drm/amdgpu: move static CSA address to top of address space v2

Move the CSA area to the top of the VA space to avoid clashing with
HMM/ATC in the lower range on GFX9.

v2: wrong sign noticed by Roger, rebase on CSA_VADDR cleanup, handle VA
hole on GFX9 as well.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Monk Liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 18 insertions, 6 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
index e7dfb7b44b4b..b832651d2137 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
@@ -24,6 +24,18 @@
 #include "amdgpu.h"
 #define MAX_KIQ_REG_WAIT        100000000 /* in usecs */
 
+uint64_t amdgpu_csa_vaddr(struct amdgpu_device *adev)
+{
+        uint64_t addr = adev->vm_manager.max_pfn << AMDGPU_GPU_PAGE_SHIFT;
+
+        addr -= AMDGPU_VA_RESERVED_SIZE;
+
+        if (addr >= AMDGPU_VA_HOLE_START)
+                addr |= AMDGPU_VA_HOLE_END;
+
+        return addr;
+}
+
 bool amdgpu_virt_mmio_blocked(struct amdgpu_device *adev)
 {
         /* By now all MMIO pages except mailbox are blocked */
@@ -55,14 +67,14 @@ void amdgpu_free_static_csa(struct amdgpu_device *adev) {
 
 /*
  * amdgpu_map_static_csa should be called during amdgpu_vm_init
- * it maps virtual address "AMDGPU_VA_RESERVED_SIZE - AMDGPU_CSA_SIZE"
- * to this VM, and each command submission of GFX should use this virtual
- * address within META_DATA init package to support SRIOV gfx preemption.
+ * it maps virtual address amdgpu_csa_vaddr() to this VM, and each command
+ * submission of GFX should use this virtual address within META_DATA init
+ * package to support SRIOV gfx preemption.
  */
-
 int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
                           struct amdgpu_bo_va **bo_va)
 {
+        uint64_t csa_addr = amdgpu_csa_vaddr(adev) & AMDGPU_VA_HOLE_MASK;
         struct ww_acquire_ctx ticket;
         struct list_head list;
         struct amdgpu_bo_list_entry pd;
@@ -90,7 +102,7 @@ int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
                 return -ENOMEM;
         }
 
-        r = amdgpu_vm_alloc_pts(adev, (*bo_va)->base.vm, AMDGPU_CSA_VADDR,
+        r = amdgpu_vm_alloc_pts(adev, (*bo_va)->base.vm, csa_addr,
                                 AMDGPU_CSA_SIZE);
         if (r) {
                 DRM_ERROR("failed to allocate pts for static CSA, err=%d\n", r);
@@ -99,7 +111,7 @@ int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
                 return r;
         }
 
-        r = amdgpu_vm_bo_map(adev, *bo_va, AMDGPU_CSA_VADDR, 0, AMDGPU_CSA_SIZE,
+        r = amdgpu_vm_bo_map(adev, *bo_va, csa_addr, 0, AMDGPU_CSA_SIZE,
                              AMDGPU_PTE_READABLE | AMDGPU_PTE_WRITEABLE |
                              AMDGPU_PTE_EXECUTABLE);