Re: lock in vsprintf(): was: Re: [PATCH] of: Add printf '%pOFm' for generating modalias

[Petr Mladek]

On Wed 2024-12-18 15:13:13, John Ogness wrote:
> On 2024-12-18, Petr Mladek <pmladek@xxxxxxxx> wrote:
> > My concern is taking a lock when processing a printf format, see
> > below for more details.
> 
> Your concern is valid! printk() uses vsnprintf() to format records for
> the kernel log. printk() may be called from contexts where locking is
> forbidden (such as NMI). If vsnprintf() can take a lock, then either:
> 
> vsnprintf() must be made to be lockless
> 
>   or
> 
> printk() must take another approach to string formatting

There were the ideas to introduce a %pX[1] or %pf[2] formats. They would
allow to pass pointer to a custom callback and parameters which
then would be used by vsprintf(). This might allow to make it more
obvious that the given vsprintf()/printk() might do some locking.

>   or
> 
> we accept that printk() can deadlock for certain format types in certain
> contexts.

I see that %pOFm does not add any new locking dependency.
The same lock (devtree_lock) is already taken, for example,
by %pOFC format.

I do not want to revert everything now just because of theoretical
problems. It somehow works because people use these formats only
in dedicated subsystems. Also lockdep is able to catch a lot of
problems.

Well, it would be nice to document the lock dependency in
Documentation/core-api/printk-formats.rst

Best Regards,
Petr