Closed vpelletier closed 7 years ago
I think this needs some testing. Enabling this option also enables:
CONFIG_TRACEPOINTS=y
CONFIG_NOP_TRACER=y
CONFIG_EVENT_TRACING=y
CONFIG_CONTEXT_SWITCH_TRACER=y
CONFIG_TRACING=y
CONFIG_GENERIC_TRACER=y
CONFIG_BINARY_PRINTF=y
I used to enable more tracing options in the kernel until I discovered it cost us about 10% in performance. If you want this in you'll need to build the kernel with it enabled and provide some evidence that doesn't cause any performance regressions (e.g. measure "free -h" and perhaps nbench before and after the change).
Sadly, I'm not near that Pi anymore (and don't want to take the risk of it not booting) and don't have aniother at hand, so I cannot run tests. It's an interesting point nevertheless, that enabling debug/trace features could slow down execution. The load from my use is very light so I could easily live with the overhead.
@vpelletier has this issue been resolved? If yes, please close this issue. Thanks.
Closing due to lack of activity. Reopen if you feel this issue is still relevant.
Block io tracing feature would help reduce SD wear by being able to know where writes happen and why (=what processes cause these writes). This is especially useful to see how often a given block is written to, such as fs superblocks.