to initialize, save, and restore a floating-point context. By switching
floating-point data and control registers in and out, each task effectively shares the
single floating-point unit.
66