On Thursday 12 July 2007 22:24:36 Joshua Wise wrote:
quoted text
> Some time after my first[1] patch to mce.c, I had continued testing using
> the test method described there (inject hundreds of thousands of single-bit
> errors while reading from /dev/mcelog), and I came across a strange issue. 
> Once every couple test cycles (i.e., every hundred-thousand injects), the
> system would begin behaving very strangely. The serial port would stop
> responding entirely, and my SSH sessions would respond, but to one keystroke
> behind what I had typed. After two minutes, the watchdog timer would reboot
> the system.

Thanks for the analysis.

I guess at some point it would make sense to model the whole thing in spin or 
similar and find even the last race. Anyone interested? @)

quoted text
> 
> At this point, mce_read() and mce_log() are blocking on each other, and will
> be for all eternity. mce_log() is waiting for the on_each_cpu() to complete
> so that the loop can complete, and on_each_cpu() (from mce_read()) is
> waiting for mce_log() to finish.

I suspect the right fix is to get rid of collect_tscs(). 

One possible way would be to get the TSCs from the mce_log table
before the synchronization instead of asking the CPUs.

Or switch it over to the NMI supporting RCU that was recently posted.


quoted text
> -- there may be other edge cases other than 
> this one. I'm actually surprised that this wasn't a ring buffer to start
> with -- it certainly seems like it wanted to be one. 

The problem with a ring buffer is that it would lose old entries; but 
for machine checks you really want the first entries because
the later ones might be just junk.

-Andi
-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/