On Wed, May 18, 2011 at 12:02:17PM -0400, Alan Stern wrote: > On Wed, 18 May 2011, Sarah Sharp wrote: > > > On Tue, May 17, 2011 at 03:12:36PM -0700, Greg KH wrote: > > > On Tue, May 17, 2011 at 02:46:26PM -0700, Sarah Sharp wrote: > > > > I see that ehci-hcd.c has a warning in when either the UHCI or OHCI > > > > driver is loaded before the EHCI driver: > > > > > > > > set_bit(USB_EHCI_LOADED, &usb_hcds_loaded); > > > > if (test_bit(USB_UHCI_LOADED, &usb_hcds_loaded) || > > > > test_bit(USB_OHCI_LOADED, &usb_hcds_loaded)) > > > > printk(KERN_WARNING "Warning! ehci_hcd should always be loaded" > > > > " before uhci_hcd and ohci_hcd, not after\n"); > > > > > > > > What actually prevents the UHCI or OHCI driver from being loaded before > > > > the EHCI driver? Is it left up to userspace? > > > > > > Yes. Hopefully all of the distros now have properly modprobe rules to > > > load these before the ehci driver, otherwise we would be getting the > > > kernel warnings sent to us :) > > > > Great. Tony Luck had mentioned he saw this warning on boot on some box > > of his, but hasn't been able to reproduce it. Where can I find these > > modprobe rules files? Is it the same for every distro, or do different > > distros store them in different places? > > Often they aren't in the form of udev rules at all. The script files > in an initramfs image may be set up to explicitly load the drivers in > the proper order. Also, most distros are building these modules into the kernel directly, solving this issue that way. If not, look in the /etc/modprobe.conf file, or in the /etc/modprobe.d/ files, it depends on the distro. > > > > Since PCI devices can be initialized in parallel, even if the EHCI > > > > driver is loaded first, that doesn't guarantee that the PCI probe for > > > > the EHCI controller actually finishes before the UHCI/OHCI PCI probe, > > > > right? > > > > > > I didn't think that PCI devices were initialized in parallel all that > > > often, so this might be pretty rare. > > > > Hmm, ok, HPA was muttering something on IRC about some PCI core rework > > to make it possible to initialize PCI devices in parallel, so it might > > be new. > > I asked about parallel probing of PCI devices at one point. Linus said > he wasn't in favor of it because of the potential for problems like > this one. I enabled this a number of years ago and found that it really didn't speed anything up, and just exposed a ton of race conditions that no one had ever noticed before. Instead I think we have been enabling multi-threaded work in the drivers that need it the most (like the libata drivers) instead of the pci core. thanks, greg k-h -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
