Intel IXC1100, IXP42X - page 129

Intel® IXP42X Product Line of Network Processors and IXC1100 Control Plane Processor
September 2006 DM
Order Number: 252480-006US 129
Intel XScale® Processor—Intel® IXP42X product line and IXC1100 control plane processors
3.6.15.2.3 Dynamic Debug Handler
On the IXP42X product line and IXC1100 control plane processors, the debug handler
and override vector tables reside in the 2-Kbyte, mini instruction cache, separate from
the main instruction cache. A “static” Debug Handler is downloaded during reset. This
is the base handler code, necessary to do common operations such as handler entry/
exit, parse commands from the debugger, read/write ARM registers, read/write
memory, etc.
Some functions may require large amounts of code or may not be used very often. As
long as there is space in the mini-instruction cache, these functions can be downloaded
as part of the static Debug Handler. However, if space is limited, the debug handler also
has a dynamic capability that allows a function to be downloaded when it is needed.
There are three methods for implementing a dynamic debug handler (using the mini
instruction cache, main instruction cache, or external memory). Each method has their
limitations and advantages. “Dynamically Loading IC After Reset” on page 123
describes how do dynamically load the mini or main instruction cache.
• Using the Mini IC
The static debug handler can support a command which can have functionality
dynamically mapped to it. This dynamic command does not have any specific
functionality associated with it until the debugger downloads a function into the
mini instruction cache. When the debugger sends the dynamic command to the
handler, new functionality can be downloaded, or the previously downloaded
functionality can be used.
There are also variations in which the debug handler supports multiple dynamic
commands, each mapped to a different dynamic function; or a single dynamic
command that can branch to one of several downloaded dynamic functions based
on a parameter passed by the debugger.
Debug Handlers that allow code to be dynamically downloaded into the mini
instruction cache must be carefully written to avoid inadvertently overwriting a
critical piece of debug handler code. Dynamic code is downloaded to the way
pointed to by the round-robin pointer. Thus, it is possible for critical debug handler
code to be overwritten, if the pointer does not select the expected way.
To avoid this problem, the debug handler should be written to avoid placing critical
code in either way of a set that is intended for dynamic code download. This allows
code to be downloaded into either way, and the only code that is overwritten is the
previously downloaded dynamic function. This method requires that space within
the mini instruction cache be allocated for dynamic download, limiting the space
available for the static Debug Handler. Also, the space available may not be suitable
for a larger dynamic function.
Once downloaded, a dynamic function essentially becomes part of the Debug
Handler. Since it is in the mini instruction cache, it does not get overwritten by
application code. It remains in the cache until it is replaced by another dynamic
function or the lines where it is downloaded are invalidated.
•Using the Main IC.
The steps for downloading dynamic functions into the main instruction cache is
similar to downloading into the mini instruction cache. However, using the main
instruction cache has its advantages.
Using the main instruction cache eliminates the problem of inadvertently
overwriting static Debug Handler code by writing to the wrong way of a set, since
the main and mini instruction caches are separate. The debug handler code does
not need to be specially mapped out to avoid this problem. Also, space for dynamic
functions does not need to be allocated in the mini instruction cache and dynamic
functions are not limited to the size allocated.
The dynamic function can actually be downloaded anywhere in the address space.
The debugger specifies the location of the dynamic function by writing the address
to RX when it signals to the handler to continue. The debug handler then does a
branch-and-link to that address.