On Tue, Apr 23, 2024 at 08:33:12AM +0300, Mika Westerberg wrote: > I think what you are looking for is that anything behind a PCIe tunnel > should not have this applied. IIRC the AMD GPU or some code there were > going to add identification of "virtual" links to the bandwidth > calculation functionality. I guess I could resurrect my correlation patch: https://lore.kernel.org/all/f53ea40a7487e145aa1a62c347cef1814072e140.1536517047.git.lukas@xxxxxxxxx/ The last time I forward-ported it was for v5.13. I still have that code running on my development machine. The problem is that it only allows lookup from tb_port to pci_dev. I'd have to add a pointer to struct pci_dev to allow lookups in the inverse direction. Though I think we have such PCI companion devices for CXL as well, so such a pointer could be useful in general. I'm knee-deep in PCI device authentication code but could probably dedicate a weekend to the correlation patch if there's interest? Once we have correlation, we can expose more precise bandwidth for virtual PCI links in sysfs. Thanks, Lukas
