On 07-08-17 00:42, Nicholas A. Bellinger wrote:
Hi Pascal, Per the thread with Lukasz, I'm curious to find out how the 'inflight' counters look for the underlying SCSI devices in the MD array, once the hung task warnings occur in your setup. As mentioned, AFAIK these counters aren't used by MD itself, just the underlying SCSI devices. But it will be helpful to at least verify if the underlying SCSI devices are misbehaving and not completing I/Os for some reason.. Also, with the additional debug in place below to print se_cmd state when I/O completion back to target-core doesn't happen for an extended period, it will be helpful to verify if COMPARE_AND_WRITE is what's triggering the scenario, similar to Lukasz's setup.
Hey, thanks for sticking with this :) I'm "sad" to say, we're having issues reproducing the issue. But we'll continue trying. And report if we get new information. Our latest setup, will have your latest debug patch applied. Regards, Pascal de Bruijn BTW, I noticed the bounces for the stable tree, I'm not sure if you already handled this off list, but your free-cmd-during-tmr-abort patch depends on the following to be applied to older (like 4.9) kernel trees: 1eb42f965cedafb700e9c902ddafb1c51e3117f7
-- To unsubscribe from this list: send the line "unsubscribe target-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
