On Fri, May 29, 2020 at 1:53 PM Doug Anderson <dianders@xxxxxxxxxxxx> wrote: > > > > I don't get it. A hypothetical machine could have literally anything > > sharing the IRQ line, right? > > It's not a real physical line, though? I don't think it's common to > have a shared interrupt between different IP blocks in a given SoC. > Even if it existed, all the drivers should disable their interrupts? I don't know, it's a hypothetical machine so it can be whatever you want. The driver requests shared irqs, if it doesn't actually support irq sharing, it shouldn't request them. > > Anyways, my screaming interrupt occurs after a a new kernel has been > > booted with kexec. In this case, it doesn't matter if the old kernel > > called disable_irq or not. As soon as the new kernel re-enables the > > interrupt line, the kernel immediately disables it again with a > > backtrace due to the unhandled screaming interrupt. That's why the > > dwc2 hardware needs to have its interrupts turned off when the old > > kernel is shutdown. > > Isn't that a bug with your new kernel? I've seen plenty of bugs where > drivers enable their interrupt before their interrupt handler is set > to handle it. You never know what state the bootloader (or previous > kernel) might have left things in and if an interrupt was pending it > shouldn't kill you. It wouldn't hurt to add disabling of the dwc2 irq early in dwc2 initialization, but why leave the irq screaming after shutdown? If there is another device using the same irq, it will generate unhandled interrupt backtraces and get its irq disabled when the new kernel requests its irq, if the device's driver is loaded before the dwc2 driver (assuming the new kernel even has a dwc2 driver). The dwc2 driver in its current state will generate unhandled interrupt backtraces by itself until it registers the right handler. -- Frank
