diff --git a/src/gallium/drivers/panfrost/pan_blit.c b/src/gallium/drivers/panfrost/pan_blit.c
index 0b621b987ea..5ec8d5950ef 100644
--- a/src/gallium/drivers/panfrost/pan_blit.c
+++ b/src/gallium/drivers/panfrost/pan_blit.c
@@ -28,6 +28,7 @@
  */
 
 #include "util/format/u_format.h"
+#include "util/perf/cpu_trace.h"
 #include "pan_context.h"
 #include "pan_resource.h"
 #include "pan_util.h"
@@ -85,6 +86,8 @@ void
 panfrost_blit_no_afbc_legalization(struct pipe_context *pipe,
                                    const struct pipe_blit_info *info)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_context *ctx = pan_context(pipe);
 
    panfrost_blitter_save(ctx, info->render_condition_enable
@@ -96,6 +99,8 @@ panfrost_blit_no_afbc_legalization(struct pipe_context *pipe,
 void
 panfrost_blit(struct pipe_context *pipe, const struct pipe_blit_info *info)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_context *ctx = pan_context(pipe);
 
    if (info->render_condition_enable && !panfrost_render_condition_check(ctx))
diff --git a/src/gallium/drivers/panfrost/pan_bo.c b/src/gallium/drivers/panfrost/pan_bo.c
index 712987edb8e..c37b3b868ff 100644
--- a/src/gallium/drivers/panfrost/pan_bo.c
+++ b/src/gallium/drivers/panfrost/pan_bo.c
@@ -38,6 +38,7 @@
 
 #include "util/u_inlines.h"
 #include "util/u_math.h"
+#include "util/perf/cpu_trace.h"
 
 /* This file implements a userspace BO cache. Allocating and freeing
  * GPU-visible buffers is very expensive, and even the extra kernel roundtrips
@@ -124,6 +125,8 @@ err_alloc:
 static void
 panfrost_bo_free(struct panfrost_bo *bo)
 {
+   MESA_TRACE_FUNC();
+
    struct pan_kmod_bo *kmod_bo = bo->kmod_bo;
    struct pan_kmod_vm *vm = bo->dev->kmod.vm;
    uint64_t gpu_va = bo->ptr.gpu;
@@ -155,6 +158,8 @@ panfrost_bo_free(struct panfrost_bo *bo)
 bool
 panfrost_bo_wait(struct panfrost_bo *bo, int64_t timeout_ns, bool wait_readers)
 {
+   MESA_TRACE_FUNC();
+
    /* If the BO has been exported or imported we can't rely on the cached
     * state, we need to call the WAIT_BO ioctl.
     */
@@ -336,6 +341,8 @@ panfrost_bo_cache_evict_all(struct panfrost_device *dev)
 int
 panfrost_bo_mmap(struct panfrost_bo *bo)
 {
+   MESA_TRACE_FUNC();
+
    if (bo->ptr.cpu)
       return 0;
 
@@ -352,6 +359,8 @@ panfrost_bo_mmap(struct panfrost_bo *bo)
 static void
 panfrost_bo_munmap(struct panfrost_bo *bo)
 {
+   MESA_TRACE_FUNC();
+
    if (!bo->ptr.cpu)
       return;
 
@@ -367,6 +376,9 @@ struct panfrost_bo *
 panfrost_bo_create(struct panfrost_device *dev, size_t size, uint32_t flags,
                    const char *label)
 {
+   assert(label);
+   MESA_TRACE_SCOPE("%s size=%zu label=\"%s\"", __func__, size, label);
+
    struct panfrost_bo *bo;
 
    if (dev->debug & PAN_DBG_DUMP) {
diff --git a/src/gallium/drivers/panfrost/pan_cmdstream.c b/src/gallium/drivers/panfrost/pan_cmdstream.c
index 274d6299aed..61fdb5ef48e 100644
--- a/src/gallium/drivers/panfrost/pan_cmdstream.c
+++ b/src/gallium/drivers/panfrost/pan_cmdstream.c
@@ -35,6 +35,7 @@
 #include "util/u_sample_positions.h"
 #include "util/u_vbuf.h"
 #include "util/u_viewport.h"
+#include "util/perf/cpu_trace.h"
 
 #include "decode.h"
 
@@ -3291,6 +3292,8 @@ panfrost_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info,
                   const struct pipe_draw_start_count_bias *draws,
                   unsigned num_draws)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_context *ctx = pan_context(pipe);
 
    if (!panfrost_render_condition_check(ctx))
@@ -3431,6 +3434,8 @@ panfrost_afbc_size(struct panfrost_batch *batch, struct panfrost_resource *src,
                    struct panfrost_bo *metadata, unsigned offset,
                    unsigned level)
 {
+   MESA_TRACE_FUNC();
+
    struct pan_image_slice_layout *slice = &src->image.layout.slices[level];
    struct panfrost_afbc_size_info consts = {
       .src =
@@ -3451,6 +3456,8 @@ panfrost_afbc_pack(struct panfrost_batch *batch, struct panfrost_resource *src,
                    struct panfrost_bo *metadata, unsigned metadata_offset,
                    unsigned level)
 {
+   MESA_TRACE_FUNC();
+
    struct pan_image_slice_layout *src_slice = &src->image.layout.slices[level];
    struct panfrost_afbc_pack_info consts = {
       .src = src->image.data.base + src->image.data.offset +
@@ -3472,6 +3479,8 @@ panfrost_afbc_pack(struct panfrost_batch *batch, struct panfrost_resource *src,
 static void
 panfrost_mtk_detile_compute(struct panfrost_context *ctx, struct pipe_blit_info *info)
 {
+   MESA_TRACE_FUNC();
+
    struct pipe_context *pipe = &ctx->base;
    struct pipe_resource *y_src = info->src.resource;
    struct pipe_resource *uv_src = y_src->next;
diff --git a/src/gallium/drivers/panfrost/pan_context.c b/src/gallium/drivers/panfrost/pan_context.c
index 51b112d9d07..778b0256422 100644
--- a/src/gallium/drivers/panfrost/pan_context.c
+++ b/src/gallium/drivers/panfrost/pan_context.c
@@ -47,6 +47,7 @@
 #include "util/u_surface.h"
 #include "util/u_upload_mgr.h"
 #include "util/u_vbuf.h"
+#include "util/perf/cpu_trace.h"
 
 #include "clc/panfrost_compile.h"
 #include "compiler/nir/nir_serialize.h"
@@ -63,6 +64,8 @@ panfrost_clear(struct pipe_context *pipe, unsigned buffers,
                const union pipe_color_union *color, double depth,
                unsigned stencil)
 {
+   MESA_TRACE_FUNC();
+
    if (!panfrost_render_condition_check(pan_context(pipe)))
       return;
 
@@ -106,11 +109,13 @@ void
 panfrost_flush(struct pipe_context *pipe, struct pipe_fence_handle **fence,
                unsigned flags)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_context *ctx = pan_context(pipe);
    struct panfrost_device *dev = pan_device(pipe->screen);
 
    /* Submit all pending jobs */
-   panfrost_flush_all_batches(ctx, NULL);
+   panfrost_flush_all_batches(ctx, "Gallium flush");
 
    if (fence) {
       struct pipe_fence_handle *f = panfrost_fence_create(ctx);
diff --git a/src/gallium/drivers/panfrost/pan_disk_cache.c b/src/gallium/drivers/panfrost/pan_disk_cache.c
index 8cc08350a0e..aae390fc102 100644
--- a/src/gallium/drivers/panfrost/pan_disk_cache.c
+++ b/src/gallium/drivers/panfrost/pan_disk_cache.c
@@ -74,6 +74,8 @@ panfrost_disk_cache_store(struct disk_cache *cache,
    if (!cache)
       return;
 
+   MESA_TRACE_FUNC();
+
    cache_key cache_key;
    panfrost_disk_cache_compute_key(cache, uncompiled, key, cache_key);
 
@@ -116,6 +118,8 @@ panfrost_disk_cache_retrieve(struct disk_cache *cache,
    if (!cache)
       return false;
 
+   MESA_TRACE_FUNC();
+
    cache_key cache_key;
    panfrost_disk_cache_compute_key(cache, uncompiled, key, cache_key);
 
diff --git a/src/gallium/drivers/panfrost/pan_job.c b/src/gallium/drivers/panfrost/pan_job.c
index c1d2909ec8a..e695edbd5d7 100644
--- a/src/gallium/drivers/panfrost/pan_job.c
+++ b/src/gallium/drivers/panfrost/pan_job.c
@@ -32,6 +32,7 @@
 #include "util/rounding.h"
 #include "util/u_framebuffer.h"
 #include "util/u_pack_color.h"
+#include "util/perf/cpu_trace.h"
 #include "pan_bo.h"
 #include "pan_context.h"
 #include "pan_util.h"
@@ -656,6 +657,8 @@ static void
 panfrost_batch_submit(struct panfrost_context *ctx,
                       struct panfrost_batch *batch)
 {
+   MESA_TRACE_FUNC();
+
    struct pipe_screen *pscreen = ctx->base.screen;
    struct panfrost_screen *screen = pan_screen(pscreen);
    bool has_frag = panfrost_has_fragment_job(batch);
@@ -724,8 +727,9 @@ out:
 void
 panfrost_flush_all_batches(struct panfrost_context *ctx, const char *reason)
 {
-   if (reason)
-      perf_debug(ctx, "Flushing everything due to: %s", reason);
+   assert(reason);
+   MESA_TRACE_SCOPE("%s reason=\"%s\"", __func__, reason);
+   perf_debug(ctx, "Flushing everything due to: %s", reason);
 
    struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
    if (!batch)
@@ -743,6 +747,9 @@ void
 panfrost_flush_writer(struct panfrost_context *ctx,
                       struct panfrost_resource *rsrc, const char *reason)
 {
+   assert(reason);
+   MESA_TRACE_SCOPE("%s reason=\"%s\"", __func__, reason);
+
    struct hash_entry *entry = _mesa_hash_table_search(ctx->writers, rsrc);
 
    if (entry) {
@@ -756,6 +763,9 @@ panfrost_flush_batches_accessing_rsrc(struct panfrost_context *ctx,
                                       struct panfrost_resource *rsrc,
                                       const char *reason)
 {
+   assert(reason);
+   MESA_TRACE_SCOPE("%s reason=\"%s\"", __func__, reason);
+
    unsigned i;
    foreach_batch(ctx, i) {
       struct panfrost_batch *batch = &ctx->batches.slots[i];
diff --git a/src/gallium/drivers/panfrost/pan_resource.c b/src/gallium/drivers/panfrost/pan_resource.c
index 249442c0504..e07addca2a1 100644
--- a/src/gallium/drivers/panfrost/pan_resource.c
+++ b/src/gallium/drivers/panfrost/pan_resource.c
@@ -45,6 +45,7 @@
 #include "util/u_surface.h"
 #include "util/u_transfer.h"
 #include "util/u_transfer_helper.h"
+#include "util/perf/cpu_trace.h"
 
 #include "decode.h"
 #include "pan_bo.h"
@@ -722,6 +723,8 @@ panfrost_resource_create_with_modifier(struct pipe_screen *screen,
                                        const struct pipe_resource *template,
                                        uint64_t modifier)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_device *dev = pan_device(screen);
 
    struct panfrost_resource *so = CALLOC_STRUCT(panfrost_resource);
@@ -900,6 +903,8 @@ panfrost_resource_create_with_modifiers(struct pipe_screen *screen,
 static void
 panfrost_resource_destroy(struct pipe_screen *screen, struct pipe_resource *pt)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_device *dev = pan_device(screen);
    struct panfrost_resource *rsrc = (struct panfrost_resource *)pt;
 
@@ -1212,6 +1217,8 @@ panfrost_ptr_map(struct pipe_context *pctx, struct pipe_resource *resource,
                  const struct pipe_box *box,
                  struct pipe_transfer **out_transfer)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_context *ctx = pan_context(pctx);
    struct panfrost_device *dev = pan_device(pctx->screen);
    struct panfrost_resource *rsrc = pan_resource(resource);
@@ -1448,6 +1455,8 @@ pan_resource_modifier_convert(struct panfrost_context *ctx,
                               struct panfrost_resource *rsrc, uint64_t modifier,
                               bool copy_resource, const char *reason)
 {
+   MESA_TRACE_FUNC();
+
    bool need_shadow = rsrc->modifier_constant;
 
    assert(!rsrc->modifier_constant || copy_resource);
@@ -1683,6 +1692,8 @@ void
 panfrost_pack_afbc(struct panfrost_context *ctx,
                    struct panfrost_resource *prsrc)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_screen *screen = pan_screen(ctx->base.screen);
    struct panfrost_device *dev = pan_device(ctx->base.screen);
    struct panfrost_bo *metadata_bo;
@@ -1811,6 +1822,8 @@ panfrost_pack_afbc(struct panfrost_context *ctx,
 static void
 panfrost_ptr_unmap(struct pipe_context *pctx, struct pipe_transfer *transfer)
 {
+   MESA_TRACE_FUNC();
+
    /* Gallium expects writeback here, so we tile */
 
    struct panfrost_context *ctx = pan_context(pctx);
diff --git a/src/gallium/drivers/panfrost/pan_screen.c b/src/gallium/drivers/panfrost/pan_screen.c
index b7958a25cdd..45582cf23c9 100644
--- a/src/gallium/drivers/panfrost/pan_screen.c
+++ b/src/gallium/drivers/panfrost/pan_screen.c
@@ -38,6 +38,7 @@
 #include "util/u_screen.h"
 #include "util/u_video.h"
 #include "util/xmlconfig.h"
+#include "util/perf/cpu_trace.h"
 
 #include <fcntl.h>
 
@@ -819,6 +820,8 @@ panfrost_create_screen(int fd, const struct pipe_screen_config *config,
    screen->max_afbc_packing_ratio = debug_get_num_option(
       "PAN_MAX_AFBC_PACKING_RATIO", DEFAULT_MAX_AFBC_PACKING_RATIO);
 
+   util_cpu_trace_init();
+
    if (panfrost_open_device(screen, fd, dev)) {
       ralloc_free(screen);
       return NULL;
diff --git a/src/gallium/drivers/panfrost/pan_shader.c b/src/gallium/drivers/panfrost/pan_shader.c
index 3c805e42c4b..5654969785e 100644
--- a/src/gallium/drivers/panfrost/pan_shader.c
+++ b/src/gallium/drivers/panfrost/pan_shader.c
@@ -32,6 +32,7 @@
 #include "nir/tgsi_to_nir.h"
 #include "util/u_memory.h"
 #include "util/u_prim.h"
+#include "util/perf/cpu_trace.h"
 #include "nir_builder.h"
 #include "nir_serialize.h"
 #include "pan_bo.h"
@@ -115,6 +116,8 @@ panfrost_shader_compile(struct panfrost_screen *screen, const nir_shader *ir,
                         unsigned fixed_varying_mask,
                         struct panfrost_shader_binary *out)
 {
+   MESA_TRACE_FUNC();
+
    struct panfrost_device *dev = pan_device(&screen->base);
 
    nir_shader *s = nir_shader_clone(NULL, ir);
@@ -452,6 +455,8 @@ static void *
 panfrost_create_shader_state(struct pipe_context *pctx,
                              const struct pipe_shader_state *cso)
 {
+   MESA_TRACE_FUNC();
+
    nir_shader *nir = (cso->type == PIPE_SHADER_IR_TGSI)
                         ? tgsi_to_nir(cso->tokens, pctx->screen, false)
                         : cso->ir.nir;
diff --git a/src/panfrost/compiler/bifrost_compile.c b/src/panfrost/compiler/bifrost_compile.c
index c1619db42b2..ddff65baae5 100644
--- a/src/panfrost/compiler/bifrost_compile.c
+++ b/src/panfrost/compiler/bifrost_compile.c
@@ -29,6 +29,7 @@
 #include "compiler/glsl_types.h"
 #include "compiler/nir/nir_builder.h"
 #include "panfrost/util/pan_ir.h"
+#include "util/perf/cpu_trace.h"
 #include "util/u_debug.h"
 
 #include "bifrost/disassemble.h"
@@ -5684,6 +5685,8 @@ bi_lower_ldexp16(nir_builder *b, nir_alu_instr *alu, UNUSED void *data)
 void
 bifrost_preprocess_nir(nir_shader *nir, unsigned gpu_id)
 {
+   MESA_TRACE_FUNC();
+
    /* Ensure that halt are translated to returns and get ride of them */
    NIR_PASS(_, nir, nir_shader_instructions_pass, bi_lower_halt_to_return,
             nir_metadata_all, NULL);
@@ -6254,6 +6257,8 @@ bifrost_compile_shader_nir(nir_shader *nir,
                            struct util_dynarray *binary,
                            struct pan_shader_info *info)
 {
+   MESA_TRACE_FUNC();
+
    bifrost_debug = debug_get_option_bifrost_debug();
 
    /* Combine stores late, to give the driver a chance to lower dual-source