Merge branch 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull perf updates from Ingo Molnar:
 "Kernel side changes mostly consist of work on x86 PMU drivers:

   - x86 Intel PT (hardware CPU tracer) improvements (Alexander
     Shishkin)

   - x86 Intel CQM (cache quality monitoring) improvements (Thomas
     Gleixner)

   - x86 Intel PEBSv3 support (Peter Zijlstra)

   - x86 Intel PEBS interrupt batching support for lower overhead
     sampling (Zheng Yan, Kan Liang)

   - x86 PMU scheduler fixes and improvements (Peter Zijlstra)

  There's too many tooling improvements to list them all - here are a
  few select highlights:

  'perf bench':

      - Introduce new 'perf bench futex' benchmark: 'wake-parallel', to
        measure parallel waker threads generating contention for kernel
        locks (hb->lock). (Davidlohr Bueso)

  'perf top', 'perf report':

      - Allow disabling/enabling events dynamicaly in 'perf top':
        a 'perf top' session can instantly become a 'perf report'
        one, i.e. going from dynamic analysis to a static one,
        returning to a dynamic one is possible, to toogle the
        modes, just press 'f' to 'freeze/unfreeze' the sampling. (Arnaldo Carvalho de Melo)

      - Make Ctrl-C stop processing on TUI, allowing interrupting the load of big
        perf.data files (Namhyung Kim)

  'perf probe': (Masami Hiramatsu)

      - Support glob wildcards for function name
      - Support $params special probe argument: Collect all function arguments
      - Make --line checks validate C-style function name.
      - Add --no-inlines option to avoid searching inline functions
      - Greatly speed up 'perf probe --list' by caching debuginfo.
      - Improve --filter support for 'perf probe', allowing using its arguments
        on other commands, as --add, --del, etc.

  'perf sched':

      - Add option in 'perf sched' to merge like comms to lat output (Josef Bacik)

  Plus tons of infrastructure work - in particular preparation for
  upcoming threaded perf report support, but also lots of other work -
  and fixes and other improvements.  See (much) more details in the
  shortlog and in the git log"

* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (305 commits)
  perf tools: Configurable per thread proc map processing time out
  perf tools: Add time out to force stop proc map processing
  perf report: Fix sort__sym_cmp to also compare end of symbol
  perf hists browser: React to unassigned hotkey pressing
  perf top: Tell the user how to unfreeze events after pressing 'f'
  perf hists browser: Honour the help line provided by builtin-{top,report}.c
  perf hists browser: Do not exit when 'f' is pressed in 'report' mode
  perf top: Replace CTRL+z with 'f' as hotkey for enable/disable events
  perf annotate: Rename source_line_percent to source_line_samples
  perf annotate: Display total number of samples with --show-total-period
  perf tools: Ensure thread-stack is flushed
  perf top: Allow disabling/enabling events dynamicly
  perf evlist: Add toggle_enable() method
  perf trace: Fix race condition at the end of started workloads
  perf probe: Speed up perf probe --list by caching debuginfo
  perf probe: Show usage even if the last event is skipped
  perf tools: Move libtraceevent dynamic list to separated LDFLAGS variable
  perf tools: Fix a problem when opening old perf.data with different byte order
  perf tools: Ignore .config-detected in .gitignore
  perf probe: Fix to return error if no probe is added
  ...

1 files changed, 19 insertions, 19 deletions

diff --git a/arch/x86/kernel/cpu/perf_event_intel_pt.c b/arch/x86/kernel/cpu/perf_event_intel_pt.c
index 123ff1b..159887c 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_pt.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_pt.c
@@ -187,15 +187,6 @@ static bool pt_event_valid(struct perf_event *event)
  * These all are cpu affine and operate on a local PT
  */
 
-static bool pt_is_running(void)
-{
-	u64 ctl;
-
-	rdmsrl(MSR_IA32_RTIT_CTL, ctl);
-
-	return !!(ctl & RTIT_CTL_TRACEEN);
-}
-
 static void pt_config(struct perf_event *event)
 {
 	u64 reg;
@@ -609,7 +600,12 @@ static unsigned int pt_topa_next_entry(struct pt_buffer *buf, unsigned int pg)
  * @handle:	Current output handle.
  *
  * Place INT and STOP marks to prevent overwriting old data that the consumer
- * hasn't yet collected.
+ * hasn't yet collected and waking up the consumer after a certain fraction of
+ * the buffer has filled up. Only needed and sensible for non-snapshot counters.
+ *
+ * This obviously relies on buf::head to figure out buffer markers, so it has
+ * to be called after pt_buffer_reset_offsets() and before the hardware tracing
+ * is enabled.
  */
 static int pt_buffer_reset_markers(struct pt_buffer *buf,
 				   struct perf_output_handle *handle)
@@ -618,9 +614,6 @@ static int pt_buffer_reset_markers(struct pt_buffer *buf,
 	unsigned long head = local64_read(&buf->head);
 	unsigned long idx, npages, wakeup;
 
-	if (buf->snapshot)
-		return 0;
-
 	/* can't stop in the middle of an output region */
 	if (buf->output_off + handle->size + 1 <
 	    sizes(TOPA_ENTRY(buf->cur, buf->cur_idx)->size))
@@ -674,7 +667,7 @@ static void pt_buffer_setup_topa_index(struct pt_buffer *buf)
 	struct topa *cur = buf->first, *prev = buf->last;
 	struct topa_entry *te_cur = TOPA_ENTRY(cur, 0),
 		*te_prev = TOPA_ENTRY(prev, prev->last - 1);
-	int pg = 0, idx = 0, ntopa = 0;
+	int pg = 0, idx = 0;
 
 	while (pg < buf->nr_pages) {
 		int tidx;
@@ -689,9 +682,9 @@ static void pt_buffer_setup_topa_index(struct pt_buffer *buf)
 			/* advance to next topa table */
 			idx = 0;
 			cur = list_entry(cur->list.next, struct topa, list);
-			ntopa++;
-		} else
+		} else {
 			idx++;
+		}
 		te_cur = TOPA_ENTRY(cur, idx);
 	}
 
@@ -703,7 +696,14 @@ static void pt_buffer_setup_topa_index(struct pt_buffer *buf)
  * @head:	Write pointer (aux_head) from AUX buffer.
  *
  * Find the ToPA table and entry corresponding to given @head and set buffer's
- * "current" pointers accordingly.
+ * "current" pointers accordingly. This is done after we have obtained the
+ * current aux_head position from a successful call to perf_aux_output_begin()
+ * to make sure the hardware is writing to the right place.
+ *
+ * This function modifies buf::{cur,cur_idx,output_off} that will be programmed
+ * into PT msrs when the tracing is enabled and buf::head and buf::data_size,
+ * which are used to determine INT and STOP markers' locations by a subsequent
+ * call to pt_buffer_reset_markers().
  */
 static void pt_buffer_reset_offsets(struct pt_buffer *buf, unsigned long head)
 {
@@ -901,6 +901,7 @@ void intel_pt_interrupt(void)
 		}
 
 		pt_buffer_reset_offsets(buf, pt->handle.head);
+		/* snapshot counters don't use PMI, so it's safe */
 		ret = pt_buffer_reset_markers(buf, &pt->handle);
 		if (ret) {
 			perf_aux_output_end(&pt->handle, 0, true);
@@ -923,7 +924,7 @@ static void pt_event_start(struct perf_event *event, int mode)
 	struct pt *pt = this_cpu_ptr(&pt_ctx);
 	struct pt_buffer *buf = perf_get_aux(&pt->handle);
 
-	if (pt_is_running() || !buf || pt_buffer_is_full(buf, pt)) {
+	if (!buf || pt_buffer_is_full(buf, pt)) {
 		event->hw.state = PERF_HES_STOPPED;
 		return;
 	}
@@ -954,7 +955,6 @@ static void pt_event_stop(struct perf_event *event, int mode)
 	event->hw.state = PERF_HES_STOPPED;
 
 	if (mode & PERF_EF_UPDATE) {
-		struct pt *pt = this_cpu_ptr(&pt_ctx);
 		struct pt_buffer *buf = perf_get_aux(&pt->handle);
 
 		if (!buf)