On Fri, Mar 12, 2021 at 12:29 PM Jiri Olsa <jo...@redhat.com> wrote:
>
> On Fri, Mar 12, 2021 at 07:34:39AM -0700, Rob Herring wrote:
> > On Fri, Mar 12, 2021 at 6:59 AM Jiri Olsa <jo...@redhat.com> wrote:
> > >
> > > On Wed, Mar 10, 2021 at 05:08:31PM -0700, Rob Herring wrote:
> > >
> > > SNIP
> > >
> > > > +
> > > > static int
> > > > sys_perf_event_open(struct perf_event_attr *attr,
> > > > pid_t pid, int cpu, int group_fd,
> > > > @@ -137,6 +147,8 @@ void perf_evsel__free_fd(struct perf_evsel *evsel)
> > > > {
> > > > xyarray__delete(evsel->fd);
> > > > evsel->fd = NULL;
> > > > + xyarray__delete(evsel->mmap);
> > > > + evsel->mmap = NULL;
> > > > }
> > > >
> > > > void perf_evsel__close(struct perf_evsel *evsel)
> > > > @@ -156,6 +168,45 @@ void perf_evsel__close_cpu(struct perf_evsel
> > > > *evsel, int cpu)
> > > > perf_evsel__close_fd_cpu(evsel, cpu);
> > > > }
> > > >
> > > > +int perf_evsel__mmap(struct perf_evsel *evsel, int pages)
> > > > +{
> > > > + int ret, cpu, thread;
> > > > + struct perf_mmap_param mp = {
> > > > + .prot = PROT_READ | PROT_WRITE,
> > > > + .mask = (pages * page_size) - 1,
> > > > + };
> > >
> > > I don't mind using evsel->fd for dimensions below,
> > > but we need to check in here that it's defined,
> > > that perf_evsel__open was called
> >
> > Right, so I'll add this here:
> >
> > if (evsel->fd == NULL)
> > return -EINVAL;
Actually, pretty much none of the API checks this.
perf_evsel__run_ioctl(), perf_evsel__enable(), perf_evsel__disable(),
perf_evsel__read() will all just deference evsel->fd. So fix all of
these or follow existing behavior?
> > Note that struct evsel has dimensions in it, but they are only set in
> > the evlist code. I couldn't tell if that was by design or mistake.
>
> we do? we have the cpus and threads in perf_evsel no?
Right, perf_evsel has cpus and threads pointers which in turn have the
sizes, but those pointers are not set within the evsel code.
> also you'd need perf_evsel not evsel, right?
>
> >
> > BTW, I just noticed perf_evsel__open is leaking memory on most of its
> > error paths.
>
> nice.. let's fix it ;-)
NM, I missed that they are static...
Rob
