When unbinding/binding a driver with DMA mapped memory, the DMA map is
not freed before the driver is reloaded. This leads to a memory leak
when the DMA map is overwritten when reprobing the driver.
This can be reproduced with a platform driver having a dma-range:
dummy {
...
#address-cells = <0x2>;
#size-cells = <0x2>;
ranges;
dma-ranges = <...>;
...
};
and then unbinding/binding it:
~# echo soc:dummy >/sys/bus/platform/drivers/<driver>/unbind
DMA map object 0xffffff800b0ae540 still being held by &pdev->dev
~# echo soc:dummy >/sys/bus/platform/drivers/<driver>/bind
~# echo scan > /sys/kernel/debug/kmemleak
~# cat /sys/kernel/debug/kmemleak
unreferenced object 0xffffff800b0ae540 (size 64):
comm "sh", pid 833, jiffies 4295174550 (age 2535.352s)
hex dump (first 32 bytes):
00 00 00 80 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 80 00 00 00 00 00 00 00 80 00 00 00 00 ................
backtrace:
[<ffffffefd1694708>] create_object.isra.0+0x108/0x344
[<ffffffefd1d1a850>] kmemleak_alloc+0x8c/0xd0
[<ffffffefd167e2d0>] __kmalloc+0x440/0x6f0
[<ffffffefd1a960a4>] of_dma_get_range+0x124/0x220
[<ffffffefd1a8ce90>] of_dma_configure_id+0x40/0x2d0
[<ffffffefd198b68c>] platform_dma_configure+0x5c/0xa4
[<ffffffefd198846c>] really_probe+0x8c/0x514
[<ffffffefd1988990>] __driver_probe_device+0x9c/0x19c
[<ffffffefd1988cd8>] device_driver_attach+0x54/0xbc
[<ffffffefd1986634>] bind_store+0xc4/0x120
[<ffffffefd19856e0>] drv_attr_store+0x30/0x44
[<ffffffefd173c9b0>] sysfs_kf_write+0x50/0x60
[<ffffffefd173c1c4>] kernfs_fop_write_iter+0x124/0x1b4
[<ffffffefd16a013c>] new_sync_write+0xdc/0x160
[<ffffffefd16a256c>] vfs_write+0x23c/0x2a0
[<ffffffefd16a2758>] ksys_write+0x64/0xec
To prevent this we should free the dma_range_map when the device is
released.
Signed-off-by: Mårten Lindahl <[email protected]>
---
drivers/base/dd.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/drivers/base/dd.c b/drivers/base/dd.c
index 9eaaff2f556c..f47cab21430f 100644
--- a/drivers/base/dd.c
+++ b/drivers/base/dd.c
@@ -629,6 +629,9 @@ static int really_probe(struct device *dev, struct device_driver *drv)
drv->remove(dev);
devres_release_all(dev);
+ arch_teardown_dma_ops(dev);
+ kfree(dev->dma_range_map);
+ dev->dma_range_map = NULL;
driver_sysfs_remove(dev);
dev->driver = NULL;
dev_set_drvdata(dev, NULL);
@@ -1209,6 +1212,8 @@ static void __device_release_driver(struct device *dev, struct device *parent)
devres_release_all(dev);
arch_teardown_dma_ops(dev);
+ kfree(dev->dma_range_map);
+ dev->dma_range_map = NULL;
dev->driver = NULL;
dev_set_drvdata(dev, NULL);
if (dev->pm_domain && dev->pm_domain->dismiss)
--
2.30.2
On Wed, Feb 16, 2022 at 3:42 AM Mårten Lindahl <[email protected]> wrote:
>
> When unbinding/binding a driver with DMA mapped memory, the DMA map is
> not freed before the driver is reloaded. This leads to a memory leak
> when the DMA map is overwritten when reprobing the driver.
>
> This can be reproduced with a platform driver having a dma-range:
>
> dummy {
> ...
> #address-cells = <0x2>;
> #size-cells = <0x2>;
> ranges;
> dma-ranges = <...>;
> ...
> };
>
> and then unbinding/binding it:
>
> ~# echo soc:dummy >/sys/bus/platform/drivers/<driver>/unbind
>
> DMA map object 0xffffff800b0ae540 still being held by &pdev->dev
>
> ~# echo soc:dummy >/sys/bus/platform/drivers/<driver>/bind
> ~# echo scan > /sys/kernel/debug/kmemleak
> ~# cat /sys/kernel/debug/kmemleak
> unreferenced object 0xffffff800b0ae540 (size 64):
> comm "sh", pid 833, jiffies 4295174550 (age 2535.352s)
> hex dump (first 32 bytes):
> 00 00 00 80 00 00 00 00 00 00 00 00 00 00 00 00 ................
> 00 00 00 80 00 00 00 00 00 00 00 80 00 00 00 00 ................
> backtrace:
> [<ffffffefd1694708>] create_object.isra.0+0x108/0x344
> [<ffffffefd1d1a850>] kmemleak_alloc+0x8c/0xd0
> [<ffffffefd167e2d0>] __kmalloc+0x440/0x6f0
> [<ffffffefd1a960a4>] of_dma_get_range+0x124/0x220
> [<ffffffefd1a8ce90>] of_dma_configure_id+0x40/0x2d0
> [<ffffffefd198b68c>] platform_dma_configure+0x5c/0xa4
> [<ffffffefd198846c>] really_probe+0x8c/0x514
> [<ffffffefd1988990>] __driver_probe_device+0x9c/0x19c
> [<ffffffefd1988cd8>] device_driver_attach+0x54/0xbc
> [<ffffffefd1986634>] bind_store+0xc4/0x120
> [<ffffffefd19856e0>] drv_attr_store+0x30/0x44
> [<ffffffefd173c9b0>] sysfs_kf_write+0x50/0x60
> [<ffffffefd173c1c4>] kernfs_fop_write_iter+0x124/0x1b4
> [<ffffffefd16a013c>] new_sync_write+0xdc/0x160
> [<ffffffefd16a256c>] vfs_write+0x23c/0x2a0
> [<ffffffefd16a2758>] ksys_write+0x64/0xec
>
> To prevent this we should free the dma_range_map when the device is
> released.
>
> Signed-off-by: Mårten Lindahl <[email protected]>
Suggested-by: Rob Herring <[email protected]>
Fixes: e0d072782c73 ("dma-mapping: introduce DMA range map,
supplanting dma_pfn_offset")
Reviewed-by: Rob Herring <[email protected]>
I've got a follow-up patch to refactor the 3 occurrences of this same
deinit code.
Rob
On Wed, Feb 16, 2022 at 09:45:09AM -0600, Rob Herring wrote:
> On Wed, Feb 16, 2022 at 3:42 AM M?rten Lindahl <[email protected]> wrote:
> >
> > When unbinding/binding a driver with DMA mapped memory, the DMA map is
> > not freed before the driver is reloaded. This leads to a memory leak
> > when the DMA map is overwritten when reprobing the driver.
> >
> > This can be reproduced with a platform driver having a dma-range:
> >
> > dummy {
> > ...
> > #address-cells = <0x2>;
> > #size-cells = <0x2>;
> > ranges;
> > dma-ranges = <...>;
> > ...
> > };
> >
> > and then unbinding/binding it:
> >
> > ~# echo soc:dummy >/sys/bus/platform/drivers/<driver>/unbind
> >
> > DMA map object 0xffffff800b0ae540 still being held by &pdev->dev
> >
> > ~# echo soc:dummy >/sys/bus/platform/drivers/<driver>/bind
> > ~# echo scan > /sys/kernel/debug/kmemleak
> > ~# cat /sys/kernel/debug/kmemleak
> > unreferenced object 0xffffff800b0ae540 (size 64):
> > comm "sh", pid 833, jiffies 4295174550 (age 2535.352s)
> > hex dump (first 32 bytes):
> > 00 00 00 80 00 00 00 00 00 00 00 00 00 00 00 00 ................
> > 00 00 00 80 00 00 00 00 00 00 00 80 00 00 00 00 ................
> > backtrace:
> > [<ffffffefd1694708>] create_object.isra.0+0x108/0x344
> > [<ffffffefd1d1a850>] kmemleak_alloc+0x8c/0xd0
> > [<ffffffefd167e2d0>] __kmalloc+0x440/0x6f0
> > [<ffffffefd1a960a4>] of_dma_get_range+0x124/0x220
> > [<ffffffefd1a8ce90>] of_dma_configure_id+0x40/0x2d0
> > [<ffffffefd198b68c>] platform_dma_configure+0x5c/0xa4
> > [<ffffffefd198846c>] really_probe+0x8c/0x514
> > [<ffffffefd1988990>] __driver_probe_device+0x9c/0x19c
> > [<ffffffefd1988cd8>] device_driver_attach+0x54/0xbc
> > [<ffffffefd1986634>] bind_store+0xc4/0x120
> > [<ffffffefd19856e0>] drv_attr_store+0x30/0x44
> > [<ffffffefd173c9b0>] sysfs_kf_write+0x50/0x60
> > [<ffffffefd173c1c4>] kernfs_fop_write_iter+0x124/0x1b4
> > [<ffffffefd16a013c>] new_sync_write+0xdc/0x160
> > [<ffffffefd16a256c>] vfs_write+0x23c/0x2a0
> > [<ffffffefd16a2758>] ksys_write+0x64/0xec
> >
> > To prevent this we should free the dma_range_map when the device is
> > released.
> >
> > Signed-off-by: M?rten Lindahl <[email protected]>
>
> Suggested-by: Rob Herring <[email protected]>
> Fixes: e0d072782c73 ("dma-mapping: introduce DMA range map,
> supplanting dma_pfn_offset")
> Reviewed-by: Rob Herring <[email protected]>
>
> I've got a follow-up patch to refactor the 3 occurrences of this same
> deinit code.
Now queued up, so feel free to send the follow-up patch too.
thanks,
greg k-h