Ashok, On Thu, Aug 20 2020 at 17:42, Ashok Raj wrote: > When offlining CPUs, fixup_irqs() migrates all interrupts away from the > outgoing CPU to an online CPU. It's always possible the device sent an > interrupt to the previous CPU destination. Pending interrupt bit in IRR in > LAPIC identifies such interrupts. apic_soft_disable() will not capture any > new interrupts in IRR. This causes interrupts from device to be lost during > CPU offline. The issue was found when explicitly setting MSI affinity to a > CPU and immediately offlining it. It was simple to recreate with a USB > ethernet device and doing I/O to it while the CPU is offlined. Lost > interrupts happen even when Interrupt Remapping is enabled. New lines exist for a reason. They help to structure information. For the content, please see below. > Current code does apic_soft_disable() before migrating interrupts. > > native_cpu_disable() > { > ... > apic_soft_disable(); > cpu_disable_common(); > --> fixup_irqs(); // Too late to capture anything in IRR. > } > > Just flipping the above call sequence seems to hit the IRR checks > and the lost interrupt is fixed for both legacy MSI and when > interrupt remapping is enabled. Seems to hit? Come on, we really want changelogs which are based on facts and not on assumptions. Aside of that, yes that's a really subtle one and thanks for tracking it down! For some reason I never looked at that ordering, but now that you stick it in front of me, it's pretty clear that this is the root cause. > /* > * Disable the local APIC. Otherwise IPI broadcasts will reach > * it. It still responds normally to INIT, NMI, SMI, and SIPI > - * messages. > + * messages. It's important to do apic_soft_disable() after > + * fixup_irqs(), because fixup_irqs() called from cpu_disable_common() > + * depends on IRR being set. That sentence does not make sense to me. > + .... After apic_soft_disable() CPU preserves > + * currently set IRR/ISR but new interrupts will not set IRR. I agree with the IRR part, but ISR is simply impossible to be set in this situation. > + * This causes interrupts sent to outgoing CPU before completion > + * of IRQ migration to be lost. Check SDM Vol 3 "10.4.7.2 Local > + * APIC State after It Has been Software Disabled" section for more > + * details. Please do not use the SDM chapter number of today. It's going to be a different one with the next version. Something like this perhaps? /* * Disable the local APIC. Otherwise IPI broadcasts will reach * it. It still responds normally to INIT, NMI, SMI, and SIPI * messages. * * Disabling the APIC must happen after cpu_disable_common() * which invokes fixup_irqs(). * * Disabling the APIC preserves already set bits in IRR, but * an interrupt arriving after disabling the local APIC does not * set the corresponding IRR bit. * * fixup_irqs() scans IRR for set bits so it can raise a not * yet handled interrupt on the new destination CPU via an IPI * but obviously it can't do so for IRR bits which are not set. * IOW, interrupts arriving after disabling the local APIC will * be lost. */ Hmm? The changelog wants to have a corresponding update. Thanks, tglx