Bug 108541

Created attachment 195601 [details]
lspci -vvv

Created attachment 195611 [details]
dmesg

Created attachment 195621 [details]
/proc/cmdline

The following report is similar in the way that acpi seems to disable the wakeup:
https://www.mail-archive.com/kernel-packages%40lists.launchpad.net/msg142829.html

I would suspect that the problem is ACPI related and due to the split of S3 state in D3hot and D3cold; i.e. it would affect all newer graphics cards being used with not ultimately new mainboards. Rafael, do you think this issue could be addressed by a kernel quirk?

(In reply to Elmar Stellnberger from comment #2)
> Created attachment 195611 [details]
> dmesg

No problems found there, the PM test all succeeded.

(In reply to Elmar Stellnberger from comment #4)
> The following report is similar in the way that acpi seems to disable the
> wakeup:
> https://www.mail-archive.com/kernel-packages%40lists.launchpad.net/msg142829.
> html

The above link described a problem that the computer failed to suspend(shutdown instead and it is capable of "waking up" from power button, except that the firmware doesn't think it is sleeping so it did a fresh boot instead), while your problem here is about not being able to wake up the system by power button, I don't think they are similar.

(In reply to Elmar Stellnberger from comment #0)
> After upgrading from an XFX Radeon R5 230 graphics card to an XFX Radeon R7
> 240 my computer does no more awake from s2ram. Blacklisting the radeon
> module does not help. If I hit a key on my USB keyboard it seems to awake at
> first the CD drive getting power (you can even open it) but then the power
> button remains to blink which indicates that s2ram is still active. There is
> no escape except a hard reset (pressing the power button does not make it
> resume either). 
>   I have also experimented with init=/bin/bash vga=0 radeon.modeset=0
> radeon.blacklist=1 no_console_suspend. Once I have tried all options for
> /sys/power/pm_test and it did always awake without intervention; even echo
> mem >/sys/power/state made it awake without intervention (see for the logs -
> they seem to tell something about wake on lan and some missing acpi awake

once you are using /sys/power/pm_test, the firmware is not involved, i.e. the OS will freeze user space/suspend devices/bring down CPUs/etc. and wait for 5 seconds, then resume. No real ACPI sleep state happened. The fact that all pm_test could pass normally suggests there is a problem in the firmware or hardware.

> code). Then I tried to re-test with the same parameters and echo mem
> >/sys/power/state but the screen turned off in spite of no_console_suspend
> and it did not awake any more; strange. 

By "same parameters" I suppose you mean the kernel cmdline options instead of the values echoed to /sys/power/pm_test? If yes, since the pm_test is not involved in this case, the firmware is taking control to enter a ACPI sleep state and the computer really sleeps.

>   Please have a very look at the log since I would plan to deploy the same
> card on different computers (it features working 4K/UHD/3840x2160).
> p.s.: the graphics mode is restored wrongly after s2disk

From the descriptions, it all seems to me the hardware/firmware loses the capability to wake up from ACPI sleep state(S3, aka suspend to mem, to be specific) after the graphics card is changed. If it's possible, you can also test Windows on it.

  Yes, it definitely is a hardware/firmware problem: I have now tested it under a Windows 7 installation on the very same machine with the same graphics card (Esprimo E5730, Fujitsu Siemens) and it did not awake either (with the same results).
  Then additionally I have removed the XFX Radeon R7 240 graphics card from the E5730 and put it into an Esprimo P5731 / D3011-A1 board (from Fujistu). s2ram did behave the same way with the same card on another Core 2 Quad computer: it simply does not work (same behaviour as described in comment 1).
  Aaron could it be as suspected in comment 5 that not any old style board will work with a 'new style' PCIe (graphics) card?

Created attachment 197191 [details]
journal/dmesg for Nvidia Geforce GT 720

Now I have purchased an Nvidia Geforce GT 720 card in the hope that it would s2ram without a problem. Unfortunately when I tested the card with the Esprimo E5730 s2ram did go awry in the very same way as with the Radeon XFX R7 240. Damn it! I would guess that absolutely no combination of a Core 2 Quad board and a K4 graphics card would work; neither from ATI nor from Nvidia.

(In reply to Elmar Stellnberger from comment #9)
>   Aaron could it be as suspected in comment 5 that not any old style board
> will work with a 'new style' PCIe (graphics) card?

I have no idea, I suppose the mainboard vendor should know this.
I'll close the bug as INVALID since this is not a OS issue and there is nothing we can do in Linux.