Hi Bjorn (and others interested) ! So I've gone back on my (still incomplete) series for factoring the resource management, and decided to try a different approach. You can see both variants here: 1) https://git.kernel.org/pub/scm/linux/kernel/git/benh/pci.git/log/?h=previous The previous variant where I pass a policy argument to pci_host_resource_survey() 2) https://git.kernel.org/pub/scm/linux/kernel/git/benh/pci.git/log/ The "new" variant where the policy is set in the host bridge instead Both variants use the PCI flags (PROBE_ONLY etc...) to establish a default policy. I tend to prefer variant 2 because in variant 1, I didn't manage to clarify the interaction between the resource policy and the new "preserve_config" flag we added for _DSM #5. It does mean I now have 4 instead of 3 policies to match the various behaviours but I think it's overall cleaner. Another advantage of variant 2 is that the policy is now established earlier (before we scan), which means we'll be able to take it into account during the scan. This will be useful if/when we try again to revive Lorenzo old patch to call pci_read_bridge_bases during probing instead of doing it from the archs pcibios_bus_fixup(). That consolidation will be needed to bring x86 and powerpc into the fold. We'll need to know the policy so we can avoid reading the bridge bases when doing a full reallocation, thus avoiding tripping on the "don't shrink" test that Yinghai added to the bus window sizing code, which breaks platforms that try to reassign everything (Note: we should understand why that was added btw, I can't make sense of it). Anyway, have a quick look, it's not well tested yet, may not even build on most archs, I'd just like to know which of the two approaches is preferred before I invest much more time one way or another. Cheers, Ben.
