On 2025-09-05 16:59:46 [+0200], Daniel Wagner wrote: > The main changes in this version are > > - merged the mapping algorithm into the existing code > - dropping a bunch of SCSI drivers update > > With the merging of the isolcpus-aware mapping code, there is a change in > how the resulting CPU–hctx mapping looks for systems with identical CPUs > (non-hyperthreaded CPUs). My understanding is that it shouldn't matter, > but the devil is in the details. …
I have been just made aware of this. It is still not merged so let me ask the questions before it is too late. What is purpose of managed_irq? Doesn't this sort of aligns with this? It was originally introduce to limit the number of nvme interrupts which can have multiple interrupts but don't need to use all of them. virtio is using it. Yet I don't see an affect if I use isolcpus=managed_irq,X. I would expect that it limits the number of used interrupts as intended by this series. The only affect I see is that once the CPU goes off/on the assigned CPU mask is restored. What do I miss? Sebastian
