From: MaJun <majun...@huawei.com> [ Upstream commit 9459a04b6a5a09967eec94a1b66f0a74312819d9 ]
The register array offset for clearing an interrupt is calculated by: offset = (hwirq - RESERVED_IRQ_PER_MBIGEN_CHIP) / 32; This is wrong because the clear register array includes the reserved interrupts. So the clear operation ends up in the wrong register. This went unnoticed so far, because the hardware clears the real bit through a timeout mechanism when the hardware is configured in debug mode. That debug mode was enabled on early generations of the hardware, so the problem was papered over. On newer hardware with updated firmware the debug mode was disabled, so the bits did not get cleared which causes the system to malfunction. Remove the subtraction of RESERVED_IRQ_PER_MBIGEN_CHIP, so the correct register is accessed. [ tglx: Rewrote changelog ] Fixes: a6c2f87b8820 ("irqchip/mbigen: Implement the mbigen irq chip operation functions") Signed-off-by: MaJun <majun...@huawei.com> Signed-off-by: Hanjun Guo <hanjun....@linaro.org> Acked-by: Marc Zyngier <marc.zyng...@arm.com> Cc: Kefeng Wang <wangkefeng.w...@huawei.com> Cc: linux...@huawei.com Cc: Wei Yongjun <weiyongj...@huawei.com> Link: http://lkml.kernel.org/r/1494561328-39514-4-git-send-email-guohan...@huawei.com Signed-off-by: Thomas Gleixner <t...@linutronix.de> Signed-off-by: Sasha Levin <alexander.le...@microsoft.com> --- drivers/irqchip/irq-mbigen.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/drivers/irqchip/irq-mbigen.c b/drivers/irqchip/irq-mbigen.c index 03b79b061d24..05d87f60d929 100644 --- a/drivers/irqchip/irq-mbigen.c +++ b/drivers/irqchip/irq-mbigen.c @@ -105,10 +105,7 @@ static inline void get_mbigen_type_reg(irq_hw_number_t hwirq, static inline void get_mbigen_clear_reg(irq_hw_number_t hwirq, u32 *mask, u32 *addr) { - unsigned int ofst; - - hwirq -= RESERVED_IRQ_PER_MBIGEN_CHIP; - ofst = hwirq / 32 * 4; + unsigned int ofst = (hwirq / 32) * 4; *mask = 1 << (hwirq % 32); *addr = ofst + REG_MBIGEN_CLEAR_OFFSET; -- 2.14.1