winsys/radeon: implement and enable 32-bit VM allocations

Reviewed-by: Samuel Pitoiset <[email protected]>

3 files changed, 64 insertions, 8 deletions

diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
index bbfe5cc7884..06842a4fbc8 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
@@ -249,6 +249,12 @@ static uint64_t radeon_bomgr_find_va(const struct radeon_info *info,
     offset = heap->start;
     waste = offset % alignment;
     waste = waste ? alignment - waste : 0;
+
+    if (offset + waste + size > heap->end) {
+        mtx_unlock(&heap->mutex);
+        return 0;
+    }
+
     if (waste) {
         n = CALLOC_STRUCT(radeon_bo_va_hole);
         n->size = waste;
@@ -261,6 +267,22 @@ static uint64_t radeon_bomgr_find_va(const struct radeon_info *info,
     return offset;
 }
 
+static uint64_t radeon_bomgr_find_va64(struct radeon_drm_winsys *ws,
+                                       uint64_t size, uint64_t alignment)
+{
+    uint64_t va = 0;
+
+    /* Try to allocate from the 64-bit address space first.
+     * If it doesn't exist (start = 0) or if it doesn't have enough space,
+     * fall back to the 32-bit address space.
+     */
+    if (ws->vm64.start)
+        va = radeon_bomgr_find_va(&ws->info, &ws->vm64, size, alignment);
+    if (!va)
+        va = radeon_bomgr_find_va(&ws->info, &ws->vm32, size, alignment);
+    return va;
+}
+
 static void radeon_bomgr_free_va(const struct radeon_info *info,
                                  struct radeon_vm_heap *heap,
                                  uint64_t va, uint64_t size)
@@ -370,7 +392,9 @@ void radeon_bo_destroy(struct pb_buffer *_buf)
             }
 	}
 
-	radeon_bomgr_free_va(&rws->info, &rws->vm64, bo->va, bo->base.size);
+	radeon_bomgr_free_va(&rws->info,
+                             bo->va < rws->vm32.end ? &rws->vm32 : &rws->vm64,
+                             bo->va, bo->base.size);
     }
 
     /* Close object. */
@@ -660,8 +684,14 @@ static struct radeon_bo *radeon_create_bo(struct radeon_drm_winsys *rws,
         unsigned va_gap_size;
 
         va_gap_size = rws->check_vm ? MAX2(4 * alignment, 64 * 1024) : 0;
-        bo->va = radeon_bomgr_find_va(&rws->info, &rws->vm64,
-                                      size + va_gap_size, alignment);
+
+        if (flags & RADEON_FLAG_32BIT) {
+            bo->va = radeon_bomgr_find_va(&rws->info, &rws->vm32,
+                                          size + va_gap_size, alignment);
+            assert(bo->va + size < rws->vm32.end);
+        } else {
+            bo->va = radeon_bomgr_find_va64(rws, size + va_gap_size, alignment);
+        }
 
         va.handle = bo->handle;
         va.vm_id = 0;
@@ -1062,8 +1092,7 @@ static struct pb_buffer *radeon_winsys_bo_from_ptr(struct radeon_winsys *rws,
     if (ws->info.has_virtual_memory) {
         struct drm_radeon_gem_va va;
 
-        bo->va = radeon_bomgr_find_va(&ws->info, &ws->vm64,
-                                      bo->base.size, 1 << 20);
+        bo->va = radeon_bomgr_find_va64(ws, bo->base.size, 1 << 20);
 
         va.handle = bo->handle;
         va.operation = RADEON_VA_MAP;
@@ -1206,8 +1235,7 @@ done:
     if (ws->info.has_virtual_memory && !bo->va) {
         struct drm_radeon_gem_va va;
 
-        bo->va = radeon_bomgr_find_va(&ws->info, &ws->vm64,
-                                      bo->base.size, 1 << 20);
+        bo->va = radeon_bomgr_find_va64(ws, bo->base.size, 1 << 20);
 
         va.handle = bo->handle;
         va.operation = RADEON_VA_MAP;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
index 4fe36dc3f25..85a186af978 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
@@ -363,6 +363,8 @@ static bool do_winsys_init(struct radeon_drm_winsys *ws)
         ws->info.max_alloc_size = MIN2(ws->info.vram_size * 0.7, ws->info.max_alloc_size);
     if (ws->info.drm_minor < 40)
         ws->info.max_alloc_size = MIN2(ws->info.max_alloc_size, 256*1024*1024);
+    /* Both 32-bit and 64-bit address spaces only have 4GB. */
+    ws->info.max_alloc_size = MIN2(ws->info.max_alloc_size, 3ull*1024*1024*1024);
 
     /* Get max clock frequency info and convert it to MHz */
     radeon_get_drm_value(ws->fd, RADEON_INFO_MAX_SCLK, NULL,
@@ -553,6 +555,7 @@ static void radeon_winsys_destroy(struct radeon_winsys *rws)
     util_hash_table_destroy(ws->bo_handles);
     util_hash_table_destroy(ws->bo_vas);
     mtx_destroy(&ws->bo_handles_mutex);
+    mtx_destroy(&ws->vm32.mutex);
     mtx_destroy(&ws->vm64.mutex);
     mtx_destroy(&ws->bo_fence_lock);
 
@@ -816,11 +819,34 @@ radeon_drm_winsys_create(int fd, const struct pipe_screen_config *config,
     ws->bo_handles = util_hash_table_create(handle_hash, handle_compare);
     ws->bo_vas = util_hash_table_create(handle_hash, handle_compare);
     (void) mtx_init(&ws->bo_handles_mutex, mtx_plain);
+    (void) mtx_init(&ws->vm32.mutex, mtx_plain);
     (void) mtx_init(&ws->vm64.mutex, mtx_plain);
     (void) mtx_init(&ws->bo_fence_lock, mtx_plain);
+    list_inithead(&ws->vm32.holes);
     list_inithead(&ws->vm64.holes);
 
-    ws->vm64.start = ws->va_start;
+    /* The kernel currently returns 8MB. Make sure this doesn't change. */
+    if (ws->va_start > 8 * 1024 * 1024) {
+        /* Not enough 32-bit address space. */
+        radeon_winsys_destroy(&ws->base);
+        mtx_unlock(&fd_tab_mutex);
+        return NULL;
+    }
+
+    ws->vm32.start = ws->va_start;
+    ws->vm32.end = 1ull << 32;
+
+    /* The maximum is 8GB of virtual address space limited by the kernel.
+     * It's obviously not enough for bigger cards, like Hawaiis with 4GB
+     * and 8GB of physical memory and 4GB of GART.
+     *
+     * Older kernels set the limit to 4GB, which is even worse, so they only
+     * have 32-bit address space.
+     */
+    if (ws->info.drm_minor >= 41) {
+        ws->vm64.start = 1ull << 32;
+        ws->vm64.end = 1ull << 33;
+    }
 
     /* TTM aligns the BO size to the CPU page size */
     ws->info.gart_page_size = sysconf(_SC_PAGESIZE);
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
index c65f5cb7108..03d96ea4c10 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
@@ -48,6 +48,7 @@ enum radeon_generation {
 struct radeon_vm_heap {
     mtx_t mutex;
     uint64_t start;
+    uint64_t end;
     struct list_head holes;
 };
 
@@ -84,6 +85,7 @@ struct radeon_drm_winsys {
     mtx_t bo_handles_mutex;
     mtx_t bo_fence_lock;
 
+    struct radeon_vm_heap vm32;
     struct radeon_vm_heap vm64;
 
     bool check_vm;