On Tue, Jan 6, 2009 at 3:37 PM, Myles Watson mylesgw@gmail.com wrote:
Marc,
Hopefully this makes everything right again. I still think some of the geode functions should be moved, but that's really a separate issue.
You have to specify to make things breadth first, which seems like the correct way.
I guess it's not really breadth first. It's just parents before siblings.
I think this should be done for all the phases unless there's some compelling reason not to.
Signed-off-by: Myles Watson mylesgw@gmail.com
I agree it is parents before siblings but I thought it looked more like a problem with the statictree. I must not understand .next.
from http://pastebin.ca/1301081
dev_apic_0 has a .sibling = &dev_domain_0 and .next = &dev_domain_0_pci_1_0
Then later in the last device domain dev_domain_0_pci_f_2 has a .next = &dev_domain_0.
I would expect .sibling and .next to point to the &dev_domain_0 and then dev_domain_0 to have .next = &dev_domain_0_pci_1_0
dev_cpus looks like I would expect .sibling = &dev_apic_0 and .next = &dev_apic_0. dev_domain_0_pci_f_0 also has the same .sibling and .next.
Can you explain the last pci dev .next pointing back to the domain?
I am not setup to test this either. I was talking with Mart on NAND device problems. Maybe he can test it tomorrow.
Marc