Re: [PATCH 16/17] uprobes: Allocate ->utask before handler_chain() for tracing handlers

[Namhyung Kim <namhyung@kernel.org>]

Hi Oleg,

On Tue, 10 Dec 2013 16:57:44 +0100, Oleg Nesterov wrote:
> On 12/10, Masami Hiramatsu wrote:
>>
>> (2013/12/09 15:20), Namhyung Kim wrote:
>> > From: Oleg Nesterov <oleg@redhat.com>
>> >
>> > uprobe_trace_print() and uprobe_perf_print() need to pass the additional
>> > info to call_fetch() methods, currently there is no simple way to do this.
>> >
>> > current->utask looks like a natural place to hold this info, but we need
>> > to allocate it before handler_chain().
>> >
>> > This is a bit unfortunate, perhaps we will find a better solution later,
>> > but this is simnple and should work right now.
>>
>> Hmm, when this will happen?
>
> Perhaps never. Perhaps it will stay forever and we remove get_utask() from
> pre_ssout() (it is not needed after this patch).
>
> However I still think we can cleanup this. And to remind, we need to clean
> the usage of utask->vaddr in trace_uprobe.c anyway. We can either try to
> find another place to pass the info, or we can create a helper(s) for the
> tracing handlers to access (and populate if NULL) utask->handler_data.
> Note that this (probably) also makes sense because we can unexport
> "struct uprobe_task" (but this needs a couple of off-topic cleanups).
>
> We will see. Lets do the minimal change which can work right now, Namhyung
> has enough more serious problems ;)

Very true. :)

>
>> and isn't it better to increment
>> miss-hit counter of the uprobe?
>
> What do you mean? This is not miss-hit and ->utask == NULL is quite normal.
> For example, on ppc it can be always NULL because ppc likely emulates the
> probed insn.

Yes, for x86, it's always NULL at first but then populated after doing
single-step.  What we try to do is just moving the allocation before
calling handler since we need to carry some information through it.

Thanks,
Namhyung

>
> Or did you mean that if get_utask() fails we should report this somehow?
> Well, GFP_KERNEL should "never" fail and even if it fails we will restart
> the same insn and retry the allocation.
>
> Or did I miss your point completely ?
>
> Oleg.