RE: [PATCH v1] kunit: fix failure to build without printk

[Tim.Bird]

> -----Original Message-----
> From: Brendan Higgins
> 
> On Tue, Aug 27, 2019 at 3:00 PM shuah <shuah@xxxxxxxxxx> wrote:
> >
> > On 8/27/19 3:36 PM, Brendan Higgins wrote:
> > > On Tue, Aug 27, 2019 at 2:09 PM Brendan Higgins
> > > <brendanhiggins@xxxxxxxxxx> wrote:
> > >>
> > >> On Tue, Aug 27, 2019 at 2:03 PM Brendan Higgins
> > >> <brendanhiggins@xxxxxxxxxx> wrote:
> > >>>
> > >>> On Tue, Aug 27, 2019 at 1:21 PM shuah <shuah@xxxxxxxxxx> wrote:
> > >>>>
> > >>>> On 8/27/19 11:49 AM, Brendan Higgins wrote:
> > >>>>> Previously KUnit assumed that printk would always be present,
> which is
> > >>>>> not a valid assumption to make. Fix that by ifdefing out functions
> which
> > >>>>> directly depend on printk core functions similar to what dev_printk
> > >>>>> does.
> > >>>>>
> > >>>>> Reported-by: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
> > >>>>> Link: https://lore.kernel.org/linux-kselftest/0352fae9-564f-4a97-
> 715a-fabe016259df@xxxxxxxxxx/T/#t
> > >>>>> Cc: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
> > >>>>> Signed-off-by: Brendan Higgins <brendanhiggins@xxxxxxxxxx>
> > >>>>> ---
> > >>>>>    include/kunit/test.h |  7 +++++++
> > >>>>>    kunit/test.c         | 41 ++++++++++++++++++++++++-----------------
> > >>>>>    2 files changed, 31 insertions(+), 17 deletions(-)
> > >>>>>
> > >>>>> diff --git a/include/kunit/test.h b/include/kunit/test.h
> > >>>>> index 8b7eb03d4971..339af5f95c4a 100644
> > >>>>> --- a/include/kunit/test.h
> > >>>>> +++ b/include/kunit/test.h
> > >>>>> @@ -339,9 +339,16 @@ static inline void *kunit_kzalloc(struct kunit
> *test, size_t size, gfp_t gfp)
> > >>> [...]
> > >>>> Okay after reviewing this, I am not sure why you need to do all
> > >>>> this.
> > >>>>
> > >>>> Why can't you just change the root function that throws the warn:
> > >>>>
> > >>>>    static int kunit_vprintk_emit(int level, const char *fmt, va_list args)
> > >>>> {
> > >>>>           return vprintk_emit(0, level, NULL, 0, fmt, args);
> > >>>> }
> > >>>>
> > >>>> You aren'r really doing anything extra here, other than calling
> > >>>> vprintk_emit()
> > >>>
> > >>> Yeah, I did that a while ago. I think it was a combination of trying
> > >>> to avoid an extra layer of adding and then removing the log level, and
> > >>> that's what dev_printk and friends did.
> > >>>
> > >>> But I think you are probably right. It's a lot of maintenance overhead
> > >>> to get rid of that. Probably best to just use what the printk people
> > >>> have.
> > >>>
> > >>>> Unless I am missing something, can't you solve this problem by
> including
> > >>>> printk.h and let it handle the !CONFIG_PRINTK case?
> > >>>
> > >>> Randy, I hope you don't mind, but I am going to ask you to re-ack my
> > >>> next revision since it basically addresses the problem in a totally
> > >>> different way.
> > >>
> > >> Actually, scratch that. Checkpatch doesn't like me calling printk
> > >> without using a KERN_<LEVEL>.
> > >>
> > >> Now that I am thinking back to when I wrote this. I think it also
> > >> might not like using a dynamic KERN_<LEVEL> either (printk("%s my
> > >> message", KERN_INFO)).
> > >>
> > >> I am going to have to do some more investigation.
> > >
> > > Alright, I am pretty sure it is safe to do printk("%smessage",
> KERN_<LEVEL>);
> > >
> > > Looking at the printk implementation, it appears to do the format
> > > before it checks the log level:
> > >
> > >
> https://elixir.bootlin.com/linux/v5.2.10/source/kernel/printk/printk.c#L1907
> > >
> > > So I am pretty sure we can do it either with the vprintk_emit or with
> printk.
> >
> > Let me see if we are on the same page first. I am asking if you can
> > just include printk.h for vprintk_emit() define for both CONFIG_PRINTK
> > and !CONFIG_PRINTK cases.
> 
> Ah sorry, I misunderstood you.
> 
> No, that doesn't work. I tried including linux/printk.h, and I get the
> same error.
> 
> The reason for this is that vprintk_emit() is only defined when
> CONFIG_PRINTK=y:
> 
> https://elixir.bootlin.com/linux/latest/ident/vprintk_emit

Ugh.  That's just a bug in include/linux/printk.h

There should be a stub definition for vprintk_emit() in the #else part 
of #ifdef CONFIG_PRINTK.

You shouldn't be dealing with whether printk is present or not
in the kunit code.  All the printk-related routines are supposed
to evaporate themselves, based on the conditional in
include/linux/printk.h

That should be fixed there instead of in your code.

Let me know if you'd like me to submit a patch for that.  I only hesitate
because your patch depends on it, and IMHO it makes more sense to
include it in your batch than separately. Otherwise there's a patch race
condition.
 -- Tim