Sync upstream changes

.gitlab-ci.yml

@@ -17,7 +17,7 @@ variables:
   #
   # Pick a pipeline on https://gitlab.freedesktop.org/mesa/mesa/-/pipelines/
   #
-  MESA_PIPELINE_ID: 1646089
+  MESA_PIPELINE_ID: 1669832
 
   MESA_PROJECT_PATH: mesa/mesa
   S3_JWT_FILE: /s3_jwt
@@ -59,7 +59,7 @@ include:
     # IMPORTANT: Use a recent Mesa Git revision
     # The commit ref must be in sync with the pipeline picked above
     # It can be found on the pipeline page below the commit message
-    ref: cc4492204894c0488e96a93f51b546345ae32f88
+    ref: 52d57d98a2948214a2d029f2f4e8b476bf39ada3
     file:
       - '/.gitlab-ci/image-tags.yml'
 

.gitlab-ci/expectations/virt/traces-virgl.yml

@@ -109,7 +109,7 @@ traces:
       label: [crash]
   godot/Material Testers.x86_64_2020.04.08_13.38_frame799.rdc:
     gl-virgl:
-      checksum: 7ba4f0c52a719d94b805653d31ce22d7
+      checksum: 465410a06d966b377bc8c41e52b21b0f
   ror/ror-default.trace:
     gl-virgl:
       label: [crash]

.gitlab-ci/expectations/virt/venus-flakes.txt

@@ -35,9 +35,12 @@ dEQP-VK.ray_query.acceleration_structures.operations.traditional_structures.comp
 dEQP-VK.ray_tracing_pipeline.acceleration_structures.flags.sparse_binding_structures.gpu_built.aabbs.identical_instances.padded.fastbuild_update_0_0_bothgeneric
 dEQP-VK.ray_tracing_pipeline.acceleration_structures.flags.traditional_structures.gpu_built.triangles.different_instances.padded.0_update_0_0_bothgeneric
 dEQP-VK.ray_tracing_pipeline.shader_binding_table.indexing_hit.sbt_offset_0.no_shaderrecord.1_5_extraSBTRecordStrideBits
+dEQP-VK.reconvergence.maximal.compute.nesting2.5.18
+dEQP-VK.reconvergence.maximal.compute.nesting2.6.22
 dEQP-VK.renderpasses.dynamic_rendering.primary_cmd_buff.random.seed88_geometry
 dEQP-VK.renderpasses.renderpass2.suballocation.attachment_allocation.roll.56
 dEQP-VK.robustness.robustness2.push.notemplate.r32ui.unroll.volatile.storage_texel_buffer.no_fmt_qual.len_36.samples_1.1d.comp
+dEQP-VK.sparse_resources.buffer.ssbo.sparse_residency.buffer_size_2_24
 dEQP-VK.sparse_resources.image_sparse_residency.mutable.2d_array.r16_unorm_r8g8_sint_r8g8_snorm
 dEQP-VK.spirv_assembly.instruction.graphics.16bit_storage.uniform_float_16_to_32.uniform_buffer_block_vector_float_const_idx_6_tesse
 dEQP-VK.spirv_assembly.instruction.graphics.16bit_storage.uniform_float_16_to_64.uniform_buffer_block_vector_float_geom
@@ -46,6 +49,7 @@ dEQP-VK.spirv_assembly.type.scalar.u8.bit_field_s_extract_offset64_count64_comp
 dEQP-VK.synchronization.cross_instance.suballocated.write_clear_color_image_read_image_compute_indirect.image_64x64x8_r32_sfloat_timeline_semaphore_zircon_handle
 dEQP-VK.synchronization.op.multi_queue.fence.write_draw_read_image_vertex.image_128x128_r16_uint_concurrent_maintenance9
 dEQP-VK.texture.swizzle.component_mapping.color.r16_snorm_2d_pot_bbbb
+dEQP-VK.wsi.direct_drm.swapchain.create.image_array_layers
 
 dEQP-VK.api.info.image_format_properties2.2d.optimal.g16_b16r16_2plane_422_unorm
 dEQP-VK.api.info.image_format_properties2.2d.optimal.g8_b8r8_2plane_420_unorm

config.h.meson

@@ -46,6 +46,7 @@
 #mesondefine ENABLE_DRM_I915
 #mesondefine ENABLE_LIBDRM
 #mesondefine ENABLE_RENDER_SERVER
+#mesondefine ENABLE_SAME_PROCESS_RENDER_SERVER
 #mesondefine ENABLE_RENDER_SERVER_WORKER_PROCESS
 #mesondefine ENABLE_RENDER_SERVER_WORKER_THREAD
 #mesondefine ENABLE_RENDER_SERVER_WORKER_MINIJAIL

meson.build

@@ -346,8 +346,14 @@ if with_check_gl_errors
 endif
 
 with_venus = get_option('venus')
-with_render_server = with_venus
+with_render_server = ''
+if with_venus
+   with_render_server = get_option('render-server-mode')
+endif
 with_render_server_worker = get_option('render-server-worker')
+if with_render_server_worker == 'auto'
+   with_render_server_worker = with_render_server
+endif
 render_server_install_dir = get_option('prefix') / get_option('libexecdir')
 if with_venus and with_host_darwin
    add_languages('objc', required: true)
@@ -361,7 +367,7 @@ if with_venus
    if get_option('vulkan-dload')
       conf_data.set('ENABLE_VULKAN_DLOAD', 1)
    else
-      venus_dep = dependency('vulkan')
+      venus_dep += dependency('vulkan')
    endif
 
    if host_machine.system() in ['freebsd', 'openbsd']
@@ -412,6 +418,13 @@ if with_venus
    else
      error('unknown render server worker ' + with_render_server_worker)
    endif
+
+   if with_render_server == 'thread'
+      conf_data.set('ENABLE_SAME_PROCESS_RENDER_SERVER', 1)
+      if with_render_server_worker != 'thread'
+         error('render-server-worker must be \'thread\' if render-server-mode is \'thread\'')
+      endif
+   endif
 endif
 
 with_video = get_option('video')
@@ -455,7 +468,7 @@ if not with_host_windows
    subdir('vtest')
 endif
 
-if with_render_server
+if with_render_server == 'process'
 subdir('server')
 endif
 
@@ -477,7 +490,8 @@ summary({'c_args': (' ').join(get_option('c_args')),
         'drm-i915': with_drm_i915,
         'drm-panfrost': with_drm_panfrost,
         'drm-msm': with_drm_msm,
-        'render server worker': with_render_server ? with_render_server_worker : 'none',
+        'render server mode': with_render_server == '' ? 'none' : with_render_server,
+        'render server worker': with_render_server == '' ? 'none' : with_render_server_worker,
         'video': with_video,
         'tests': with_tests,
         'fuzzer': with_fuzzer,

meson_options.txt

@@ -86,10 +86,18 @@ option(
 )
 
 option(
-  'render-server-worker',
+  'render-server-mode',
   type : 'combo',
   value : 'process',
-  choices : ['process', 'thread', 'minijail'],
+  choices : ['process', 'thread'],
+  description : 'how render server itself is spawned'
+)
+
+option(
+  'render-server-worker',
+  type : 'combo',
+  value : 'auto',
+  choices : ['auto', 'process', 'thread', 'minijail'],
   description : 'how a context in render server is serviced'
 )
 

server/render_common.c

@@ -4,6 +4,9 @@
  */
 
 #include "render_common.h"
+#ifdef ENABLE_SAME_PROCESS_RENDER_SERVER
+#include "virgl_util.h"
+#endif
 
 #include <stdarg.h>
 #include <stdio.h>
@@ -21,6 +24,10 @@ render_log(const char *fmt, ...)
    va_list va;
 
    va_start(va, fmt);
+#ifdef ENABLE_SAME_PROCESS_RENDER_SERVER
+   virgl_prefixed_logv("server", VIRGL_LOG_LEVEL_INFO, fmt, va);
+#else
    vsyslog(LOG_DEBUG, fmt, va);
+#endif
    va_end(va);
 }

server/render_state.c

@@ -93,13 +93,15 @@ render_state_lookup_context(uint32_t ctx_id)
    return ctx;
 }
 
+#ifndef ENABLE_SAME_PROCESS_RENDER_SERVER
 static void
 render_state_cb_debug_logger(UNUSED enum virgl_log_level_flags log_level,
                              const char *message,
                              UNUSED void* user_data)
 {
    render_log(message);
 }
+#endif
 
 static void
 render_state_cb_retire_fence(uint32_t ctx_id, uint32_t ring_idx, uint64_t fence_id)
@@ -112,7 +114,9 @@ render_state_cb_retire_fence(uint32_t ctx_id, uint32_t ring_idx, uint64_t fence_
 }
 
 static const struct vkr_renderer_callbacks render_state_cbs = {
+#ifndef ENABLE_SAME_PROCESS_RENDER_SERVER
    .debug_logger = render_state_cb_debug_logger,
+#endif
    .retire_fence = render_state_cb_retire_fence,
 };
 

server/render_worker.c

@@ -33,7 +33,7 @@
 #include <sys/types.h>
 #include <sys/wait.h>
 #ifdef ENABLE_RENDER_SERVER_WORKER_THREAD
-#include <threads.h>
+#include "c11/threads.h"
 #endif
 #include <unistd.h>
 

src/drm/amdgpu/amdgpu_renderer.c

@@ -259,7 +259,7 @@ amdgpu_renderer_attach_resource(struct virgl_context *vctx, struct virgl_resourc
          obj->bo = import.buf_handle;
          amdgpu_bo_export(obj->bo, amdgpu_bo_handle_type_kms, &obj->base.handle);
          amdgpu_object_set_res_id(ctx, obj, res->res_id);
-         print(1, "imported dmabuf -> res_id=%u" PRIx64, res->res_id);
+         print(1, "imported dmabuf -> res_id=%u", res->res_id);
       } else {
          print(2, "Ignored res_id: %d (fd_type = %d)", res->res_id, fd_type);
          if (fd_type != VIRGL_RESOURCE_FD_INVALID)
@@ -411,7 +411,7 @@ amdgpu_renderer_get_blob(struct virgl_context *vctx, uint32_t res_id, uint64_t b
 
    /* If GEM_NEW fails, we can end up here without a backing obj or if it's a dumb buffer. */
    if (!obj) {
-      print(0, "No object with blob_id=%ld", blob_id);
+      print(0, "No object with blob_id=%" PRIu64, blob_id);
       return -ENOENT;
    }
 
@@ -424,7 +424,7 @@ amdgpu_renderer_get_blob(struct virgl_context *vctx, uint32_t res_id, uint64_t b
     * to the same storage.
     */
    if (obj->exported) {
-      print(0, "Already exported! blob_id:%ld", blob_id);
+      print(0, "Already exported! blob_id:%" PRIu64, blob_id);
       return -EINVAL;
    }
 
@@ -436,7 +436,7 @@ amdgpu_renderer_get_blob(struct virgl_context *vctx, uint32_t res_id, uint64_t b
       ret = amdgpu_bo_export(obj->bo, amdgpu_bo_handle_type_dma_buf_fd, (uint32_t *)&fd);
 
       if (ret) {
-         print(0, "Export to fd failed for blob_id:%ld r=%d (%s)", blob_id, ret, strerror(errno));
+         print(0, "Export to fd failed for blob_id:%" PRIu64 " r=%d (%s)", blob_id, ret, strerror(errno));
          return ret;
       }
 
@@ -470,7 +470,7 @@ amdgpu_ccmd_query_info(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
    struct amdgpu_ccmd_query_info_rsp *rsp;
    unsigned rsp_len;
    if (__builtin_add_overflow(sizeof(*rsp), req->info.return_size, &rsp_len)) {
-      print(1, "%s: Request size overflow: %zu + %" PRIu32 " > %u",
+      print(1, "%s: Request size overflow: %zu + %u > %u",
             __FUNCTION__, sizeof(*rsp), req->info.return_size, UINT_MAX);
       return -EINVAL;
    }
@@ -510,12 +510,12 @@ amdgpu_ccmd_gem_new(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
 
    if (req->r.__pad) {
       print(0, "Invalid value for struct %s_req::r::__pad: "
-            "0x%" PRIx32, __FUNCTION__, req->r.__pad);
+            "0x%x", __FUNCTION__, req->r.__pad);
       ret = -EINVAL;
       goto alloc_failed;
    }
    if (!drm_context_blob_id_valid(dctx, req->blob_id)) {
-      print(0, "Invalid blob_id %ld", req->blob_id);
+      print(0, "Invalid blob_id %" PRIu64, req->blob_id);
       ret = -EINVAL;
       goto alloc_failed;
    }
@@ -553,7 +553,7 @@ amdgpu_ccmd_gem_new(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
 
    drm_context_object_set_blob_id(dctx, &obj->base, req->blob_id);
 
-   print(2, "new object blob_id: %ld heap: %08x flags: %lx size: %ld",
+   print(2, "new object blob_id: %" PRIu64 " heap: %08x flags: %" PRIx64" size: %" PRIu64,
          req->blob_id, req->r.preferred_heap, req->r.flags, req->r.alloc_size);
 
    return 0;
@@ -562,7 +562,7 @@ amdgpu_ccmd_gem_new(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
    amdgpu_bo_free(bo_handle);
 
 alloc_failed:
-   print(2, "ERROR blob_id: %ld heap: %08x flags: %lx",
+   print(2, "ERROR blob_id: %" PRIu64 " heap: %08x flags: %" PRIx64,
          req->blob_id, req->r.preferred_heap, req->r.flags);
    if (ctx->shmem)
       ctx->shmem->async_error++;
@@ -593,7 +593,10 @@ amdgpu_ccmd_bo_va_op(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
    } else {
       obj = amdgpu_get_object_from_res_id(ctx, req->res_id, __FUNCTION__);
       if (!obj) {
-         print(0, "amdgpu_bo_va_op_raw failed: op: %d res_id: %d offset: 0x%lx size: 0x%lx va: %" PRIx64 " r=%d",
+         print(0,
+            "amdgpu_bo_va_op_raw failed: "
+               "op: %d res_id: %d offset: 0x%" PRIx64 " "
+               "size: 0x%" PRIx64 " va: %" PRIx64 " r=%d",
             req->op, obj->base.res_id, req->offset, req->vm_map_size, req->va, rsp->ret);
 
          /* This is ok. This means the guest closed the GEM already. */
@@ -609,11 +612,16 @@ amdgpu_ccmd_bo_va_op(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
       if (ctx->shmem)
          ctx->shmem->async_error++;
 
-      print(0, "amdgpu_bo_va_op_raw failed: op: %d res_id: %d offset: 0x%lx size: 0x%lx va: %" PRIx64 " r=%d",
+      print(0,
+         "amdgpu_bo_va_op_raw failed: "
+            "op: %d res_id: %d offset: 0x%" PRIx64 " "
+            "size: 0x%" PRIx64 " va: %" PRIx64 " r=%d",
          req->op, req->res_id, req->offset, req->vm_map_size, req->va, rsp->ret);
    } else {
-      print(2, "va_op %d res_id: %u va: [0x%" PRIx64 ", 0x%" PRIx64 "] @offset 0x%" PRIx64,
-            req->op, req->res_id, req->va, req->va + req->vm_map_size - 1, req->offset);
+      print(2,
+         "va_op %d res_id: %u "
+            "va: [0x%" PRIx64 ", 0x%" PRIx64 "] @offset 0x%" PRIx64,
+         req->op, req->res_id, req->va, req->va + req->vm_map_size - 1, req->offset);
    }
 
    return 0;
@@ -646,7 +654,7 @@ amdgpu_ccmd_set_metadata(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
    metadata.size_metadata = req->size_metadata;
    if (req->size_metadata) {
       if (req->size_metadata > sizeof(metadata.umd_metadata)) {
-         print(0, "Metadata size is too large for target buffer: %" PRIu32 " > %zu",
+         print(0, "Metadata size is too large for target buffer: %u > %zu",
                req->size_metadata, sizeof(metadata.umd_metadata));
          rsp->ret = -EINVAL;
          return -1;
@@ -655,7 +663,7 @@ amdgpu_ccmd_set_metadata(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
                                        offsetof(struct amdgpu_ccmd_set_metadata_req,
                                                 umd_metadata));
       if (requested_size > hdr->len) {
-         print(0, "Metadata size is too large for source buffer: %zu > %" PRIu32,
+         print(0, "Metadata size is too large for source buffer: %zu > %u",
                requested_size, hdr->len);
          rsp->ret = -EINVAL;
          return -1;
@@ -736,7 +744,7 @@ amdgpu_ccmd_create_ctx(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
    }
 
    if (req->flags & ~AMDGPU_CCMD_CREATE_CTX_DESTROY) {
-      print(0, "Invalid flags 0x%" PRIu32, req->flags);
+      print(0, "Invalid flags 0x%u", req->flags);
       rsp->hdr.ret = -EINVAL;
       return -1;
    }
@@ -827,7 +835,7 @@ amdgpu_ccmd_cs_submit(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
    }
    /* Do not allocate arbitrarily large buffer. */
    if (req->num_chunks > AMDGPU_CCMD_CS_SUBMIT_MAX_NUM_CHUNKS) {
-      print(1, "%s: Invalid num_chunks: %" PRIu32 " > %d",
+      print(1, "%s: Invalid num_chunks: %u > %d",
             __FUNCTION__, req->num_chunks, AMDGPU_CCMD_CS_SUBMIT_MAX_NUM_CHUNKS);
       rsp->ret = -EINVAL;
       return -1;
@@ -842,7 +850,7 @@ amdgpu_ccmd_cs_submit(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
    chunks = malloc((req->num_chunks + 1 /* syncobj_in */ + 1 /* syncobj_out */) *
                    sizeof(*chunks));
    if (chunks == NULL) {
-      print(0, "Failed to allocate %" PRIu32 " chunks", req->num_chunks + 2);
+      print(0, "Failed to allocate %u chunks", req->num_chunks + 2);
       r = -EINVAL;
       goto end;
    }
@@ -858,7 +866,7 @@ amdgpu_ccmd_cs_submit(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
    size_t descriptors_len = size_add(offsetof(struct amdgpu_ccmd_cs_submit_req, payload),
                                      size_mul(req->num_chunks, sizeof(struct desc)));
    if (descriptors_len > hdr->len) {
-      print(0, "Descriptors are out of bounds: %zu + %zu * %" PRIu32 " > %" PRIu32,
+      print(0, "Descriptors are out of bounds: %zu + %zu * %u > %u",
             offsetof(struct amdgpu_ccmd_cs_submit_req, payload),
             sizeof(struct desc), req->num_chunks, hdr->len);
       r = -EINVAL;
@@ -875,7 +883,7 @@ amdgpu_ccmd_cs_submit(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
       chunks[num_chunks].chunk_id = chunk_id;
       /* Validate input. */
       if (end > hdr->len) {
-         print(0, "Descriptors are out of bounds: %zu > %" PRIu32, end, hdr->len);
+         print(0, "Descriptors are out of bounds: %zu > %u", end, hdr->len);
          r = -EINVAL;
          goto end;
       }
@@ -936,7 +944,7 @@ amdgpu_ccmd_cs_submit(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
          }
          in = input;
          if (in->offset % sizeof(uint64_t)) {
-            print(0, "Invalid chunk offset %" PRIu32 " (not multiple of 8)", in->offset);
+            print(0, "Invalid chunk offset %u (not multiple of 8)", in->offset);
             r = -EINVAL;
             goto end;
          }
@@ -1043,7 +1051,7 @@ amdgpu_ccmd_cs_submit(struct drm_context *dctx, struct vdrm_ccmd_req *hdr)
 
    drmSyncobjDestroy(amdgpu_device_get_fd(ctx->dev), syncobj_out.handle);
 
-   print(3, "ctx: %d -> seqno={v=%d a=%ld} r=%d", req->ctx_id, hdr->seqno, seqno, r);
+   print(3, "ctx: %d -> seqno={v=%d a=%" PRIu64 "} r=%d", req->ctx_id, hdr->seqno, seqno, r);
 
 end:
    if (bo_list)
@@ -1158,7 +1166,7 @@ amdgpu_renderer_submit_fence(struct virgl_context *vctx, uint32_t flags,
 
    /* timeline is ring_idx-1 (because ring_idx 0 is host CPU timeline) */
    if (ring_idx > AMDGPU_HW_IP_NUM) {
-      print(0, "invalid ring_idx: %" PRIu32, ring_idx);
+      print(0, "invalid ring_idx: %u", ring_idx);
       return -EINVAL;
    }
    /* ring_idx zero is used for the guest to synchronize with host CPU,
@@ -1170,7 +1178,7 @@ amdgpu_renderer_submit_fence(struct virgl_context *vctx, uint32_t flags,
       return 0;
    }
 
-   print(3, "ring_idx: %d fence_id: %lu", ring_idx, fence_id);
+   print(3, "ring_idx: %d fence_id: %" PRIu64, ring_idx, fence_id);
    return drm_timeline_submit_fence(&ctx->timelines[ring_idx - 1], flags, fence_id);
 }