Re: [RFC] printk/sysrq: Don't play with console_loglevel

From: Dmitry Safonov
Date: Wed Jun 12 2019 - 10:52:20 EST

On 6/12/19 1:00 PM, Petr Mladek wrote:
> On Wed 2019-06-12 17:36:43, Sergey Senozhatsky wrote:
>> On (06/06/19 09:10), Petr Mladek wrote:
>>> Just to be sure. I wanted to say that I like the idea with
>>> KERN_UNSUPRESSED. So, I think that we are on the same page.
>> I understand. All I wanted to say is that KERN_UNSUPRESSED is
>> per-message, while the most interesting (and actually broken)
>> cases, IMHO, are per-context, IOW things like this one
>> console_loglevel = NEW
>> foo()
>> dump_stack()
>> printk
>> ...
>> printk
>> console_loglevel = OLD
>> KERN_UNSUPRESSED does not help here. We probably can't convert
>> dump_stack() to KERN_UNSUPRESSED.
> I agree. I take KERN_UNSUPRESSED like a nice trick how to pass
> the information about the unsupressed printk context via
> printk_safe and printk_nmi per-CPU buffers.
> The single line in sysrq __handle_sysrq() seems to be the only
> location where KERN_UNSUPRESSED can be used directly.

I likely don't understand all the reasons why it's not possible.

Looking at kdb - prints those can't be converted straight-away are
show_regs() and show_stack().. could we add helpers those take a
loglevel to show the info? (having as an example show_trace_log_lvl()
that can "eat" a modifier already).

Not that I want to sell the idea of KERN_UNSUPRESSED, but to my mind the
alternative patches were kind of too intricate and bring more complexity
to printk.. and there are only ~3(?) places those manipulate
console_loglevel directly in the kernel tree (so we might want something
simple and dumb as hell).