Patent Application
20080126877
The present invention pertains to microprocessors, in general, and to a microprocessor debug feature that provides trace capability, in particular.
To debug micro-processor code it is desirable to provide trace functionality. A call trace involves providing the target of every call instruction and return instruction. This provides basic program flow data to the user for program analysis.
A branch trace involves providing the target of every branch instruction in addition to the calls and returns. This provides detail to the user for program analysis.
In addition to call and branch data, it is desirable to provide context switch and interrupt data for user. For a context switch, it is desirable to provide the context number.
The mechanism for obtaining this data on a standard microprocessor is to instrument the customer's code, libraries, etc. Problems with this approach include the fact that code is modified in order to capture the data necessary for the analysis.
Some micro-processors provide a trace feature that does not have a performance overhead. Such micro-processors require a dedicated trace interface on the micro-processor chip and a dedicated hardware tool connected to the micro-processor any time the trace function is to be used.
In accordance with the principles of the invention, a method for providing trace functionality for a microprocessor comprising an execution unit comprises: providing a memory coupled to the execution unit, utilizing the memory to store trace data during a trace operation; and providing hardware utilizable during a trace operation to assist in the trace operation.
In accordance with the principles of the invention the method is operable with a microprocessor comprising an execution unit operable in one or a plurality of contexts. The method comprises: providing a buffer memory coupled to the execution unit. The buffer memory is utilized to store trace data during operation of the execution unit. The method includes providing a trace control coupled to the execution unit and to the buffer memory; providing a breakpoint control coupled to the execution unit, the trace control and the buffer memory; providing a plurality of sets of programmable registers, and each set comprising a base address register, a data register, a data mask register and a control register; and providing in the trace control a programmable trace control register and a programmable base write address register, the trace control register being programmable to determine one of a plurality of trace functions.
In accordance with an aspect of the invention the buffer may be filled with data trace rather than code execution trace. In this case the user specifies a memory location or range of locations and, optionally, a data mask to filter the contents of writes to the memory, when a write occurs to the specified location(s) the data written is mirrored to the trace buffer. An example application would be to trace on all data received over a UART to try to debug associated driver code.
In accordance with another aspect of the invention, a timed trace function is also provided. In accordance with the timed trace, in addition to writing data to the trace buffer, the contents of a timer register are also written out. This provides a timestamp for each value/address written to the buffer, giving the user detailed execution timing for either code or a communications interface.
The invention will be better understood from a reading of the following detailed description of an illustrative embodiment of the invention in conjunction with the drawing figures in which:
A feature of the debug environment in accordance with the principles of the invention is that trace is saved to the standard address space. That is, the user can enable call trace or branch trace, and the results are written to any memory in the system as specified by the user.
For example, if a user enables call trace into external RAM, then each time that an instruction that causes a subroutine call is executed, the address of the routine being called is written to the memory range specified when setting up the trace function. When a return from subroutine instruction is executed, the address that is returned to is written into the trace buffer. These two elements enable a user with the help of appropriate tools to examine program flow after the fact by dumping and inspecting the trace buffer.
A branch trace involves storing the target of every branch instruction in addition to the calls and returns described above. This provides additional detail to a user for analysis.
In addition to calls and branches, context switches and interrupt data are stored so that a user can know about these kinds of changes of code flow as well. On a context switch, the context number is also stored.
As noted above one mechanism for obtaining trace data on a standard microprocessor is to instrument the customer's code, libraries, etc., but a problem with this approach is that that code is modified in order to capture the data necessary for the analysis. In accordance with the principles of the invention no modification of the code or reloading of code is necessary. In addition the inventive approach also entails less overhead on the processor.
A block diagram of a microprocessor in accordance with the principles of the invention that provides the trace functionalities described above is shown in
Execution unit 901 has connections to a memory interface 903. The connections include Write and Read control lines, Address and Data buses. The memory interface 903 provides connections to any existing memory utilized by execution unit 901 including a trace buffer memory, that may be a designated portion of an existing memory or a separate trace buffer memory.
A memory 911 is shown coupled to memory interface 903. Memory 911 may be a memory existing on the same chip as the execution unit 901, another memory associated with execution unit 901 or a separate memory dedicated as a trace buffer.
The Write and Read control lines, Address and Data buses are coupled to a Trace Control 905 and a Breakpoint Control 907.
Trace data can be written to memory in one of three ways:
Another option for using the buffer memory 911 is to fill it with data trace rather than code execution trace. In this case the user specifies a memory location or range of locations and, optionally, a data mask to filter the contents of writes to the memory 911, when a write occurs to the specified location(s) the data written is mirrored to the trace buffer. An example application would be to trace on all data received over a UART to try to debug associated driver code.
The trace feature is controlled by a group of registers shown in
One register is the Trace Control Register 101. The Trace Control Register 101 shown in
The Enable field is a one bit field enabled by software, break/watch point and disabled by software, break/watch point or when the trace buffer memory is full. The two states that this field defines are:
The Full Flag field comprises one bit that is used by simple buffer mode. The bit is set when the buffer fills and is cleared by software writing a 1 to this bit. If the bit is set, hitting break/watch point will not re-enable tracing.
The Timing Mode field is a one bit field. When the bit=0, the Timing is disabled; when the bit=1, the Timing is enabled. This enables a timer value to be written out for each write to buffer (A7).
The Trace Mode field is a two-bit field that defines four states by the binary code:
The Buffer Mode field specifies a trace buffer mode. Three of the codes are used as follows:
The Size field is a four bit field used to define the buffer size (used in the case of the buffer modes). The codes and associated sizes are:
A Base Address Register 201 is provided as a 32 bit register as shown in
For a Single address mode, all data is written to the same address. This would likely be captured by a logic analyzer and post processed as needed.
For a Circular Buffer mode, data is written to a memory buffer and when the memory buffer is full, it loops around to the beginning. The Base Address Register 201 always points to the next location to be written, i.e., it is incremented as the buffer is written, then the lower bits are cleared when the end of the buffer is reached. In circular buffer mode, when the end of the buffer is reached, a Full Flag is set and remains set until cleared by software.
For a Simple Buffer mode, once the buffer is filled to capacity, tracing is disabled, i.e., the Enable bit is cleared, and break/watch points cannot re-enable until the Full Flag is cleared.
The contents of the Base Address Register 201 always points to the next location to be written to, i.e., it is incremented as the buffer is written, then the lower bits are cleared when the end of the buffer is reached. In Simple Buffer mode, when the end of the buffer is reached, the Full Flag is set and remains set until cleared by software.
A set of hardware break/trace registers shown in
The break/watch Point Control Register 307 shown in
The R/W field specifies a watch point memory cycle type. The four states that this field defines are:
The Context Aware field is a two bit field. The two states that this field defines are:
A three bit Context field is used if break/watch points are context specific.
A Type field is a one bit field that defines the functionality type. The two states that this field defines are:
A Mode field is a two bit field that defines an operational mode. The four states that this field defines are:
A Block Size field is a four bit field. The watch point will “watch” memory in the block sizes:
To enable call trace a user sets up registers as follows:
Once this is set up, the following behaviors are enabled:
A Change of hardware context will cause the number of the new context to be written to the trace buffer (as a 32-bit integer).
To enable branch trace the user would set up registers as follows:
The Trace Control Register of
Once the registers are set up, the following behaviors are enabled:
All behaviors activated during Call Trace (see above);
A Data Trace requires the setup of both the Trace Control Register 101 of
The Base Write Address Register 201 of
Trace Control Register 101 of
The Break Point Address/Watch Point Base Address Register 301 of
The Watch Point Data Register 303 of
The Watch Point Data Mask Register 305 of
The Break/Watch Point Control Register 307 of
With the above register arrangement, when a memory access meets the parameters set up in the Watchpoint registers 301, 303, 305, 307 of
For any of the trace modes, i.e., Call Trace, Branch Trace, or Data Trace, if the timing field flag of the Trace Control Register 101 of
Where “Data” is either an address in Call Trace or Branch Trace mode or the data captured in Data Trace. This allows the user to determine what the timing is on various accesses including the time spent in a given subroutine or other program.
The timer value can come from a number of sources of different periods or precision.
Using the breakpoint registers described above, either in breakpoint mode or data trace mode, tracing can be enabled or disabled by various system events.
By way of example, in the register setups set out above if the enable flag in the Trace Control Register 101 of
Similarly, if the mode of the Break/Watchpoint Control Register 307 of
Turning now to
Turning now to
Turning now to
Turning now to
After the context number is written to the Trace Buffer 911 at step 709, or if at step 703 it is determined that a context switch is not required, or if at step 707 it is determined that the Trace Control Register 101 is not set to enable a Call or Branch trace, then a vector address for a new context/interrupt is obtained at step 711. After obtaining a vector address for the new context/interrupt at step 711, a determination is made at step 713 as to whether a Call Trace or a Branch Trace is enabled.
If it is determined at step 713 that neither a Call Trace nor a Branch Trace is enabled for that context, then the exception execution is completed at step 721. If, at step 713, it is determined that a Call Trace or Branch Trace is enabled, then a determination is made at step 715 as to whether or not the Trace Timing mode is called out by the Trace Control Register 101. If not, the target address is written to the Trace Buffer 911 at step 719. If Trace Timing is called for at step 715, then at step 717 the timer value is written to Trace Buffer 911. After the target address is written to the Trace Buffer, the exception execution is completed at step 721.
Turning now to
The invention has been described in conjunction with the illustrative embodiment. As will be appreciated by those skilled in the art, the invention is not limited to the specific embodiment shown. Various changes and modifications may be made to the embodiment without departing from the spirit or scope of the invention. It is intended that the invention be limited only by the claims appended hereto. It is further intended that the claims be given the broadest scope to which they are permitted.
