Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <5FBCBE569E134E4CA167B91C0A77FD610198F851AC@EXMBX-SZMAIL022.tencent.com>
 <20181109182654.ojzc7wae3jcivlup@linux-r8p5>
In-Reply-To: <20181109182654.ojzc7wae3jcivlup@linux-r8p5>
From:   Todd Kjos <tkjos@google.com>
Date:   Fri, 9 Nov 2018 11:03:45 -0800
Message-ID: <CAHRSSEzc96qbDOQt8L+J4JMAazYv-d2xdyLT2nyPe7_t+8_5=A@mail.gmail.com>
Subject: Re: [PATCH V3] binder: ipc namespace support for android binder
To:     dave@stgolabs.net
Cc:     chouryzhou@tencent.com,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        =?UTF-8?B?QXJ2ZSBIasO4bm5ldsOlZw==?= <arve@android.com>,
        Todd Kjos <tkjos@android.com>, akpm@linux-foundation.org,
        "open list:ANDROID DRIVERS" <devel@driverdev.osuosl.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Nov 9, 2018 at 10:27 AM Davidlohr Bueso <dave@stgolabs.net> wrote:
>
> On Thu, 08 Nov 2018, chouryzhou(??????) wrote:
>
> >+#ifdef CONFIG_ANDROID_BINDER_IPC
> >+       /* next fields are for binder */
> >+       struct mutex      binder_procs_lock;
> >+       struct hlist_head binder_procs;
> >+       struct hlist_head binder_contexts;
> >+#endif
>
> Now, I took a look at how the binder_procs list is used; and no, what
> follows isn't really related to this patch, but a general observation.
>
> I think that a mutex is also an overkill and you might wanna replace it
> with a spinlock/rwlock. Can anything block while holding the binder_procs_lock?
> I don't see anything... you mainly need it for consulting the hlist calling
> print_binder_proc[_stat]() - which will take the proc->inner_lock anyway, so
> no blocking there.

print_binder_proc() drops proc->inner_lock and calls
binder_alloc_print_allocated() which acquires proc->alloc->mutex.
Likewise, print_binder_stats() calls print_binder_proc_stats() which
drops its locks to call binder_alloc_print_pages() which also acquires
proc->alloc->mutex. So binder_procs_lock needs to be a mutex since it
can block on proc->alloc->mutex.

> Also, if this is perhaps because of long hold times, dunno,
> the rb_first_cached primitive might reduce some of it, although I don't know
> how big the rbtrees in binder can get and if it matters at all.
>
> Anyway, that said and along with addressing Todd's comments, the ipc/ bits look
> good. Feel free to add my:
>
> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net>
>
> Thanks,
> Davidlohr