On 2014/11/14 5:28, Marc Zyngier wrote: > On 13/11/14 21:11, Thomas Gleixner wrote: >> On Thu, 13 Nov 2014, Marc Zyngier wrote: >>> With the new stacked irq domains, it becomes pretty tempting >>> to allocate an MSI domain per PCI bus, which would remove >>> the requirement of either relying on arch-specific code, or >>> a default PCI MSI domain. >> >> Right. That's what I roughly had in mind. And that would solve the >> multi-iommu issue on x86 nicely as well. We establish the association >> at the time where the bus gets populated. So the whole lookup magic >> simply goes away. > > Great. I've pushed the whole thing out with this patch, the couple > of fixes I mentioned earlier, as well as the whole ITS code: > > git://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git irq/stacked-its-v2 Hi Marc, I have looked at the code, and found some issues. 1) With my next version, no need to implemeent its_pci_msi_free() anymore. msi_domain_free_irqs() will reset desc->irq to zero for all architectures. 2) This piece of code in its_msi_prepare() may run into trouble for PCI device with both MSI and MSIX capability. I will change msi_prepare() prototype to pass in the "nvec" parameter. And you may access first_pci_msi_entry()->msi_attrib.is_msix to get allocation type if needed. nvec = pci_msix_vec_count(pdev); if (nvec < 0) nvec = pci_msi_vec_count(pdev); if (nvec < 0) return nvec; 3) Do we need to increase the default of NUM_MSI_ALLOC_SCRATCHPAD_REGS to 4? 2 is a little too limited. Regards! Gerry > > Time to go home. > > M. > -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html