[PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Vineeth Pillai (Google)]

When a caller already guards a tracepoint with an explicit enabled check:

  if (trace_foo_enabled() && cond)
      trace_foo(args);

trace_foo() internally re-evaluates the static_branch_unlikely() key.
Since static branches are patched binary instructions the compiler cannot
fold the two evaluations, so every such site pays the cost twice.

This series introduces trace_invoke_##name() as a companion to
trace_##name().  It calls __do_trace_##name() directly, bypassing the
redundant static-branch re-check, while preserving all other correctness
properties of the normal path (RCU-watching assertion, might_fault() for
syscall tracepoints).  The internal __do_trace_##name() symbol is not
leaked to call sites; trace_invoke_##name() is the only new public API.

  if (trace_foo_enabled() && cond)
      trace_invoke_foo(args);   /* calls __do_trace_foo() directly */

The first patch adds the three-location change to
include/linux/tracepoint.h (__DECLARE_TRACE, __DECLARE_TRACE_SYSCALL,
and the !TRACEPOINTS_ENABLED stub).  The remaining 14 patches
mechanically convert all guarded call sites found in the tree:
kernel/, io_uring/, net/, accel/habanalabs, cpufreq/, devfreq/,
dma-buf/, fsi/, drm/, HID, i2c/, spi/, scsi/ufs/, and btrfs/.

This series is motivated by Peter Zijlstra's observation in the discussion
around Dmitry Ilvokhin's locking tracepoint instrumentation series, where
he noted that compilers cannot optimize static branches and that guarded
call sites end up evaluating the static branch twice for no reason, and
by Steven Rostedt's suggestion to add a proper API instead of exposing
internal implementation details like __do_trace_##name() directly to
call sites:

  https://lore.kernel.org/linux-trace-kernel/8298e098d3418cb446ef396f119edac58a3414e9.1772642407.git.d@ilvokhin.com

Suggested-by: Steven Rostedt <rostedt@goodmis.org>
Suggested-by: Peter Zijlstra <peterz@infradead.org>

Vineeth Pillai (Google) (15):
  tracepoint: Add trace_invoke_##name() API
  kernel: Use trace_invoke_##name() at guarded tracepoint call sites
  io_uring: Use trace_invoke_##name() at guarded tracepoint call sites
  net: Use trace_invoke_##name() at guarded tracepoint call sites
  accel/habanalabs: Use trace_invoke_##name() at guarded tracepoint call
    sites
  cpufreq: Use trace_invoke_##name() at guarded tracepoint call sites
  devfreq: Use trace_invoke_##name() at guarded tracepoint call sites
  dma-buf: Use trace_invoke_##name() at guarded tracepoint call sites
  fsi: Use trace_invoke_##name() at guarded tracepoint call sites
  drm: Use trace_invoke_##name() at guarded tracepoint call sites
  HID: Use trace_invoke_##name() at guarded tracepoint call sites
  i2c: Use trace_invoke_##name() at guarded tracepoint call sites
  spi: Use trace_invoke_##name() at guarded tracepoint call sites
  scsi: ufs: Use trace_invoke_##name() at guarded tracepoint call sites
  btrfs: Use trace_invoke_##name() at guarded tracepoint call sites

 drivers/accel/habanalabs/common/device.c          | 12 ++++++------
 drivers/accel/habanalabs/common/mmu/mmu.c         |  3 ++-
 drivers/accel/habanalabs/common/pci/pci.c         |  4 ++--
 drivers/cpufreq/amd-pstate.c                      | 10 +++++-----
 drivers/cpufreq/cpufreq.c                         |  2 +-
 drivers/cpufreq/intel_pstate.c                    |  2 +-
 drivers/devfreq/devfreq.c                         |  2 +-
 drivers/dma-buf/dma-fence.c                       |  4 ++--
 drivers/fsi/fsi-master-aspeed.c                   |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c            |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c            |  4 ++--
 drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c |  2 +-
 drivers/gpu/drm/scheduler/sched_entity.c          |  4 ++--
 drivers/hid/intel-ish-hid/ipc/pci-ish.c           |  2 +-
 drivers/i2c/i2c-core-slave.c                      |  2 +-
 drivers/spi/spi-axi-spi-engine.c                  |  4 ++--
 drivers/ufs/core/ufshcd.c                         | 12 ++++++------
 fs/btrfs/extent_map.c                             |  4 ++--
 fs/btrfs/raid56.c                                 |  4 ++--
 include/linux/tracepoint.h                        | 11 +++++++++++
 io_uring/io_uring.h                               |  2 +-
 kernel/irq_work.c                                 |  2 +-
 kernel/sched/ext.c                                |  2 +-
 kernel/smp.c                                      |  2 +-
 net/core/dev.c                                    |  2 +-
 net/core/xdp.c                                    |  2 +-
 net/openvswitch/actions.c                         |  2 +-
 net/openvswitch/datapath.c                        |  2 +-
 net/sctp/outqueue.c                               |  2 +-
 net/tipc/node.c                                   |  2 +-
 30 files changed, 62 insertions(+), 50 deletions(-)

-- 
2.53.0

[PATCH 01/15] tracepoint: Add trace_invoke_##name() API

[Vineeth Pillai (Google)]

Add trace_invoke_##name() as a companion to trace_##name().  When a
caller already guards a tracepoint with an explicit enabled check:

  if (trace_foo_enabled() && cond)
      trace_foo(args);

trace_foo() internally repeats the static_branch_unlikely() test, which
the compiler cannot fold since static branches are patched binary
instructions.  This results in two static-branch evaluations for every
guarded call site.

trace_invoke_##name() calls __do_trace_##name() directly, skipping the
redundant static-branch re-check.  This avoids leaking the internal
__do_trace_##name() symbol into call sites while still eliminating the
double evaluation:

  if (trace_foo_enabled() && cond)
      trace_invoke_foo(args);   /* calls __do_trace_foo() directly */

Three locations are updated:
- __DECLARE_TRACE: invoke form omits static_branch_unlikely, retains
  the LOCKDEP RCU-watching assertion.
- __DECLARE_TRACE_SYSCALL: same, plus retains might_fault().
- !TRACEPOINTS_ENABLED stub: empty no-op so callers compile cleanly
  when tracepoints are compiled out.

Suggested-by: Steven Rostedt <rostedt@goodmis.org>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Vineeth Pillai (Google) <vineeth@bitbyteword.org>
Assisted-by: Claude:claude-sonnet-4-6
---
 include/linux/tracepoint.h | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h
index 22ca1c8b54f32..07219316a8e14 100644
--- a/include/linux/tracepoint.h
+++ b/include/linux/tracepoint.h
@@ -294,6 +294,10 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
 			WARN_ONCE(!rcu_is_watching(),			\
 				  "RCU not watching for tracepoint");	\
 		}							\
+	}								\
+	static inline void trace_invoke_##name(proto)			\
+	{								\
+		__do_trace_##name(args);				\
 	}
 
 #define __DECLARE_TRACE_SYSCALL(name, proto, args, data_proto)		\
@@ -313,6 +317,11 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
 			WARN_ONCE(!rcu_is_watching(),			\
 				  "RCU not watching for tracepoint");	\
 		}							\
+	}								\
+	static inline void trace_invoke_##name(proto)			\
+	{								\
+		might_fault();						\
+		__do_trace_##name(args);				\
 	}
 
 /*
@@ -398,6 +407,8 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
 #define __DECLARE_TRACE_COMMON(name, proto, args, data_proto)		\
 	static inline void trace_##name(proto)				\
 	{ }								\
+	static inline void trace_invoke_##name(proto)			\
+	{ }								\
 	static inline int						\
 	register_trace_##name(void (*probe)(data_proto),		\
 			      void *data)				\
-- 
2.53.0

Re: [PATCH 01/15] tracepoint: Add trace_invoke_##name() API

[Steven Rostedt]

On Thu, 12 Mar 2026 11:04:56 -0400
"Vineeth Pillai (Google)" <vineeth@bitbyteword.org> wrote:

> Add trace_invoke_##name() as a companion to trace_##name().  When a
> caller already guards a tracepoint with an explicit enabled check:
> 
>   if (trace_foo_enabled() && cond)
>       trace_foo(args);
> 
> trace_foo() internally repeats the static_branch_unlikely() test, which
> the compiler cannot fold since static branches are patched binary
> instructions.  This results in two static-branch evaluations for every
> guarded call site.
> 
> trace_invoke_##name() calls __do_trace_##name() directly, skipping the
> redundant static-branch re-check.  This avoids leaking the internal
> __do_trace_##name() symbol into call sites while still eliminating the
> double evaluation:
> 
>   if (trace_foo_enabled() && cond)
>       trace_invoke_foo(args);   /* calls __do_trace_foo() directly */
> 
> Three locations are updated:
> - __DECLARE_TRACE: invoke form omits static_branch_unlikely, retains
>   the LOCKDEP RCU-watching assertion.
> - __DECLARE_TRACE_SYSCALL: same, plus retains might_fault().
> - !TRACEPOINTS_ENABLED stub: empty no-op so callers compile cleanly
>   when tracepoints are compiled out.
> 
> Suggested-by: Steven Rostedt <rostedt@goodmis.org>
> Suggested-by: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Vineeth Pillai (Google) <vineeth@bitbyteword.org>
> Assisted-by: Claude:claude-sonnet-4-6

I'm guessing Claude helped with the other patches. Did it really help with this one?

-- Steve


> ---
>  include/linux/tracepoint.h | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h
> index 22ca1c8b54f32..07219316a8e14 100644
> --- a/include/linux/tracepoint.h
> +++ b/include/linux/tracepoint.h
> @@ -294,6 +294,10 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
>  			WARN_ONCE(!rcu_is_watching(),			\
>  				  "RCU not watching for tracepoint");	\
>  		}							\
> +	}								\
> +	static inline void trace_invoke_##name(proto)			\
> +	{								\
> +		__do_trace_##name(args);				\
>  	}
>  
>  #define __DECLARE_TRACE_SYSCALL(name, proto, args, data_proto)		\
> @@ -313,6 +317,11 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
>  			WARN_ONCE(!rcu_is_watching(),			\
>  				  "RCU not watching for tracepoint");	\
>  		}							\
> +	}								\
> +	static inline void trace_invoke_##name(proto)			\
> +	{								\
> +		might_fault();						\
> +		__do_trace_##name(args);				\
>  	}
>  
>  /*
> @@ -398,6 +407,8 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
>  #define __DECLARE_TRACE_COMMON(name, proto, args, data_proto)		\
>  	static inline void trace_##name(proto)				\
>  	{ }								\
> +	static inline void trace_invoke_##name(proto)			\
> +	{ }								\
>  	static inline int						\
>  	register_trace_##name(void (*probe)(data_proto),		\
>  			      void *data)				\

Re: [PATCH 01/15] tracepoint: Add trace_invoke_##name() API

[Vineeth Remanan Pillai]

On Thu, Mar 12, 2026 at 11:13 AM Steven Rostedt <rostedt@goodmis.org> wrote:
>
> On Thu, 12 Mar 2026 11:04:56 -0400
> "Vineeth Pillai (Google)" <vineeth@bitbyteword.org> wrote:
>
> > Add trace_invoke_##name() as a companion to trace_##name().  When a
> > caller already guards a tracepoint with an explicit enabled check:
> >
> >   if (trace_foo_enabled() && cond)
> >       trace_foo(args);
> >
> > trace_foo() internally repeats the static_branch_unlikely() test, which
> > the compiler cannot fold since static branches are patched binary
> > instructions.  This results in two static-branch evaluations for every
> > guarded call site.
> >
> > trace_invoke_##name() calls __do_trace_##name() directly, skipping the
> > redundant static-branch re-check.  This avoids leaking the internal
> > __do_trace_##name() symbol into call sites while still eliminating the
> > double evaluation:
> >
> >   if (trace_foo_enabled() && cond)
> >       trace_invoke_foo(args);   /* calls __do_trace_foo() directly */
> >
> > Three locations are updated:
> > - __DECLARE_TRACE: invoke form omits static_branch_unlikely, retains
> >   the LOCKDEP RCU-watching assertion.
> > - __DECLARE_TRACE_SYSCALL: same, plus retains might_fault().
> > - !TRACEPOINTS_ENABLED stub: empty no-op so callers compile cleanly
> >   when tracepoints are compiled out.
> >
> > Suggested-by: Steven Rostedt <rostedt@goodmis.org>
> > Suggested-by: Peter Zijlstra <peterz@infradead.org>
> > Signed-off-by: Vineeth Pillai (Google) <vineeth@bitbyteword.org>
> > Assisted-by: Claude:claude-sonnet-4-6
>
> I'm guessing Claude helped with the other patches. Did it really help with this one?
>

Claude wrote and build tested the whole series based on my guidance
and prompt :-). I verified the series before sending it out, but
claude did the initial work.

Thanks,
Vineeth

Re: [PATCH 01/15] tracepoint: Add trace_invoke_##name() API

[Peter Zijlstra]

On Thu, Mar 12, 2026 at 11:39:06AM -0400, Vineeth Remanan Pillai wrote:
> On Thu, Mar 12, 2026 at 11:13 AM Steven Rostedt <rostedt@goodmis.org> wrote:
> >
> > On Thu, 12 Mar 2026 11:04:56 -0400
> > "Vineeth Pillai (Google)" <vineeth@bitbyteword.org> wrote:
> >
> > > Add trace_invoke_##name() as a companion to trace_##name().  When a
> > > caller already guards a tracepoint with an explicit enabled check:
> > >
> > >   if (trace_foo_enabled() && cond)
> > >       trace_foo(args);
> > >
> > > trace_foo() internally repeats the static_branch_unlikely() test, which
> > > the compiler cannot fold since static branches are patched binary
> > > instructions.  This results in two static-branch evaluations for every
> > > guarded call site.
> > >
> > > trace_invoke_##name() calls __do_trace_##name() directly, skipping the
> > > redundant static-branch re-check.  This avoids leaking the internal
> > > __do_trace_##name() symbol into call sites while still eliminating the
> > > double evaluation:
> > >
> > >   if (trace_foo_enabled() && cond)
> > >       trace_invoke_foo(args);   /* calls __do_trace_foo() directly */
> > >
> > > Three locations are updated:
> > > - __DECLARE_TRACE: invoke form omits static_branch_unlikely, retains
> > >   the LOCKDEP RCU-watching assertion.
> > > - __DECLARE_TRACE_SYSCALL: same, plus retains might_fault().
> > > - !TRACEPOINTS_ENABLED stub: empty no-op so callers compile cleanly
> > >   when tracepoints are compiled out.
> > >
> > > Suggested-by: Steven Rostedt <rostedt@goodmis.org>
> > > Suggested-by: Peter Zijlstra <peterz@infradead.org>
> > > Signed-off-by: Vineeth Pillai (Google) <vineeth@bitbyteword.org>
> > > Assisted-by: Claude:claude-sonnet-4-6
> >
> > I'm guessing Claude helped with the other patches. Did it really help with this one?
> >
> 
> Claude wrote and build tested the whole series based on my guidance
> and prompt :-). I verified the series before sending it out, but
> claude did the initial work.

That seems like an unreasonable waste of energy. You could've had claude
write a Coccinelle script for you and saved a ton of tokens.

Re: [PATCH 01/15] tracepoint: Add trace_invoke_##name() API

[Vineeth Remanan Pillai]

On Thu, Mar 12, 2026 at 11:53 AM Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Thu, Mar 12, 2026 at 11:39:06AM -0400, Vineeth Remanan Pillai wrote:
> > On Thu, Mar 12, 2026 at 11:13 AM Steven Rostedt <rostedt@goodmis.org> wrote:
> > >
> > > On Thu, 12 Mar 2026 11:04:56 -0400
> > > "Vineeth Pillai (Google)" <vineeth@bitbyteword.org> wrote:
> > >
> > > > Add trace_invoke_##name() as a companion to trace_##name().  When a
> > > > caller already guards a tracepoint with an explicit enabled check:
> > > >
> > > >   if (trace_foo_enabled() && cond)
> > > >       trace_foo(args);
> > > >
> > > > trace_foo() internally repeats the static_branch_unlikely() test, which
> > > > the compiler cannot fold since static branches are patched binary
> > > > instructions.  This results in two static-branch evaluations for every
> > > > guarded call site.
> > > >
> > > > trace_invoke_##name() calls __do_trace_##name() directly, skipping the
> > > > redundant static-branch re-check.  This avoids leaking the internal
> > > > __do_trace_##name() symbol into call sites while still eliminating the
> > > > double evaluation:
> > > >
> > > >   if (trace_foo_enabled() && cond)
> > > >       trace_invoke_foo(args);   /* calls __do_trace_foo() directly */
> > > >
> > > > Three locations are updated:
> > > > - __DECLARE_TRACE: invoke form omits static_branch_unlikely, retains
> > > >   the LOCKDEP RCU-watching assertion.
> > > > - __DECLARE_TRACE_SYSCALL: same, plus retains might_fault().
> > > > - !TRACEPOINTS_ENABLED stub: empty no-op so callers compile cleanly
> > > >   when tracepoints are compiled out.
> > > >
> > > > Suggested-by: Steven Rostedt <rostedt@goodmis.org>
> > > > Suggested-by: Peter Zijlstra <peterz@infradead.org>
> > > > Signed-off-by: Vineeth Pillai (Google) <vineeth@bitbyteword.org>
> > > > Assisted-by: Claude:claude-sonnet-4-6
> > >
> > > I'm guessing Claude helped with the other patches. Did it really help with this one?
> > >
> >
> > Claude wrote and build tested the whole series based on my guidance
> > and prompt :-). I verified the series before sending it out, but
> > claude did the initial work.
>
> That seems like an unreasonable waste of energy. You could've had claude
> write a Coccinelle script for you and saved a ton of tokens.

Yeah true, Steve also mentioned this to me offline. Haven't used
Coccinelle before, but now I know :-)

Thanks,
Vineeth

Re: [PATCH 01/15] tracepoint: Add trace_invoke_##name() API

[Keith Busch]

On Thu, Mar 12, 2026 at 12:05:37PM -0400, Vineeth Remanan Pillai wrote:
> On Thu, Mar 12, 2026 at 11:53 AM Peter Zijlstra <peterz@infradead.org> wrote:
> >
> > That seems like an unreasonable waste of energy. You could've had claude
> > write a Coccinelle script for you and saved a ton of tokens.
> 
> Yeah true, Steve also mentioned this to me offline. Haven't used
> Coccinelle before, but now I know :-)

[+ Chris Mason]

At the risk of creating a distraction...

This discussion got me thinking the right skill loaded should have the
AI implicitly use coccinelle to generate the patchset rather than do it
by hand. You could prompt with simple language for a pattern
substitution rather than explicitly request coccinelle, and it should
generate a patch set using a script rather than spending tokens on doing
it "by hand".

I sent such a "skill" to Chris' kernel "review-prompts":

  https://github.com/masoncl/review-prompts/pull/35

I used patch one from this series as the starting point and let the AI
figure the rest out. The result actually found additional patterns that
could take advantage of the optimisation that this series did not
include. The resulting kernel tree that the above github pull request
references cost 2.8k tokens to create with the skill.

[PATCH 03/15] io_uring: Use trace_invoke_##name() at guarded tracepoint call sites

[Vineeth Pillai (Google)]

Replace trace_foo() with the new trace_invoke_foo() at sites already
guarded by trace_foo_enabled(), avoiding a redundant
static_branch_unlikely() re-evaluation inside the tracepoint.
trace_invoke_foo() calls the tracepoint callbacks directly without
utilizing the static branch again.

Suggested-by: Steven Rostedt <rostedt@goodmis.org>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Vineeth Pillai (Google) <vineeth@bitbyteword.org>
Assisted-by: Claude:claude-sonnet-4-6
---
 io_uring/io_uring.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
index 0fa844faf2871..68b7656e1547a 100644
--- a/io_uring/io_uring.h
+++ b/io_uring/io_uring.h
@@ -299,7 +299,7 @@ static __always_inline bool io_fill_cqe_req(struct io_ring_ctx *ctx,
 	}
 
 	if (trace_io_uring_complete_enabled())
-		trace_io_uring_complete(req->ctx, req, cqe);
+		trace_invoke_io_uring_complete(req->ctx, req, cqe);
 	return true;
 }
 
-- 
2.53.0

Re: [PATCH 03/15] io_uring: Use trace_invoke_##name() at guarded tracepoint call sites

[Keith Busch]

On Thu, Mar 12, 2026 at 11:04:58AM -0400, Vineeth Pillai (Google) wrote:
>  	if (trace_io_uring_complete_enabled())
> -		trace_io_uring_complete(req->ctx, req, cqe);
> +		trace_invoke_io_uring_complete(req->ctx, req, cqe);

Curious, this one doesn't follow that pattern of "if (enabed && cond)"
that this cover letter said it was addressing, so why doesn't this call
just drop the 'if' check and go straight to trace_io_uring_complete()? I
followed this usage to commit a0730c738309a06, which says that the
compiler was generating code to move args before checking if the trace
was enabled. That commit was a while ago though, and suggests to remove
the check if that problem is solved. Is it still a problem?

Re: [PATCH 03/15] io_uring: Use trace_invoke_##name() at guarded tracepoint call sites

[Steven Rostedt]

On Thu, 12 Mar 2026 09:24:21 -0600
Keith Busch <kbusch@kernel.org> wrote:

> On Thu, Mar 12, 2026 at 11:04:58AM -0400, Vineeth Pillai (Google) wrote:
> >  	if (trace_io_uring_complete_enabled())
> > -		trace_io_uring_complete(req->ctx, req, cqe);
> > +		trace_invoke_io_uring_complete(req->ctx, req, cqe);  
> 
> Curious, this one doesn't follow that pattern of "if (enabed && cond)"
> that this cover letter said it was addressing, so why doesn't this call
> just drop the 'if' check and go straight to trace_io_uring_complete()? I
> followed this usage to commit a0730c738309a06, which says that the

You mean 'a0727c738309a06'? As I could not find the above 'a0730c738309a06'

> compiler was generating code to move args before checking if the trace
> was enabled. That commit was a while ago though, and suggests to remove

It was only 2023.

> the check if that problem is solved. Is it still a problem?

We should check. But also, tracepoints should never be in a header file.
That really should be:

#include <linux/tracepoint-defs.h>

DECLARE_TRACEPOINT(io_uring_complete);

[..]

	if (tracepoint_enabled(io_uring_complete))
		do_trace_io_uring_complete(...);

And in a C file, that should be:

void do_io_uring_complete(...)
{
	trace_inovke_io_uring_complete(...);
}


Which reminds me. There's other places that have that tracepoint_enabled()
in header files that do the above. The C wrapper functions should also
convert the callback to the trace_invoke_<event>() call.

-- Steve

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Mathieu Desnoyers]

On 2026-03-12 11:04, Vineeth Pillai (Google) wrote:
> When a caller already guards a tracepoint with an explicit enabled check:
> 
>    if (trace_foo_enabled() && cond)
>        trace_foo(args);
> 
> trace_foo() internally re-evaluates the static_branch_unlikely() key.
> Since static branches are patched binary instructions the compiler cannot
> fold the two evaluations, so every such site pays the cost twice.
> 
> This series introduces trace_invoke_##name() as a companion to
> trace_##name().  It calls __do_trace_##name() directly, bypassing the
> redundant static-branch re-check, while preserving all other correctness
> properties of the normal path (RCU-watching assertion, might_fault() for
> syscall tracepoints).  The internal __do_trace_##name() symbol is not
> leaked to call sites; trace_invoke_##name() is the only new public API.
> 
>    if (trace_foo_enabled() && cond)
>        trace_invoke_foo(args);   /* calls __do_trace_foo() directly */

FYI, we have a similar concept in LTTng-UST for userspace
instrumentation already:

if (lttng_ust_tracepoint_enabled(provider, name))
         lttng_ust_do_tracepoint(provider, name, ...);

Perhaps it can provide some ideas about API naming.

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Steven Rostedt]

On Thu, 12 Mar 2026 11:12:41 -0400
Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:

> >    if (trace_foo_enabled() && cond)
> >        trace_invoke_foo(args);   /* calls __do_trace_foo() directly */  
> 
> FYI, we have a similar concept in LTTng-UST for userspace
> instrumentation already:
> 
> if (lttng_ust_tracepoint_enabled(provider, name))
>          lttng_ust_do_tracepoint(provider, name, ...);
> 
> Perhaps it can provide some ideas about API naming.

I find the word "invoke" sounding more official than "do" ;-)

Note, Vineeth came up with the naming. I would have done "do" but when I
saw "invoke" I thought it sounded better.

-- Steve

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Mathieu Desnoyers]

On 2026-03-12 11:23, Steven Rostedt wrote:
> On Thu, 12 Mar 2026 11:12:41 -0400
> Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> 
>>>     if (trace_foo_enabled() && cond)
>>>         trace_invoke_foo(args);   /* calls __do_trace_foo() directly */
>>
>> FYI, we have a similar concept in LTTng-UST for userspace
>> instrumentation already:
>>
>> if (lttng_ust_tracepoint_enabled(provider, name))
>>           lttng_ust_do_tracepoint(provider, name, ...);
>>
>> Perhaps it can provide some ideas about API naming.
> 
> I find the word "invoke" sounding more official than "do" ;-)
> 
> Note, Vineeth came up with the naming. I would have done "do" but when I
> saw "invoke" I thought it sounded better.

It works as long as you don't have a tracing subsystem called
"invoke", then you get into identifier clash territory.

Thanks,

Mathieu



-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Steven Rostedt]

On Thu, 12 Mar 2026 11:28:07 -0400
Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:

> > Note, Vineeth came up with the naming. I would have done "do" but when I
> > saw "invoke" I thought it sounded better.  
> 
> It works as long as you don't have a tracing subsystem called
> "invoke", then you get into identifier clash territory.

True. Perhaps we should do the double underscore trick.

Instead of:  trace_invoke_foo()

use:  trace_invoke__foo()


Which will make it more visible to what the trace event is.

Hmm, we probably should have used: trace__foo() for all tracepoints, as
there's still functions that are called trace_foo() that are not
tracepoints :-p

-- Steve

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Mathieu Desnoyers]

On 2026-03-12 11:40, Steven Rostedt wrote:
> On Thu, 12 Mar 2026 11:28:07 -0400
> Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> 
>>> Note, Vineeth came up with the naming. I would have done "do" but when I
>>> saw "invoke" I thought it sounded better.
>>
>> It works as long as you don't have a tracing subsystem called
>> "invoke", then you get into identifier clash territory.
> 
> True. Perhaps we should do the double underscore trick.
> 
> Instead of:  trace_invoke_foo()
> 
> use:  trace_invoke__foo()
> 
> 
> Which will make it more visible to what the trace event is.
> 
> Hmm, we probably should have used: trace__foo() for all tracepoints, as
> there's still functions that are called trace_foo() that are not
> tracepoints :-p

One certain way to eliminate identifier clash would be to go for a
prefix to "trace_", e.g.

do_trace_foo()
call_trace_foo()
emit_trace_foo()
__trace_foo()
invoke_trace_foo()
dispatch_trace_foo()

Thanks,

Mathieu



-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Peter Zijlstra]

On Thu, Mar 12, 2026 at 11:49:23AM -0400, Mathieu Desnoyers wrote:
> On 2026-03-12 11:40, Steven Rostedt wrote:
> > On Thu, 12 Mar 2026 11:28:07 -0400
> > Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> > 
> > > > Note, Vineeth came up with the naming. I would have done "do" but when I
> > > > saw "invoke" I thought it sounded better.
> > > 
> > > It works as long as you don't have a tracing subsystem called
> > > "invoke", then you get into identifier clash territory.
> > 
> > True. Perhaps we should do the double underscore trick.
> > 
> > Instead of:  trace_invoke_foo()
> > 
> > use:  trace_invoke__foo()
> > 
> > 
> > Which will make it more visible to what the trace event is.
> > 
> > Hmm, we probably should have used: trace__foo() for all tracepoints, as
> > there's still functions that are called trace_foo() that are not
> > tracepoints :-p
> 
> One certain way to eliminate identifier clash would be to go for a
> prefix to "trace_", e.g.

Oh, I know!, call them __do_trace_##foo().

/me runs like hell

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Mathieu Desnoyers]

On 2026-03-12 11:54, Peter Zijlstra wrote:
> On Thu, Mar 12, 2026 at 11:49:23AM -0400, Mathieu Desnoyers wrote:
>> On 2026-03-12 11:40, Steven Rostedt wrote:
>>> On Thu, 12 Mar 2026 11:28:07 -0400
>>> Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
>>>
>>>>> Note, Vineeth came up with the naming. I would have done "do" but when I
>>>>> saw "invoke" I thought it sounded better.
>>>>
>>>> It works as long as you don't have a tracing subsystem called
>>>> "invoke", then you get into identifier clash territory.
>>>
>>> True. Perhaps we should do the double underscore trick.
>>>
>>> Instead of:  trace_invoke_foo()
>>>
>>> use:  trace_invoke__foo()
>>>
>>>
>>> Which will make it more visible to what the trace event is.
>>>
>>> Hmm, we probably should have used: trace__foo() for all tracepoints, as
>>> there's still functions that are called trace_foo() that are not
>>> tracepoints :-p
>>
>> One certain way to eliminate identifier clash would be to go for a
>> prefix to "trace_", e.g.
> 
> Oh, I know!, call them __do_trace_##foo().
> 
> /me runs like hell

So s/__do_trace_/do_trace_/g and call it a day ?

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Vineeth Remanan Pillai]

On Thu, Mar 12, 2026 at 11:49 AM Mathieu Desnoyers
<mathieu.desnoyers@efficios.com> wrote:
>
> On 2026-03-12 11:40, Steven Rostedt wrote:
> > On Thu, 12 Mar 2026 11:28:07 -0400
> > Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> >
> >>> Note, Vineeth came up with the naming. I would have done "do" but when I
> >>> saw "invoke" I thought it sounded better.
> >>
> >> It works as long as you don't have a tracing subsystem called
> >> "invoke", then you get into identifier clash territory.
> >
> > True. Perhaps we should do the double underscore trick.
> >
> > Instead of:  trace_invoke_foo()
> >
> > use:  trace_invoke__foo()
> >
> >
> > Which will make it more visible to what the trace event is.
> >
> > Hmm, we probably should have used: trace__foo() for all tracepoints, as
> > there's still functions that are called trace_foo() that are not
> > tracepoints :-p
>
> One certain way to eliminate identifier clash would be to go for a
> prefix to "trace_", e.g.
>
> do_trace_foo()
> call_trace_foo()

This was the initial idea, but it had conflict in the existing source:
call_trace_sched_update_nr_running. do_trace_##name also had
collisions when I checked. So, went with trace_invoke_##name. Did not
check rest of the suggestions here though.

Thanks,
Vineeth

> emit_trace_foo()
> __trace_foo()
> invoke_trace_foo()
> dispatch_trace_foo()
>
> Thanks,
>
> Mathieu
>
>
>
> --
> Mathieu Desnoyers
> EfficiOS Inc.
> https://www.efficios.com

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Andrii Nakryiko]

On Thu, Mar 12, 2026 at 9:15 AM Vineeth Remanan Pillai
<vineeth@bitbyteword.org> wrote:
>
> On Thu, Mar 12, 2026 at 11:49 AM Mathieu Desnoyers
> <mathieu.desnoyers@efficios.com> wrote:
> >
> > On 2026-03-12 11:40, Steven Rostedt wrote:
> > > On Thu, 12 Mar 2026 11:28:07 -0400
> > > Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> > >
> > >>> Note, Vineeth came up with the naming. I would have done "do" but when I
> > >>> saw "invoke" I thought it sounded better.
> > >>
> > >> It works as long as you don't have a tracing subsystem called
> > >> "invoke", then you get into identifier clash territory.
> > >
> > > True. Perhaps we should do the double underscore trick.
> > >
> > > Instead of:  trace_invoke_foo()
> > >
> > > use:  trace_invoke__foo()
> > >
> > >
> > > Which will make it more visible to what the trace event is.
> > >
> > > Hmm, we probably should have used: trace__foo() for all tracepoints, as
> > > there's still functions that are called trace_foo() that are not
> > > tracepoints :-p
> >
> > One certain way to eliminate identifier clash would be to go for a
> > prefix to "trace_", e.g.
> >
> > do_trace_foo()
> > call_trace_foo()
>
> This was the initial idea, but it had conflict in the existing source:
> call_trace_sched_update_nr_running. do_trace_##name also had
> collisions when I checked. So, went with trace_invoke_##name. Did not
> check rest of the suggestions here though.
>
> Thanks,
> Vineeth
>
> > emit_trace_foo()
> > __trace_foo()

this seems like the best approach, IMO. double-underscored variants
are usually used for some specialized/internal version of a function
when we know that some conditions are correct (e.g., lock is already
taken, or something like that). Which fits here: trace_xxx() will
check if tracepoint is enabled, while __trace_xxx() will not check and
just invoke the tracepoint? It's short, it's distinct, and it says "I
know what I am doing".

> > invoke_trace_foo()
> > dispatch_trace_foo()
> >
> > Thanks,
> >
> > Mathieu
> >
> >
> >
> > --
> > Mathieu Desnoyers
> > EfficiOS Inc.
> > https://www.efficios.com
>

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Steven Rostedt]

On Thu, 12 Mar 2026 09:54:29 -0700
Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:

> > > emit_trace_foo()
> > > __trace_foo()  
> 
> this seems like the best approach, IMO. double-underscored variants
> are usually used for some specialized/internal version of a function
> when we know that some conditions are correct (e.g., lock is already
> taken, or something like that). Which fits here: trace_xxx() will
> check if tracepoint is enabled, while __trace_xxx() will not check and
> just invoke the tracepoint? It's short, it's distinct, and it says "I
> know what I am doing".

Honestly, I consider double underscore as internal only and not something
anyone but the subsystem maintainers use.

This, is a normal function where it's just saying: If you have it already
enabled, then you can use this. Thus, I don't think it qualifies as a "you
know what you are doing".

Perhaps: call_trace_foo() ?

-- Steve

Re: [PATCH 00/15] tracepoint: Avoid double static_branch evaluation at guarded call sites

[Vineeth Remanan Pillai]

On Thu, Mar 12, 2026 at 1:03 PM Steven Rostedt <rostedt@goodmis.org> wrote:
>
> On Thu, 12 Mar 2026 09:54:29 -0700
> Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
>
> > > > emit_trace_foo()
> > > > __trace_foo()
> >
> > this seems like the best approach, IMO. double-underscored variants
> > are usually used for some specialized/internal version of a function
> > when we know that some conditions are correct (e.g., lock is already
> > taken, or something like that). Which fits here: trace_xxx() will
> > check if tracepoint is enabled, while __trace_xxx() will not check and
> > just invoke the tracepoint? It's short, it's distinct, and it says "I
> > know what I am doing".
>
> Honestly, I consider double underscore as internal only and not something
> anyone but the subsystem maintainers use.
>
> This, is a normal function where it's just saying: If you have it already
> enabled, then you can use this. Thus, I don't think it qualifies as a "you
> know what you are doing".
>
> Perhaps: call_trace_foo() ?
>
call_trace_foo has one collision with the tracepoint
sched_update_nr_running and a function
call_trace_sched_update_nr_running. I had considered this and later
moved to trace_invoke_foo() because of the collision. But I can rename
call_trace_sched_update_nr_running to something else if call_trace_foo
is the general consensus.

Thanks,
Vineeth