On 11/12/19 4:25 AM, Sergey Senozhatsky wrote:

On (19/11/12 02:40), Dmitry Safonov wrote:
[..]

quoted

In my point of view the cost of one-time [mostly build] testing every
architecture is cheaper than introducing some new smart code that will
live forever.

Well, there may be the need to pass loglevel deeper due to "hey __show_stack()
on that arch invokes bar(), which invokes foo() and now foo() does printk(),
but we don't see it". The context which decided to backtaraces decided
to do so for a reason, probably, so I guess we can look at it as "a special
error reporting code block".

The proposed patch set passes loglevel via slightly unusual channel -
via sprintf(). We probably can do it, but I would prefer to minimize
the number of such printk-s in the kernel. The code snippet which I
posted also does pretty unusual thing w.r.t loglevel. Both approaches
are "non-standard" from that POV.

I don't strongly disagree, but if you look at those results:
git grep 'printk("%s.*", \(lvl\|level\)'

it seems to be used in quite a few places.

Thanks,
          Dmitry