| Message ID | 18143736.brt1iGhlQ1@al |
|---|---|
| State | Accepted, archived |
| Delegated to: | David Miller |
| Headers | show |
From: Lekensteyn <lekensteyn@gmail.com> Date: Tue, 17 Jul 2012 18:29:34 +0200 > From: Peter Wu <lekensteyn@gmail.com> > > This patch adds the netpoll function to support netconsole. Tested and works > fine on my "JMC250 PCI Express Gigabit Ethernet Controller" (PCI ID 0250). > > Signed-off-by: Peter Wu <lekensteyn@gmail.com> Applied to net-next I really wonder if this driver works on SMP systems at all. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, 18 Jul 2012 09:45:24 -0700 (PDT), David Miller wrote > From: Lekensteyn <lekensteyn@gmail.com> > Date: Tue, 17 Jul 2012 18:29:34 +0200 > > > From: Peter Wu <lekensteyn@gmail.com> > > > > This patch adds the netpoll function to support netconsole. Tested and works > > fine on my "JMC250 PCI Express Gigabit Ethernet Controller" (PCI ID 0250). > > > > Signed-off-by: Peter Wu <lekensteyn@gmail.com> > > Applied to net-next > > I really wonder if this driver works on SMP systems at all. Hi davem: I wrote/tested this driver on SMP system at very beginning. In face some of the design was aimed for SMP systems. Is there any obvious bug or something I should be aware of? Guo-Fu Tseng -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
From: "Guo-Fu Tseng" <cooldavid@cooldavid.org> Date: Thu, 19 Jul 2012 10:00:10 +0800 > On Wed, 18 Jul 2012 09:45:24 -0700 (PDT), David Miller wrote >> From: Lekensteyn <lekensteyn@gmail.com> >> Date: Tue, 17 Jul 2012 18:29:34 +0200 >> >> > From: Peter Wu <lekensteyn@gmail.com> >> > >> > This patch adds the netpoll function to support netconsole. Tested and works >> > fine on my "JMC250 PCI Express Gigabit Ethernet Controller" (PCI ID 0250). >> > >> > Signed-off-by: Peter Wu <lekensteyn@gmail.com> >> >> Applied to net-next >> >> I really wonder if this driver works on SMP systems at all. > Hi davem: > > I wrote/tested this driver on SMP system at very beginning. > In face some of the design was aimed for SMP systems. > Is there any obvious bug or something I should be aware of? Well, you do no locking in your IRQ handler. So how can you be sure in jme_free_irq() during jme_close() that the handler is not running on some cpu somewhere? That's why drivers like drivers/net/ethernet/broadcom/tg3.c have a function like tg3_irq_quiesce(), to guarentee that no code path is executing in the interrupt handler. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, 2012-07-18 at 20:24 -0700, David Miller wrote: > From: "Guo-Fu Tseng" <cooldavid@cooldavid.org> > Date: Thu, 19 Jul 2012 10:00:10 +0800 > > > On Wed, 18 Jul 2012 09:45:24 -0700 (PDT), David Miller wrote > >> From: Lekensteyn <lekensteyn@gmail.com> > >> Date: Tue, 17 Jul 2012 18:29:34 +0200 > >> > >> > From: Peter Wu <lekensteyn@gmail.com> > >> > > >> > This patch adds the netpoll function to support netconsole. Tested and works > >> > fine on my "JMC250 PCI Express Gigabit Ethernet Controller" (PCI ID 0250). > >> > > >> > Signed-off-by: Peter Wu <lekensteyn@gmail.com> > >> > >> Applied to net-next > >> > >> I really wonder if this driver works on SMP systems at all. > > Hi davem: > > > > I wrote/tested this driver on SMP system at very beginning. > > In face some of the design was aimed for SMP systems. > > Is there any obvious bug or something I should be aware of? > > Well, you do no locking in your IRQ handler. > > So how can you be sure in jme_free_irq() during jme_close() that the > handler is not running on some cpu somewhere? > > That's why drivers like drivers/net/ethernet/broadcom/tg3.c have > a function like tg3_irq_quiesce(), to guarentee that no code path > is executing in the interrupt handler. What on earth are you talking about? You can never be sure of that with shared IRQ handlers; so free_irq() wait if necessary. Ben.
diff --git a/drivers/net/ethernet/jme.c b/drivers/net/ethernet/jme.c index 4ea6580..c911d88 100644 --- a/drivers/net/ethernet/jme.c +++ b/drivers/net/ethernet/jme.c @@ -2743,6 +2743,17 @@ jme_set_features(struct net_device *netdev, netdev_features_t features) return 0; } +#ifdef CONFIG_NET_POLL_CONTROLLER +static void jme_netpoll(struct net_device *dev) +{ + unsigned long flags; + + local_irq_save(flags); + jme_intr(dev->irq, dev); + local_irq_restore(flags); +} +#endif + static int jme_nway_reset(struct net_device *netdev) { @@ -2944,6 +2955,9 @@ static const struct net_device_ops jme_netdev_ops = { .ndo_tx_timeout = jme_tx_timeout, .ndo_fix_features = jme_fix_features, .ndo_set_features = jme_set_features, +#ifdef CONFIG_NET_POLL_CONTROLLER + .ndo_poll_controller = jme_netpoll, +#endif }; static int __devinit