On Wed, Jan 29, 2025, at 12:48, oushixiong wrote:
> 在 2025/1/28 17:39, Amit Shah 写道:
>> On Tue, 2025-01-28 at 13:52 +0800, oushixiong1025@xxxxxxx wrote:
>>
>> Hm, I'm not entirely sure about this - the devm_ interface is better,
>> but to me that just says that it's an extra safety net that avoids
>> memleaks when we forget to kfree, and not that we deliberately do not
>> free and get lax about managing allocated memory.
>>
>> So I'd prefer a patch that keeps all current frees as they are, but
>> with the added advantage of using the devm interfaces.
>>
>
> Hi, I don't think driver need to call kfree() by manual if driver use
> the devm_ interface
Calling the plain kfree() on memory allocated by devm_kmalloc()
is a bug, it has to be freed when the device is released or
explicitly using devm_kfree(). The explicit devm_kfree() in
turn is what you'd use only if memory has to be freed early
while the device is still going to be used.
> in the probe() phase. This can simplifies the code.
>
> This is similar to the commit "55dd7378ba94 (fbdev: imxfb: Convert to
> devm_kmalloc_array())"
I don't think it's worth converting drivers that correctly use
kmalloc()/kfree() already. For new drivers, the devm_* functions
are usually the easiest to use and review, and it makes sense to
convert drivers that have known memory leaks from incorrect
cleanup paths, but I don't think it's a good use of developer
or reviewer time to convert drivers that are not wrong for the
sake of simplicity. There is a memory overhead in the devm_*
functions and there is a risk of introducing bugs if the conversion
is done incorrectly, e.g. when a regular kfree() is left behind,
or when changing allocations that are meant to be freed before
.release() is called.
Arnd
