On 12/01/2014 07:48 PM, Timur Tabi wrote:
On 12/01/2014 10:49 AM, Lars-Peter Clausen wrote:
The driver creates the mapping by calling irq_of_parse_and_map(), so it also has to dispose the mapping.
I agree with Markus, this does seem weird. It sounds like you're saying that irq_of_parse_and_map() and devm_request_irq() are incompatible.
They probably are. You have to create the mapping before you request the IRQ and if devm_request_irq() is used the IRQ is only freed again after the remove function of the driver been called. Yet if a driver creates a mapping in its probe function it should also dispose it in its remove function. So you are stuck with either freeing the mapping before freeing the IRQ or leaking the mapping.
My opinion on this is that devices should not create mappings and should leave that to the core. This quite easily solves the dilemma.
A quick grep shows the following drivers that call both functions:
Most of these drivers will probably work fine without irq_of_parse_and_map().
ata/pata_mpc52xx.c built-in.o cpufreq/exynos5440-cpufreq.c crypto/omap-sham.c dma/moxart-dma.c edac/mpc85xx_edac.c hsi/clients/nokia-modem.c i2c/busses/i2c-wmt.c input/serio/apbps2.c mmc/host/omap_hsmmc.c mmc/host/moxart-mmc.c mtd/nand/mpc5121_nfc.c net/ethernet/arc/emac_main.c net/ethernet/moxa/moxart_ether.c pci/host/pcie-rcar.c pinctrl/samsung/pinctrl-exynos5440.c pinctrl/samsung/pinctrl-exynos.c pinctrl/pinctrl-bcm2835.c spi/spi-bcm2835.c spi/spi-mpc512x-psc.c staging/xillybus/xillybus_of.c thermal/samsung/exynos_tmu.c