On Wed, Jan 07, 2026 at 11:08:14AM -0500, Steven Rostedt wrote:
> On Wed, 7 Jan 2026 14:16:05 +0000
> Will Deacon <[email protected]> wrote:
>
> > I still don't understand why we need anything new in the arch code for this.
> >
> > We've selected HAVE_ARCH_FTRACE_REGS and we implement
> > ftrace_regs_set_instruction_pointer() and ftrace_regs_set_return_value()
> > so the core code already has everything it needs to make this work
> > without additional arch support.
>
> I believe the issue is that the BPF code takes a pt_regs and does the
> update directly with that, and not the ftrace_regs.
>
> I'm guessing this is due to BPF programs modifying the pt_regs directly,
> and BPF programs do not yet understand ftrace_regs?
>
> Because arm64 requires making a copy of pt_regs as the ftrace_regs has a
> different layout, and the ftrace_regs is what does the changes, if the
> pt_regs passed to the BPF program modifies the values it needs a way to
> propagate that back to the ftrace_regs.
Sure, but isn't that exactly what selecting HAVE_ARCH_FTRACE_REGS says,
and so any other architecture (i.e. riscv) using a separate structure
will run into the same problem?
In other words, add the helper to the core code along the lines of the
diff below (purely untested, just to illustrate what I'm trying to get
at).
Will
--->8
diff --git a/include/linux/ftrace_regs.h b/include/linux/ftrace_regs.h
index 15627ceea9bc..3ebd8cdac7c6 100644
--- a/include/linux/ftrace_regs.h
+++ b/include/linux/ftrace_regs.h
@@ -33,6 +33,15 @@ struct ftrace_regs;
#define ftrace_regs_get_frame_pointer(fregs) \
frame_pointer(&arch_ftrace_regs(fregs)->regs)
+#else
+
+static __always_inline void
+ftrace_partial_regs_update(const struct ftrace_regs *fregs, struct pt_regs
*regs)
+{
+ ftrace_regs_set_instruction_pointer(fregs, instruction_pointer(regs));
+ ftrace_regs_set_return_value(fregs, regs_return_value(regs));
+}
+
#endif /* HAVE_ARCH_FTRACE_REGS */
/* This can be overridden by the architectures */
