Re: [PATCH] io_uring: Add to traces the req pointer when available

[Olivier Langlois]

Hi Pavel,

On Tue, 2021-05-25 at 09:33 +0100, Pavel Begunkov wrote:
> On 5/25/21 9:21 AM, Pavel Begunkov wrote:
> > On 5/25/21 6:54 AM, Olivier Langlois wrote:
> > > The req pointer uniquely identify a specific request.
> > > Having it in traces can provide valuable insights that is not
> > > possible
> > > to have if the calling process is reusing the same user_data value.
> > 
> > How about hashing kernel pointers per discussed? Even if it's better
> > to have it done by tracing or something as you mentioned, there is no
> > such a thing at the moment, so should be done by hand.
> 
> Or do you mean that it's already the case? Can anyone
> confirm if so?

I did consider your option but then I did some research on the pointer
hashing idea.

It turns out to be already addressed by the trace subsystem.

Here is what I have found:

whippet2 /sys/kernel/tracing # cat trace_options 
print-parent
nosym-offset
nosym-addr
noverbose
noraw
nohex
nobin
noblock
trace_printk
annotate
nouserstacktrace
nosym-userobj
noprintk-msg-only
context-info
nolatency-format
record-cmd
norecord-tgid
overwrite
nodisable_on_free
irq-info
markers
noevent-fork
nopause-on-trace
hash-ptr
function-trace
nofunction-fork
nodisplay-graph
nostacktrace
notest_nop_accept
notest_nop_refuse

hash-ptr option is enabled by default.

I am not 100% sure to understand why I am getting naked pointer values
when I am getting the traces with 'sudo perf':

  9287.369 test/625 io_uring:io_uring_task_run(ctx: 0xffff8fbf9a834800,
opcode: 22, user_data: 216454257090494477, result: 195)
  9287.386 test/625 io_uring:io_uring_task_run(ctx: 0xffff8fbf9a834800,
opcode: 22, user_data: 216454257090494477, result: 195)

but the pointers should be hashed by trace.

That would be nice if someone more knowledgeable about the tracing
system could jump in and comment about the hash-ptr option and tell
when it is applied and when it is not...