Re: [RFC 1/4] perf kvm: Enable 'record' on powerpc

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Tue, Mar 22, 2016 at 11:19:21PM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Tue, Mar 22, 2016 at 04:12:11PM -0300, Arnaldo Carvalho de Melo escreveu:
> > Em Wed, Feb 24, 2016 at 02:37:42PM +0530, Ravi Bangoria escreveu:
> > > 'perf kvm record' is not available on powerpc because 'perf' relies on
> > > the 'cycles' event (a PMU event) to profile the guest. However, for
> > > powerpc, this can't be used from the host because the PMUs are controlled
> > > by the guest rather than the host.
> > > 
> > > There exists a tracepoint 'kvm_hv:kvm_guest_exit' in powerpc which is
> > > hit whenever any of the threads exit the guest context. The guest
> > > instruction pointer dumped along with this tracepoint data in the field
> > > 'pc', can be used as guest instruction pointer.
> > > 
> > > This patch changes default event as kvm_hv:kvm_guest_exit for recording
> > > guest data in host on powerpc. As we are using host event to record guest
> > > data, this approach will enable only --guest option of 'perf kvm'. Still
> > > --host --guest together won't work.
> > 
> > It should, i.e. --host --guest should translate to:
> > 
> >    -e cycles:H,kvm_hv:kvm_guest_exit
> > 
> > I.e. both collect cycles only in the host, and also the tracepoint that
> > will allow us to get the guest approximation for the unavailable cycles
> > event, no?
> > 
> > I'm putting the infrastructure work needed for this the perf/cpumode
> > branch. More work will be put there soon.
> 
> So I took a different path and made perf_evsel__parse_sample set a new
> perf_sample.cpumode field, this way we'll end up having just to set a
> per-evsel ->post_parse_sample() callback for the event that replaces
> "cycles" for PPC guests where we'll just set data->ip and data->cpumode,
> the rest of the code remains unchanged.
> 
> The changes I made looks useful in itself, as, IIRC more code was
> removed than added.
> 
> I'll continue tomorrow and will test with the kvm:kvm_exit on x86_64 for
> testing, that has:

Ok, so the infrastructure got merged already and from there the next
steps are in running with:

 perf kvm --guest record -a -e cycles:H,kvm:kvm_exit

And then, with the patch below applied, try:

perf kvm --guestkallsyms kallsyms.guest --guestmodules modules.guest report -i perf.data.guest --munge-ppc-guest-sample kvm:kvm_exit

I'm almost there, it is still not resolving to the kernel DSO, etc, so I
get:

Samples: 1K of event 'kvm:kvm_exit', Event count (approx.): 1924
Overhead  Command  Shared Object     Symbol
  34.77%  :5343    [unknown]         [g] 0xffffffff81043158
  16.84%  :5345    [unknown]         [g] 0xffffffff813f3d5a
  16.84%  :5345    [unknown]         [g] 0xffffffff813f43ec
  13.83%  :5345    [unknown]         [g] 0xffffffff81043158
   9.62%  :5343    [unknown]         [g] 0xffffffff8104301a
   3.79%  :5345    [unknown]         [g] 0xffffffff8104301a
   1.77%  :5345    [unknown]         [u] 0x0000003ae6c75dc9
   0.52%  :5343    [unknown]         [g] 0xffffffff812a29b1
   0.16%  :5343    [unknown]         [g] 0xffffffff8100bc00
   0.10%  :5343    [unknown]         [g] 0xffffffff8104315a
   0.10%  :5343    [unknown]         [g] 0xffffffff8106306f
   0.10%  :5343    [unknown]         [g] 0xffffffff8153b7fc
   0.10%  :5345    [unknown]         [g] 0xffffffff8106306f
   0.05%  :5343    [unknown]         [g] 0xffffffff8100b720

[root@jouet ~]# cat /proc/*/task/5343/comm
qemu-system-x86
[root@jouet ~]#

The patch does several of the things you did per sample, but only right after
opening the perf.data file, and I'll break it down in multiple patches, this is
just a heads up, please review it if you have the time, in the end we should
have a mechanism useful not just for PPC and that affects just 'perf kvm' in
this specific case.

- Arnaldo

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index bff666458b28..b7b6527446f8 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -1480,6 +1480,86 @@ perf_stat:
 }
 #endif /* HAVE_KVM_STAT_SUPPORT */
 
+#ifdef __powerpc__
+#define PPC_HV_DECREMENTER 2432
+#define PPC_HV_BIT 3
+#define PPC_PR_BIT 49
+#define PPC_MAX 63
+
+static bool perf_sample__is_hv_dec_trap(struct perf_sample *sample, struct perf_evsel *evsel)
+{
+	int trap = perf_evsel__intval(evsel, sample, "trap");
+	return trap == PPC_HV_DECREMENTER;
+}
+
+static void perf_kvm__munge_ppc_guest_sample(struct perf_evsel *evsel, struct perf_sample *sample)
+{
+	unsigned long msr, hv, pr;
+
+	if (!perf_sample__is_hv_dec_trap(sample, evsel))
+		return;
+
+	sample->ip = perf_evsel__intval(evsel, sample, "pc");
+	sample->cpumode = PERF_RECORD_MISC_GUEST_KERNEL;
+
+	msr = perf_evsel__intval(evsel, sample, "msr");
+	hv = msr & ((unsigned long)1 << (PPC_MAX - PPC_HV_BIT));
+	pr = msr & ((unsigned long)1 << (PPC_MAX - PPC_PR_BIT));
+	if (!hv && pr)
+		sample->cpumode = PERF_RECORD_MISC_GUEST_USER;
+}
+
+static bool perf_evlist__recorded_on_ppc(const struct perf_evlist *evlist)
+{
+	if (evlist->env && evlist->env->arch) {
+		return !strcmp(evlist->env->arch, "ppc64") ||
+		       !strcmp(evlist->env->arch, "ppc64le");
+	}
+	return false;
+}
+#elif defined(__i386__) || defined(__x86_64__)
+/*
+ * For testing with kvm:kvm_exit
+ */
+static void perf_kvm__munge_ppc_guest_sample(struct perf_evsel *evsel, struct perf_sample *sample)
+{
+	sample->ip = perf_evsel__intval(evsel, sample, "guest_rip");
+	sample->cpumode = PERF_RECORD_MISC_GUEST_KERNEL;
+
+	if (sample->ip < 0xffffffff00000000)
+		sample->cpumode = PERF_RECORD_MISC_GUEST_USER;
+
+	if (0) {
+		fprintf(stderr, "%s: %s, cpumode=%d ip=%" PRIx64 "\n",
+			__func__, perf_evsel__name(evsel), sample->cpumode, sample->ip);
+	}
+}
+
+static bool perf_evlist__recorded_on_ppc(const struct perf_evlist *evlist __maybe_unused)
+{
+	return true;
+}
+#endif
+
+int perf_kvm__setup_munge_ppc_guest_event(struct perf_evlist *evlist, const char *name)
+{
+	struct perf_evsel *evsel;
+
+	if (!perf_evlist__recorded_on_ppc(evlist))
+		return 0;
+
+	evsel = perf_evlist__find_tracepoint_by_name(evlist, name);
+	if (evsel == NULL)
+		return -1;
+
+	if (0)
+		fprintf(stderr, "%s: %s\n", __func__, perf_evsel__name(evsel));
+
+	evsel->munge_sample = perf_kvm__munge_ppc_guest_sample;
+
+	return 0;
+}
+
 static int __cmd_record(const char *file_name, int argc, const char **argv)
 {
 	int rec_argc, i = 0, j;
diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 160ea23b45aa..ba8786d12423 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -667,6 +667,7 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 	int branch_mode = -1;
 	bool branch_call_mode = false;
 	char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
+	const char *munge_ppc_guest_event = NULL;
 	const char * const report_usage[] = {
 		"perf report [<options>]",
 		NULL
@@ -814,6 +815,8 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 		    "Show raw trace event output (do not use print fmt or plugins)"),
 	OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
 		    "Show entries in a hierarchy"),
+	OPT_STRING(0, "munge-ppc-guest-sample", &munge_ppc_guest_event, "event name",
+		   "Use a different event to simulate guest cycles, not accessible on PPC hosts"),
 	OPT_END()
 	};
 	struct perf_data_file file = {
@@ -880,6 +883,12 @@ repeat:
 	if (session == NULL)
 		return -1;
 
+	if (munge_ppc_guest_event &&
+	    perf_kvm__setup_munge_ppc_guest_event(session->evlist, munge_ppc_guest_event)) {
+		pr_err("PPC event not present in %s file\n", input_name);
+		goto error;
+	}
+
 	if (report.queue_size) {
 		ordered_events__set_alloc_size(&session->ordered_events,
 					       report.queue_size);
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 738ce226002b..1665171a6e9c 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -216,6 +216,7 @@ void perf_evsel__init(struct perf_evsel *evsel,
 	evsel->sample_size = __perf_evsel__sample_size(attr->sample_type);
 	perf_evsel__calc_id_pos(evsel);
 	evsel->cmdline_group_boundary = false;
+	evsel->munge_sample = NULL;
 }
 
 struct perf_evsel *perf_evsel__new_idx(struct perf_event_attr *attr, int idx)
@@ -1887,6 +1888,9 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		}
 	}
 
+	if (evsel->munge_sample != NULL)
+		evsel->munge_sample(evsel, data);
+
 	return 0;
 }
 
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 501ea6e565f1..46379451f641 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -12,6 +12,7 @@
 #include "counts.h"
 
 struct perf_evsel;
+struct perf_sample;
 
 /*
  * Per fd, to map back from PERF_SAMPLE_ID to evsel, only used when there are
@@ -92,6 +93,8 @@ struct perf_evsel {
 	double			scale;
 	const char		*unit;
 	struct event_format	*tp_format;
+	void			(*munge_sample)(struct perf_evsel *evsel,
+						struct perf_sample *sample);
 	off_t			id_offset;
 	void			*priv;
 	u64			db_id;
@@ -236,8 +239,6 @@ int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
 		     struct thread_map *threads);
 void perf_evsel__close(struct perf_evsel *evsel, int ncpus, int nthreads);
 
-struct perf_sample;
-
 void *perf_evsel__rawptr(struct perf_evsel *evsel, struct perf_sample *sample,
 			 const char *name);
 u64 perf_evsel__intval(struct perf_evsel *evsel, struct perf_sample *sample,
diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
index 8298d607c738..12bc51450d56 100644
--- a/tools/perf/util/util.h
+++ b/tools/perf/util/util.h
@@ -356,4 +356,9 @@ typedef void (*print_binary_t)(enum binary_printer_ops,
 void print_binary(unsigned char *data, size_t len,
 		  size_t bytes_per_line, print_binary_t printer,
 		  void *extra);
+
+struct perf_evlist;
+
+int perf_kvm__setup_munge_ppc_guest_event(struct perf_evlist *evlist, const char *name);
+
 #endif /* GIT_COMPAT_UTIL_H */