Return-Path: Sender: "Gustavo F. Padovan" Date: Wed, 10 Nov 2010 13:47:35 -0200 From: "Gustavo F. Padovan" To: Marcel Holtmann Cc: Ville Tervo , "linux-bluetooth@vger.kernel.org" Subject: Re: [PATCH 1/7] Bluetooth: Hold the lock inside l2cap_get_sock_by_addr() Message-ID: <20101110154735.GB3275@vigoh> References: <1288710198-6108-1-git-send-email-padovan@profusion.mobi> <20101105134935.GD19903@null> <20101105143711.GB9116@vigoh> <1289367595.9615.240.camel@aeonflux> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <1289367595.9615.240.camel@aeonflux> List-ID: Hi Marcel, * Marcel Holtmann [2010-11-10 14:39:55 +0900]: > Hi Gustavo, > > > > On Tue, Nov 02, 2010 at 04:03:12PM +0100, ext Gustavo F. Padovan wrote: > > > > It also have to change the name of the function to > > > > l2cap_get_sock_by_addr() because we do hold the lock inside it now. > > > > > > > > Signed-off-by: Gustavo F. Padovan > > > > --- > > > > net/bluetooth/l2cap.c | 17 ++++++----------- > > > > 1 files changed, 6 insertions(+), 11 deletions(-) > > > > > > > > diff --git a/net/bluetooth/l2cap.c b/net/bluetooth/l2cap.c > > > > index 6f931cc..3d48867 100644 > > > > --- a/net/bluetooth/l2cap.c > > > > +++ b/net/bluetooth/l2cap.c > > > > @@ -728,15 +728,18 @@ static inline void l2cap_chan_add(struct l2cap_conn *conn, struct sock *sk, stru > > > > } > > > > > > > > /* ---- Socket interface ---- */ > > > > -static struct sock *__l2cap_get_sock_by_addr(__le16 psm, bdaddr_t *src) > > > > +static struct sock *l2cap_get_sock_by_addr(__le16 psm, bdaddr_t *src) > > > > { > > > > struct sock *sk; > > > > struct hlist_node *node; > > > > + > > > > + write_lock_bh(&l2cap_sk_list.lock); > > > > > > Code is only reading so read_lock_bh would be enough? > > > > Sure, I didn't looked to that, I just keept the same code that we were > > using before. I'll fix it. > > we might also not just bother with read/write locks. Since they are not > always the right thing to do. In a lot of cases a pure spinlock is just > better. And in case of Bluetooth I think we would be just fine with > using a pure spinlock. You might run some tests with this. Better than spinlock is move everything to RCU, than we will just need a mutex or spinlock in the write operations. Using a RCU is a lot simpler. These patches are a clean up to add RCU to sk_list, so I propose to upstream the new version of my patches and do the RCU change on top of them. After all these changes we will be ready for the l2cap.c split. -- Gustavo F. Padovan http://profusion.mobi