blk_classic tracing horribly broken

blk_classic tracing horribly broken

[Jeff Moyer]

Hi,

I've stared at the code for a while, and I've come up empty.  Would
anyone care to take a stab at fixing this?

To reproduce:

mount -t debugfs none /sys/kernel/debug
cd /sys/kernel/debug/tracing
echo blk > current_tracer
echo 1 > events/block/enable
echo 1 > options/blk_classic
echo 1 > tracing_on

dd if=/dev/<device> of=/dev/null bs=1M count=1 iflag=direct
cat trace

This is what I see:

# tracer: blk
#
# DEV   CPU TIMESTAMP     PID ACT FLG
#  |     |     |           |   |   |
Unknown action 52
Unknown action 52
Unknown action bce
Unknown action 52
Unknown action 52
Unknown action 52
Unknown action 52
Unknown action 52
Unknown action 52
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
  0,0    2   110.941937315  2584  F   R [pgrep]
Unknown action 52
Unknown action 52
Unknown action 231
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
  0,0    0   110.949383287  2584  F   R [pgrep]
Unknown action 52
Unknown action 52
Unknown action 7fd1
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
  0,0    3   110.984565328  2581  F   R [ps]
Unknown action 52
Unknown action 52
Unknown action 5ad1
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0
  0,0    3   111.003843579  2581  F   R [ps]
Unknown action 52
Unknown action 52
Unknown action 7551
Unknown action 0
Unknown action 0
Unknown action 0
Unknown action 0

Cheers,
Jeff

Re: blk_classic tracing horribly broken

[Christoph Hellwig]

On Thu, Sep 22, 2011 at 01:57:11PM -0400, Jeff Moyer wrote:
> Hi,
> 
> I've stared at the code for a while, and I've come up empty.  Would
> anyone care to take a stab at fixing this?
> 
> To reproduce:
> 
> mount -t debugfs none /sys/kernel/debug
> cd /sys/kernel/debug/tracing
> echo blk > current_tracer
> echo 1 > events/block/enable
> echo 1 > options/blk_classic
> echo 1 > tracing_on
> 
> dd if=/dev/<device> of=/dev/null bs=1M count=1 iflag=direct
> cat trace

Why are you using this interface anyway?  We actually had planned to
drop it long time ago, but for some reason that never happened.

Re: blk_classic tracing horribly broken

[Jeff Moyer]

Christoph Hellwig <hch@infradead.org> writes:

> On Thu, Sep 22, 2011 at 01:57:11PM -0400, Jeff Moyer wrote:
>> Hi,
>> 
>> I've stared at the code for a while, and I've come up empty.  Would
>> anyone care to take a stab at fixing this?
>> 
>> To reproduce:
>> 
>> mount -t debugfs none /sys/kernel/debug
>> cd /sys/kernel/debug/tracing
>> echo blk > current_tracer
>> echo 1 > events/block/enable
>> echo 1 > options/blk_classic
>> echo 1 > tracing_on
>> 
>> dd if=/dev/<device> of=/dev/null bs=1M count=1 iflag=direct
>> cat trace
>
> Why are you using this interface anyway?  We actually had planned to
> drop it long time ago, but for some reason that never happened.

I was using it when testing the flush+fua reporting in blktrace, to be
sure we didn't break it.  Since it was already broken, I didn't have to
worry, but I put it on my todo list to investigate why.  Dropping broken
code is fine with me (given that there are two alternatives to use).

Cheers,
Jeff