2018-05-08 20:50:22

by Alexey Khoroshilov

[permalink] [raw]
Subject: [PATCH] mtd: nxp-spifi: decrement flash_np refcnt on error paths

nxp_spifi_probe() increments refcnt of SPI flash device node by
of_get_next_available_child() and then it passes the node
to mtd device in nxp_spifi_setup_flash().
But if a failure happens before mtd_device_register() succeed,
the refcnt is left undecremented.

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <[email protected]>
---
drivers/mtd/spi-nor/nxp-spifi.c | 16 +++++++++++-----
1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/mtd/spi-nor/nxp-spifi.c b/drivers/mtd/spi-nor/nxp-spifi.c
index 15374216d4d9..8919e31f2ab8 100644
--- a/drivers/mtd/spi-nor/nxp-spifi.c
+++ b/drivers/mtd/spi-nor/nxp-spifi.c
@@ -294,7 +294,8 @@ static int nxp_spifi_setup_flash(struct nxp_spifi *spifi,
break;
default:
dev_err(spifi->dev, "unsupported rx-bus-width\n");
- return -EINVAL;
+ ret = -EINVAL;
+ goto err_node_put;
}
}

@@ -328,7 +329,8 @@ static int nxp_spifi_setup_flash(struct nxp_spifi *spifi,
break;
default:
dev_err(spifi->dev, "only mode 0 and 3 supported\n");
- return -EINVAL;
+ ret = -EINVAL;
+ goto err_node_put;
}

writel(ctrl, spifi->io_base + SPIFI_CTRL);
@@ -356,22 +358,26 @@ static int nxp_spifi_setup_flash(struct nxp_spifi *spifi,
ret = spi_nor_scan(&spifi->nor, NULL, &hwcaps);
if (ret) {
dev_err(spifi->dev, "device scan failed\n");
- return ret;
+ goto err_node_put;
}

ret = nxp_spifi_setup_memory_cmd(spifi);
if (ret) {
dev_err(spifi->dev, "memory command setup failed\n");
- return ret;
+ goto err_node_put;
}

ret = mtd_device_register(&spifi->nor.mtd, NULL, 0);
if (ret) {
dev_err(spifi->dev, "mtd device parse failed\n");
- return ret;
+ goto err_node_put;
}

return 0;
+
+err_node_put:
+ of_node_put(np);
+ return ret;
}

static int nxp_spifi_probe(struct platform_device *pdev)
--
2.7.4



2018-05-09 09:44:04

by Boris Brezillon

[permalink] [raw]
Subject: Re: [PATCH] mtd: nxp-spifi: decrement flash_np refcnt on error paths

On Tue, 8 May 2018 23:47:36 +0300
Alexey Khoroshilov <[email protected]> wrote:

> nxp_spifi_probe() increments refcnt of SPI flash device node by
> of_get_next_available_child() and then it passes the node
> to mtd device in nxp_spifi_setup_flash().
> But if a failure happens before mtd_device_register() succeed,
> the refcnt is left undecremented.

Why not doing that in the error path of the probe function? Also, you
probably want to call of_node_put() in the ->remove() function.

>
> Found by Linux Driver Verification project (linuxtesting.org).
>
> Signed-off-by: Alexey Khoroshilov <[email protected]>
> ---
> drivers/mtd/spi-nor/nxp-spifi.c | 16 +++++++++++-----
> 1 file changed, 11 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/mtd/spi-nor/nxp-spifi.c b/drivers/mtd/spi-nor/nxp-spifi.c
> index 15374216d4d9..8919e31f2ab8 100644
> --- a/drivers/mtd/spi-nor/nxp-spifi.c
> +++ b/drivers/mtd/spi-nor/nxp-spifi.c
> @@ -294,7 +294,8 @@ static int nxp_spifi_setup_flash(struct nxp_spifi *spifi,
> break;
> default:
> dev_err(spifi->dev, "unsupported rx-bus-width\n");
> - return -EINVAL;
> + ret = -EINVAL;
> + goto err_node_put;
> }
> }
>
> @@ -328,7 +329,8 @@ static int nxp_spifi_setup_flash(struct nxp_spifi *spifi,
> break;
> default:
> dev_err(spifi->dev, "only mode 0 and 3 supported\n");
> - return -EINVAL;
> + ret = -EINVAL;
> + goto err_node_put;
> }
>
> writel(ctrl, spifi->io_base + SPIFI_CTRL);
> @@ -356,22 +358,26 @@ static int nxp_spifi_setup_flash(struct nxp_spifi *spifi,
> ret = spi_nor_scan(&spifi->nor, NULL, &hwcaps);
> if (ret) {
> dev_err(spifi->dev, "device scan failed\n");
> - return ret;
> + goto err_node_put;
> }
>
> ret = nxp_spifi_setup_memory_cmd(spifi);
> if (ret) {
> dev_err(spifi->dev, "memory command setup failed\n");
> - return ret;
> + goto err_node_put;
> }
>
> ret = mtd_device_register(&spifi->nor.mtd, NULL, 0);
> if (ret) {
> dev_err(spifi->dev, "mtd device parse failed\n");
> - return ret;
> + goto err_node_put;
> }
>
> return 0;
> +
> +err_node_put:
> + of_node_put(np);
> + return ret;
> }
>
> static int nxp_spifi_probe(struct platform_device *pdev)


2018-05-09 14:40:04

by Alexey Khoroshilov

[permalink] [raw]
Subject: Re: [PATCH] mtd: nxp-spifi: decrement flash_np refcnt on error paths

On 09.05.2018 12:42, Boris Brezillon wrote:
> On Tue, 8 May 2018 23:47:36 +0300
> Alexey Khoroshilov <[email protected]> wrote:
>
>> nxp_spifi_probe() increments refcnt of SPI flash device node by
>> of_get_next_available_child() and then it passes the node
>> to mtd device in nxp_spifi_setup_flash().
>> But if a failure happens before mtd_device_register() succeed,
>> the refcnt is left undecremented.
>
> Why not doing that in the error path of the probe function? Also, you
> probably want to call of_node_put() in the ->remove() function.
>


You are right.

I believed that after successful mtd_device_register()
the node is managed by mtd device. I missed that it calls of_node_get()
in add_mtd_device() by itself.

I will prepare v2.
But I guess there is no need to have of_node_put() in ->remove(), since
probe() finishes its own usage of flash_np, while mtd_device incremented
refcnt by itself and will decrement it in ->remove() in
mtd_device_unregister(&spifi->nor.mtd). So, I would propose
of_node_put() on both successful and error path.

Thank you,
Alexey

2018-05-09 14:41:38

by Boris Brezillon

[permalink] [raw]
Subject: Re: [PATCH] mtd: nxp-spifi: decrement flash_np refcnt on error paths

On Wed, 9 May 2018 17:35:41 +0300
Alexey Khoroshilov <[email protected]> wrote:

> On 09.05.2018 12:42, Boris Brezillon wrote:
> > On Tue, 8 May 2018 23:47:36 +0300
> > Alexey Khoroshilov <[email protected]> wrote:
> >
> >> nxp_spifi_probe() increments refcnt of SPI flash device node by
> >> of_get_next_available_child() and then it passes the node
> >> to mtd device in nxp_spifi_setup_flash().
> >> But if a failure happens before mtd_device_register() succeed,
> >> the refcnt is left undecremented.
> >
> > Why not doing that in the error path of the probe function? Also, you
> > probably want to call of_node_put() in the ->remove() function.
> >
>
>
> You are right.
>
> I believed that after successful mtd_device_register()
> the node is managed by mtd device. I missed that it calls of_node_get()
> in add_mtd_device() by itself.
>
> I will prepare v2.
> But I guess there is no need to have of_node_put() in ->remove(), since
> probe() finishes its own usage of flash_np, while mtd_device incremented
> refcnt by itself and will decrement it in ->remove() in
> mtd_device_unregister(&spifi->nor.mtd). So, I would propose
> of_node_put() on both successful and error path.

Sounds good.


2018-05-09 14:57:45

by Alexey Khoroshilov

[permalink] [raw]
Subject: [PATCH v2] mtd: nxp-spifi: release flash_np in nxp_spifi_probe()

nxp_spifi_probe() increments refcnt of SPI flash device node by
of_get_next_available_child() and leaves it undecremented on both
successful and error paths.

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <[email protected]>
---
drivers/mtd/spi-nor/nxp-spifi.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/mtd/spi-nor/nxp-spifi.c b/drivers/mtd/spi-nor/nxp-spifi.c
index 15374216d4d9..7b047951d0a2 100644
--- a/drivers/mtd/spi-nor/nxp-spifi.c
+++ b/drivers/mtd/spi-nor/nxp-spifi.c
@@ -438,11 +438,15 @@ static int nxp_spifi_probe(struct platform_device *pdev)
ret = nxp_spifi_setup_flash(spifi, flash_np);
if (ret) {
dev_err(&pdev->dev, "unable to setup flash chip\n");
- goto dis_clks;
+ goto put_np;
}

+ of_node_put(flash_np);
+
return 0;

+put_np:
+ of_node_put(flash_np);
dis_clks:
clk_disable_unprepare(spifi->clk_spifi);
dis_clk_reg:
--
2.7.4


2018-05-09 15:04:04

by Boris Brezillon

[permalink] [raw]
Subject: Re: [PATCH v2] mtd: nxp-spifi: release flash_np in nxp_spifi_probe()

On Wed, 9 May 2018 17:56:46 +0300
Alexey Khoroshilov <[email protected]> wrote:

> nxp_spifi_probe() increments refcnt of SPI flash device node by
> of_get_next_available_child() and leaves it undecremented on both
> successful and error paths.
>
> Found by Linux Driver Verification project (linuxtesting.org).
>
> Signed-off-by: Alexey Khoroshilov <[email protected]>
> ---
> drivers/mtd/spi-nor/nxp-spifi.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/mtd/spi-nor/nxp-spifi.c b/drivers/mtd/spi-nor/nxp-spifi.c
> index 15374216d4d9..7b047951d0a2 100644
> --- a/drivers/mtd/spi-nor/nxp-spifi.c
> +++ b/drivers/mtd/spi-nor/nxp-spifi.c
> @@ -438,11 +438,15 @@ static int nxp_spifi_probe(struct platform_device *pdev)
> ret = nxp_spifi_setup_flash(spifi, flash_np);

Just put the of_node_put() here and that's the only change you'll need.

> if (ret) {
> dev_err(&pdev->dev, "unable to setup flash chip\n");
> - goto dis_clks;
> + goto put_np;
> }
>
> + of_node_put(flash_np);
> +
> return 0;
>
> +put_np:
> + of_node_put(flash_np);
> dis_clks:
> clk_disable_unprepare(spifi->clk_spifi);
> dis_clk_reg:


2018-05-09 15:12:10

by Alexey Khoroshilov

[permalink] [raw]
Subject: [PATCH v3] mtd: nxp-spifi: release flash_np in nxp_spifi_probe()

nxp_spifi_probe() increments refcnt of SPI flash device node by
of_get_next_available_child() and leaves it undecremented on both
successful and error paths.

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <[email protected]>
---
v3: Move of_node_put() before return value check as Boris Brezillon suggested.

drivers/mtd/spi-nor/nxp-spifi.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/drivers/mtd/spi-nor/nxp-spifi.c b/drivers/mtd/spi-nor/nxp-spifi.c
index 15374216d4d9..0c9094ec5966 100644
--- a/drivers/mtd/spi-nor/nxp-spifi.c
+++ b/drivers/mtd/spi-nor/nxp-spifi.c
@@ -436,6 +436,7 @@ static int nxp_spifi_probe(struct platform_device *pdev)
}

ret = nxp_spifi_setup_flash(spifi, flash_np);
+ of_node_put(flash_np);
if (ret) {
dev_err(&pdev->dev, "unable to setup flash chip\n");
goto dis_clks;
--
2.7.4


2018-07-07 07:00:37

by Boris Brezillon

[permalink] [raw]
Subject: Re: [PATCH v3] mtd: nxp-spifi: release flash_np in nxp_spifi_probe()

On Wed, 9 May 2018 18:11:20 +0300
Alexey Khoroshilov <[email protected]> wrote:

> nxp_spifi_probe() increments refcnt of SPI flash device node by
> of_get_next_available_child() and leaves it undecremented on both
> successful and error paths.
>
> Found by Linux Driver Verification project (linuxtesting.org).
>
> Signed-off-by: Alexey Khoroshilov <[email protected]>

Applied after fixing the subject prefix.

> ---
> v3: Move of_node_put() before return value check as Boris Brezillon suggested.
>
> drivers/mtd/spi-nor/nxp-spifi.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/mtd/spi-nor/nxp-spifi.c b/drivers/mtd/spi-nor/nxp-spifi.c
> index 15374216d4d9..0c9094ec5966 100644
> --- a/drivers/mtd/spi-nor/nxp-spifi.c
> +++ b/drivers/mtd/spi-nor/nxp-spifi.c
> @@ -436,6 +436,7 @@ static int nxp_spifi_probe(struct platform_device *pdev)
> }
>
> ret = nxp_spifi_setup_flash(spifi, flash_np);
> + of_node_put(flash_np);
> if (ret) {
> dev_err(&pdev->dev, "unable to setup flash chip\n");
> goto dis_clks;