Re: buffer allocation: was: [PATCH v3 3/3] printk: use the lockless ringbuffer

[Petr Mladek]

On Mon 2020-06-29 23:57:59, John Ogness wrote:
> On 2020-06-29, Petr Mladek <pmladek@xxxxxxxx> wrote:
> >> @@ @@ void __init setup_log_buf(int early)
> >> +	prb_init(&printk_rb_dynamic,
> >> +		 new_log_buf, order_base_2(new_log_buf_len),
> >> +		 new_dict_buf, order_base_2(new_log_buf_len),
> >> +		 new_descs, order_base_2(new_descs_count));
> >
> > order_base_2() is safe. But the result might be tat some allocated
> > space is not used.
> >
> > I would prefer to make sure that new_log_buf_len is rounded, e.g.
> > by roundup_pow_of_two(), at the beginning of the function. Then we
> > could use ilog2() here.
> 
> new_log_buf_len can only be set within log_buf_len_update(), and it
> is already doing exactly what you want:
> 
>         if (size)
>                 size = roundup_pow_of_two(size);
>         if (size > log_buf_len)
>                 new_log_buf_len = (unsigned long)size;

I know. I though about repeating this in setup_log_bug() to make it
error proof. But it is not really needed. I do not resist on it.

> I can switch to ilog2() instead of the more conservative order_base_2().

Yes, please change it to ilog2().

Best Regards,
Petr