In case the nvmem DT node contains "wp-gpios" DT property, the device
currently defaults to read-write and the force_ro sysfs attribute reads
0. Switch to the default read-only, which is both safer, and aligned
with eMMC HW BOOT partition force_ro sysfs attribute behavior, which
also defaults to read-only.

The adjustment of nvmem->read_only value to read-only in case wp-gpios
DT property is present must be done only after the device_add() got
called because device_add() does internally call nvmem_bin_attr_get_umode(),
which configures the permissions of 'nvmem' bin attr based on the value
of nvmem->read_only that is only parsed from DT property 'read-only',
without any adjustment. This way, if DT property 'read-only' is present,
the 'nvmem' attribute is always read-only. Otherwise, if the device is
writeable, then 'nvmem' attribute is writeable, and nvmem->read_only
defaults to read-only, but can be switched to read-write at runtime via
the 'force_ro' attribute.

The updated behavior can be tested as follows:

Current content:
"
$ cat /sys/bus/nvmem/devices/logging7/force_ro
1

$ hexdump -C /sys/bus/nvmem/devices/logging7/nvmem
00000000  66 6f 6f 0a ff ff ff ff
"

Write into default-read-only device:
"
$ echo bar > /sys/bus/nvmem/devices/logging7/nvmem
bash: echo: write error: Operation not permitted
$ cat /sys/bus/nvmem/devices/logging7/force_ro
1
"

Unlock and write into device:
"
$ echo 0 > /sys/bus/nvmem/devices/logging7/force_ro
$ cat /sys/bus/nvmem/devices/logging7/force_ro
0

$ echo bar > /sys/bus/nvmem/devices/logging7/nvmem
$ hexdump -C /sys/bus/nvmem/devices/logging7/nvmem
00000000  62 61 72 0a ff ff ff ff
"

Relock and write into device, fails because device is read-only again:
"
$ echo 1 > /sys/bus/nvmem/devices/logging7/force_ro
$ echo baz > /sys/bus/nvmem/devices/logging7/nvmem
bash: echo: write error: Operation not permitted

$ hexdump -C /sys/bus/nvmem/devices/logging7/nvmem
00000000  62 61 72 0a ff ff ff ff
"

Signed-off-by: Marek Vasut <marex@nabladev.com>
---
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Bartosz Golaszewski <brgl@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Srinivas Kandagatla <srini@kernel.org>
Cc: linux-i2c@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
---
V2: Simplify the implementation, make devices with wp-gpios default
    to read-only and let users unlock those devices via force_ro
    sysfs attribute, just like eMMC HW BOOT partitions
V3: Drop unnecessary gpiod_set_value_cansleep()
---
 drivers/nvmem/core.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
index 311cb2e5a5c02..be28a366f6031 100644
--- a/drivers/nvmem/core.c
+++ b/drivers/nvmem/core.c
@@ -1019,6 +1019,10 @@ struct nvmem_device *nvmem_register(const struct nvmem_config *config)
 	if (rval)
 		goto err_remove_dev;
 
+	/* If the device has WP GPIO, default to read-only */
+	if (nvmem->wp_gpio)
+		nvmem->read_only = true;
+
 #ifdef CONFIG_NVMEM_SYSFS
 	rval = nvmem_populate_sysfs_cells(nvmem);
 	if (rval)
-- 
2.53.0

On Mon, 27 Apr 2026 17:19:29 +0200, Marek Vasut <marex@nabladev.com> said:
> In case the nvmem DT node contains "wp-gpios" DT property, the device
> currently defaults to read-write and the force_ro sysfs attribute reads
> 0. Switch to the default read-only, which is both safer, and aligned
> with eMMC HW BOOT partition force_ro sysfs attribute behavior, which
> also defaults to read-only.
>
> The adjustment of nvmem->read_only value to read-only in case wp-gpios
> DT property is present must be done only after the device_add() got
> called because device_add() does internally call nvmem_bin_attr_get_umode(),
> which configures the permissions of 'nvmem' bin attr based on the value
> of nvmem->read_only that is only parsed from DT property 'read-only',
> without any adjustment. This way, if DT property 'read-only' is present,
> the 'nvmem' attribute is always read-only. Otherwise, if the device is
> writeable, then 'nvmem' attribute is writeable, and nvmem->read_only
> defaults to read-only, but can be switched to read-write at runtime via
> the 'force_ro' attribute.

This slightly changes the behavior for user-space but I'm willing to give it
a go and see if anyone complains.

Reviewed-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>

Bart