From mboxrd@z Thu Jan 1 00:00:00 1970 From: mika.westerberg@iki.fi (Mika Westerberg) Date: Thu, 22 Mar 2012 22:03:39 +0200 Subject: [PATCH] ep93xx: Implement double buffering for M2M DMA channels In-Reply-To: References: <4F683B36.8090101@metasoft.pl> <20120321193231.GA3740@mwesterb-mobl.ger.corp.intel.com> <20120322073750.GA637@mwesterb-mobl.ger.corp.intel.com> Message-ID: <20120322200339.GA5812@mwesterb-mobl.ger.corp.intel.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Thu, Mar 22, 2012 at 01:52:52PM -0500, H Hartley Sweeten wrote: > I hacked this back to the handle_one_vic function. It looks like every time > I get the message above the stat read in that function shows an active > interrupt for DMAM2M0 (the dma rx channel) and DMAM2M1 (the dma > tx channel). Because of the way handle_one_vic processes the irq's, the > DMAM2M0 interrupt gets handled first. When it's done it must be clearing > the DMAM2M1 interrupt automatically for some reason. So, when it's > handled there is no interrupt pending so we get the message. > > The "cleanest" fix I can think of is the following patch. This is against > linux-next. I can't figure out any better fix unless we move ep93xx back to use !MULTI_IRQ_HANDLER which obviously is not good thing to do. > From: H Hartley Sweeten > > dma: ep93xx: check for spurious interrupts > > The ep93xx dma controller generates spurious interrupts on dma1chan1 when > used with the mmc_spi driver. Catch these early by making sure there is an > interrupt to actually process. > > Signed-off-by: H Hartley Sweeten Tested-by: Mika Westerberg Thanks for investigating this. You missed one thing though, see below. Otherwise it doesn't compile :) diff --git a/drivers/dma/ep93xx_dma.c b/drivers/dma/ep93xx_dma.c index e1cc87a..614bff4 100644 --- a/drivers/dma/ep93xx_dma.c +++ b/drivers/dma/ep93xx_dma.c @@ -751,7 +751,7 @@ static irqreturn_t ep93xx_dma_interrupt(int irq, void *dev_id) return IRQ_NONE; } - switch (edmac->edma->hw_interrupt(edmac)) { + switch (edmac->edma->hw_interrupt(edmac, irq_status)) { case INTERRUPT_DONE: desc->complete = true; tasklet_schedule(&edmac->tasklet);