Re: [PATCH] x86/tracing: introduce enter/exit tracepoint pairs for page faults

[Steven Rostedt]

On Mon, 14 Apr 2025 13:42:23 -0700
Dave Hansen <dave.hansen@xxxxxxxxx> wrote:

> On 4/14/25 13:35, Junxuan Liao wrote:
> > Rename page_fault_{user,kernel} to page_fault_{user,kernel}_enter, and
> > add the exit counterparts. This might be useful for measuring page fault
> > handling latencies.  
> 
> Is there a reason kprobes don't work for this?

Kprobes is not always easy to add, and it does add more overhead.

I use to have measurements by using function graph tracing of all timings
into the kernel, but when the noinstr was added, that broke. I still do
timings but that's by manually adding hacks into the kernel. I haven't done
timings on a vanilla kernel for some time. It would be nice to be able to
do that again.

-- Steve