Re: [PATCH] tty: use printk_safe context at tty_msg()

[Tetsuo Handa]

On 2021/04/06 16:10, Greg Kroah-Hartman wrote:
> On Tue, Apr 06, 2021 at 02:31:43PM +0900, Tetsuo Handa wrote:
>> On 2021/04/06 13:51, Jiri Slaby wrote:
>>> On 03. 04. 21, 6:14, Tetsuo Handa wrote:
>>>> --- a/include/linux/tty.h
>>>> +++ b/include/linux/tty.h
>>>> @@ -14,6 +14,7 @@
>>>>   #include <uapi/linux/tty.h>
>>>>   #include <linux/rwsem.h>
>>>>   #include <linux/llist.h>
>>>> +#include <../../kernel/printk/internal.h>
>>>
>>> Including printk's internal header in linux/tty.h doesn't look correct to me.
>>>
>>
>> This is because this patch wants __printk_safe_enter()/__printk_safe_exit()
>> without #ifdef'ing CONFIG_PRINTK.
> 
> Then those functions need to be "properly" exported, not placed only in
> an "internal.h" file that obviously should not be included from anywhere
> like this.
> 
>> Peter and Sergey, what should we do?
>> Can we move printk_safe_enter_irqsave()/printk_safe_exit_irqrestore() to include/linux/printk.h ?
> 
> Are you sure that is the only way to resolve this?

Publishing printk_safe_enter_irqsave() etc. was once proposed at
https://lkml.kernel.org/r/20181016050428.17966-3-sergey.senozhatsky@xxxxxxxxx and
Peter Zijlstra did not like such change. But we reconfirmed that "tty_port lock must
switch to printk_safe" at https://lkml.kernel.org/r/20190219013254.GA20023@jagdpanzerIV .

Two years has elapsed since a completely new printk implementation which would not
use printk_safe context was proposed, but printk_safe context is still remaining.
Therefore, we might need to tolerate, with a warning that printk_safe_enter_irqsave()
etc. are not intended for general use, use of printk_safe context for tty_msg() and
tty_buffer_alloc().