Message ID | 14385191E87B904DBD836449AA30269D021A42@MORGANITE.micrel.com |
---|---|
State | Rejected, archived |
Delegated to: | David Miller |
Headers | show |
From: "Ha, Tristram" <Tristram.Ha@Micrel.Com> Date: Thu, 3 Dec 2009 13:30:03 -0800 > From: Tristram Ha <Tristram.Ha@micrel.com> > > The Micrel KSZ8851 SNL Ethernet chip is used in the OMAP Beagle Zippy combo board. Requesting interrupt using level triggering flag hangs the system because the interrupt handling routine is called continually. Using edge triggering avoids this problem. As a result, disabling interrupt may not be necessary. > > Signed-off-by: Tristram Ha <Tristram.Ha@micrel.com> You say "may not be necessary", do you actually know for sure? This seems questionable, if the interrupt disabling was there for a reason what has changed to make it such that it is no longer necessary? -- 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
David Miller wrote: > From: "Ha, Tristram" <Tristram.Ha@Micrel.Com> > Date: Thu, 3 Dec 2009 13:30:03 -0800 > >> From: Tristram Ha <Tristram.Ha@micrel.com> >> >> The Micrel KSZ8851 SNL Ethernet chip is used in the OMAP Beagle Zippy combo board. Requesting >> interrupt using level triggering flag hangs the system because the interrupt handling routine >> is called continually. Using edge triggering avoids this problem. As a result, disabling >> interrupt may not be necessary. >> >> Signed-off-by: Tristram Ha <Tristram.Ha@micrel.com> > > You say "may not be necessary", do you actually know for sure? > > This seems questionable, if the interrupt disabling was there for a reason what has changed to > make it such that it is no longer necessary? I have tested the KSZ8851 SNL driver using the OMAP Beagle Zippy combo board, developed by Tin Can Tools. Ben Dooks, who implemented the original driver, has his own KSZ8851 SNL evaluation board, but he probably did not run any performance test on the driver. I did sent him the changes before so that he can verify on his board, but I have not received any response. So I really do not know if interrupt disabling is necessary or not. But from my understanding of the SPI network driver model, it is almost required the interrupt be edge triggered, as the hardware interrupt cannot be disabled inside the actual interrupt handling routine. So if the interrupt is level triggered, the interrupt routine will be called continually, and there is no chance for the workqueue to disable the interrupt. As the interrupt is edge triggered, I do not think the interrupt needs to be disabled. -- 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
Ha, Tristram wrote: > From: Tristram Ha <Tristram.Ha@micrel.com> > > The Micrel KSZ8851 SNL Ethernet chip is used in the OMAP Beagle Zippy combo board. Requesting interrupt using level triggering flag hangs the system because the interrupt handling routine is called continually. Using edge triggering avoids this problem. As a result, disabling interrupt may not be necessary. > > Signed-off-by: Tristram Ha <Tristram.Ha@micrel.com> Making this sort of change could impact other users. I'll submit a patch to get the interrupt type properly and push everything else i've got, -- 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
Ha, Tristram wrote: > David Miller wrote: >> >> You say "may not be necessary", do you actually know for sure? >> >> This seems questionable, if the interrupt disabling was there for a >> reason what has changed to make it such that it is no longer necessary? > > I have tested the KSZ8851 SNL driver using the OMAP Beagle Zippy combo board, developed by Tin > Can Tools. Ben Dooks, who implemented the original driver, has his own KSZ8851 SNL evaluation > board, but he probably did not run any performance test on the driver. I did sent him the > changes before so that he can verify on his board, but I have not received any response. > > So I really do not know if interrupt disabling is necessary or not. But from my understanding > of the SPI network driver model, it is almost required the interrupt be edge triggered, as the > hardware interrupt cannot be disabled inside the actual interrupt handling routine. So if the > interrupt is level triggered, the interrupt routine will be called continually, and there is no > chance for the workqueue to disable the interrupt. As the interrupt is edge triggered, I do not > think the interrupt needs to be disabled. The "[PATCH 2.6.33] net: Fix ks8851 snl receive problem" I just submitted turns off hardware interrupt during interrupt handling. That makes the disable_irq() call really unnecessary when the interrupt is edge-triggered. The code also works when the interrupt is level-triggered under Beagle Zippy combo board, but theoretically the disable_irq() is required when the interrupt is level-triggered. When I first worked on the ks8851 driver I was using Linux 2.6.22, and the level-triggered interrupt code did not work. That is why I changed the code to edge-triggered. -- 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
diff -urpN linux-2.6.32.old/drivers/net/ks8851.c linux-2.6.32.new/drivers/net/ks8851.c --- linux-2.6.32.old/drivers/net/ks8851.c 2009-11-03 11:37:49.000000000 -0800 +++ linux-2.6.32.new/drivers/net/ks8851.c 2009-12-02 15:31:39.000000000 -0800 @@ -398,7 +398,6 @@ static irqreturn_t ks8851_irq(int irq, v { struct ks8851_net *ks = pw; - disable_irq_nosync(irq); schedule_work(&ks->irq_work); return IRQ_HANDLED; } @@ -611,8 +610,6 @@ static void ks8851_irq_work(struct work_ } mutex_unlock(&ks->lock); - - enable_irq(ks->netdev->irq); } /** @@ -1284,7 +1281,7 @@ static int __devinit ks8851_probe(struct ks8851_init_mac(ks); ks->tx_space = ks8851_rdreg16(ks, KS_TXMIR); - ret = request_irq(spi->irq, ks8851_irq, IRQF_TRIGGER_LOW, + ret = request_irq(spi->irq, ks8851_irq, IRQF_TRIGGER_FALLING, ndev->name, ks); if (ret < 0) { dev_err(&spi->dev, "failed to get irq\n");