Merge lp:~cypressyew/checkbox/gpu_test_fix1321995 into lp:checkbox

[Po-Hsu Lin]

Parse the kernel output message from dmesg instead of the kern.log.
To make sure this GPU hung issue is happened during this session.

[Daniel Manrique (roadmr)]

Looks OK to me, of course this won't work if the gpu hang causes the system to crash and the dmesg is lost, do you think we could do something in that case?

I'll approve this anyway as it sounds more accurate than parsing the whole historical log.

review: Approve

[Po-Hsu Lin (cypressyew)]

oh... indeed, I missed that spot.

It's pity that I didn't notice this issue when it happen.
Otherwise I could try to get some debugging information from it.
There is not much we can get from the submission itself:
https://certification.canonical.com/hardware/201208-11587/submission/98817/

This one is lucky enough to finish all tests after a GPU lockup.
From the Internet, I also found many people still able to get this GPU hung pattern from their dmesg output, so I think this tweak is okay for those non-crashing systems.

We need something else to deal with the crashed one.
(when it crashes, will it log the event to kern.log?)

=== modified file 'providers/plainbox-provider-checkbox/bin/gpu_test'
--- providers/plainbox-provider-checkbox/bin/gpu_test	2014-04-07 16:42:01 +0000
+++ providers/plainbox-provider-checkbox/bin/gpu_test	2014-05-22 06:14:10 +0000
@@ -140,11 +140,10 @@
 
 def check_gpu(log=None):
     if not log:
-        log = '/var/log/kern.log'
-    with open(log, 'rb') as f:
-        if re.findall(r'gpu\s+hung', str(f.read()), flags=re.I):
-            print("GPU hung Detected")
-            return 1
+        log = subprocess.check_output(['dmesg'], universal_newlines=True)
+    if re.findall(r'gpu\s+hung', log, flags=re.I):
+        print("GPU hung Detected")
+        return 1
 
 
 def main():