[PATCH v2 1/6] reboot-mode: Use actual cell size instead of sizeof(u32)

[Sean Anderson]

On 2/13/26 10:46, Sean Anderson wrote:
> On 2/13/26 06:27, Aswin Murugan wrote:
>> The nvmem reboot-mode driver hardcodes sizeof(u32) for NVMEM read/write
>> operations. This fails on platforms where the reboot reason storage is
>> smaller than 4 bytes.
>>
>> On Qualcomm platforms with single-byte PMIC register storage, attempting
>> to read 4 bytes from a 1-byte cell causes the NVMEM subsystem to return
>> -EINVAL, breaking reboot-mode functionality.
>>
>> Replace sizeof(*mode) and sizeof(mode) with priv->cell.size to use the
>> actual NVMEM cell size. Also initialize *mode to 0 in reboot_mode_get()
>> to ensure a clean state for cells smaller than u32.
>>
>> This allows the driver to work with cells of any size, from single bytes
>> to multi-byte words.
>>
>> Signed-off-by: Aswin Murugan <aswin.murugan at oss.qualcomm.com>
>> ---
>>   drivers/reboot-mode/reboot-mode-nvmem.c | 5 +++--
>>   1 file changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/reboot-mode/reboot-mode-nvmem.c b/drivers/reboot-mode/reboot-mode-nvmem.c
>> index b9af242520a..60307cce49f 100644
>> --- a/drivers/reboot-mode/reboot-mode-nvmem.c
>> +++ b/drivers/reboot-mode/reboot-mode-nvmem.c
>> @@ -18,15 +18,16 @@ struct nvmem_reboot_mode_priv {
>>   static int reboot_mode_get(struct udevice *dev, u32 *mode)
>>   {
>>       struct nvmem_reboot_mode_priv *priv = dev_get_priv(dev);
>> +    *mode = 0;
>> -    return nvmem_cell_read(&priv->cell, mode, sizeof(*mode));
>> +    return nvmem_cell_read(&priv->cell, mode, priv->cell.size);
>>   }
>>   static int reboot_mode_set(struct udevice *dev, u32 mode)
>>   {
>>       struct nvmem_reboot_mode_priv *priv = dev_get_priv(dev);
>> -    return nvmem_cell_write(&priv->cell, &mode, sizeof(mode));
>> +    return nvmem_cell_write(&priv->cell, &mode, priv->cell.size);
>>   }
>>   static const struct reboot_mode_ops nvmem_reboot_mode_ops = {
> 
> I'm not sure about this. drivers/power/reset/nvmem-reboot-mode.c in Linux
> always writes sizeof(unsigned long), and this is checked against the nvmem
> cell length in __nvmem_cell_entry_write. We should try to keep our behavior
> close to Linux's.
> 
> --Sean

Actually, NACK since this smashes the stack with a cell size that is larger than 4.

--Sean