Re: acpi suspend debugging techniques?

On 31/08/2015 11:53, Adrian Chadd wrote:
> Try disabling hardware one at a time. Ie, unload usb; unload wifi;
> leave kms loaded for mostly obvious reasons.

Adrian, Garrett,

thank you very much for your tips.
Turned out that it was radeonkms that was causing the problem :-)

BTW, here is another tool for the toolkit: on sufficiently recent system devctl
suspend and devctl resume can be used to test individual drivers.

So, I noticed that I could suspend/resume drmn0 device just fine but with
vgapci0 I had a trouble suspending.  One thing led to another and here is a
patch that seems to fix the problem for me:
-------------------------------------------------------------------------------
commit fecb5e8a90631f06600d87165cc8b6de3e035dfc
Author: Andriy Gapon <avg_at_icyb.net.ua>
Date:   Thu Sep 3 17:24:23 2015 +0300

    radeon_suspend_kms: don't mess with pci state that's managed by the bus

    The pci bus driver handles the power state and configuration state saving
    and restoring for its child devices.

diff --git a/sys/dev/drm2/radeon/radeon_device.c
b/sys/dev/drm2/radeon/radeon_device.c
index e5c676b11ed47..73b2f4c51ada2 100644
--- a/sys/dev/drm2/radeon/radeon_device.c
+++ b/sys/dev/drm2/radeon/radeon_device.c
_at__at_ -1342,14 +1342,10 _at__at_ int radeon_suspend_kms(struct drm_device *dev)

 	radeon_agp_suspend(rdev);

-	pci_save_state(device_get_parent(dev->dev));
 #ifdef FREEBSD_WIP
 	if (state.event == PM_EVENT_SUSPEND) {
 		/* Shut down the device */
 		pci_disable_device(dev->pdev);
-#endif /* FREEBSD_WIP */
-		pci_set_powerstate(dev->dev, PCI_POWERSTATE_D3);
-#ifdef FREEBSD_WIP
 	}
 	console_lock();
 #endif /* FREEBSD_WIP */
_at__at_ -1380,10 +1376,6 _at__at_ int radeon_resume_kms(struct drm_device *dev)

 #ifdef FREEBSD_WIP
 	console_lock();
-#endif /* FREEBSD_WIP */
-	pci_set_powerstate(device_get_parent(dev->dev), PCI_POWERSTATE_D0);
-	pci_restore_state(device_get_parent(dev->dev));
-#ifdef FREEBSD_WIP
 	if (pci_enable_device(dev->pdev)) {
 		console_unlock();
 		return -1;
-------------------------------------------------------------------------------

However, I am not sure about an exact mechanism of the hard system hang that I
experienced without the patch.

BTW, I noticed that only very few drivers make explicit calls to
pci_set_powerstate and pci_save_state/pci_restore_state.
sys/dev/usb/controller/ohci_pci.c looks like a good use of pci_set_powerstate.
sys/dev/ixgbe/if_ix.c looks like an incorrect / redundant use of the functions.

-- 
Andriy Gapon