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

[Brendan Higgins]

On Tue, Aug 27, 2019 at 3:55 PM shuah <shuah@xxxxxxxxxx> wrote:
>
> On 8/27/19 4:16 PM, Brendan Higgins wrote:
> > 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:
> >
>
> This is the real problem here. printk.h defines several for
> !CONFIG_PRINTK case.

Yeah, Tim pointed that out.

I think both of you are right, I should be filing my fix against them.

> > https://elixir.bootlin.com/linux/latest/ident/vprintk_emit
> >
> >> I am not asking you to use printk() in place of vprintk_emit().
> >> It is perfectly fine to use vprintk_emit()
> >
> > Okay, cool.
> >
> >>>
> >>> So it appears that we have to weigh the following trade-offs:
> >>>
> >>> Using vprintk_emit:
> >>>
> >>> Pros:
> >>>    - That's what dev_printk uses.
> >>
> >> Not sure what you mean by this. I am suggesting if you can just
> >> call vprintk_emit() and include printk.h and not have to ifdef
> >> around all the other callers of kunit_vprintk_emit()
> >
> > Oh, I was just saying that I heavily based my implementation of
> > kunit_printk on dev_printk. So I have a high degree of confidence that
> > it is okay to use it the way that I am using it.
> >
> >> Yes. There is the other issue of why do you need the complexity
> >> of having kunit_vprintk_emit() at all.
> >
> > Right, and the problem with the alternative, is there is no good
> > kernel API for logging with the log level set dynamically. printk
> > prefers to have it as a string prefix on the format string, but I
> > cannot do that because I need to add my own prefix to the format
> > string.
> >
> > So, I guess I should just go ahead and address the earlier comments on
> > this patch?
> >
>
> So what does your code do in the case of !CONFIG_PRINTK. From my read of
> it, it returns 0 from kunit_printk_emit(0 from my read of it. What I am
> saying is this is a lot of indirection instead of fixing the leaf
> function which is kunit_vprintk_emit().

Agreed. My apologies, as I mentioned in response to Tim, I just
assumed I was using it wrong.

> +#else /* CONFIG_PRINTK */
> +static inline int kunit_printk_emit(int level, const char *fmt, ...)
> +{
> +       return 0;
> +}
> +#endif /* CONFIG_PRINTK */
>
> Does the following work?
>
> #if defined CONFIG_PRINTK
> static int kunit_vprintk_emit(int level, const char *fmt, va_list args)
> {
>          return vprintk_emit(0, level, NULL, 0, fmt, args);
> }
> #else
> static int kunit_vprintk_emit(int level, const char *fmt, va_list args)
> {
>         return 0;
> }
> #endif
>
> I think the real problem is in the printk.h with its missing define for
> vprintk_emit() for !CONFIG_PRINTK case. There seem to only one call for
> this in drivers/base/core.c in CONFIG_PRINTK path. Unless I am totally
> missing some context for why there is no stub for vprintk_emit() for
> !CONFIG_PRINTK case

Agreed.

Sorry again for the confusion.

Thanks!