On 01/22/2016 05:48 PM, Peter Hurley wrote: > Hi John, Hi Peter,
> On 01/22/2016 02:27 AM, John Ogness wrote: >> It has been seen that spurious interrupts are generated when the >> DMA engine is in use. By disabling timeout interrupts (~IER_RDI) >> this phenomenon goes away, but this driver relies on the timeout >> interrupts, so we just consume the spurious interrupts. >> >> Since we are consuming spurious interrupts, the irq cannot be >> shared with other drivers. (It is never really shared anyway.) > > Does this fix the spurious irqs referred to by Sekhar in > this email chain https://lkml.org/lkml/2015/12/3/442 ? >From what I remember is that if you use DMA (sdma or edma) for RX and the DMA fetches the data from FIFO (the programmed 48 bytes) you receive a DMA interrupt for transfer complete _and_ an UART interrupt which returns UART_IIR_NO_INT (since the FIFO is empty). You usually don't notice this at 115200. At 3Mbaud those IRQ_NONE are enough to get the IRQ line shutdown by the IRQ-core. However instead of assuming that DMA is used in case of UART_IIR_NO_INT we could limit this only if DMA is actually used (the 8250 DMA ifdef and maybe set a flag if RX-DMA is successfully programmed). >> Signed-off-by: John Ogness <[email protected]> >> --- >> patch against next-20160122 >> >> drivers/tty/serial/8250/8250_omap.c | 10 ++++++++-- >> 1 file changed, 8 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/tty/serial/8250/8250_omap.c >> b/drivers/tty/serial/8250/8250_omap.c >> index ef7a60b..004b85a 100644 >> --- a/drivers/tty/serial/8250/8250_omap.c >> +++ b/drivers/tty/serial/8250/8250_omap.c >> @@ -630,7 +630,7 @@ static int omap_8250_startup(struct uart_port *port) >> } >> #endif >> >> - ret = request_irq(port->irq, omap8250_irq, IRQF_SHARED, >> + ret = request_irq(port->irq, omap8250_irq, 0, >> dev_name(port->dev), port); >> if (ret < 0) >> goto err; >> @@ -1112,8 +1112,14 @@ static int omap_8250_dma_handle_irq(struct uart_port >> *port, unsigned int iir) >> unsigned char status; >> int dma_err; >> >> + /* >> + * It has been seen that spurious interrupts are generated when the >> + * DMA engine is in use. By disabling timeout interrupts (~IER_RDI) >> + * this phenomenon goes away, but this driver relies on the timeout >> + * interrupts, so we just consume the spurious interrupts. >> + */ >> if (iir & UART_IIR_NO_INT) >> - return 0; >> + return 1; >> >> spin_lock(&port->lock); >> >> Sebastian
