From mboxrd@z Thu Jan 1 00:00:00 1970 From: Bartlomiej Zolnierkiewicz Subject: Re: [PATCH 3/9] ide: move IRQ clearing from ack_intr() method to clear_irq() method Date: Fri, 12 Jun 2009 20:48:08 +0200 Message-ID: <200906122048.08585.bzolnier@gmail.com> References: <200702140101.26639.sshtylyov@ru.mvista.com> <200906121818.52719.bzolnier@gmail.com> <4A329D7A.2080305@ru.mvista.com> Mime-Version: 1.0 Content-Type: Text/Plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Return-path: Received: from mail-fx0-f216.google.com ([209.85.220.216]:41924 "EHLO mail-fx0-f216.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750941AbZFLSms (ORCPT ); Fri, 12 Jun 2009 14:42:48 -0400 Received: by fxm12 with SMTP id 12so815755fxm.37 for ; Fri, 12 Jun 2009 11:42:48 -0700 (PDT) In-Reply-To: <4A329D7A.2080305@ru.mvista.com> Content-Disposition: inline Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Sergei Shtylyov Cc: linux-ide@vger.kernel.org On Friday 12 June 2009 20:24:58 Sergei Shtylyov wrote: > Hello. > > Bartlomiej Zolnierkiewicz wrote: > > >> There are now two methods that clear the port interrupt: ack_intr() method, > >> implemented only on M680x0 machines, that is called at the start of ide_intr(), > >> and clear_irq() method, that is called somewhat later in this function. In > >> order to stop this duplication, delegate the task of clearing the interrupt > >> to clear_irq() method, only leaving to ack_intr() the task of testing for the > >> port interrupt. This involves moving clear_irq() method call in ide_intr() > >> closer to the beginning of the function and removing ack_intr() method call > >> in ide_timer_expiry(), now becoming useless... > >> > >> Signed-off-by: Sergei Shtylyov > >> > >> --- > >> The patch is atop of ide-2.6.git 'for-next' branch. > >> > >> drivers/ide/gayle.c | 23 +++++++++++------------ > >> drivers/ide/ide-io.c | 11 ++++------- > >> drivers/ide/macide.c | 18 ++++++++++++++---- > >> 3 files changed, 29 insertions(+), 23 deletions(-) > >> > >> Index: ide-2.6/drivers/ide/gayle.c > >> =================================================================== > >> --- ide-2.6.orig/drivers/ide/gayle.c > >> +++ ide-2.6/drivers/ide/gayle.c > >> @@ -66,7 +66,7 @@ MODULE_PARM_DESC(doubler, "enable suppor > >> * Check and acknowledge the interrupt status > >> */ > >> > >> -static int gayle_ack_intr_a4000(ide_hwif_t *hwif) > >> +static int gayle_ack_intr(ide_hwif_t *hwif) > >> { > >> unsigned char ch; > >> > >> @@ -76,16 +76,12 @@ static int gayle_ack_intr_a4000(ide_hwif > >> return 1; > >> } > >> > >> -static int gayle_ack_intr_a1200(ide_hwif_t *hwif) > >> +static void gayle_a1200_clear_irq(ide_drive_t *drive) > >> { > >> - unsigned char ch; > >> + ide_hwif_t *hwif = drive->hwif; > >> > >> - ch = z_readb(hwif->io_ports.irq_addr); > >> - if (!(ch & GAYLE_IRQ_IDE)) > >> - return 0; > >> (void)z_readb(hwif->io_ports.status_addr); > >> z_writeb(0x7c, hwif->io_ports.irq_addr); > >> - return 1; > >> } > >> > > > > buddha.c needs a similar treatment > > > > Do you mean this fragment of xsurf_ack_intr()? Yes. > /* X-Surf needs a 0 written to IRQ register to ensure ISA bit A11 stays at 0 */ > z_writeb(0, hwif->io_ports.irq_addr); > > > I felt doubtful about it and decided to leave it as is. > > >> --- ide-2.6.orig/drivers/ide/ide-io.c > >> +++ ide-2.6/drivers/ide/ide-io.c > >> > > > > > >> @@ -791,6 +789,10 @@ irqreturn_t ide_intr (int irq, void *dev > >> goto out; > >> > >> handler = hwif->handler; > >> + drive = hwif->cur_dev; > >> + > >> + if (hwif->port_ops && hwif->port_ops->clear_irq) > >> + hwif->port_ops->clear_irq(drive); > >> > > > > We need to check for valid ->handler before using ->cur_dev > > (it may contain a stale value otherwise). > > > > Hm... > > > Moreover I somehow miss the point of moving ->clear_irq call here > > > > I moved it here because ack_intr() was clearing the interrupt at > exactly *this* point. ack_intr() is a m68k-specific thing -- please take a look at things from this perspective. > > (it should be done after we know that it is really our IRQ). > > > > I don't think that really matters much... We can have ->clear_irq without ->test_irq (i.e. piix) and thus end up incorrectly clearing IRQ in case of shared IRQs..