Ilpo Järvinen wrote:

I suppose it flooded the log so that head of the log got lost already 
before userspace could consume it... (I think it's ringbuffer or something 
though I'm not sure), it might be wise to enlarge it (I think it's 
controlled through CONFIG_LOG_BUF_SHIFT).

Since I've not yet found any other path that could lead to it except the 
yesterday's false-positive, getting the write queue state more accurately 
captured by the debug patch on the very first occassion might help. Here's 
less spammy version of the debug patch (I added one printout as well to 
get more usefulness to transitional state reported).

OK, increased CONFIG_LOG_BUF_SHIFT from 14 to 18 and reversed old / 
applied new patch.

Andy.