panfrost: add tiler-heap driconfs

We currently allocate 10 MB up-front for tiler heaps. That can sometimes
be a lot, depending on the system. So let's add some DRIconf variables to
allow tweaking these.

Until we have incremental-rendering implemented, it's a bit risky to lower
these too much. Be warned.

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28594>

src/gallium/drivers/panfrost/driinfo_panfrost.h

@@ -1,4 +1,12 @@
 /* panfrost specific driconf options */
 DRI_CONF_SECTION_PERFORMANCE
    DRI_CONF_OPT_B(pan_force_afbc_packing, false, "Use AFBC-P for textures")
+
+   /* 2M chunks. */
+   DRI_CONF_OPT_I(pan_csf_chunk_size, 2 * 1024 * 1024, 256 * 1024, 8 * 1024 * 1024, "CSF Tiler Chunk Size")
+   DRI_CONF_OPT_I(pan_csf_initial_chunks, 5, 1, 65535, "CSF Tiler Initial Chunks")
+   /* 64 x 2M = 128M, which matches the tiler_heap BO allocated in
+    * panfrost_open_device() for pre-v10 HW.
+    */
+   DRI_CONF_OPT_I(pan_csf_max_chunks, 64, 1, 65535, "CSF Tiler Max Chunks")
 DRI_CONF_SECTION_END

src/gallium/drivers/panfrost/pan_csf.c

@@ -974,14 +974,9 @@ GENX(csf_init_context)(struct panfrost_context *ctx)
    /* Get tiler heap */
    struct drm_panthor_tiler_heap_create thc = {
       .vm_id = pan_kmod_vm_handle(dev->kmod.vm),
-      /* 2M chunks. */
-      .chunk_size = 2 * 1024 * 1024,
-      .initial_chunk_count = 5,
-
-      /* 64 x 2M = 128M, which matches the tiler_heap BO allocated in
-       * panfrost_open_device() for pre-v10 HW.
-       */
-      .max_chunks = 64,
+      .chunk_size = pan_screen(ctx->base.screen)->csf_tiler_heap.chunk_size,
+      .initial_chunk_count = pan_screen(ctx->base.screen)->csf_tiler_heap.initial_chunks,
+      .max_chunks = pan_screen(ctx->base.screen)->csf_tiler_heap.max_chunks,
       .target_in_flight = 65535,
    };
    ret = drmIoctl(panfrost_device_fd(dev), DRM_IOCTL_PANTHOR_TILER_HEAP_CREATE,
@@ -994,7 +989,7 @@ GENX(csf_init_context)(struct panfrost_context *ctx)
    ctx->csf.heap.desc_bo =
       panfrost_bo_create(dev, pan_size(TILER_HEAP), 0, "Tiler Heap");
    pan_pack(ctx->csf.heap.desc_bo->ptr.cpu, TILER_HEAP, heap) {
-      heap.size = 2 * 1024 * 1024;
+      heap.size = pan_screen(ctx->base.screen)->csf_tiler_heap.chunk_size;
       heap.base = thc.first_heap_chunk_gpu_va;
       heap.bottom = heap.base + 64;
       heap.top = heap.base + heap.size;

src/gallium/drivers/panfrost/pan_screen.c

@@ -861,6 +861,13 @@ panfrost_create_screen(int fd, const struct pipe_screen_config *config,
       screen->force_afbc_packing = driQueryOptionb(config->options,
                                                    "pan_force_afbc_packing");
 
+   screen->csf_tiler_heap.chunk_size = driQueryOptioni(config->options,
+                                                       "pan_csf_chunk_size");
+   screen->csf_tiler_heap.initial_chunks = driQueryOptioni(config->options,
+                                                           "pan_csf_initial_chunks");
+   screen->csf_tiler_heap.max_chunks = driQueryOptioni(config->options,
+                                                       "pan_csf_max_chunks");
+
    dev->ro = ro;
 
    screen->base.destroy = panfrost_destroy_screen;

src/gallium/drivers/panfrost/pan_screen.h

@@ -118,6 +118,12 @@ struct panfrost_screen {
    struct disk_cache *disk_cache;
    unsigned max_afbc_packing_ratio;
    bool force_afbc_packing;
+
+   struct {
+      unsigned chunk_size;
+      unsigned initial_chunks;
+      unsigned max_chunks;
+   } csf_tiler_heap;
 };
 
 static inline struct panfrost_screen *