Am Dienstag, 20. Februar 2018, 14:01:17 CET schrieb Marc Zyngier:
This small series fixes a number of issues that I found while trying to get kexec working on the Chromebook Plus (aka rk3399-gru-kevin) in order to use it as some sort of interactive bootloader.
The main issue is that the vop driver expects the interrupts to be cleared and disabled when booting. Nothing could be more wrong. The device should be expected to be alive and screaming, and it is the driver's job to put it back into a sane state.
This is what the first patch does, making sure the interrupt is requested only when the device has been put back into a known state. Given that this is an observable bug that has been around for a while, I've tagged it with a Cc: stable.
The two following patches are less important: Using memcpy on MMIO ranges is plain wrong, and using spin_lock_irqsave in irq context is slightly pointless.
With these patches in, I'm able to get kexec to work. There is still some funny issues at the iommu level, but that's for another day.
Patches on top of 4.16-rc2.
Marc Zyngier (3): drm/rockchip: Clear all interrupts before requesting the IRQ drm/rockchip: Do not use memcpy for MMIO addresses drm/rockchip: Don't use spin_lock_irqsave in interrupt context
Tested on rk3036 (hdmi), rk3288 (hdmi+edp) and rk3399 (edp) and applied to drm-misc-next after slightly fixing patch1 for a recent change to the code around the old request_irq position, so it applies.
Thanks Heiko