I've been able to identify the origin of the issue. It happens since the below commit.
Do you know what should be done to fix that ?
Christophe
From 7a40054361162d2f78f332aa868fed137beb7901 Mon Sep 17 00:00:00 2001
From: Axel Lin <redacted>
Date: Sun, 30 Mar 2014 16:42:57 +0800
Subject: spi: fsl-spi: Fix memory leak
mpc8xxx_spi_probe() has set master->cleanup = mpc8xxx_spi_cleanup,
however current code overrides the setting in fsl_spi_probe() and set
master->cleanup = fsl_spi_cleanup.
Thus the memory allocated for cs is not freed anywhere.
Convert to use devm_kzalloc to fix the memory leak.
Signed-off-by: Axel Lin <redacted>
Signed-off-by: Mark Brown <redacted>
On Thu, Aug 28, 2014 at 5:53 PM, leroy christophe [off-list ref]
wrote:
I've been able to identify the origin of the issue. It happens since the
below commit.
Do you know what should be done to fix that ?
Christophe
Actually, more things are wrong with what the driver is doing.
If inside spi_add_device() the call to device_add() fails, then that
code bails out without any call to spi_cleanup() and the same
memory will leak (Is this intended?).
Basically, fsl_spi_setup allocates memory using devm_kzalloc, while
device_add expects that any memory allocated via this way is only
done in the device's probe function.
The simple fix would be to do a normal allocation (revert the patch) and
add a free to the cleanup() function. Unfortunately that doesn't fix the
memleak I mentioned above.
So, some outside-the-box thinking brings me to conclude that another way
to fix this problem is to allocate the devm_kzalloc not on the device's
resource
list but on the controller's resources (it's controller state after all...).
Regards,
Stijn
On Thu, Aug 28, 2014 at 5:53 PM, leroy christophe
[off-list ref] wrote:
I've been able to identify the origin of the issue. It happens since the below commit.
Do you know what should be done to fix that ?
Christophe
Actually, more things are wrong with what the driver is doing.
If inside spi_add_device() the call to device_add() fails, then that
code bails out without any call to spi_cleanup() and the same
memory will leak (Is this intended?).
Basically, fsl_spi_setup allocates memory using devm_kzalloc, while
device_add expects that any memory allocated via this way is only
done in the device's probe function.
The simple fix would be to do a normal allocation (revert the patch) and
add a free to the cleanup() function. Unfortunately that doesn't fix the
memleak I mentioned above.
So, some outside-the-box thinking brings me to conclude that another way
to fix this problem is to allocate the devm_kzalloc not on the device's
resource
list but on the controller's resources (it's controller state after all...).
Regards,
Stijn