On Mon, 2012-12-17 at 13:28 -0800, Sarah Sharp wrote: > On Sun, Dec 16, 2012 at 12:37:00AM +0000, Ben Hutchings wrote: > > Since Linux 3.2.16, my desktop with an Asus P8Z68-V LX motherboard > > always wakes up a few seconds after I shutdown. I then have to switch > > it off a second time. This still occurs in Linux 3.6.9 (haven't tried > > 3.7 yet). > > > > I initially bisected this to: > > > > commit 13a307e1a4a10502d688ce058141bc503933214e > > Author: Lin Ming <ming.m.lin@xxxxxxxxx> > > Date: Mon Nov 28 09:46:02 2011 +0800 > > > > ACPICA: Fix to allow region arguments to reference other scopes > > > > commit 8931d9ea78848b073bf299594f148b83abde4a5e upstream. > > > > However, that seems to be a perfectly good bug fix (the error mentioned > > in the commit message did previously appear on this system). I > > therefore tried cherry-picking it onto earlier versions, and bisected > > to: > > > > commit 9777e3ce907d4cb5a513902a87ecd03b52499569 > > Author: Andiry Xu <andiry.xu@xxxxxxx> > > Date: Thu Oct 14 07:23:03 2010 -0700 > > > > USB: xHCI: bus power management implementation > > Hmm. So if we enable PCI bus suspend for this xHCI host controller, > then it will immediately wakeup the system on shutdown? So it seems. > What happens when runtime PM is enabled for the host (and any USB > devices attached to it)? Does the host controller stay in D3hot, or > does it immediately pop back into D0? It stays in D3hot. Also, this fixes the problem. > Can you turn on CONFIG_USB_XHCI_HCD_DEBUGGING and CONFIG_USB_DEBUG for > your kernel and send me the dmesg? I'm wondering if the xHCI host > controller is either failing to complete the host suspend command, or > perhaps failing to be halted. This happens in S5 (soft-off), so there is no way to read dmesg afterwards! I just checked S3 (mem) and S4 (disk), and the system stays properly suspended in those states. I set sysctl kernel.printk=8 and recorded the screen during shutdown, from which I transcribe the following: pcieport 0000:00:1c.6: wake-up capability enabled by ACPI xhci_hcd 0000:04:00.0: // Halt the HC xhci_hcd 0000:04:00.0: xhci_shutdown completed - status = 1 ACPI: Preparing to enter system sleep state S5 Disabling non-boot CPUs ... Power down. acpi_power_off called The first line is related to wake-on-LAN as the specified device is the downstream port connected to the network controller. [...] > > No devices are connected to the xHCI. > > > > 'echo disabled > /sys/bus/pci/devices/0000:04:00.0/power/wakeup' does > > *not* work around this. > > Does unloading the xHCI driver before shutdown fix this? [...] Yes. Also, I tried plugging devices into the USB 3.0-capable ports. Plugging in a mouse fixes this. Plugging in a super-speed SATA enclosure doesn't. So, as far as I can see this problem is specific to: - Sleep in S5 - Driver loaded - Run-time power management disabled - No devices or super-speed device plugged in Any of these avoid the problem: - Sleep in S3 or S4 - Driver unloaded - Run-time power management enabled - Low-speed device plugged in Ben. -- Ben Hutchings Life is like a sewer: what you get out of it depends on what you put into it.
Attachment:
signature.asc
Description: This is a digitally signed message part
