Most recent kernel where this bug did not occur: always Distribution: Ubuntu (edgy) Hardware Environment: Sony Vaio Intel Core Duo Software Environment: Problem Description: MSI interrupt in sky2 driver is not enabled correctly on resume. The problem is that PCI configuration is restored correctly in device, but somewhere between end of device resume and when USB hub is detected, the PCI command register INTX_DISABLE bit is cleared. Suspect some code after the resume is doing a PCI bus reset. Steps to reproduce: Suspend laptop Resume laptop Network IRQ is routed to 17 rather than MSI so first network interrupt cause IRQ code to trap and disable, since no IRQ's happen network device is deaf.
Created attachment 9447 [details] Console output from suspend/resume Console log showing IRQ misrouting Some additional instrumentation of MSI save/restore, but problem is not there.
This is a regression. 2.6.18 works okay
I have a similar bug here. Hardware Environment: IBM R60e 06588PC UP Software Environment: 2.6.19~2.6.20 MP kernel Problem Description: I got "irq 20: nobody cared" after S3 resume, and network became unavailable. It's caused by the net driver "tg3", which is in "MSI_PCI" mode. I tried different workarounds and finally found that it can work in one of the following situations: 1.build a UP kernel, even with lapic enabled. 2.MP kernel with pci=nomsi.(tg3 is is routed to 16 rather than "MSI_PCI") 3.rmmod tg3 driver before suspend and modprobe it after resume. Steps to reproduce: Suspend the laptop and resume.
Created attachment 10310 [details] cat /proc/interrupts
Is it still a problem in 2.6.22-rc7+? Thanks.
Natilie this should be solved, or if it remains it is a driver bug. pci_save_state, pci_restore_state were reworked to be more robust and generally more correct as of commit: 392ee1e6dd901db6c4504617476f6442ed91f72d Which git-name-rev tell me went in about 2.6.21-rc4
Great! - can you test with latest kernel Stephen? Thanks.
I gave the Sony laptop away to another developer. I will ping him to retest...
Any updates on this? We can probably close it per #6 since no more complaints in a while now.
Mark it as fixed, if it shows up again reopen or file new bug.