Low-overhead method and apparatus for collecting function call trace data

• US 8,910,124 B1
• Filed: 10/31/2011
• Issued: 12/09/2014
• Est. Priority Date: 10/31/2011
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Bhansali, Sanjay, Chopra, Harshit
• Primary Examiner(s): Tecklu, Isaac T
• Assistant Examiner(s): Morshed, Hossain
• Application Number:
  US13/285,438
• Time in Patent Office: 1,135 Days
• Field of Search: None
• US Class Current: 717/128
• CPC Class Codes: G06F 11/3466 Performance evaluation by tracing or monitoring
• CPC Class Codes: G06F 11/3612 by runtime analysis performance monitoring G06F11/3466

Abstract

An apparatus and method for aiding in debugging and optimizing software is presented, in particular for instrumenting software to trace function calls. Source code is compiled with placeholder memory ranges. While executing the compiled binary object, a call-trace profiling function is inserted into at least certain of the placeholder memory ranges to log call-trace profiling data. Then, the compiled binary object is restored to its original state by removing the call-trace profiling function(s). A NOP function can be inserted in each placeholder memory range during compilation.

Claims

1. A method for on-demand call-trace profiling an application having source code to be executed on a system, the method comprising:

compiling a binary object from the source code using a processor while including a plurality of placeholder memory ranges in the compiled binary object, each placeholder memory range being of a size suitable for inserting a call-trace profiling function call and each placeholder memory range including a no operation (NOP) instruction; and

providing a set of code modification instructions configured to modify the compiled binary object while it is being executed, the instructions including;

inserting a call to a first call-trace profiling function in at least one of the plurality of placeholder memory ranges in place of the NOP instruction;

replacing a return instruction and a subsequent NOP instruction with a jump instruction associated with a call to a second call-trace profiling function;

logging call-trace profiling data using the inserted call-trace profiling function calls;

and removing the inserted call-trace profiling function call after logging the call-trace profiling data including replacing the removed call-trace profiling function call with the NOP instruction and replacing the jump instruction with the return instruction and the subsequent NOP instruction.

2. The method of claim 1 wherein a runtime overhead due to including the plurality of placeholder memory ranges in the compiled binary object is substantially undetectable as compared to that of the source code compiled without including the plurality of placeholder memory ranges.

3. The method of claim 1 wherein the compiled binary object comprises at least one remote procedure call.

4. The method of claim 3 wherein the compiled binary object includes a first one of the plurality of placeholder memory ranges adjacent an address of the at least one remote procedure call and a second one of the plurality of placeholder memory ranges adjacent a return point of the at least one remote procedure call.

5. A method of on-demand call-trace profiling an application having source code to be executed on a system, the method comprising:

inserting, while executing a compiled binary object, a call to a first call-trace profiling function in at least one of a plurality of placeholder memory ranges included in the compiled binary object, the compiled binary object compiled from the source code and stored in a memory and each placeholder memory range being of a size suitable for the inserted call-trace profiling function call and including a NOP instruction inserted during compilation of the compiled binary object, wherein inserting the call-trace profiling function call comprises replacing the NOP function in the at least one of the plurality of placeholder memory ranges with a respective call-trace profiling function call and replacing a return instruction and a subsequent NOP instruction with a jump instruction associated with a call to a second call-trace profiling function;

logging call-trace profiling data using the inserted call-trace profiling function calls; and

removing the inserted call-trace profiling function call after logging the call-trace profiling data including replacing the removed call-trace profiling function call with the NOP function and replacing the jump instruction with the return instruction and the subsequent NOP instruction.

6. The method of claim 5 wherein a runtime overhead due to executing the inserted call-trace profiling function is small enough to permit normal operation of the compiled binary object while it is being executed as compared to executing the source code compiled without the plurality of placeholder memory ranges.

7. The method of claim 5, further comprising:

compressing the call-trace profiling data to minimize data storage requirements.

8. The method of claim 5 wherein the inserted call-trace profiling function call comprises a function call.

9. The method of claim 5 wherein the call-tracer profiling data comprises a function identifier and a time.

10. An apparatus for on-demand call-trace profiling an application having source code to be executed on a system, the apparatus comprising:

a memory; and

a processor configured to execute instructions stored in the memory to;

insert, while executing the compiled binary object, a call to a first call-trace profiling function in at least one of a plurality of placeholder memory ranges included in a compiled binary object, the compiled binary object compiled from the source code and stored in a memory and each placeholder memory range being of a size suitable for the inserted call-trace profiling function call and including a NOP instruction inserted during compilation of the compiled binary object, wherein the processor is configured to insert the call-trace profiling function call by replacing the NOP function in the at least one of the plurality of placeholder memory ranges with a respective call-trace profiling function call and replacing a return instruction and a subsequent NOP instruction with a jump instruction associated with a call to a second call-trace profiling function;

log call-trace profiling data using the inserted call-trace profiling function calls; and

remove the inserted call-trace profiling function call after logging the call-trace profiling data including replacing the removed call-trace profiling function call with the NOP function and replacing the jump instruction with the return instruction and the subsequent NOP instruction.

11. The apparatus of claim 10 wherein the processor is configured to compile the compiled binary object from the source code while including the plurality of placeholder memory ranges in the compiled binary object.

12. The apparatus of claim 10 wherein the processor is configured to insert the call-trace profiling function call by inserting a respective call-trace profiling function in at least two of the plurality of placeholder memory ranges.

13. The apparatus of claim 10 wherein a runtime overhead of the plurality of placeholder memory ranges in the compiled binary object is substantially undetectable.

14. The apparatus of claim 10 wherein the processor is further configured to compress the call-trace profiling data.

15. The apparatus of claim 14 wherein the call trace-profiling data comprises a function identifier and a time.

16. A method for on-demand call-trace profiling an application having source code to be executed on a system, the method comprising:

compiling a binary object from the source code using a processor while including a plurality of NOP functions in the compiled binary object, each NOP function defining a placeholder memory range of a size suitable for inserting a call-trace profiling function call; and

providing a set of code modification instructions configured to modify the compiled binary object while it is being executed, the instructions including;

replacing at least one of the NOP functions with a call to a first call-trace profiling function;

replacing a return instruction and a subsequent NOP instruction with a jump instruction associated with a call to a second call-trace profiling function;

logging call-trace profiling data using the inserted call-trace profiling function calls; and

subsequent to logging call-trace profiling data, removing the inserted call-trace profiling function call and restoring the NOP function and replacing the jump instruction with the return instruction and the subsequent NOP instruction.