On Fri, Dec 06, 2024 at 11:02:29AM +0100, Björn Töpel wrote: > Adding Robbin for input, who's doing much more crazy text patching in > JVM, than what we do in the kernel. ;-) > > Let's say we're tracing "f". Previously w/ stop_machine() it was > something like: > > f: > 1: nop > nop > ... > ... > > ftrace_caller: > ... > auipc a2, function_trace_op > ld a2, function_trace_op(a2) > ... > 2: auipc ra, ftrace_stub > jalr ftrace_stub(ra) > > The text was patched by ftrace in 1 and 2. > > ...and now: > f: > auipc t0, ftrace_caller > A: nop > ... > ... > > ftrace_caller: > ... > auipc a2, function_trace_op > ld a2, function_trace_op(a2) > ... > auipc ra, ftrace_call_dest > ld ra, ftrace_call_dest(ra) > jalr ra > > The text is only patched in A, and the tracer func is loaded via > ftrace_call_dest.
Previously the operation was no-op, right? Confused... -- An old man doll... just what I always wanted! - Clara
signature.asc
Description: PGP signature
