libpci: use "brute force" scan for devices

[abrandnewusername]

To handle cases where PCI or PCIe buses are not accessible via
a bridge from bus 0, We use "brute force" scan instead of recursive
scan here. Details of these methods can be found on
https://wiki.osdev.org/PCI#.22Brute_Force.22_Scan.

Signed-off-by: Jingyao Zhou jingyao.zhou@unsw.edu.au

This is not the ideal solution, but a workaround for this issue: #131

[kent-mcleod]

This is not the ideal solution, but a workaround for this issue: #131

Would it be possible to implement this alternate scanning approach as a separate function to maintain the existing functionality for clients that already depend on it?

[wom-bat]

Would it be possible to implement this alternate scanning approach as a separate function to maintain the existing functionality for clients that already depend on it?

The existing functionality is just an optimisation to avoid scanning everything. The new code brute-force scans every possible bus, and will pick up any that have been configured by the BIOS.

The existing code is wrong when:
--- some buses are not attached via bridges but have a separate root complex
--- the BIOS doesn't enable the bus (the new code won't work for this case either)
--- where there is more than one PCI domain.

[kent-mcleod]

The existing functionality is just an optimisation to avoid scanning everything. The new code brute-force scans every possible bus, and will pick up any that have been configured by the BIOS.

Is the brute force approach noticeably slower to execute?

[wom-bat]

Is the brute force approach noticeably slower to execute?

Not that I can tell with a single PCI domain (and we don't deal with more than a single PCI domain at the moment).
Of course, to do it properly, you'd parse the DSDT and use the full 4k config space for each device.

[wom-bat]

Looks good to me now.