Trace management during aborted speculative operations

Patent metadata

Field	Value
Publication number	US-11061807-B2
Application number	US-201816235489-A
Country	US
Kind code	B2
Filing date	Dec 28, 2018
Priority date	Dec 28, 2018
Publication date	Jul 13, 2021
Grant date	Jul 13, 2021

Abstract

A method for tracing software code executing on a core of a processor is described. The method includes generating a set of packets for a trace packet stream based on a main cycle counter, which maintains a count of cycles elapsing in the core since a packet was emitted into the trace packet stream, and a commit cycle counter, which maintains a cycle count in the core since the last commit operation, wherein the generating comprises (1) storing a value of the main cycle counter in the commit cycle counter in response to detecting a commit operation and (2) storing a value of the commit cycle counter in the main cycle counter in response to detecting an abort in the core; and emitting the set of packets from the processor into the trace packet stream for tracing execution of the software code.

First claim

What is claimed is:

1. A method for tracing software code executing on a core of a processor, the method comprising:
generating a set of packets for a trace packet stream based on a main cycle counter, which maintains a count of cycles elapsing in the core since a packet was emitted into the trace packet stream, and a commit cycle counter, which maintains a cycle count in the core since the last commit operation, wherein the generating comprises (1) storing a value of the main cycle counter in the commit cycle counter in response to detecting a commit operation and (2) storing a value of the commit cycle counter in the main cycle counter in response to detecting an abort in the core;
emitting the set of packets from the processor into the trace packet stream for tracing execution of the software code;
updating, during each cycle of the core, a main take-not-taken (TNT) buffer with indications of all conditional branches that retired during the cycle;
setting, in response to a commit, a commit TNT buffer equal to the main TNT buffer; and
setting, in response to an abort, the main TNT buffer equal to the commit TNT buffer.

2. The method of claim 1 , wherein generating the set of packets for the trace packet stream further comprises:
generating a cycle packet for the trace packet stream with a value of the main cycle counter, in response to retiring an instruction of the software code; and
resetting the main cycle counter in response to generating the cycle packet.

3. The method of claim 1 , wherein generating the set of packets for the trace packet stream further comprises:
determining whether a cycle has elapsed in the core; and
setting, in response to determining that a cycle has elapsed in the core, the value of the main cycle counter to the minimum of (1) one more than a current value of the main cycle counter and (2) a limit of the main cycle counter.

4. The method of claim 3 , wherein generating the set of packets for the trace packet stream further comprises:
setting, in response to determining that a cycle has elapsed in the core, the value of the commit cycle counter to the minimum of (1) one more than a current value of the commit cycle counter and (2) a limit of the commit cycle counter.

5. The method of claim 4 , wherein generating the set of packets for the trace packet stream further comprises:
determining whether the core is performing speculative instruction execution, wherein the setting the value of the commit cycle counter is to be performed in response to determining that the core is performing speculative instruction execution.

6. The method of claim 4 , wherein generating the set of packets for the trace packet stream further comprises:
generating an overflow packet for the trace packet stream in response to setting the commit cycle counter to the limit of the commit cycle counter and thereafter detecting an abort.

7. The method of claim 1 , further comprising:
storing, in a trace information buffer, the set of packets of the trace packet stream prior to emission from the processor;
setting a commit pointer to point to a last packet in the set of packets that has been committed, wherein packets in the set of packets older than the last packet are available for emission from the processor and packets in the set of packets younger than the last packet are unavailable for emission from the processor;
setting a write pointer to point to a next write point in the trace information buffer; and
setting the write pointer to a position of the commit pointer upon an abort.

8. A non-transitory computer readable medium that stores instructions, which when executed by a processor, cause the processor to:
generate a set of packets for a trace packet stream based on a main cycle counter, which maintains a count of cycles elapsing in a core of the processor since a packet was emitted into the trace packet stream, and a commit cycle counter, which maintains a cycle count in the core since the last commit operation, wherein the generating comprises (1) storing a value of the main cycle counter in the commit cycle counter in response to detecting a commit operation and (2) storing a value of the commit cycle counter in the main cycle counter in response to detecting an abort in the core, wherein generating the set of packets includes:
determining whether a cycle has elapsed in the core,
setting, in response to determining that a cycle has elapsed in the core, the value of the commit cycle counter to the minimum of (1) one more than a current value of the commit cycle counter and (2) a limit of the commit cycle counter, and
generating an overflow packet for the trace packet stream in response to setting the commit cycle counter to the limit of the commit cycle counter and thereafter detecting an abort; and

emit the set of packets from the processor into the trace packet stream for tracing execution of the software code.

9. The non-transitory computer readable medium of claim 8 , wherein, for generating the set of packets for the trace packet stream, the instructions further cause the processor to:
generate a cycle packet for the trace packet stream with a value of the main cycle counter, in response to retiring an instruction of the software code; and
reset the main cycle counter in response to generating the cycle packet.

10. The non-transitory computer readable medium of claim 8 , wherein, for generating the set of packets for the trace packet stream, the instructions further cause the processor to:
set, in response to determining that a cycle has elapsed in the core, the value of the main cycle counter to the minimum of (1) one more than a current value of the main cycle counter and (2) a limit of the main cycle counter.

11. The non-transitory computer readable medium of claim 8 , wherein, for generating the set of packets for the trace packet stream, the instructions further cause the processor to:
determine whether the core is performing speculative instruction execution, wherein the setting the value of the commit cycle counter is to be performed in response to determining that the core is performing speculative instruction execution.

12. The non-transitory computer readable medium of claim 8 , wherein the instructions further cause the processor to:
store, in a trace information buffer, the set of packets of the trace packet stream prior to emission from the processor;
set a commit pointer to point to a last packet in the set of packets that has been committed, wherein packets in the set of packets older than the last packet are available for emission from the processor and packets in the set of packets younger than the last packet are unavailable for emission from the processor;
set a write pointer to point to a next write point in the trace information buffer; and
set the write pointer to a position of the commit pointer upon an abort.

13. The non-transitory computer readable medium of claim 8 , wherein the instructions further cause the processor to:
update, during each cycle of the core, a main take-not-taken (TNT) buffer with indications of all conditional branches that retired during the cycle;
set, in response to a commit, a commit TNT buffer equal to the main TNT buffer; and
set, in response to an abort, the main TNT buffer equal to the commit TNT buffer.

14. A system comprising:
a trace decoder; and
a processing core, including trace logic to manage a main cycle counter, which maintains a count of cycles elapsing in the processing core since a packet was emitted into a trace packet stream, and a commit cycle counter, which maintains a cycle count in the processing core sinc

Assignees

• Intel Corp

Inventors

• Strong Beeman
• Merten Matthew C
• Agron Jason

Classifications

• G06F2201/835

  Timestamp · CPC title

• G06F11/3466

  Performance evaluation by tracing or monitoring · CPC title

• G06F11/348

  Circuit details, i.e. tracer hardware · CPC title

• G06F2201/87

  Monitoring of transactions · CPC title

• G06F11/3636 Primary

  by tracing the execution of the program · CPC title