drm/nouveau/vm: take subdev mutex, not the mm, protects against race with vm/nvc0
authorBen Skeggs <bskeggs@redhat.com>
Mon, 13 May 2013 12:26:26 +0000 (22:26 +1000)
committerBen Skeggs <bskeggs@redhat.com>
Mon, 1 Jul 2013 03:45:03 +0000 (13:45 +1000)
nvc0_vm_flush() accesses the pgd list, which will soon be able to race
with vm_unlink() during channel destruction.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drivers/gpu/drm/nouveau/core/subdev/vm/base.c

index 77c67fc970e64f187d0b37d533fd3bef3cc8139c..6fc38916353256728244ed36705450bb5819bb8c 100644 (file)
@@ -236,9 +236,9 @@ nouveau_vm_unmap_pgt(struct nouveau_vm *vm, int big, u32 fpde, u32 lpde)
                        vmm->map_pgt(vpgd->obj, pde, vpgt->obj);
                }
 
-               mutex_unlock(&vm->mm.mutex);
+               mutex_unlock(&nv_subdev(vmm)->mutex);
                nouveau_gpuobj_ref(NULL, &pgt);
-               mutex_lock(&vm->mm.mutex);
+               mutex_lock(&nv_subdev(vmm)->mutex);
        }
 }
 
@@ -256,18 +256,18 @@ nouveau_vm_map_pgt(struct nouveau_vm *vm, u32 pde, u32 type)
        pgt_size  = (1 << (vmm->pgt_bits + 12)) >> type;
        pgt_size *= 8;
 
-       mutex_unlock(&vm->mm.mutex);
+       mutex_unlock(&nv_subdev(vmm)->mutex);
        ret = nouveau_gpuobj_new(nv_object(vm->vmm), NULL, pgt_size, 0x1000,
                                 NVOBJ_FLAG_ZERO_ALLOC, &pgt);
-       mutex_lock(&vm->mm.mutex);
+       mutex_lock(&nv_subdev(vmm)->mutex);
        if (unlikely(ret))
                return ret;
 
        /* someone beat us to filling the PDE while we didn't have the lock */
        if (unlikely(vpgt->refcount[big]++)) {
-               mutex_unlock(&vm->mm.mutex);
+               mutex_unlock(&nv_subdev(vmm)->mutex);
                nouveau_gpuobj_ref(NULL, &pgt);
-               mutex_lock(&vm->mm.mutex);
+               mutex_lock(&nv_subdev(vmm)->mutex);
                return 0;
        }
 
@@ -289,11 +289,11 @@ nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
        u32 fpde, lpde, pde;
        int ret;
 
-       mutex_lock(&vm->mm.mutex);
+       mutex_lock(&nv_subdev(vmm)->mutex);
        ret = nouveau_mm_head(&vm->mm, page_shift, msize, msize, align,
                             &vma->node);
        if (unlikely(ret != 0)) {
-               mutex_unlock(&vm->mm.mutex);
+               mutex_unlock(&nv_subdev(vmm)->mutex);
                return ret;
        }
 
@@ -314,11 +314,11 @@ nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
                        if (pde != fpde)
                                nouveau_vm_unmap_pgt(vm, big, fpde, pde - 1);
                        nouveau_mm_free(&vm->mm, &vma->node);
-                       mutex_unlock(&vm->mm.mutex);
+                       mutex_unlock(&nv_subdev(vmm)->mutex);
                        return ret;
                }
        }
-       mutex_unlock(&vm->mm.mutex);
+       mutex_unlock(&nv_subdev(vmm)->mutex);
 
        vma->vm     = vm;
        vma->offset = (u64)vma->node->offset << 12;
@@ -338,10 +338,10 @@ nouveau_vm_put(struct nouveau_vma *vma)
        fpde = (vma->node->offset >> vmm->pgt_bits);
        lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits;
 
-       mutex_lock(&vm->mm.mutex);
+       mutex_lock(&nv_subdev(vmm)->mutex);
        nouveau_vm_unmap_pgt(vm, vma->node->type != vmm->spg_shift, fpde, lpde);
        nouveau_mm_free(&vm->mm, &vma->node);
-       mutex_unlock(&vm->mm.mutex);
+       mutex_unlock(&nv_subdev(vmm)->mutex);
 }
 
 int
@@ -405,24 +405,25 @@ nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
 
        nouveau_gpuobj_ref(pgd, &vpgd->obj);
 
-       mutex_lock(&vm->mm.mutex);
+       mutex_lock(&nv_subdev(vmm)->mutex);
        for (i = vm->fpde; i <= vm->lpde; i++)
                vmm->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj);
        list_add(&vpgd->head, &vm->pgd_list);
-       mutex_unlock(&vm->mm.mutex);
+       mutex_unlock(&nv_subdev(vmm)->mutex);
        return 0;
 }
 
 static void
 nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *mpgd)
 {
+       struct nouveau_vmmgr *vmm = vm->vmm;
        struct nouveau_vm_pgd *vpgd, *tmp;
        struct nouveau_gpuobj *pgd = NULL;
 
        if (!mpgd)
                return;
 
-       mutex_lock(&vm->mm.mutex);
+       mutex_lock(&nv_subdev(vmm)->mutex);
        list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
                if (vpgd->obj == mpgd) {
                        pgd = vpgd->obj;
@@ -431,7 +432,7 @@ nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *mpgd)
                        break;
                }
        }
-       mutex_unlock(&vm->mm.mutex);
+       mutex_unlock(&nv_subdev(vmm)->mutex);
 
        nouveau_gpuobj_ref(NULL, &pgd);
 }
This page took 0.030492 seconds and 5 git commands to generate.