Some functions are known to potentially return -EPROBE_DEFER. In such a
case, it is likely that dev_err_probe() is a better choice than err_err().

dev_err_probe():
  - is usually less verbose
  - generates smaller .o files
  - handles -EPROBE_DEFER so that logs are not spammed
  - automatically log the error code in a human readable way

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
---
This patch is only a PoC to see if there is some interest in such a new
check.
The hard coded '5 lines of context' has been chosen because a typical
pattern is:

	clk = devm_clk_get(dev, "clk_lcd");
	if (IS_ERR(clk) {
		dev_err(dev, "Error meesage\n");
		return PTR_ERR(clk);
	}
---
 scripts/checkpatch.pl | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
index 2737e4ced574..88365749ed2e 100755
--- a/scripts/checkpatch.pl
+++ b/scripts/checkpatch.pl
@@ -2625,6 +2625,9 @@ sub process {
 	my $last_blank_line = 0;
 	my $last_coalesced_string_linenr = -1;
 
+	my $last_function_that_return_defer = "";
+	my $last_function_that_return_defer_linenr = 0;
+
 	our @report = ();
 	our $cnt_lines = 0;
 	our $cnt_error = 0;
@@ -7459,6 +7462,17 @@ sub process {
 			WARN("DUPLICATED_SYSCTL_CONST",
 				"duplicated sysctl range checking value '$1', consider using the shared one in include/linux/sysctl.h\n" . $herecurr);
 		}
+
+# check for places where dev_err_probe() would likely be better than dev_err()
+		if ($line =~ /((?:devm_)?clk_get)s*\(/) {
+			$last_function_that_return_defer = $1;
+			$last_function_that_return_defer_linenr = $linenr;
+		}
+		if ($last_function_that_return_defer_linenr >= ($linenr - 5) &&
+		    $line =~ /dev_err[^_]/) {
+			WARN("LIKELY_DEV_ERR_PROBE",
+				"dev_err_probe() is likely a better choice than err_err() after a " . $last_function_that_return_defer . "() call\n" . $herecurr);
+		}
 	}
 
 	# If we have no input at all, then there is nothing to report on
-- 
2.34.1

Le 11/09/2022 à 15:15, Christophe JAILLET a écrit :
> Some functions are known to potentially return -EPROBE_DEFER. In such a
> case, it is likely that dev_err_probe() is a better choice than err_err().
> 
> dev_err_probe():
>    - is usually less verbose
>    - generates smaller .o files
>    - handles -EPROBE_DEFER so that logs are not spammed
>    - automatically log the error code in a human readable way
> 
> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
> ---
> This patch is only a PoC to see if there is some interest in such a new
> check.
> The hard coded '5 lines of context' has been chosen because a typical
> pattern is:
> 
> 	clk = devm_clk_get(dev, "clk_lcd");
> 	if (IS_ERR(clk) {
> 		dev_err(dev, "Error meesage\n");
> 		return PTR_ERR(clk);
> 	}

(adding Linus Walleij)


I forgot to say that this patch is a try to address the comment from 
Linus Walleij at [1].

It would not help "fixing a gazillion dev_err_probe()", but it could 
help not having more to fix later :)

CJ

[1]: 
https://lore.kernel.org/all/CACRpkdZEcTD1A3tR=d4fDF89ECMDfchVPW921v6X6ARiPXHEMQ@mail.gmail.com/


> ---
>   scripts/checkpatch.pl | 14 ++++++++++++++
>   1 file changed, 14 insertions(+)
> 
> diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
> index 2737e4ced574..88365749ed2e 100755
> --- a/scripts/checkpatch.pl
> +++ b/scripts/checkpatch.pl
> @@ -2625,6 +2625,9 @@ sub process {
>   	my $last_blank_line = 0;
>   	my $last_coalesced_string_linenr = -1;
>   
> +	my $last_function_that_return_defer = "";
> +	my $last_function_that_return_defer_linenr = 0;
> +
>   	our @report = ();
>   	our $cnt_lines = 0;
>   	our $cnt_error = 0;
> @@ -7459,6 +7462,17 @@ sub process {
>   			WARN("DUPLICATED_SYSCTL_CONST",
>   				"duplicated sysctl range checking value '$1', consider using the shared one in include/linux/sysctl.h\n" . $herecurr);
>   		}
> +
> +# check for places where dev_err_probe() would likely be better than dev_err()
> +		if ($line =~ /((?:devm_)?clk_get)s*\(/) {
> +			$last_function_that_return_defer = $1;
> +			$last_function_that_return_defer_linenr = $linenr;
> +		}
> +		if ($last_function_that_return_defer_linenr >= ($linenr - 5) &&
> +		    $line =~ /dev_err[^_]/) {
> +			WARN("LIKELY_DEV_ERR_PROBE",
> +				"dev_err_probe() is likely a better choice than err_err() after a " . $last_function_that_return_defer . "() call\n" . $herecurr);
> +		}
>   	}
>   
>   	# If we have no input at all, then there is nothing to report on

On Sun, Sep 11, 2022 at 3:21 PM Christophe JAILLET
<christophe.jaillet@wanadoo.fr> wrote:
> Le 11/09/2022 à 15:15, Christophe JAILLET a écrit :
> > Some functions are known to potentially return -EPROBE_DEFER. In such a
> > case, it is likely that dev_err_probe() is a better choice than err_err().
> >
> > dev_err_probe():
> >    - is usually less verbose
> >    - generates smaller .o files
> >    - handles -EPROBE_DEFER so that logs are not spammed
> >    - automatically log the error code in a human readable way
> >
> > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
> > ---
> > This patch is only a PoC to see if there is some interest in such a new
> > check.
> > The hard coded '5 lines of context' has been chosen because a typical
> > pattern is:
> >
> >       clk = devm_clk_get(dev, "clk_lcd");
> >       if (IS_ERR(clk) {
> >               dev_err(dev, "Error meesage\n");
> >               return PTR_ERR(clk);
> >       }
>
> (adding Linus Walleij)
>
>
> I forgot to say that this patch is a try to address the comment from
> Linus Walleij at [1].
>
> It would not help "fixing a gazillion dev_err_probe()", but it could
> help not having more to fix later :)

Needless to say I am a big fan of this patch!
Acked-by: Linus Walleij <linus.walleij@linaro.org>

Yours,
Linus Walleij