On Saturday 31 January 2009, Linus Torvalds wrote:

On Sat, 31 Jan 2009, Rafael J. Wysocki wrote:

quoted

I wonder if this change makes any difference:

--- linux-2.6.orig/drivers/pci/pci-driver.c
+++ linux-2.6/drivers/pci/pci-driver.c

@@ -501,6 +501,9 @@ static int pci_pm_suspend(struct device
 	if (pci_has_legacy_pm_support(pci_dev))
 		return pci_legacy_suspend(dev, PMSG_SUSPEND);
 
+	if (!drv || !drv->pm)
+		return 0;
+
 	if (drv && drv->pm && drv->pm->suspend) {
 		error = drv->pm->suspend(dev);
 		suspend_report_result(drv->pm->suspend, error);

I don't think that's right. Now you don't end up calling 
pci_pm_default_suspend_generic() at all, and this no pci_save_state().

But I think it could easily be the call to pci_disable_enabled_device(). 
It does that

	if (atomic_read(&dev->enable_cnt))
		do_pci_disable_device(dev);

and that ends up disabling PCI_COMMAND_MASTER and then calling 
pcibios_disable_device().

pci_disable_enabled_device() is not called for the PCIe port driver, because
it has the legacy PM support.

What happens is

pci_pm_suspend(port) ->
 		pci_legacy_suspend(port) ->
 		 		pcie_portdrv_suspend(port) [this doesn't save the state]
 		 		pci_save_state(port)

and then, with interrupts off

pci_pm_suspend_noirq(port) ->
 		pci_legacy_suspend_late(port) ->
 		 		pcie_portdrv_suspend_late(port) ->
 		 		 		pci_save_state(port)

and I suspect this last pci_save_state() breaks things.  I'm not sure why,
though.

Thanks,
Rafael