On Thu, Jul 2, 2020 at 2:44 PM Greg Kroah-Hartman
[off-list ref] wrote:

On Mon, Jun 29, 2020 at 08:50:08AM +0200, Bartosz Golaszewski wrote:

quoted

From: Bartosz Golaszewski <redacted>

Shrink pmbus code by using devm_hwmon_device_register_with_groups()
and devm_krealloc() instead of their non-managed variants.

Signed-off-by: Bartosz Golaszewski <redacted>
---
 drivers/hwmon/pmbus/pmbus_core.c | 28 +++++++++-------------------
 1 file changed, 9 insertions(+), 19 deletions(-)

diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c
index a420877ba533..225d0ac162c7 100644
--- a/drivers/hwmon/pmbus/pmbus_core.c
+++ b/drivers/hwmon/pmbus/pmbus_core.c

@@ -1022,9 +1022,9 @@ static int pmbus_add_attribute(struct pmbus_data *data, struct attribute *attr)
 {
      if (data->num_attributes >= data->max_attributes - 1) {
              int new_max_attrs = data->max_attributes + PMBUS_ATTR_ALLOC_SIZE;
-             void *new_attrs = krealloc(data->group.attrs,
-                                        new_max_attrs * sizeof(void *),
-                                        GFP_KERNEL);
+             void *new_attrs = devm_krealloc(data->dev, data->group.attrs,
+                                             new_max_attrs * sizeof(void *),
+                                             GFP_KERNEL);

dynamic sysfs attributes in a devm-allocated chunk of memory?  What
could go wrong...

So what *can* go wrong, which it couldn't before this patch? The
drivers in this directory kfree() this memory anyway on driver detach.
Using devm here is equivalent to the previous behavior - only that the
memory is freed after remove() not inside it.

Anyway, is this the only in-kernel user that you could find for this
function?  If so, it feels like it's a lot of extra work for no real
gain.

No. There are around 100 calls to krealloc() in drivers/. I assume
that at least half of these are called with an attached struct device.
I chose this driver, because it has a commit in its history that
explicitly says that it would use devm_krealloc() if it were available
(commit 85cfb3a83536 ("hwmon: (pmbus) Use krealloc to allocate
attribute memory"). I didn't want to spend a lot of time on converting
other users in case this patch gets rejected.

Bartosz