Summary: ACPI sleep/resume is fine, but only if 3d isn't enabled. If 3d accel is
enabled (by loading the dri module in X), even if X is subsequently stopped(!),
then the thinkpad (x22) doesn't resume from sleep. 

I have had this problem with kernels 2.6.12.x (Mandriva 2006 distro), and with
unpatched kernel.org 2.6.16.20 and 2.6.17.1. It occurs in exactly the same way 
whether I use the older echo mem > /proc/acpi/sleep  or the newer s2ram -f -a3.

I've tried all the fixes on google and thinkwiki, to no avail:  
  permutations of: noapic; nolapic; ec_intr=0,
  vbetool,
  saving and restoring the video device state. 
  killall -STOP X; [suspend]; killall -CONT X
  all other permutations of s2ram -f,-p,-a1-3,s 

All tests are done from VT1. ACPI is otherwise working fine.

 
The hardware is a Thinkpad X22 laptop, with an ATI Radeon Mobility M6 LY.
Xorg is version 6.9.0-11  using the ati driver.

When it crashes, the suspend itself is fine, but on pressing 'Fn' to resume, 
the HDD powers up, the floppy clunks once, but the crescent LED doesn't go out;
thereafter, it is totally dead and responds only to the reset button.


Here are some more diagnostics:

TEST 1)

Boot up with "linux single". Very few modules are loaded.
echo 3 > /proc/acpi/sleep.
It sleeps (and resumes) happily, and repeatedly.

TEST 2)

Restart, to avoid bias. Again, boot up with "linux single".
modprobe drm; modprobe radeon
echo 3 > /proc/acpi/sleep.
It sleeps (and resumes) happily, and repeatedly.


TEST 3)

linux single
modprobe drm; modprobe radeon
startx
STOP x
[drm and radeon modules are automatically removed when X shuts down]
echo 3 > /proc/acpi/sleep
Result: it suspends fine, but will not wake up!
Pressing Fn-F4 spins up the HDD and CPU fan, but nothing else.


TEST 4)

Running X, but without the "Load dri" line is fine.

Test 5)

Running X, with "load dri", AND with Mesa installed is fine.



I repeated this with both kernel 2.6.16.20 and 2.6.17.1 and 2.6.12. The results
are identical (except that 2.6.16 needs to use echo mem > /sys/power/state
instead of echo 3 > /proc/acpi/sleep)

My conclusion is that something in Xorg does something to the graphics card
which interferes with resume.


----
Thanks very much for your help - Richard