Re: [PATCH 02/10] perf, tools: Support handling complete branch stacks as histograms

[Andi Kleen <andi@firstfloor.org>]

On Thu, Nov 13, 2014 at 05:08:33PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Thu, Nov 13, 2014 at 08:52:08PM +0100, Andi Kleen escreveu:
> > On Thu, Nov 13, 2014 at 04:14:17PM -0300, Arnaldo Carvalho de Melo wrote:
> > > Em Wed, Nov 12, 2014 at 06:05:20PM -0800, Andi Kleen escreveu:
> > > > +static int remove_loops(struct branch_entry *l, int nr)
> > > > +{
> > > > +	int i, j, off;
> > > > +	unsigned char chash[CHASHSZ];
> > > > +
> > > > +	memset(chash, NO_ENTRY, sizeof(chash));
> > > > +
> > > > +	BUG_ON(nr >= 256);
> > > 
> > > What is wrong with return -1 and propagating the error, so that the user
> > > is informed if the data being processed is bogus, stop processing with a
> > > warning or continue processing if finding the next valid record is
> > > possible?
> > 
> > The error doesn't depend on the record. There is a check for the record
> > being < 127 in front of this. This is merely to catch that 
> > if someone increases PERF_MAX_BRANCH_DEPTH to below 255 they need
> > to increase the type of the hash table from u8.
> 
> Ok, so this would be better as a BUILD_BUG_ON? 

Yes that should work.

-Andi

> 
> Like:
> 
> #define CHASHSZ 127
> #define CHASHBITS 7
> #define NO_ENTRY 0xff
> 
> #define PERF_MAX_BRANCH_DEPTH 127
> 
> /* Remove loops. */
> static int remove_loops(struct branch_entry *l, int nr)
> {
> 	int i, j, off;
> 	unsigned char chash[CHASHSZ];
> 
> 	memset(chash, NO_ENTRY, sizeof(chash));
> 
> 	/* Change the type of the chash table, u8 is not enough now */
> 	BUILD_BUG_ON(PERF_MAX_BRANCH_DEPTH >= 256);
> 	for (i = 0; i < nr; i++) {
> 		int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
> 
> -------------------------------------
> 
> - Arnaldo
> 

-- 
ak@linux.intel.com -- Speaking for myself only.