2.2.7 with 2.6.17.6. gives me on resume the following:

BUG: sleeping function called from invalid context at mm/slab.c:2793
in_atomic():0, irqs_disabled():1

The reason is the following callchain is called with disabled IRQs:

suspend2_suspend -> device_power_up -> sysdev_resume -> __sysdev_resume ->
irqrouter_resume -> acpi_pci_link_resume -> acpi_pci_link_set ->
acpi_set_current_resources -> acpi_rs_set_srs_method_data ->
acpi_ut_create_internal_object_dbg -> acpi_ut_allocate_object_desc_dbg ->
acpi_os_acquire_object -> kmem_cache_alloc with GFP_KERNEL flag.

I don't see a reason why device_power_up() is being called with disabled
IRQs, as ACPI definitely doesn't request it (or acpi_pci_link_resume, to
be concrete). I haven't checked with APM or other archs, though.

Submitted to swsup2-devel on 22 Jun 2006.

						--- JiKos

== addon

This patch is bogus. The problem has to be solved inside ACPI layer.
See my other acpi-pci-link-allocation.patch
						
--- atomic_copy.c.orig	2006-07-20 13:27:34.000000000 +0200
+++ atomic_copy.c	2006-07-22 17:15:17.000000000 +0200
@@ -379,11 +379,16 @@
 		printk(KERN_ERR "Error %d suspending\n", error);
 	/* Restore control flow appears here */
 	restore_processor_state();
-	if (!suspend2_in_suspend)
+	if (!suspend2_in_suspend) {
 		copyback_high();
+		local_irq_enable();
+		device_power_up();
+		goto ret;
+	}
 	device_power_up();
 enable_irqs:
 	local_irq_enable();
+ret:
 	return error;
 }