Received-SPF: pass (google.com: domain of linux-kernel+bounces-186961-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Precedence: bulk
MIME-Version: 1.0
References: <20240514084724.557100-1-visitorckw@gmail.com> <20240514084724.557100-5-visitorckw@gmail.com>
 <20240522161048.8d8bbc7b153b4ecd92c50666@linux-foundation.org> <gch7c2merdegxvqwpbb5ean7uj35tnh62fyviggjglftlfjs3i@jhbnllco5kau>
In-Reply-To: <gch7c2merdegxvqwpbb5ean7uj35tnh62fyviggjglftlfjs3i@jhbnllco5kau>
From: Ian Rogers <irogers@google.com>
Date: Wed, 22 May 2024 20:51:06 -0700
Message-ID: <CAP-5=fWn8AmE=0Y15Y3fm71zK0jsdgvH3UdmKUq5Pe6SAAsKCQ@mail.gmail.com>
Subject: Re: [RESEND PATCH v5 04/16] lib min_heap: Add type safe interface
To: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Andrew Morton <akpm@linux-foundation.org>, Kuan-Wei Chiu <visitorckw@gmail.com>, colyli@suse.de, 
	msakai@redhat.com, peterz@infradead.org, mingo@redhat.com, acme@kernel.org, 
	namhyung@kernel.org, bfoster@redhat.com, mark.rutland@arm.com, 
	alexander.shishkin@linux.intel.com, jolsa@kernel.org, adrian.hunter@intel.com, 
	bagasdotme@gmail.com, jserv@ccns.ncku.edu.tw, linux-bcache@vger.kernel.org, 
	linux-kernel@vger.kernel.org, dm-devel@lists.linux.dev, 
	linux-bcachefs@vger.kernel.org, linux-perf-users@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Wed, May 22, 2024 at 4:33=E2=80=AFPM Kent Overstreet
<kent.overstreet@linux.dev> wrote:
>
> On Wed, May 22, 2024 at 04:10:48PM -0700, Andrew Morton wrote:
> > On Tue, 14 May 2024 16:47:12 +0800 Kuan-Wei Chiu <visitorckw@gmail.com>=
 wrote:
> >
> > > Implement a type-safe interface for min_heap using strong type
> > > pointers instead of void * in the data field. This change includes
> > > adding small macro wrappers around functions, enabling the use of
> > > __minheap_cast and __minheap_obj_size macros for type casting and
> > > obtaining element size. This implementation removes the necessity of
> > > passing element size in min_heap_callbacks. Additionally, introduce t=
he
> > > MIN_HEAP_PREALLOCATED macro for preallocating some elements.
> >
> > I guess it's a nice change, although it's unclear how valuable all this
> > is.  Being able to remove the local implementations in bcache and
> > bcachefs is good.
>
> Yes, and now the standard kernel version is type safe. I suggested it to
> Kuan as a good starting project, I find (well picked) refactorings and
> cleanups are a great place for people to start.
>
> >
> > > --- a/drivers/md/dm-vdo/repair.c
> > > +++ b/drivers/md/dm-vdo/repair.c
> > > @@ -51,6 +51,8 @@ struct recovery_point {
> > >     bool increment_applied;
> > >  };
> > >
> > > +MIN_HEAP(struct numbered_block_mapping, replay_heap);
> > >
> > > ...
> > >
> > > -struct min_heap {
> > > -   void *data;
> > > -   int nr;
> > > -   int size;
> > > -};
> > > +#define MIN_HEAP_PREALLOCATED(_type, _name, _nr)   \
> > > +struct _name {     \
> > > +   int nr; \
> > > +   int size;       \
> > > +   _type *data;    \
> > > +   _type preallocated[_nr];        \
> > > +}
> >
> > I think that the MIN_HEAP() macro would be better named
> > DEFINE_MIN_HEAP().  There's a lot of precedent for this and it's
> > unclear whether "MIN_HEAP" actually implements storage.  I looked at
> > the repair.c addition and thought "hm, shouldn't that be static"!
>
> Agreed
>
> > >  /* Sift the element at pos down the heap. */
> > >  static __always_inline
> > > -void min_heapify(struct min_heap *heap, int pos,
> > > +void __min_heapify(min_heap_char *heap, int pos, size_t elem_size,
> > >             const struct min_heap_callbacks *func)
> > >  {
> >
> > It's a separate project, but min_heap.h has some crazily huge inlined
> > functions.  I expect we'd have a smaller and faster kernel if those
> > were moved into a new lib/min_heap.c.
>
> good thought - Kuan, want to do that too? Since we missed the merge
> window, may as well :)

Just to mention that the perf events use of the min heap is during a
context switch, so it is somewhat performance sensitive for people
context switching with perf events opened/enabled.

Thanks,
Ian