Re: [PATCH] driver core: Make dev_err_probe() silent for -ENOMEM

[Uwe Kleine-König]

Hello,

On Fri, Apr 12, 2024 at 08:07:29PM +0300, Andy Shevchenko wrote:
> On Fri, Apr 12, 2024 at 07:03:01PM +0200, Uwe Kleine-König wrote:
> > On Fri, Apr 12, 2024 at 07:51:48PM +0300, Andy Shevchenko wrote:
> > > On Fri, Apr 12, 2024 at 06:44:05PM +0200, Uwe Kleine-König wrote:
> > > > For an out-of-memory error there should be no additional output. Adapt
> > > > dev_err_probe() to not emit the error message when err is -ENOMEM.
> > > > This simplifies handling errors that might among others be -ENOMEM.
> 
> > > ...
> > > 
> > > 	BUILD_BUG_ON(err == -ENOMEM);
> > > 
> > > Done!
> > 
> > Well no, that doesn't do the trick. Consider for example device_add().
> > That function can return (at least) -EINVAL and -ENOMEM. To properly
> > ensure that the error handling is silent with the current
> > dev_err_probe(), we'd need to do:
> > 
> > 	ret = device_add(...);
> > 	if (ret) {
> > 		if (ret != -ENOMEM)
> > 			return dev_err_probe(...);
> > 		else
> > 			return ret;
> > 	}
> > 
> > With my suggested patch this can be reduced to:
> > 
> > 	ret = device_add(...);
> > 	if (ret)
> > 		return dev_err_probe(...);
> 
> Fair enough, but these two should be combined.
> Mine is for the rejecting a dead code on the phase of the submission.

So something like the following on top of my change (only compile
tested):

diff --git a/drivers/base/core.c b/drivers/base/core.c
index 0bc04dcf31d5..02b6ec9a26c1 100644
--- a/drivers/base/core.c
+++ b/drivers/base/core.c
@@ -5033,7 +5033,7 @@ define_dev_printk_level(_dev_info, KERN_INFO);
  *
  * Returns @err.
  */
-int dev_err_probe(const struct device *dev, int err, const char *fmt, ...)
+int __dev_err_probe(const struct device *dev, int err, const char *fmt, ...)
 {
 	struct va_format vaf;
 	va_list args;
@@ -5064,7 +5064,7 @@ int dev_err_probe(const struct device *dev, int err, const char *fmt, ...)
 
 	return err;
 }
-EXPORT_SYMBOL_GPL(dev_err_probe);
+EXPORT_SYMBOL_GPL(__dev_err_probe);
 
 static inline bool fwnode_is_primary(struct fwnode_handle *fwnode)
 {
diff --git a/include/linux/dev_printk.h b/include/linux/dev_printk.h
index 6bfe70decc9f..a7573355ff90 100644
--- a/include/linux/dev_printk.h
+++ b/include/linux/dev_printk.h
@@ -14,6 +14,7 @@
 #include <linux/compiler.h>
 #include <linux/types.h>
 #include <linux/ratelimit.h>
+#include <linux/build_bug.h>
 
 #ifndef dev_fmt
 #define dev_fmt(fmt) fmt
@@ -274,6 +275,7 @@ do {									\
 	WARN_ONCE(condition, "%s %s: " format, \
 			dev_driver_string(dev), dev_name(dev), ## arg)
 
-__printf(3, 4) int dev_err_probe(const struct device *dev, int err, const char *fmt, ...);
+__printf(3, 4) int __dev_err_probe(const struct device *dev, int err, const char *fmt, ...);
+#define dev_err_probe(dev, err, ...) ( { BUILD_BUG_ON(__builtin_constant_p(err) && err == -ENOMEM); __dev_err_probe((dev), (err), __VA_ARGS__); })
 
 #endif /* _DEVICE_PRINTK_H_ */


I don't know if the kernel doc for dev_err_probe() should move to
include/linux/dev_printk.h then?!

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |

Attachment: signature.asc
Description: PGP signature