Currently the driver calls pm_runtime_put_autosuspend but without ever having done a pm_runtime_get, this causes the reference count in the pm runtime core to become -1. The bad reference count causes the core to sometimes suspend whilst an active SPI transfer is in progress. arizona spi0.1: SPI transfer timed out spi_master spi0: failed to transfer one message from queue The correct proceedure is to do all the initialisation that requires the hardware to be powered up before enabling the PM runtime, then enable the PM runtime having called pm_runtime_set_active to inform it that the hardware is currently powered up. The core will then power it down at it's leisure and no explicit pm_runtime_put is required. Fixes: d36ccd9f7ea4 ("spi: cadence: Runtime pm adaptation") Signed-off-by: Charles Keepax <ckeepax@xxxxxxxxxxxxxxxxxxxxx> --- drivers/spi/spi-cadence.c | 11 ++++------- 1 file changed, 4 insertions(+), 7 deletions(-) diff --git a/drivers/spi/spi-cadence.c b/drivers/spi/spi-cadence.c index 7c88f74f7f470..94cc0a152449f 100644 --- a/drivers/spi/spi-cadence.c +++ b/drivers/spi/spi-cadence.c @@ -584,11 +584,6 @@ static int cdns_spi_probe(struct platform_device *pdev) goto clk_dis_apb; } - pm_runtime_use_autosuspend(&pdev->dev); - pm_runtime_set_autosuspend_delay(&pdev->dev, SPI_AUTOSUSPEND_TIMEOUT); - pm_runtime_set_active(&pdev->dev); - pm_runtime_enable(&pdev->dev); - ret = of_property_read_u32(pdev->dev.of_node, "num-cs", &num_cs); if (ret < 0) master->num_chipselect = CDNS_SPI_DEFAULT_NUM_CS; @@ -603,8 +598,10 @@ static int cdns_spi_probe(struct platform_device *pdev) /* SPI controller initializations */ cdns_spi_init_hw(xspi); - pm_runtime_mark_last_busy(&pdev->dev); - pm_runtime_put_autosuspend(&pdev->dev); + pm_runtime_set_active(&pdev->dev); + pm_runtime_enable(&pdev->dev); + pm_runtime_use_autosuspend(&pdev->dev); + pm_runtime_set_autosuspend_delay(&pdev->dev, SPI_AUTOSUSPEND_TIMEOUT); irq = platform_get_irq(pdev, 0); if (irq <= 0) { -- 2.11.0