Closed Bug 895175 Opened 11 years ago Closed 11 years ago

Investigate rate of PC samples in epilogue/prologue to decide if "second-chance hack" is necessary

Categories

(Firefox OS Graveyard :: General, defect, P5)

Product:
Firefox OS Graveyard
Component:
General
Platform:
ARM
Gonk (Firefox OS)
Type:
defect

Tracking

(Not tracked)

Status:
RESOLVED WONTFIX
Milestone:
1.3 C3/1.4 S3(31jan)

People

(Reporter: huseby, Unassigned)

References

Details

(Keywords: perf, Whiteboard: [c=profiling p= s=2014.01.31 u=]) 

So sometimes the profile sig handler fires when the PC is in the prologue/epilogue of a function.  This is a situation the arm exception handling index can't handle and if we run the frame unpacking microcode, we get a bogus PC sometimes.

Because the prologue/epilogue is usually just two instructions, push/pop multiple and an SP adjustment, we can sometimes just remove the SP offset instruction from the microcode we got back from the arm exception handling index and try again.  This is usually enough to get a good PC and further reduce the number of bad samples in our profiling data.
Assignee: jld → nobody
Summary: Add "hack" to do 'second attempt' stack walk if the PC is bogus in the first frame → Investigate rate of PC samples in epilogue/prologue to decide if "second-chance hack" is necessary
Whiteboard: c=profiling → [c=profiling s=2013.09.06 p=3]
jld and I decided that I should take this bug.
Assignee: nobody → dhuseby
Status: NEW → ASSIGNED
Depends on: 810526
Whiteboard: [c=profiling s=2013.09.06 p=3] → [c=profiling s= p=3]
Assignee: dhuseby → nobody
Status: ASSIGNED → RESOLVED
Closed: 11 years ago
Priority: -- → P5
Resolution: --- → WONTFIX
Whiteboard: [c=profiling s= p=3] → [c=profiling p= s=2014.01.31 u=]
Target Milestone: --- → 1.3 C3/1.4 S3(31jan)
You need to log in before you can comment on or make changes to this bug.