On Wed, 2018-10-03 at 16:12 -0600, Jens Axboe wrote: +AD4 On 10/3/18 3:42 PM, Bart Van Assche wrote: +AD4 +AD4 On Fri, 2018-01-12 at 22:11 +-0000, Bart Van Assche wrote: +AD4 +AD4 +AD4 /+ACo +AD4 +AD4 +AD4 +- +ACo Show +ACI-busy+ACI requests - these are the requests owned by the block driver. +AD4 +AD4 +AD4 +- +ACo The test list+AF8-empty(+ACY-rq-+AD4-queuelist) is used to figure out whether or not +AD4 +AD4 +AD4 +- +ACo a request is owned by the block driver. That test works because the block +AD4 +AD4 +AD4 +- +ACo layer core uses list+AF8-del+AF8-init() consistently to remove a request from one +AD4 +AD4 +AD4 +- +ACo of the request lists. +AD4 +AD4 +AD4 +- +ACo +AD4 +AD4 +AD4 +ACo Note: the state of a request may change while this function is in progress, +AD4 +AD4 +AD4 +ACo e.g. due to a concurrent blk+AF8-mq+AF8-finish+AF8-request() call. +AD4 +AD4 +AD4 +ACo-/ +AD4 +AD4 +AD4 +AEAAQA -402,7 +-408,7 +AEAAQA static void hctx+AF8-show+AF8-busy+AF8-rq(struct request +ACo-rq, void +ACo-data, bool reserved) +AD4 +AD4 +AD4 const struct show+AF8-busy+AF8-params +ACo-params +AD0 data+ADs +AD4 +AD4 +AD4 +AD4 +AD4 +AD4 if (blk+AF8-mq+AF8-map+AF8-queue(rq-+AD4-q, rq-+AD4-mq+AF8-ctx-+AD4-cpu) +AD0APQ params-+AD4-hctx +ACYAJg +AD4 +AD4 +AD4 - blk+AF8-mq+AF8-rq+AF8-state(rq) +ACEAPQ MQ+AF8-RQ+AF8-IDLE) +AD4 +AD4 +AD4 +- list+AF8-empty(+ACY-rq-+AD4-queuelist)) +AD4 +AD4 +AD4 +AF8AXw-blk+AF8-mq+AF8-debugfs+AF8-rq+AF8-show(params-+AD4-m, +AD4 +AD4 +AD4 list+AF8-entry+AF8-rq(+ACY-rq-+AD4-queuelist))+ADs +AD4 +AD4 +AD4 +AH0 +AD4 +AD4 +AD4 +AD4 Hello Jens, +AD4 +AD4 +AD4 +AD4 Can you share your opinion about the above patch? +AD4 +AD4 I just convince myself that the list check is super useful. The request +AD4 could be on any number of lists, either not yet seen by the driver, or +AD4 maybe sitting in dispatch. You're only going to be finding these +AD4 requests if the tag is allocated, which means that it's already in some +AD4 sort of non-idle state. +AD4 +AD4 So enlighten me why we need the list check at all? Wouldn't it be better +AD4 to simply remove it, and ensure that the debug print includes things +AD4 like list state, rq state, etc? Hello Jens, I have tried to leave the list+AF8-empty() check out but if I do that then requests that have the state +ACI-idle+ACI (allocated but not yet started) also show up. I think these should be left out from the output produced by reading the +ACI-busy+ACI attribute because such requests are not interesting when analyzing an I/O lockup: nullb0/hctx1/busy:00000000abe67123 +AHs.op+AD0-READ, .cmd+AF8-flags+AD0, .rq+AF8-flags+AD0-IO+AF8-STAT+AHw-STATS, .s tate+AD0-idle, .tag+AD0-63, .internal+AF8-tag+AD0--1+AH0 Thanks, Bart.
