pan/kmod,panvk: rewrite how alignment for an allocation is chosen

The current code assumes 2M and 4k pages. If an allocation request
is larger than 2M, allocation will be aligned to 2M, otherwise to
4k.

The new code is informed by pgsize_bitmap. The allocation is
aligned to a particular page boundary, if the request is at least
as large as that page boundary. Note that given

	pgsize_bitmap = PGSIZE_4K | PGSIZE_2M

2M requests will now be aligned to a 2M boundary, rather than 4k,
which is the case with the current code.

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37288>

src/panfrost/lib/kmod/panthor_kmod.c

@@ -21,6 +21,7 @@
 #include "drm-uapi/panthor_drm.h"
 
 #include "pan_kmod_backend.h"
+#include "pan_props.h"
 
 /* Maximum kmod BO label length, including NUL-terminator */
 #define PANTHOR_BO_LABEL_MAXLEN 4096
@@ -829,7 +830,7 @@ panthor_kmod_vm_alloc_va(struct panthor_kmod_vm *panthor_vm, uint64_t size)
    simple_mtx_lock(&panthor_vm->auto_va.lock);
    panthor_kmod_vm_collect_freed_vas(panthor_vm);
    va = util_vma_heap_alloc(&panthor_vm->auto_va.heap, size,
-                            size > 0x200000 ? 0x200000 : 0x1000);
+      pan_choose_gpu_va_alignment(&panthor_vm->base, size));
    simple_mtx_unlock(&panthor_vm->auto_va.lock);
 
    return va;

src/panfrost/lib/pan_props.c

@@ -302,3 +302,18 @@ pan_clamp_to_usable_va_range(const struct pan_kmod_dev *dev, uint64_t va)
 
    return va;
 }
+
+uint64_t
+pan_choose_gpu_va_alignment(const struct pan_kmod_vm *vm, uint64_t size)
+{
+   assert(vm->pgsize_bitmap != 0);
+
+   uint64_t align = 0;
+   u_foreach_bit64(pgsize_bit, vm->pgsize_bitmap) {
+      uint64_t pgsize = (uint64_t)1 << pgsize_bit;
+      if (align > 0 && pgsize > size)
+         break;
+      align = pgsize;
+   }
+   return align;
+}

src/panfrost/lib/pan_props.h

@@ -34,6 +34,7 @@
 
 struct pan_kmod_dev;
 struct pan_kmod_dev_props;
+struct pan_kmod_vm;
 
 /** Implementation-defined tiler features */
 struct pan_tiler_features {
@@ -155,6 +156,9 @@ pan_query_optimal_z_tib_size(unsigned arch, const struct pan_model *model)
 uint64_t pan_clamp_to_usable_va_range(const struct pan_kmod_dev *dev,
                                       uint64_t va);
 
+uint64_t pan_choose_gpu_va_alignment(const struct pan_kmod_vm *vm,
+                                     uint64_t size);
+
 unsigned pan_compute_max_thread_count(const struct pan_kmod_dev_props *props,
                                       unsigned work_reg_count);
 

src/panfrost/vulkan/panvk_device_memory.c

@@ -12,6 +12,8 @@
 #include "panvk_device_memory.h"
 #include "panvk_entrypoints.h"
 
+#include "pan_props.h"
+
 #include "vk_log.h"
 
 static void *
@@ -118,9 +120,8 @@ panvk_AllocateMemory(VkDevice _device,
    };
 
    if (!(device->kmod.vm->flags & PAN_KMOD_VM_FLAG_AUTO_VA)) {
-      op.va.start =
-         panvk_as_alloc(device, op.va.size,
-                        op.va.size > 0x200000 ? 0x200000 : 0x1000);
+      op.va.start = panvk_as_alloc(device, op.va.size,
+         pan_choose_gpu_va_alignment(device->kmod.vm, op.va.size));
       if (!op.va.start) {
          result = panvk_error(device, VK_ERROR_OUT_OF_DEVICE_MEMORY);
          goto err_put_bo;

src/panfrost/vulkan/panvk_priv_bo.c

@@ -12,6 +12,7 @@
 #include "panvk_priv_bo.h"
 
 #include "kmod/pan_kmod.h"
+#include "pan_props.h"
 
 #include "genxml/decode.h"
 
@@ -59,8 +60,8 @@ panvk_priv_bo_create(struct panvk_device *dev, uint64_t size, uint32_t flags,
    };
 
    if (!(dev->kmod.vm->flags & PAN_KMOD_VM_FLAG_AUTO_VA)) {
-      op.va.start = panvk_as_alloc(dev,
-         op.va.size, op.va.size > 0x200000 ? 0x200000 : 0x1000);
+      op.va.start = panvk_as_alloc(dev, op.va.size,
+         pan_choose_gpu_va_alignment(dev->kmod.vm, op.va.size));
       if (!op.va.start) {
          result = panvk_error(dev, VK_ERROR_OUT_OF_DEVICE_MEMORY);
          goto err_munmap_bo;