intel/perf: deal with OA reports timestamp values on DG2

OA reports on XeHP have their timestamp shifted to the left by 1. To
get that back in the same time domain as the REG_READ you need to
shift it back to the right and you're loosing the top bit.

v2: use ull for 64bit constant (Ian)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16144>

src/intel/ds/intel_pps_driver.cc

@@ -161,6 +161,7 @@ void IntelDriver::enable_perfcnt(uint64_t sampling_period_ns)
 {
    this->sampling_period_ns = sampling_period_ns;
 
+   gpu_timestamp_udw = intel_read_gpu_timestamp(drm_device.fd) & ~perf->cfg->oa_timestamp_mask;
    if (!perf->open(sampling_period_ns, selected_query)) {
       PPS_LOG_FATAL("Failed to open intel perf");
    }
@@ -212,10 +213,10 @@ std::vector<PerfRecord> IntelDriver::parse_perf_records(const std::vector<uint8_
           * do it now. If we see a roll over the lower 32bits capture it
           * again.
           */
-         if (gpu_timestamp_udw == 0 || (gpu_timestamp_udw + gpu_timestamp_ldw) < last_gpu_timestamp)
-            gpu_timestamp_udw = intel_read_gpu_timestamp(drm_device.fd) & 0xffffffff00000000;
+         if (gpu_timestamp_udw == 0 || (gpu_timestamp_udw | gpu_timestamp_ldw) < last_gpu_timestamp)
+            gpu_timestamp_udw = intel_read_gpu_timestamp(drm_device.fd) & ~perf->cfg->oa_timestamp_mask;
 
-         uint64_t gpu_timestamp = gpu_timestamp_udw + gpu_timestamp_ldw;
+         uint64_t gpu_timestamp = gpu_timestamp_udw | gpu_timestamp_ldw;
 
          auto duration = intel_device_info_timebase_scale(&perf->devinfo,
                                                           gpu_timestamp - prev_gpu_timestamp);

src/intel/perf/intel_perf.c

@@ -714,6 +714,13 @@ oa_metrics_available(struct intel_perf_config *perf, int fd,
    perf->i915_query_supported = i915_query_perf_config_supported(perf, fd);
    perf->i915_perf_version = i915_perf_version(fd);
 
+   /* TODO: We should query this from i915 */
+   if (intel_device_info_is_dg2(devinfo))
+      perf->oa_timestamp_shift = 1;
+
+   perf->oa_timestamp_mask =
+      0xffffffffffffffffull >> (32 + perf->oa_timestamp_shift);
+
    /* Record the default SSEU configuration. */
    i915_get_sseu(fd, &perf->sseu);
 
@@ -1042,7 +1049,7 @@ uint64_t
 intel_perf_report_timestamp(const struct intel_perf_query_info *query,
                             const uint32_t *report)
 {
-   return report[1];
+   return report[1] >> query->perf->oa_timestamp_shift;
 }
 
 void

src/intel/perf/intel_perf.h

@@ -326,6 +326,18 @@ struct intel_perf_config {
    /* Version of the i915-perf subsystem, refer to i915_drm.h. */
    int i915_perf_version;
 
+   /* Number of bits to shift the OA timestamp values by to match the ring
+    * timestamp.
+    */
+   int oa_timestamp_shift;
+
+   /* Mask of bits valid from the OA report (for instance you might have the
+    * lower 31 bits [30:0] of timestamp value). This is useful if you want to
+    * recombine a full timestamp value captured from the CPU with OA
+    * timestamps captured on the device but that only include 31bits of data.
+    */
+   uint64_t oa_timestamp_mask;
+
    /* Powergating configuration for the running the query. */
    struct drm_i915_gem_context_param_sseu sseu;