Re: [PATCH 06/22] tools lib traceevent: Add kmem plugin

[Steven Rostedt]

On Fri, 22 Nov 2013 23:38:17 +0900
Namhyung Kim <namhyung@xxxxxxxxxx> wrote:

> 2013-11-21 (ë), 12:01 +0100, Jiri Olsa:
> > Backporting kmem plugin.
> > 
> > Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
> > git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git
> > 
> > This plugin adds call_site field resolving for following
> > tracepoint events:
> >   kmem:kfree
> >   kmem:kmalloc
> >   kmem:kmalloc_node
> >   kmem:kmem_cache_alloc
> >   kmem:kmem_cache_alloc_node
> >   kmem:kmem_cache_free
> > 
> > The diff of 'perf script' output generated by old and new code:
> > (data was generated by 'perf record -e 'kmem:*' -a')
> > 
> > --- script.kmem.old
> > +++ script.kmem.new
> > -            perf 27846 [001] 29643.403319: kmem:kfree: call_site=ffffffff810e64f6 ptr=(nil)
> > +            perf 27846 [001] 29643.403238: kmem:kfree: (__audit_syscall_exit+0x1f6) call_site=ffffffff810e64f6 ptr=(nil)
> > -            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
> > +            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: (radix_tree_preload+0x35) call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
> > -            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
> > +            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: (jbd2_journal_stop+0x221) call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
> > -         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
> > +         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: (__alloc_skb+0x4e) call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
> > -            perf 27846 [001] 29643.445510: kmem:kmalloc: call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
> > +            perf 27846 [001] 29643.445510: kmem:kmalloc: (ext4_ext_find_extent+0x362) call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
> > -               X   888 [002] 29643.445546: kmem:kmalloc_node: call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1
> > +               X   888 [002] 29643.445546: kmem:kmalloc_node: (__alloc_skb+0x7e) call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1
> 
> It'd be great if the "call_site" in the output changes to display
> function names instead of hex addresses directly.
> 

Actually, that's what's in the (). 

   kmem:kmalloc_node: (__alloc_skb+0x7e) call_site=ffffffff8153c67e

This uses a short cut, where we don't overwrite the entire handler, in
case the TP_printk() gets new fields.

If the registered handler for an event, like "call_site_handler" (see
how we use it for all of tracepoints) returns >0, that tells the
library that we only added extra information, and to print the
tracepoint as it is normally.

The better solution here is to use "%pS" or something in the actual
tracepoint instead.

-- Steve
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/