Messages in this thread | | | Date | Tue, 24 Mar 2009 09:39:05 +0100 | From | Ingo Molnar <> | Subject | Re: [PATCH 5/5] blktrace: print human-readable act_mask |
| |
* Li Zefan <lizf@cn.fujitsu.com> wrote:
> Print stringified act_mask instead of hex value: > # cat act_mask > read,write,barrier,sync,queue,requeue,issue,complete,fs,pc,ahead,meta, > discard,drv_data > # echo "meta,write" > act_mask > # cat act_mask > write,meta
Nice!
It would also be nice to activate trace filters for the blktrace tracepoints - i.e. to convert them to the TRACE_EVENT() enumeration format. Beyond user-space parseable field enumeration and filter, that will also speed up tracing and allows binary record streaming with splice() zero-copy.
Via that "act_mask" can become a filterable field and you can define expressions to filter. All other fields like sector become in-kernel filterable too.
See a few examples here:
include/trace/irq_event_types.h include/trace/sched_event_types.h
Note, blktrace tracepoints are certainly more complex than the tracepoints above - you can embedd C statements in TRACE_EVENT()'s TP_fast_assign() bit.
It was specifically designed to allow the support of blktrace tracepoints, so you can embedd the blk_pc_request() and disk_devt() translation for the block_rq_complete event or block_rq_requeue/issue tracepoints.
> +static struct { > + int mask; > + const char *str; > +} mask_maps[] = { > + { BLK_TC_READ, "read" }, > + { BLK_TC_WRITE, "write" }, > + { BLK_TC_BARRIER, "barrier" }, > + { BLK_TC_SYNC, "sync" }, > + { BLK_TC_QUEUE, "queue" }, > + { BLK_TC_REQUEUE, "requeue" }, > + { BLK_TC_ISSUE, "issue" }, > + { BLK_TC_COMPLETE, "complete" }, > + { BLK_TC_FS, "fs" }, > + { BLK_TC_PC, "pc" }, > + { BLK_TC_AHEAD, "ahead" }, > + { BLK_TC_META, "meta" }, > + { BLK_TC_DISCARD, "discard" }, > + { BLK_TC_DRV_DATA, "drv_data" }, > +};
(minor nit: this should be a const.)
Jens, Arnaldo, Steve, the series from Li looks good to me - Ack?
Ingo
| |