Re: [PATCH] mm/highmem: Add notes about conversions from kmap{,_atomic}()

[Fabio M. De Francesco]

On mercoledì 7 dicembre 2022 09:00:29 CET Sebastian Andrzej Siewior wrote:
> On 2022-12-06 20:12:13 [+0100], Fabio M. De Francesco wrote:
> > >   Furthermore, code between the kmap_atomic() and kunmap_atomic()
> > >   functions may implicitly depended
> > 
> > I suppose it should be "depend"? Shouldn't it?
> 
> Ehm, yes, correct.
> 
> > >   on the side effects of kmap_atomic()
> > >   namely disabling pagefaults or preemption or both.
> > 
> > I agree with you for rephrasing, mainly because it is
> > written in poor English.
> > 
> > However, I still have doubts about why you deleted "migration".
> > AFAIK, __kmap_local_pfn_prot() always takes care of disabling migration 
for
> > HIGHMEM enabled kernels.
> 
> That is correct. Historically kmap_atomic() never had a
> migrate_disable() statement - only preempt_disable(). With disabled
> preemption the task migration is implicitly disabled.

Sure, I understand this mechanism: task migration is implicitly disabled with 
disabled preemption.

> 
> > How about !HIGHMEM, where kmap_local_page() is an indirect call to
> > page_address()? Did you mean that, if the code between kmap_atomic() and
> > kunmap_atomic() depended on migrate_disable() (in PREEMPT_RT) we should
> > always just stay safe and call preempt_disable() together with conversion
> > to kmap_local_page()?
> 
> Even in the !HIGHMEM case it always uses preempt_disable().

With the only exception of PREEMPT_RT kernels, which instead use 
migrate_disable().

> With
> PREEMPT_RT it is different as it never disabled preemption and always
> did a migrate_disable() instead.

OK, I see that I'm recalling correctly :-) 

> If you talk about what needs to be
> considered while migrating away from kmap_atomic()

Yes, I'm trying to explain what needs to be considered while converting from 
kmap_atomic() by looking at all the different cases.

> then I wouldn't add
> the PREEMPT_RT bits to it since it was never in the picture while the
> code (using kmap_atomic()) was originally written.

Ah, OK. Now I understand why you changed my last phrase.
I agree with you, so I won't add anything about the special PREEMPT_RT case.

> > If so, I understand and I again agree with you. If not, I'm missing
> > something; so please let me understand properly.
> > 
> > Aside from the above, I'm not sure whether you deleted the last phrase
> > before
> > your suggestion. What about making it to become "For the above-mentioned
> > cases, conversions should also explicitly disable page-faults and/or
> > preemption"?
> 
> They need to disable preemption or page-faults or both if it is needed
> (not unconditionally) and where it is needed. This means not
> unconditionally over the whole kmap-ed section.

I never meant to suggest to _unconditionally_ disable page-faults 
and/or preemption. I was only trying to say that developers must carefully 
check whether or not the whole kmap-ed section depended on those side effects.

If so, they must _explicitly_ disable preemption or page-faults or both 
together with the use of kmap_local_page(). Instead, if the section doesn't 
depend on preemption and/or page-faults disabling, they must only replace 
kmap_atomic() with kmap_local_page().

I had probably used a bad wording when trying to say the same things that you 
wrote much more clearly.

Thanks,

Fabio

> 
> > Thanks again for noticing my mistakes.
> > 
> > Fabio
> 
> Sebastian