Re: [PATCH 1/2 v2] tracing/events: provide string with undefined size support

[Frédéric Weisbecker]

Le 17 avril 2009 08:22, Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> a écrit :
> On Thu, 2009-04-16 at 22:28 +0200, Frédéric Weisbecker wrote:
>> 2009/4/16 Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>:
>> > On Thu, 2009-04-16 at 22:00 +0200, Frederic Weisbecker wrote:
>> >> Impact: less memory usage for tracing
>> >>
>> >> This patch provides the support for dynamic size strings on
>> >> event tracing.
>> >>
>> >> The key concept is to use a structure with an ending char array field of
>> >> undefined size and use such ability to allocate the minimal size on the ring
>> >> buffer to make the entry fit inside as opposite to a fixed length strings with
>> >> upper bound.
>> >>
>> >> This patch provides one new macro:
>> >>
>> >> -__ending_string(name, src)
>> >>
>> >> This one declares the string to the structure inside TP_STRUCT__entry.
>> >> You need to provide the name of the string field and the source that will be
>> >> copied inside.
>> >> Two constraints: only one __ending_string() per TRACE_EVENT can be added and
>> >> it must be the last field to be declared. Hence the __ending prefix.
>> >>
>> >> This macro will declare the necessary field and will also add the dynamic
>> >> size of the string needed for the ring buffer entry allocation.
>> >>
>> >> It also support filtering because these strings behave essentially
>> >> like usual fixed length string.
>> >
>> > can't we simply do __string(name, src) and output something like:
>> >
>> >  struct {
>> >   u16 size;
>> >   char str[0];
>> >  } name;
>> >
>> > That would get rid of this __ending_ wart.
>> >
>> >
>>
>> Hmm, I don't understand.
>> Such a thing doesn't seem to work. Once we fill the string it would
>> override the fields that
>> follow it if it's not at the end.
>
> Just grow the thing to fit whatever string length -- rather common
> pattern:
>
> struct foo {
>  int length;
>  char data[0];
> };
>
> struct foo *bar = kmalloc(sizeof(struct foo) + data_size);
>
> and bob's your uncle.
>
>


Ok, but I can't do that.
What we do is allocating towards the ring buffer before filling:

struct foo {
   field1;
   field2;
   ...
};

struct foo *f;

event = ring_buffer_lock_reserve(sizeof(*f), ...);
f = ring_buffer_event_data(event);

I can't use a kmalloc here. We are tracing a random event which can happen
at a random frequency, random context, etc...
--
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/