On Fri, 13 Apr 2018 18:36:20 +0200 Steffen Maier <ma...@linux.ibm.com> wrote:
> I had the need to understand I/O request processing in detail. > But I also had the need to enrich block traces with other trace events > including my own dynamic kprobe events. So I preferred block trace events > over blktrace to get everything nicely sorted into one ftrace output. > However, I missed device filtering for (un)plug events and also > the difference between the two flavors of unplug. > > The first two patches bring block trace events closer to their > counterpart in blktrace tooling. > > The last patch is just an RFC. I still kept it in this patch set because > it is inspired by PATCH 2/2. > > Changes since v1: > [PATCH v2 1/2] > Use 0 and 1 instead of false and true for __print_symbolic table. > Now "trace-cmd report" can decode it. [Steven Rostedt] > > Steffen Maier (3): > tracing/events: block: track and print if unplug was explicit or > schedule > tracing/events: block: dev_t via driver core for plug and unplug > events > tracing/events: block: also try to get dev_t via driver core for some > events > > include/trace/events/block.h | 33 ++++++++++++++++++++++++++++----- > 1 file changed, 28 insertions(+), 5 deletions(-) > >From just the tracing point of view: Reviewed-by: Steven Rostedt (VMware) <rost...@goodmis.org> as for the race conditions in the block layer, I'll let someone else decided that. -- Steve
