switchtec_dma_create() takes a reference on the PCI device with
get_device() and stores it in dma->dev. The reference is normally
dropped by switchtec_dma_release(), the device_release() callback,
which only runs after a successful dma_async_device_register(). If
the registration fails, the code frees the dma device directly from
the err_chans_release_exit label without ever dropping that extra
reference, leaking the reference taken on the PCI device.

Drop the reference with put_device() on the error path.

Fixes: d9587042b50f ("dmaengine: switchtec-dma: Introduce Switchtec DMA engine skeleton")
Cc: stable@vger.kernel.org
Signed-off-by: Wentao Liang <vulab@iscas.ac.cn>
---
 drivers/dma/switchtec_dma.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/dma/switchtec_dma.c b/drivers/dma/switchtec_dma.c
index 3ef928640615..b07bdd0ac540 100644
--- a/drivers/dma/switchtec_dma.c
+++ b/drivers/dma/switchtec_dma.c
@@ -1297,6 +1297,7 @@ static int switchtec_dma_create(struct pci_dev *pdev)
 
 err_chans_release_exit:
 	switchtec_dma_chans_release(pdev, swdma_dev);
+	put_device(dma->dev);
 
 err_exit:
 	if (swdma_dev->chan_status_irq)
-- 
2.34.1

On Wed, Sep 16, 2026 at 09:33:32AM +0000, Wentao Liang wrote:
> switchtec_dma_create() takes a reference on the PCI device with
> get_device() and stores it in dma->dev. The reference is normally
> dropped by switchtec_dma_release(), the device_release() callback,
> which only runs after a successful dma_async_device_register(). If
> the registration fails, the code frees the dma device directly from
> the err_chans_release_exit label without ever dropping that extra
> reference, leaking the reference taken on the PCI device.
>
> Drop the reference with put_device() on the error path.
>
> Fixes: d9587042b50f ("dmaengine: switchtec-dma: Introduce Switchtec DMA engine skeleton")
> Cc: stable@vger.kernel.org
> Signed-off-by: Wentao Liang <vulab@iscas.ac.cn>
> ---
>  drivers/dma/switchtec_dma.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/drivers/dma/switchtec_dma.c b/drivers/dma/switchtec_dma.c
> index 3ef928640615..b07bdd0ac540 100644
> --- a/drivers/dma/switchtec_dma.c
> +++ b/drivers/dma/switchtec_dma.c
> @@ -1297,6 +1297,7 @@ static int switchtec_dma_create(struct pci_dev *pdev)
>
>  err_chans_release_exit:
>  	switchtec_dma_chans_release(pdev, swdma_dev);
> +	put_device(dma->dev);

Does switchtec_dma_remove() also need put_device()?

Frank

>
>  err_exit:
>  	if (swdma_dev->chan_status_irq)
> --
> 2.34.1
>

On 2026-09-16 10:05, Frank Li wrote:
> On Wed, Sep 16, 2026 at 09:33:32AM +0000, Wentao Liang wrote:
>> switchtec_dma_create() takes a reference on the PCI device with
>> get_device() and stores it in dma->dev. The reference is normally
>> dropped by switchtec_dma_release(), the device_release() callback,
>> which only runs after a successful dma_async_device_register(). If
>> the registration fails, the code frees the dma device directly from
>> the err_chans_release_exit label without ever dropping that extra
>> reference, leaking the reference taken on the PCI device.
>>
>> Drop the reference with put_device() on the error path.
>>
>> Fixes: d9587042b50f ("dmaengine: switchtec-dma: Introduce Switchtec DMA engine skeleton")
>> Cc: stable@vger.kernel.org
>> Signed-off-by: Wentao Liang <vulab@iscas.ac.cn>
>> ---
>>  drivers/dma/switchtec_dma.c | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/dma/switchtec_dma.c b/drivers/dma/switchtec_dma.c
>> index 3ef928640615..b07bdd0ac540 100644
>> --- a/drivers/dma/switchtec_dma.c
>> +++ b/drivers/dma/switchtec_dma.c
>> @@ -1297,6 +1297,7 @@ static int switchtec_dma_create(struct pci_dev *pdev)
>>
>>  err_chans_release_exit:
>>  	switchtec_dma_chans_release(pdev, swdma_dev);
>> +	put_device(dma->dev);
> 
> Does switchtec_dma_remove() also need put_device()?

No, the release already happens in switchtec_dma_release() after the
device is removed. It can't be done in swtichtec_dma_remove() or it will
be potentially used if the device still exists after
switchtec_dma_remove() completes.

The patch looks correct to me but it will conflict with the patches in
my series waiting for merging:

https://lore.kernel.org/all/20260902062153.31048-1-logang@deltatee.com/T/#u

I'd appreciate it if my series could go in first and this gets rebased
on top of it.

Other than that:

Reviewed-by: Logan Gunthorpe <logang@deltatee.com>

Thanks!

Logan