Re: [PATCH rcu 5/5] checkpatch: Complain about unexpected uses of RCU Tasks Trace

[Paul E. McKenney]

On Wed, Jul 19, 2023 at 07:51:58AM -0400, Joel Fernandes wrote:
> 
> 
> On 7/17/23 19:34, Paul E. McKenney wrote:
> > On Mon, Jul 17, 2023 at 03:34:14PM -0700, Joe Perches wrote:
> > > On Mon, 2023-07-17 at 11:04 -0700, Paul E. McKenney wrote:
> > > > RCU Tasks Trace is quite specialized, having been created specifically
> > > > for sleepable BPF programs.  Because it allows general blocking within
> > > > readers, any new use of RCU Tasks Trace must take current use cases into
> > > > account.  Therefore, update checkpatch.pl to complain about use of any of
> > > > the RCU Tasks Trace API members outside of BPF and outside of RCU itself.
> > > > 
> > > > Cc: Andy Whitcroft <apw@xxxxxxxxxxxxx> (maintainer:CHECKPATCH)
> > > > Cc: Joe Perches <joe@xxxxxxxxxxx> (maintainer:CHECKPATCH)
> > > > Cc: Dwaipayan Ray <dwaipayanray1@xxxxxxxxx> (reviewer:CHECKPATCH)
> > > > Cc: Lukas Bulwahn <lukas.bulwahn@xxxxxxxxx>
> > > > Cc: Alexei Starovoitov <ast@xxxxxxxxxx>
> > > > Cc: Daniel Borkmann <daniel@xxxxxxxxxxxxx>
> > > > Cc: John Fastabend <john.fastabend@xxxxxxxxx>
> > > > Cc: <bpf@xxxxxxxxxxxxxxx>
> > > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > > > ---
> > > >   scripts/checkpatch.pl | 18 ++++++++++++++++++
> > > >   1 file changed, 18 insertions(+)
> > > > 
> > > > diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
> > > []
> > > > @@ -7457,6 +7457,24 @@ sub process {
> > > >   			}
> > > >   		}
> > > > +# Complain about RCU Tasks Trace used outside of BPF (and of course, RCU).
> > > > +		if ($line =~ /\brcu_read_lock_trace\s*\(/ ||
> > > > +		    $line =~ /\brcu_read_lock_trace_held\s*\(/ ||
> > > > +		    $line =~ /\brcu_read_unlock_trace\s*\(/ ||
> > > > +		    $line =~ /\bcall_rcu_tasks_trace\s*\(/ ||
> > > > +		    $line =~ /\bsynchronize_rcu_tasks_trace\s*\(/ ||
> > > > +		    $line =~ /\brcu_barrier_tasks_trace\s*\(/ ||
> > > > +		    $line =~ /\brcu_request_urgent_qs_task\s*\(/) {
> > > > +			if ($realfile !~ m@^kernel/bpf@ &&
> > > > +			    $realfile !~ m@^include/linux/bpf@ &&
> > > > +			    $realfile !~ m@^net/bpf@ &&
> > > > +			    $realfile !~ m@^kernel/rcu@ &&
> > > > +			    $realfile !~ m@^include/linux/rcu@) {
> > > 
> > > Functions and paths like these tend to be accreted.
> > > 
> > > Please use a variable or 2 like:
> > > 
> > > our $rcu_trace_funcs = qr{(?x:
> > > 	rcu_read_lock_trace |
> > > 	rcu_read_lock_trace_held |
> > > 	rcu_read_unlock_trace |
> > > 	call_rcu_tasks_trace |
> > > 	synchronize_rcu_tasks_trace |
> > > 	rcu_barrier_tasks_trace |
> > > 	rcu_request_urgent_qs_task
> > > )};
> > > our $rcu_trace_paths = qr{(?x:
> > > 	kernel/bfp/ |
> > > 	include/linux/bpf |
> > > 	net/bpf/ |
> > > 	kernel/rcu/ |
> > > 	include/linux/rcu
> > > )};
> > 
> > Like this?
> > 
> > # Complain about RCU Tasks Trace used outside of BPF (and of course, RCU).
> > 		our $rcu_trace_funcs = qr{(?x:
> > 			rcu_read_lock_trace |
> > 			rcu_read_lock_trace_held |
> > 			rcu_read_unlock_trace |
> > 			call_rcu_tasks_trace |
> > 			synchronize_rcu_tasks_trace |
> > 			rcu_barrier_tasks_trace |
> > 			rcu_request_urgent_qs_task
> > 		)};
> > 		our $rcu_trace_paths = qr{(?x:
> > 			kernel/bfp/ |
> > 			include/linux/bpf |
> > 			net/bpf/ |
> > 			kernel/rcu/ |
> > 			include/linux/rcu
> > 		)};
> > 		if ($line =~ /$rcu_trace_funcs/) {
> > 			if ($realfile !~ m@^$rcu_trace_paths@) {
> > 				WARN("RCU_TASKS_TRACE",
> > 				     "use of RCU tasks trace is incorrect outside BPF or core RCU code\n" . $herecurr);
> > 			}
> > 		}
> > 
> > No, that is definitely wrong.  It has lost track of the list of pathnames,
> > thus complaining about uses of those functions in files where their use
> > is permitted.
> > 
> > But this seems to work:
> > 
> > # Complain about RCU Tasks Trace used outside of BPF (and of course, RCU).
> > 		our $rcu_trace_funcs = qr{(?x:
> > 			rcu_read_lock_trace |
> > 			rcu_read_lock_trace_held |
> > 			rcu_read_unlock_trace |
> > 			call_rcu_tasks_trace |
> > 			synchronize_rcu_tasks_trace |
> > 			rcu_barrier_tasks_trace |
> > 			rcu_request_urgent_qs_task
> > 		)};
> > 		if ($line =~ /\b$rcu_trace_funcs\s*\(/) {
> > 			if ($realfile !~ m@^kernel/bpf@ &&
> > 			    $realfile !~ m@^include/linux/bpf@ &&
> > 			    $realfile !~ m@^net/bpf@ &&
> > 			    $realfile !~ m@^kernel/rcu@ &&
> > 			    $realfile !~ m@^include/linux/rcu@) {
> > 				WARN("RCU_TASKS_TRACE",
> > 				     "use of RCU tasks trace is incorrect outside BPF or core RCU code\n" . $herecurr);
> > 			}
> > 		}
> > 
> > Maybe the "^" needs to be distributed into $rcu_trace_paths?
> > 
> > # Complain about RCU Tasks Trace used outside of BPF (and of course, RCU).
> > 		our $rcu_trace_funcs = qr{(?x:
> > 			rcu_read_lock_trace |
> > 			rcu_read_lock_trace_held |
> > 			rcu_read_unlock_trace |
> > 			call_rcu_tasks_trace |
> > 			synchronize_rcu_tasks_trace |
> > 			rcu_barrier_tasks_trace |
> > 			rcu_request_urgent_qs_task
> > 		)};
> > 		our $rcu_trace_paths = qr{(?x:
> > 			^kernel/bfp/ |
> > 			^include/linux/bpf |
> > 			^net/bpf/ |
> > 			^kernel/rcu/ |
> > 			^include/linux/rcu
> > 		)};
> > 		if ($line =~ /\b$rcu_trace_funcs\s*\(/) {
> > 			if ($realfile !~ m@$rcu_trace_paths@) {
> > 				WARN("RCU_TASKS_TRACE",
> > 				     "use of RCU tasks trace is incorrect outside BPF or core RCU code\n" . $herecurr);
> > 			}
> > 		}
> > 
> > But no joy here, either.  Which is no surprise, given that perl is
> > happy to distribute the "\b" and the "\s*\(" across the elements of
> > $rcu_trace_funcs.  I tried a number of other variations, including
> > inverting the "if" condition "(!(... =~ ...))", inverting the "if"
> > condition via an empty "then" clause, replacing the "m@" with "/",
> > replacing the "|" in the "qr{}" with "&", and a few others.
> > 
> > Again, listing the pathnames explicitly in the second "if" condition
> > works just fine.
> > 
> 
> Not a perl expert but I wonder if the following are any options at all:
> 
> 1. Instead of having a complex list of strings in a regex variable, it might
> be easier to hold the strings as a perl array, and then iterate over that
> array checking each element of the array on every iteration, against the
> line.
> 
> 2. Roll the "\b" and/or "^" in into the regex variable itself than trying
> make them play with the variable later.
> 
> 3. Use parentheses around the variable? Not sure if that will work but I
> wonder if it has something to do with operator precedence.
> 
> 4. Instead of a list of paths, maybe it is better to look for "rcu" or "bpf"
> in the regex itself? That way new paths don't require script updates (at the
> expense though of false-positives (highly unlikely, IMHO)).

Given perl's tendency to have corner cases in its corner cases, I
am guessing that the "^" character combined with the "/" character is
causing trouble here.  Especially given that I don't see any use of such
a pattern anywhere in checkpatch.pl except directly in a "~" expression,
and there are a lot of those.

So I will keep it as is unless I hear otherwise from Joe Perches.

							Thanx, Paul