i965: Expose OA counters via INTEL_performance_query

This adds support for exposing basic Observation Architecture
performance counters on Haswell.

This support is based on the i915 perf kernel interface which is used
to configure the OA unit, allowing Mesa to emit MI_REPORT_PERF_COUNT
commands around queries to collect counter snapshots.

To take into account the small chance that some of the 32bit counters
could wrap around for long queries (~50 milliseconds for a GT3 Haswell @
1.1GHz) the implementation also collects periodic metrics.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

src/mesa/drivers/dri/i965/brw_context.h

@@ -1114,7 +1114,59 @@ struct brw_context
       struct brw_perf_query_info *queries;
       int n_queries;
 
+      /* The i915 perf stream we open to setup + enable the OA counters */
+      int oa_stream_fd;
+
+      /* An i915 perf stream fd gives exclusive access to the OA unit that will
+       * report counter snapshots for a specific counter set/profile in a
+       * specific layout/format so we can only start OA queries that are
+       * compatible with the currently open fd...
+       */
+      int current_oa_metrics_set_id;
+      int current_oa_format;
+
+      /* List of buffers containing OA reports */
+      struct exec_list sample_buffers;
+
+      /* Cached list of empty sample buffers */
+      struct exec_list free_sample_buffers;
+
+      int n_active_oa_queries;
       int n_active_pipeline_stats_queries;
+
+      /* The number of queries depending on running OA counters which
+       * extends beyond brw_end_perf_query() since we need to wait until
+       * the last MI_RPC command has parsed by the GPU.
+       *
+       * Accurate accounting is important here as emitting an
+       * MI_REPORT_PERF_COUNT command while the OA unit is disabled will
+       * effectively hang the gpu.
+       */
+      int n_oa_users;
+
+      /* To help catch an spurious problem with the hardware or perf
+       * forwarding samples, we emit each MI_REPORT_PERF_COUNT command
+       * with a unique ID that we can explicitly check for...
+       */
+      int next_query_start_report_id;
+
+      /**
+       * An array of queries whose results haven't yet been assembled
+       * based on the data in buffer objects.
+       *
+       * These may be active, or have already ended.  However, the
+       * results have not been requested.
+       */
+      struct brw_perf_query_object **unaccumulated;
+      int unaccumulated_elements;
+      int unaccumulated_array_size;
+
+      /* The total number of query objects so we can relinquish
+       * our exclusive access to perf if the application deletes
+       * all of its objects. (NB: We only disable perf while
+       * there are no active queries)
+       */
+      int n_query_instances;
    } perfquery;
 
    int num_atoms[BRW_NUM_PIPELINES];