[ltt-dev] [PATCH 07/13] implement fmt align

[Mathieu Desnoyers]

* Lai Jiangshan (laijs at cn.fujitsu.com) wrote:
> Mathieu Desnoyers wrote:
> > * Lai Jiangshan (laijs at cn.fujitsu.com) wrote:
> >> we need know the align for accessing the binary data from
> >> event.(skip the padding data, calculate offset).
> >>
> >> It can also be used for write-side.
> >>
> > 
> > Yes, but I prefer not, because I like to do a single pass in the format
> > string to calculate both the type sizes individually and the largest
> > alignment. It makes sense to implement this for the read-side though.
> > 
> > Please see comments below,
> > 
> >> Signed-off-by: Lai Jiangshan <laijs at cn.fujitsu.com>
> >> ---
> >>  b/ltt/ltt-serialize.c      |   60 +++++++++++++++++++++++++++++++++++++++++++++
> >>  include/linux/ltt-tracer.h |    9 ++++++
> >>  2 files changed, 69 insertions(+)
> >> diff --git a/include/linux/ltt-tracer.h b/include/linux/ltt-tracer.h
> >> index ece6819..b49cac2 100644
> >> --- a/include/linux/ltt-tracer.h
> >> +++ b/include/linux/ltt-tracer.h
> >> @@ -311,6 +316,9 @@ static inline unsigned int ltt_align(size_t align_drift, size_t size_of_type)
> >>  	size_t alignment = min(sizeof(void *), size_of_type);
> >>  	return (alignment - align_drift) & (alignment - 1);
> >>  }
> >> +
> >> +extern unsigned int ltt_fmt_largest_align(size_t align_drift, const char *fmt);
> >> +
> >>  /* Default arch alignment */
> >>  #define LTT_ALIGN
> >>  
> >> @@ -327,6 +335,12 @@ static inline unsigned int ltt_align(size_t align_drift,
> >>  	return 0;
> >>  }
> >>  
> >> +static inline unsigned int ltt_fmt_largest_align(size_t align_drift,
> >> +		const char *fmt)
> >> +{
> >> +	return 0;
> >> +}
> >> +
> >>  #define LTT_ALIGN __attribute__((packed))
> >>  
> >>  static inline int ltt_get_alignment(void)
> >> diff --git a/ltt/ltt-serialize.c b/ltt/ltt-serialize.c
> >> index 5c34899..deb9106 100644
> >> --- a/ltt/ltt-serialize.c
> >> +++ b/ltt/ltt-serialize.c
> >> @@ -531,6 +538,66 @@
> >>  }
> >>  EXPORT_SYMBOL_GPL(ltt_serialize_printf);
> >>  
> >> +#ifdef CONFIG_LTT_ALIGNMENT
> >> +
> >> +unsigned int ltt_fmt_largest_align(size_t align_drift, const char *fmt)
> >> +{
> >> +	char trace_size = 0, c_size = 0;
> >> +	enum ltt_type trace_type = LTT_TYPE_NONE, c_type = LTT_TYPE_NONE;
> >> +	unsigned long attributes = 0;
> >> +	int largest_align = 1;
> >> +
> >> +	for (; *fmt ; ++fmt) {
> >> +		switch (*fmt) {
> >> +		case '#':
> >> +			/* tracetypes (#) */
> >> +			++fmt;			/* skip first '#' */
> >> +			if (*fmt == '#')	/* Escaped ## */
> >> +				break;
> >> +			attributes = 0;
> >> +			fmt = parse_trace_type(fmt, &trace_size, &trace_type,
> >> +				&attributes);
> >> +
> >> +			largest_align = max_t(int, largest_align, trace_size);
> >> +			if (largest_align >= ltt_get_alignment())
> >> +				goto exit;
> >> +			break;
> >> +		case '%':
> >> +			/* c types (%) */
> >> +			++fmt;			/* skip first '%' */
> >> +			if (*fmt == '%')	/* Escaped %% */
> >> +				break;
> >> +			fmt = parse_c_type(fmt, &c_size, &c_type, NULL);
> >> +			/*
> >> +			 * Output c types if no trace types has been
> >> +			 * specified.
> >> +			 */
> >> +			if (!trace_size)
> >> +				trace_size = c_size;
> >> +			if (trace_type == LTT_TYPE_NONE)
> >> +				trace_type = c_type;
> >> +			if (c_type == LTT_TYPE_STRING)
> >> +				trace_type = LTT_TYPE_STRING;
> >> +
> >> +			largest_align = max_t(int, largest_align, trace_size);
> >> +			if (largest_align >= ltt_get_alignment())
> >> +				goto exit;
> >> +
> >> +			trace_size = 0;
> >> +			c_size = 0;
> >> +			trace_type = LTT_TYPE_NONE;
> >> +			c_size = LTT_TYPE_NONE;
> >> +			break;
> >> +		}
> >> +	}
> >> +
> >> +exit:
> >> +	return (largest_align - align_drift) & (largest_align - 1);
> > 
> > Hrm, I think there is something wrong here.
> > 
> > On x86_32, the max alignment is sizeof(u32).
> > 
> > However, if you get a u64 type here, the largest_align will become
> > sizeof(u64). This is why, in ltt_vtrace(), I do a 
> > 
> >         largest_align = 1;      /* must be non-zero for ltt_align */
> >         data_size = ltt_get_data_size(&closure, serialize_private,
> >                                         &largest_align, fmt, &args_copy);
> >         largest_align = min_t(int, largest_align, sizeof(void *));
> > 
> > So we should document in the ltt_fmt_largest_align header to always do a
> > largest_align = min_t(int, largest_align, sizeof(void *)); on the return
> > value (and check that we do it) or we simply do this operation before
> > returning.
> > 
> > Mathieu
> 
> I haven't seen this line:
> 			if (largest_align >= ltt_get_alignment())
> 				goto exit;
> 
> We shouldn't hardcode "sizeof(void *)", we should use ltt_get_alignment().
> 

My comment was that it's necessary to do a 

largest_align = min_t(int, largest_align, sizeof(void *));

to make sure the alignment value is never bigger that sizeof(void *). I
don't understand how your answer here relates to my concern. (maybe I
have problem understanding your answer on my side..). Can you elaborate
more please ?

Mathieu


> 
> > 
> >> +}
> >> +EXPORT_SYMBOL_GPL(ltt_fmt_largest_align);
> >> +
> >> +#endif
> >> +
> >>  /*
> >>   * Calculate data size
> >>   * Assume that the padding for alignment starts at a sizeof(void *) address.
> >>
> >>
> >>
> >>
> >> _______________________________________________
> >> ltt-dev mailing list
> >> ltt-dev at lists.casi.polymtl.ca
> >> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
> >>
> > 
> 
> 
> 
> _______________________________________________
> ltt-dev mailing list
> ltt-dev at lists.casi.polymtl.ca
> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68