Re: [PATCH v2 12/17] md/raid5-cache: Move struct r5l_log definition to raid5-log.h

[Song Liu]

On Mon, May 30, 2022 at 8:48 AM Logan Gunthorpe <logang@xxxxxxxxxxxx> wrote:
>
>
>
> On 2022-05-29 23:59, Christoph Hellwig wrote:
> > On Thu, May 26, 2022 at 10:35:59AM -0600, Logan Gunthorpe wrote:
> >> Move struct r5l_log definition to raid5-log.h. While this reduces
> >> encapsulation, it is necessary for the definition of r5l_log to be
> >> public so that rcu_access_pointer() can be used on conf-log in the
> >> next patch.
> >>
> >> rcu_access_pointer(p) doesn't technically dereference the log pointer
> >> however, it does use typeof(*p) and some older GCC versions (anything
> >> earlier than gcc-10) will wrongly try to dereference the structure:
> >>
> >>     include/linux/rcupdate.h:384:9: error: dereferencing pointer to
> >>                  incomplete type ‘struct r5l_log’
> >>
> >>       typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
> >>            ^
> >>
> >>     include/linux/rcupdate.h:495:31: note: in expansion of
> >>                   macro ‘__rcu_access_pointer’
> >>
> >>        #define rcu_access_pointer(p) __rcu_access_pointer((p),
> >>        __UNIQUE_ID(rcu), __rcu)
> >>
> >> To prevent this, simply provide the definition where
> >> rcu_access_pointer() may be used.
> >
> > What about just moving any code that does the rcu_access_pointer on
> > conf->log to raid5-cache.c and doing an out of line call for it
> > instead?
>
> I guess we could do that. All the inline functions in raid5-log.h are
> there to choose between the r5l or the ppl implementaiton. So it that
> would mean the r5l implementation would probably be inlined and ppl
> would be doing a second out of line call. Not sure if that matters, but
> it seems a little odd.

I like the current version better. raid5-log.h is not used in many files anyway.

Thanks,
Song