Re: [RFC 1/2] lib/vsprintf: Add support for pte_t

[Petr Mladek <pmladek@suse.com>]

On Wed 2025-06-18 19:16:00, Pedro Falcato wrote:
> On Wed, Jun 18, 2025 at 10:44:20AM +0200, David Hildenbrand wrote:
> > On 18.06.25 10:37, Anshuman Khandual wrote:
> > > 
> > > 
> > > On 18/06/25 1:48 PM, David Hildenbrand wrote:
> > > > On 18.06.25 06:12, Anshuman Khandual wrote:
> > > > > Add a new format for printing page table entries.
> > > > > 
> > > > > Cc: Petr Mladek <pmladek@suse.com>
> > > > > Cc: Steven Rostedt <rostedt@goodmis.org>
> > > > > Cc: Jonathan Corbet <corbet@lwn.net>
> > > > > Cc: Andrew Morton <akpm@linux-foundation.org>
> > > > > Cc: David Hildenbrand <david@redhat.com>
> > > > > Cc: linux-doc@vger.kernel.org
> > > > > Cc: linux-kernel@vger.kernel.org
> > > > > Cc: linux-mm@kvack.org
> > > > > Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
> > > > > ---
> > > > >    Documentation/core-api/printk-formats.rst | 14 ++++++++++++++
> > > > >    lib/vsprintf.c                            | 20 ++++++++++++++++++++
> > > > >    mm/memory.c                               |  5 ++---
> > > > >    scripts/checkpatch.pl                     |  2 +-
> > > > >    4 files changed, 37 insertions(+), 4 deletions(-)
> > > > > 
> > > > > diff --git a/Documentation/core-api/printk-formats.rst b/Documentation/core-api/printk-formats.rst
> > > > > index 4b7f3646ec6ce..75a110b059ee1 100644
> > > > > --- a/Documentation/core-api/printk-formats.rst
> > > > > +++ b/Documentation/core-api/printk-formats.rst
> > > > > @@ -689,6 +689,20 @@ Rust
> > > > >    Only intended to be used from Rust code to format ``core::fmt::Arguments``.
> > > > >    Do *not* use it from C.
> > > > >    +Page Table Entry
> > > > > +----------------
> > > > > +
> > > > > +::
> > > > > +        %ppte
> > > > > +
> > > > > +Print standard page table entry pte_t.
> > > > > +
> > > > > +Passed by reference.
> > > > 
> > > > Curious, why the decision to pass by reference?
> > > 
> > > Just to make this via %p<> based address mechanism. But wondering
> > > will it be better for the pte to be represented via value instead
> > > of reference ?
> > 
> > We commonly pass ptes to functions through value, not reference, that's why
> > I am asking.
> 
> 
> All printf/printk extensions in the kernel follow %p<some letters> and use
> pointers because %p takes pointers, so it lets us use -Wformat with no issues.
> 
> So yes, taking a pte_t * is required.

Correct. But the pointer is usually needed because the %pxx format
need to access a structure.

Passing a pointer is another potential source of errors. I mean that
the callers might pass an invalid pointer by mistake...

Another aspect is performance. It is likely not a big deal for classic
printk() which is a slow path. But trace_printk() tries to optimize
the speed by deferred formatting where possible, see vbin_printf()
and bstr_printf().

I think that this is not a blocker for this patchset. But you should
know that using %pxx has a cost.

Best Regards,
Petr