On Tue, 8 Sep 2015, Nikolay Borisov wrote: > Unfortunately I haven't found a way to reproduce it so the only option > would be to do this on a live server. However, the performance impact I > believe is going to be very prohibitive :(. Alternatively what I could > do is probably leave merging on but enable debugging only for the > kmalloc-32 slab cache. Do you think this would provide enough > information to help track the corruption when it happens, without > impacting performance?
You have read https://www.kernel.org/doc/Documentation/vm/slub.txt? The problem now is that merging is on so it could be that the corruption happens in one of the aliased caches. So maybe only kmalloc-32 wont do much good. Run slabinfo -a (slabinfo.c is a tool in the kernel tree.) to see the list of aliases for kmalloc-32. You can also use slabinfo to enable some debugging at runtime. Just enabling sanity checks may catch something that allows us to track this to the subsystem. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/