Re: [PATCH] iommu/amd: Fix extended features logging

[Petr Mladek <pmladek@suse.com>]

On Sun 2021-04-11 14:08:14, Joe Perches wrote:
> On Sun, 2021-04-11 at 21:52 +0200, John Ogness wrote:
> > I'd rather fix dev_info callers to allow pr_cont and then fix any code
> > that is using this workaround.
> 
> Assuming you mean all dev_<level>() uses, me too.
> 
> > And if the print maintainers agree it is OK to encourage
> > pr_cont(LOGLEVEL "...") usage, then people should really start using
> > that if the loglevel on those pieces is important.
> 
> I have no stong feeling about the use of pr_cont(<KERN_LEVEL>
> as valuable or not.  I think it's just a trivial bit that
> could be somewhat useful when interleaving occurs.
> 
> A somewhat better mechanism would be to have an explicit
> cookie use like:
> 
> 	cookie = printk_multipart_init(KERN_LEVEL, fmt, ...);
> 	while (<condition>)
> 		printk_multipart_cont(cookie, fmt, ...);
> 	printk_multipark_end(cookie, fmt, ...);
> 
> And separately, there should be a pr_debug_cont or equivalent.

We would need to implement this a lockless way. It is doable,
for example, using some per-CPU X per-context buffers. Which would
require to disable preemption in the section.

But I think that using dev_cont_info() would be easier after all.

That said, some printk_*_init()/end() API would be useful
for storing the pieces in a temporary buffer. It would allow
to store the entire lines without the risk of interleaving
with other messages.

Best Regards,
Petr