On Fri, 19 Nov 2021 15:46:31 -0700 jim.cromie@xxxxxxxxx wrote: > > So I could see us supporting subsystem specific trace buffer output > > via dynamic debug here. We could add new dev_debug() variants that > > allow say a trace buffer to be supplied. So in that way subsystems > > could 'opt-out' of having their data put into the global trace buffer. > > And perhaps some subsystems we would want to allow output to both > > buffers? The subsystem specific one and the global one? > > > > * trace_array_printk - Print a message to a specific instance > * @tr: The instance trace_array descriptor > * @ip: The instruction pointer that this is called from. > * @fmt: The format to print (printf format) > * > > what happens when @tr == NULL ? It does nothing, but perhaps crash the kernel. > It could allow up-flow of events to the global instance Absolutely not! Then it's just a reimplementation of trace_printk(). Which I refuse to have. Nothing should just dump to the main instance. Once we allow that, then everyone will be dumping there and you will no longer be able to trace anything because it will be filled with noise. What is allowed is an event that acts like a trace_printk() but is an event, which you can turn off (have default off), and even pick which instance to go to. > > > Thanks, > > > > -Jason > > > > > > So I wonder, is there any conceptual utility to this ? > > echo 1 > instances/foo/filter_up # enable event upflow (or query-time merging?) > > Maybe enabling this causes other files (the ones missing from > instances/foo) to magically appear > so all those filtering capacities also appear. I've been busy doing other things so I haven't been keeping up with this thread (which I need to go back and read). Perhaps it was already stated, but I don't know why you want that. trace-cmd can read several instances (including the top level one) and interleave them nicely, if that is what you are looking for. So can KernelShark. -- Steve
