Received-SPF: pass (google.com: domain of linux-kernel+bounces-83892-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: <20240214063708.972376-1-irogers@google.com> <20240214063708.972376-5-irogers@google.com>
 <CAM9d7cjuv2VAVfGM6qQEMYO--WvgPvAvmnF73QrS_PzGzCF32w@mail.gmail.com>
 <CAP-5=fUUSpHUUAc3jvJkPAUuuJAiSAO4mjCxa9qUppnqk76wWg@mail.gmail.com> <CAM9d7chXtmfaC73ykiwn+RqJmy5jZFWFaV_QNs10c_Td+zmLBQ@mail.gmail.com>
In-Reply-To: <CAM9d7chXtmfaC73ykiwn+RqJmy5jZFWFaV_QNs10c_Td+zmLBQ@mail.gmail.com>
From: Ian Rogers <irogers@google.com>
Date: Tue, 27 Feb 2024 11:02:57 -0800
Message-ID: <CAP-5=fVmrrkMdNwYPqYbK_M3AKQMqoXEi4whbzoUeoj-ROxzeA@mail.gmail.com>
Subject: Re: [PATCH v1 4/6] perf threads: Move threads to its own files
To: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>, Ingo Molnar <mingo@redhat.com>, 
	Arnaldo Carvalho de Melo <acme@kernel.org>, Mark Rutland <mark.rutland@arm.com>, 
	Alexander Shishkin <alexander.shishkin@linux.intel.com>, Jiri Olsa <jolsa@kernel.org>, 
	Adrian Hunter <adrian.hunter@intel.com>, Oliver Upton <oliver.upton@linux.dev>, 
	Yang Jihong <yangjihong1@huawei.com>, linux-kernel@vger.kernel.org, 
	linux-perf-users@vger.kernel.org, bpf@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Tue, Feb 27, 2024 at 9:31=E2=80=AFAM Namhyung Kim <namhyung@kernel.org> =
wrote:
>
> On Mon, Feb 26, 2024 at 11:24=E2=80=AFPM Ian Rogers <irogers@google.com> =
wrote:
> >
> > On Mon, Feb 26, 2024 at 11:07=E2=80=AFPM Namhyung Kim <namhyung@kernel.=
org> wrote:
> > >
> > > On Tue, Feb 13, 2024 at 10:37=E2=80=AFPM Ian Rogers <irogers@google.c=
om> wrote:
> > > >
> > > > Move threads out of machine and move thread_rb_node into the C
> > > > file. This hides the implementation of threads from the rest of the
> > > > code allowing for it to be refactored.
> > > >
> > > > Locking discipline is tightened up in this change.
> > >
> > > Doesn't look like a simple code move.  Can we split the locking
> > > change from the move to make the reviewer's life a bit easier? :)
> >
> > Not sure I follow. Take threads_nr as an example.
> >
> > The old code is in machine.c, so:
> > -static size_t machine__threads_nr(const struct machine *machine)
> > -{
> > -       size_t nr =3D 0;
> > -
> > -       for (int i =3D 0; i < THREADS__TABLE_SIZE; i++)
> > -               nr +=3D machine->threads[i].nr;
> > -
> > -       return nr;
> > -}
> >
> > The new code is in threads.c:
> > +size_t threads__nr(struct threads *threads)
> > +{
> > +       size_t nr =3D 0;
> > +
> > +       for (int i =3D 0; i < THREADS__TABLE_SIZE; i++) {
> > +               struct threads_table_entry *table =3D &threads->table[i=
];
> > +
> > +               down_read(&table->lock);
> > +               nr +=3D table->nr;
> > +               up_read(&table->lock);
> > +       }
> > +       return nr;
> > +}
> >
> > So it is a copy paste from one file to the other. The only difference
> > is that the old code failed to take a lock when reading "nr" so the
> > locking is added. I wanted to make sure all the functions in threads.c
> > were properly correct wrt locking, semaphore creation and destruction,
> > etc.  We could have a broken threads.c and fix it in the next change,
> > but given that's a bug it could make bisection more difficult.
> > Ultimately I thought the locking changes were small enough to not
> > warrant being on their own compared to the advantages of having a sane
> > threads abstraction.
>
> I can see some other differences like machine__findnew_thread()
> which I think is due to the locking change.  Maybe we can fix the
> problem before moving the code and let the code move simple.

I'll see what I can split out in v2. I don't think findnew will change
and the nr change is trivial. In the previous code the lock is taken
before calling __machine__findnew_thread, which doesn't make sense
when we try to abstract inside of threads, where it should
take/release the lock in the threads and not the machine code. Moving
the lock to __machine__findnew_thread doesn't really make sense as the
__ implies the lock is already held.

Thanks,
Ian

> Thanks,
> Namhyung