intel/perf: repurpose INTEL_DEBUG=no-oaconfig

We initially used this debug option to mean "don't bother registering
the OA configuration into the kernel".

This change makes this option suppress any interaction with the
i915/perf interface. This is useful when debugging self modifying
batches with performance queries while running on the intel_mi_runner.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2775>

diff --git a/src/intel/perf/gen_perf.c b/src/intel/perf/gen_perf.c
index e9ad6dd570f25dd2fcaa214d1d84fb57c7a28184..6fd66a1d2dd3dc1bbe2777424c2977d1ad1428e8 100644 (file)
--- a/src/intel/perf/gen_perf.c
+++ b/src/intel/perf/gen_perf.c
@@ -80,6 +80,9 @@ get_sysfs_dev_dir(struct gen_perf_config *perf, int fd)
 
    perf->sysfs_dev_dir[0] = '\0';
 
+   if (unlikely(INTEL_DEBUG & DEBUG_NO_OACONFIG))
+      return true;
+
    if (fstat(fd, &sb)) {
       DBG("Failed to stat DRM fd\n");
       return false;
@@ -232,6 +235,16 @@ enumerate_sysfs_metrics(struct gen_perf_config *perf,
    closedir(metricsdir);
 }
 
+static void
+add_all_metrics(struct gen_perf_config *perf,
+                const struct gen_device_info *devinfo)
+{
+   hash_table_foreach(perf->oa_metrics_table, entry) {
+      const struct gen_perf_query_info *query = entry->data;
+      register_oa_config(perf, devinfo, query, 0);
+   }
+}
+
 static bool
 kernel_has_dynamic_config_support(struct gen_perf_config *perf, int fd)
 {
@@ -394,11 +407,16 @@ init_oa_sys_vars(struct gen_perf_config *perf, const struct gen_device_info *dev
 {
    uint64_t min_freq_mhz = 0, max_freq_mhz = 0;
 
-   if (!read_sysfs_drm_device_file_uint64(perf, "gt_min_freq_mhz", &min_freq_mhz))
-      return false;
+   if (likely(!(INTEL_DEBUG & DEBUG_NO_OACONFIG))) {
+      if (!read_sysfs_drm_device_file_uint64(perf, "gt_min_freq_mhz", &min_freq_mhz))
+         return false;
 
-   if (!read_sysfs_drm_device_file_uint64(perf,  "gt_max_freq_mhz", &max_freq_mhz))
-      return false;
+      if (!read_sysfs_drm_device_file_uint64(perf,  "gt_max_freq_mhz", &max_freq_mhz))
+         return false;
+   } else {
+      min_freq_mhz = 300;
+      max_freq_mhz = 1000;
+   }
 
    memset(&perf->sys_vars, 0, sizeof(perf->sys_vars));
    perf->sys_vars.gt_min_freq = min_freq_mhz * 1000000;
@@ -679,11 +697,14 @@ load_oa_metrics(struct gen_perf_config *perf, int fd,
     */
    oa_register(perf);
 
-   if (likely((INTEL_DEBUG & DEBUG_NO_OACONFIG) == 0) &&
-       kernel_has_dynamic_config_support(perf, fd))
-      init_oa_configs(perf, fd, devinfo);
-   else
-      enumerate_sysfs_metrics(perf, devinfo);
+   if (likely(!(INTEL_DEBUG & DEBUG_NO_OACONFIG))) {
+      if (kernel_has_dynamic_config_support(perf, fd))
+         init_oa_configs(perf, fd, devinfo);
+      else
+         enumerate_sysfs_metrics(perf, devinfo);
+   } else {
+      add_all_metrics(perf, devinfo);
+   }
 
    build_unique_counter_list(perf);
 

diff --git a/src/intel/vulkan/anv_batch_chain.c b/src/intel/vulkan/anv_batch_chain.c
index f820a69ceec5851ecd0f0f230392946d5a3c4587..acbc07816b18e7c9ef1e056883c2e507d8d7a80d 100644 (file)
--- a/src/intel/vulkan/anv_batch_chain.c
+++ b/src/intel/vulkan/anv_batch_chain.c
@@ -1774,8 +1774,9 @@ anv_queue_execbuf_locked(struct anv_queue *queue,
       /* Some performance queries just the pipeline statistic HW, no need for
        * OA in that case, so no need to reconfigure.
        */
-      if (query_info->kind == GEN_PERF_QUERY_TYPE_OA ||
-          query_info->kind == GEN_PERF_QUERY_TYPE_RAW) {
+      if (likely((INTEL_DEBUG & DEBUG_NO_OACONFIG) == 0) &&
+          (query_info->kind == GEN_PERF_QUERY_TYPE_OA ||
+           query_info->kind == GEN_PERF_QUERY_TYPE_RAW)) {
          int ret = gen_ioctl(device->perf_fd, I915_PERF_IOCTL_CONFIG,
                              (void *)(uintptr_t) query_info->oa_metrics_set_id);
          if (ret < 0) {

diff --git a/src/intel/vulkan/anv_perf.c b/src/intel/vulkan/anv_perf.c
index e8575b1bd702378c15b95b786f0f39043174bc08..ff47317fa68bd39431b6fce714b7eb8bcff8a95d 100644 (file)
--- a/src/intel/vulkan/anv_perf.c
+++ b/src/intel/vulkan/anv_perf.c
@@ -180,18 +180,21 @@ VkResult anv_AcquirePerformanceConfigurationINTEL(
     VkPerformanceConfigurationINTEL*            pConfiguration)
 {
    ANV_FROM_HANDLE(anv_device, device, _device);
-
-   struct gen_perf_registers *perf_config =
-      gen_perf_load_configuration(device->physical->perf, device->fd,
-                                  GEN_PERF_QUERY_GUID_MDAPI);
-   if (!perf_config)
-      return VK_INCOMPLETE;
-
-   int ret = gen_perf_store_configuration(device->physical->perf, device->fd,
-                                          perf_config, NULL /* guid */);
-   if (ret < 0) {
-      ralloc_free(perf_config);
-      return VK_INCOMPLETE;
+   int ret = -1;
+
+   if (likely(!(INTEL_DEBUG & DEBUG_NO_OACONFIG))) {
+      struct gen_perf_registers *perf_config =
+         gen_perf_load_configuration(device->physical->perf, device->fd,
+                                     GEN_PERF_QUERY_GUID_MDAPI);
+      if (!perf_config)
+         return VK_INCOMPLETE;
+
+      ret = gen_perf_store_configuration(device->physical->perf, device->fd,
+                                         perf_config, NULL /* guid */);
+      if (ret < 0) {
+         ralloc_free(perf_config);
+         return VK_INCOMPLETE;
+      }
    }
 
    *pConfiguration = (VkPerformanceConfigurationINTEL) (uint64_t) ret;
@@ -206,7 +209,8 @@ VkResult anv_ReleasePerformanceConfigurationINTEL(
    ANV_FROM_HANDLE(anv_device, device, _device);
    uint64_t config = (uint64_t) _configuration;
 
-   gen_ioctl(device->fd, DRM_IOCTL_I915_PERF_REMOVE_CONFIG, &config);
+   if (likely(!(INTEL_DEBUG & DEBUG_NO_OACONFIG)))
+      gen_ioctl(device->fd, DRM_IOCTL_I915_PERF_REMOVE_CONFIG, &config);
 
    return VK_SUCCESS;
 }
@@ -219,15 +223,17 @@ VkResult anv_QueueSetPerformanceConfigurationINTEL(
    struct anv_device *device = queue->device;
    uint64_t configuration = (uint64_t) _configuration;
 
-   if (device->perf_fd < 0) {
-      device->perf_fd = anv_device_perf_open(device, configuration);
-      if (device->perf_fd < 0)
-         return VK_ERROR_INITIALIZATION_FAILED;
-   } else {
-      int ret = gen_ioctl(device->perf_fd, I915_PERF_IOCTL_CONFIG,
+   if (likely(!(INTEL_DEBUG & DEBUG_NO_OACONFIG))) {
+      if (device->perf_fd < 0) {
+         device->perf_fd = anv_device_perf_open(device, configuration);
+         if (device->perf_fd < 0)
+            return VK_ERROR_INITIALIZATION_FAILED;
+      } else {
+         int ret = gen_ioctl(device->perf_fd, I915_PERF_IOCTL_CONFIG,
                           (void *)(uintptr_t) _configuration);
-      if (ret < 0)
-         return anv_device_set_lost(device, "i915-perf config failed: %m");
+         if (ret < 0)
+            return anv_device_set_lost(device, "i915-perf config failed: %m");
+      }
    }
 
    return VK_SUCCESS;
@@ -342,12 +348,15 @@ VkResult anv_AcquireProfilingLockKHR(
    ANV_FROM_HANDLE(anv_device, device, _device);
    struct gen_perf_config *perf = device->physical->perf;
    struct gen_perf_query_info *first_metric_set = &perf->queries[0];
+   int fd = -1;
 
    assert(device->perf_fd == -1);
 
-   int fd = anv_device_perf_open(device, first_metric_set->oa_metrics_set_id);
-   if (fd < 0)
-      return VK_TIMEOUT;
+   if (likely(!(INTEL_DEBUG & DEBUG_NO_OACONFIG))) {
+      fd = anv_device_perf_open(device, first_metric_set->oa_metrics_set_id);
+      if (fd < 0)
+         return VK_TIMEOUT;
+   }
 
    device->perf_fd = fd;
    return VK_SUCCESS;
@@ -358,8 +367,10 @@ void anv_ReleaseProfilingLockKHR(
 {
    ANV_FROM_HANDLE(anv_device, device, _device);
 
-   assert(device->perf_fd >= 0);
-   close(device->perf_fd);
+   if (likely(!(INTEL_DEBUG & DEBUG_NO_OACONFIG))) {
+      assert(device->perf_fd >= 0);
+      close(device->perf_fd);
+   }
    device->perf_fd = -1;
 }