asahi: Update to UAPI 10011

Incompatible changes:
- Make VM layout more flexible to allow for SVM with rusticl
  (eventually, hopefully)

Compatible changes:
- Expose soft fault state to userspace as a flag

Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30633>

src/asahi/drm-shim/asahi_noop.c

@@ -18,8 +18,8 @@ static const struct drm_asahi_params_global params = {
    .gpu_revision = 0,
    .vm_user_start = 0x1000000,
    .vm_user_end = 0x5000000,
-   .vm_shader_start = 0x8000000,
-   .vm_shader_end = 0x9000000,
+   .vm_usc_start = 0,
+   .vm_usc_end = 0,
    .vm_page_size = 4096,
 };
 

src/asahi/lib/agx_device.c

@@ -551,7 +551,19 @@ agx_open_device(void *memctx, struct agx_device *dev)
    }
 
    dev->guard_size = dev->params.vm_page_size;
-   dev->shader_base = dev->params.vm_shader_start;
+   if (dev->params.vm_usc_start) {
+      dev->shader_base = dev->params.vm_usc_start;
+   } else {
+      // Put the USC heap at the bottom of the user address space, 4GiB aligned
+      dev->shader_base = ALIGN_POT(dev->params.vm_user_start, 0x100000000ull);
+   }
+
+   uint64_t shader_size = 0x100000000ull;
+   // Put the user heap after the USC heap
+   uint64_t user_start = dev->shader_base + shader_size;
+
+   assert(dev->shader_base >= dev->params.vm_user_start);
+   assert(user_start < dev->params.vm_user_end);
 
    util_sparse_array_init(&dev->bo_map, sizeof(struct agx_bo), 512);
    pthread_mutex_init(&dev->bo_map_lock, NULL);
@@ -562,7 +574,16 @@ agx_open_device(void *memctx, struct agx_device *dev)
    for (unsigned i = 0; i < ARRAY_SIZE(dev->bo_cache.buckets); ++i)
       list_inithead(&dev->bo_cache.buckets[i]);
 
-   struct drm_asahi_vm_create vm_create = {};
+   // Put the kernel heap at the top of the address space.
+   // Give it 32GB of address space, should be more than enough for any
+   // reasonable use case.
+   uint64_t kernel_size = MAX2(dev->params.vm_kernel_min_size, 32ull << 30);
+   struct drm_asahi_vm_create vm_create = {
+      .kernel_start = dev->params.vm_user_end - kernel_size,
+      .kernel_end = dev->params.vm_user_end,
+   };
+
+   uint64_t user_size = vm_create.kernel_start - user_start;
 
    int ret = asahi_simple_ioctl(dev, DRM_IOCTL_ASAHI_VM_CREATE, &vm_create);
    if (ret) {
@@ -572,11 +593,8 @@ agx_open_device(void *memctx, struct agx_device *dev)
    }
 
    simple_mtx_init(&dev->vma_lock, mtx_plain);
-   util_vma_heap_init(&dev->main_heap, dev->params.vm_user_start,
-                      dev->params.vm_user_end - dev->params.vm_user_start + 1);
-   util_vma_heap_init(
-      &dev->usc_heap, dev->params.vm_shader_start,
-      dev->params.vm_shader_end - dev->params.vm_shader_start + 1);
+   util_vma_heap_init(&dev->main_heap, user_start, user_size);
+   util_vma_heap_init(&dev->usc_heap, dev->shader_base, shader_size);
 
    dev->vm_id = vm_create.vm_id;
 

src/asahi/vulkan/hk_queue.c

@@ -79,6 +79,8 @@ asahi_fill_cdm_command(struct hk_device *dev, struct hk_cs *cs,
       .sampler_count = dev->samplers.table.alloc,
       .sampler_max = dev->samplers.table.alloc + 1,
 
+      .usc_base = dev->dev.shader_base,
+
       .encoder_id = agx_get_global_id(&dev->dev),
       .cmd_id = agx_get_global_id(&dev->dev),
       .unk_mask = 0xffffffff,
@@ -119,6 +121,9 @@ asahi_fill_vdm_command(struct hk_device *dev, struct hk_cs *cs,
    c->cmd_ta_id = cmd_ta_id;
    c->ppp_ctrl = 0x202;
 
+   c->fragment_usc_base = dev->dev.shader_base;
+   c->vertex_usc_base = c->fragment_usc_base;
+
    c->fb_width = cs->cr.width;
    c->fb_height = cs->cr.height;
 

src/gallium/drivers/asahi/agx_pipe.c

@@ -1276,6 +1276,9 @@ agx_cmdbuf(struct agx_device *dev, struct drm_asahi_cmd_render *c,
    c->cmd_3d_id = cmd_3d_id;
    c->cmd_ta_id = cmd_ta_id;
 
+   c->fragment_usc_base = dev->shader_base;
+   c->vertex_usc_base = dev->shader_base;
+
    /* bit 0 specifies OpenGL clip behaviour. Since ARB_clip_control is
     * advertised, we don't set it and lower in the vertex shader.
     */
@@ -1620,6 +1623,7 @@ agx_flush_compute(struct agx_context *ctx, struct agx_batch *batch,
       .encoder_ptr = batch->cdm.bo->ptr.gpu,
       .encoder_end = batch->cdm.bo->ptr.gpu +
                      (batch->cdm.current - (uint8_t *)batch->cdm.bo->ptr.cpu),
+      .usc_base = dev->shader_base,
       .helper_arg = 0,
       .helper_cfg = 0,
       .helper_program = 0,