We are trying to achieve WiFi6e and WiFi7 speeds over PCIe on a i.MX8M Mini based single board computer. So far we have been able to achieve 800 megabits per second, but typically should be able to reach higher, even up to 2800 mbps. We are wondering if the limitation is via a interrupt bottleneck. We were hoping to distribute the interrupts across multiple cores to achieve a higher throughput, but tools such as smp_affinity don't seem to provide that option. Is it true that 'smp affinity' works only for interrupts being handled by the ARM GIC (Interrupt Controller) which is why this cannot be changed for PCI interrupts?
Is this just a limitation of NXP's PCI controller and implementation? Or are we missing any kernel configuration here?
See below output of interrupts.
root@jammy-ubuntu:~# cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3
228: 4 0 0 0 PCI-MSI 1572864 Edge bhi
229: 5 0 0 0 PCI-MSI 1572865 Edge mhi
230: 39 0 0 0 PCI-MSI 1572866 Edge mhi
231: 77 0 0 0 PCI-MSI 1572867 Edge ce0, ce7
232: 77 0 0 0 PCI-MSI 1572868 Edge ce1, ce14
233: 6217 0 0 0 PCI-MSI 1572869 Edge ce2
234: 1 0 0 0 PCI-MSI 1572870 Edge ce3
235: 2 0 0 0 PCI-MSI 1572871 Edge ce5
236: 4896 0 0 0 PCI-MSI 1572872 Edge DP_EXT_IRQ
237: 2 0 0 0 PCI-MSI 1572873 Edge DP_EXT_IRQ
238: 586 0 0 0 PCI-MSI 1572874 Edge DP_EXT_IRQ
239: 488757 0 0 0 PCI-MSI 1572875 Edge DP_EXT_IRQ
240: 1333 0 0 0 PCI-MSI 1572876 Edge DP_EXT_IRQ
241: 6688 0 0 0 PCI-MSI 1572877 Edge DP_EXT_IRQ
242: 1333 0 0 0 PCI-MSI 1572878 Edge DP_EXT_IRQ
243: 1333 0 0 0 PCI-MSI 1572879 Edge DP_EXT_IRQ
