Hello, Shaohua. On Fri, Jan 06, 2012 at 04:15:57PM +0800, Shaohua Li wrote: > it's not related to the recursive merge. I forgot the detail of the > workload why we add INSERT_SORT_MERGE, it's added several months ago > anyway. > Obviously we shouldn't do complex things inside the merge window. I just > didn't agree to disable INSERT_SORT_MERGE, which will bring performance > regression for sure. Can we just change CFQ like your first path? But that is exactly what was broken and papering it over is exactly the thing we shouldn't do, for now or for ever. Think about it. Your commit f1f8cc94651 "block, cfq: fix empty queue crash caused by request merge" fixed one symptom caused by cross cfqq merging happening behind cfq's back without properly root-causing it. I'm not saying it was your fault but in the end all it did was obscuring the root cause of the problem further, making the next more subtle failure much more difficult to diagnose. So, if you think cross-cfqq merging is a good idea, please go ahead and do it properly. It shouldn't happen as a side effect of two bugs folded together. As for quicker fix than revamping merge infrastructure, adding another callback to explicitly ask elevator whether two requests chosen by block core can be merged. I didn't go that route because I wasn't sure about the benefit of doing so - I still can't see how that would make a lot of difference without cross cfqq merging, and wanted to stay on the safer side given the intricacy of the problem. So, if you wanna do that and have workload which is benefited enough to justify the extra step, let's do it after the merge window, merge as fixes and send it back via -stable. But we shouldn't be doing that inside merge window when all other trees are getting pushed upstream and we don't have much time for testing ourselves or in linux-next. Thanks. -- tejun -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
