With PERF_SAMPLE_STACK_USER and PERF_SAMPLE_REGS_USER, we can do stack unwinding for the ARM Exception Handling ABI. Supporting the former on ARM is just setting a config flag (and fixing a bug that breaks it on all non-x86 platforms, because it's only ever been tested on x86 before now, but I already have the fix); the latter will need some glue but should be simple. Also these changes, and the ones that added those sampling options in the first place (and anything that they depend on), need to be backported to the old kernel versions we're using on devices. I've done this for keon, and hopefully the other 3.0.x kernels floating around shouldn't be too different. Work in progress: https://github.com/jld/gp-keon-kernel/tree/perf-stackcopy-gp
Whiteboard: [c=performance p=5] → [c=profiling p=5]
Summary: Port Linux perf_event register/stack sampling to ARM (and to the old kernel versions we're using) → [META] Port Linux perf_event register/stack sampling to ARM (and to the old kernel versions we're using)
making this a meta bug because it was broken down into 904907, 904908, 904909, and 904912. as per discussion with :mlee this will have an s= value for the soonest sprint that one of the dependent bugs is due to be delivered. this bug will get carried forward to subsequent sprints until all dependent bugs are closed. then this will be closed.
Whiteboard: [c=profiling p=5] → [c=profiling s=2013.09.06 p=5]
Whiteboard: [c=profiling s=2013.09.06 p=5] → [c=profiling s= p=5]
Assignee: jld → nobody
Status: ASSIGNED → RESOLVED
Last Resolved: 5 years ago
Priority: -- → P3
Resolution: --- → WONTFIX
Whiteboard: [perf-reviewed] → [c=profiling p= s=2014.03.14 u=]
Target Milestone: --- → 1.4 S3 (14mar)
You need to log in before you can comment on or make changes to this bug.