On Tue, Aug 6, 2019 at 3:29 AM laokz <laokz@xxxxxxxxxxx> wrote:
>
> Sorry for the late reply. I didn't submit any patch, for as newbie I wasn't
> quite sure about the severity of this issue. The developers may do their
> choice.
It does seem pretty bad.
Also, INIT_KFIFO() and DECLARE_KFIFO() should probably have a
BUILD_BUG_ON(!__is_kfifo_ptr && !is_power_of_2(ARRAY_SIZE(__tmp->buf)));
or something. Probably worth indirection through a helper macro to set
the ".mask" field.
And yes, commit dfe2a77fd243 ("kfifo: fix kfifo_alloc() and
kfifo_init()") seems wrong, It's fine for the __kfifo_alloc() case
where we actually allocate the rounded-up size, but it's completely
wrong for the __kfifo_init() case where somebody _else_ allocated the
size.
So the kfifo_init() case needs to just use a round_down, and possibly
add a WARN_ON_ONCE() too.
Anybody?
Linus
