Re: [PATCH 5/9] perf utils: add support for arch standard events

[Jiri Olsa]

On Thu, Feb 08, 2018 at 02:45:37PM +0000, John Garry wrote:
> On 08/02/2018 13:55, Jiri Olsa wrote:
> > On Wed, Feb 07, 2018 at 01:45:00AM +0800, John Garry wrote:
> > 
> > SNIP
> > 
> > > +	char *perpkg;
> > > +	char *unit;
> > > +	char *metric_expr;
> > > +	char *metric_name;
> > > +	char *metric_group;
> > > +	struct list_head list;
> > > +	char strings[];
> > > +};
> > > +
> > > +static LIST_HEAD(arch_std_events);
> > > +
> > > +#define ADD_EVENT_STRING(string) do { if (string) {		\
> > > +	es->string = strings;					\
> > > +	strings += snprintf(strings, len, "%s", string) + 1;	\
> > > +} } while (0)
> > > +
> > > +static int save_arch_std_events(void *data, char *name, char *event,
> > > +				char *desc, char *long_desc, char *pmu,
> > > +				char *unit, char *perpkg, char *metric_expr,
> > > +				char *metric_name, char *metric_group)
> > > +{
> > > +	struct event_struct *es;
> > > +	struct stat *sb = data;
> > > +	int len;
> > > +	char *strings;
> > > +
> > > +	/*
> > > +	 * Lazily allocate size of the json file to hold the
> > > +	 * strings, which would be more than large enough.
> > > +	 */
> > > +	len = sb->st_size;
> > > +
> > > +	es = malloc(sizeof(*es) + len);
> > 
> > hum, so for single event you allocate buffer of the size
> > of the entire file this event is defined in?
> > 
> > what do I miss? I assume there're more of those arch-defined
> > events defined in the single file..
> 
> Hi Jirka,
> 
> Yes, allocating the file size per event was just to make the code more
> concise (instead of finding each string length), but obviously it is an
> inefficient practice in terms of memory usage.
> 
> But since the JSONs are generally not huge, and in practice we would only be
> accessing a fraction of the buffer's physical memory to save the event
> strings, I thought it ok.
> 
> Anyway, I'll see if there is something more efficient I can do.

maybe the json parser could provide the overall lenght?

jirka