drivers/gpu/drm/nouveau/core/subdev/vm/nv50.c

   1 /*
   2  * Copyright 2010 Red Hat Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20  * OTHER DEALINGS IN THE SOFTWARE.
  21  *
  22  * Authors: Ben Skeggs
  23  */
  24
  25 #include <core/device.h>
  26 #include <core/gpuobj.h>
  27
  28 #include <subdev/timer.h>
  29 #include <subdev/fb.h>
  30 #include <subdev/vm.h>
  31
  32 struct nv50_vmmgr_priv {
  33         struct nouveau_vmmgr base;
  34         spinlock_t lock;
  35 };
  36
  37 void
  38 nv50_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 pde,
  39                 struct nouveau_gpuobj *pgt[2])
  40 {
  41         u64 phys = 0xdeadcafe00000000ULL;
  42         u32 coverage = 0;
  43
  44         if (pgt[0]) {
  45                 phys = 0x00000003 | pgt[0]->addr; /* present, 4KiB pages */
  46                 coverage = (pgt[0]->size >> 3) << 12;
  47         } else
  48         if (pgt[1]) {
  49                 phys = 0x00000001 | pgt[1]->addr; /* present */
  50                 coverage = (pgt[1]->size >> 3) << 16;
  51         }
  52
  53         if (phys & 1) {
  54                 if (coverage <= 32 * 1024 * 1024)
  55                         phys |= 0x60;
  56                 else if (coverage <= 64 * 1024 * 1024)
  57                         phys |= 0x40;
  58                 else if (coverage <= 128 * 1024 * 1024)
  59                         phys |= 0x20;
  60         }
  61
  62         nv_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys));
  63         nv_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys));
  64 }
  65
  66 static inline u64
  67 vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
  68 {
  69         phys |= 1; /* present */
  70         phys |= (u64)memtype << 40;
  71         phys |= target << 4;
  72         if (vma->access & NV_MEM_ACCESS_SYS)
  73                 phys |= (1 << 6);
  74         if (!(vma->access & NV_MEM_ACCESS_WO))
  75                 phys |= (1 << 3);
  76         return phys;
  77 }
  78
  79 void
  80 nv50_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
  81             struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
  82 {
  83         u32 comp = (mem->memtype & 0x180) >> 7;
  84         u32 block, target;
  85         int i;
  86
  87         /* IGPs don't have real VRAM, re-target to stolen system memory */
  88         target = 0;
  89         if (nouveau_fb(vma->vm->vmm)->ram.stolen) {
  90                 phys += nouveau_fb(vma->vm->vmm)->ram.stolen;
  91                 target = 3;
  92         }
  93
  94         phys  = vm_addr(vma, phys, mem->memtype, target);
  95         pte <<= 3;
  96         cnt <<= 3;
  97
  98         while (cnt) {
  99                 u32 offset_h = upper_32_bits(phys);
 100                 u32 offset_l = lower_32_bits(phys);
 101
 102                 for (i = 7; i >= 0; i--) {
 103                         block = 1 << (i + 3);
 104                         if (cnt >= block && !(pte & (block - 1)))
 105                                 break;
 106                 }
 107                 offset_l |= (i << 7);
 108
 109                 phys += block << (vma->node->type - 3);
 110                 cnt  -= block;
 111                 if (comp) {
 112                         u32 tag = mem->tag->offset + ((delta >> 16) * comp);
 113                         offset_h |= (tag << 17);
 114                         delta    += block << (vma->node->type - 3);
 115                 }
 116
 117                 while (block) {
 118                         nv_wo32(pgt, pte + 0, offset_l);
 119                         nv_wo32(pgt, pte + 4, offset_h);
 120                         pte += 8;
 121                         block -= 8;
 122                 }
 123         }
 124 }
 125
 126 void
 127 nv50_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
 128                struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
 129 {
 130         u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 3 : 2;
 131         pte <<= 3;
 132         while (cnt--) {
 133                 u64 phys = vm_addr(vma, (u64)*list++, mem->memtype, target);
 134                 nv_wo32(pgt, pte + 0, lower_32_bits(phys));
 135                 nv_wo32(pgt, pte + 4, upper_32_bits(phys));
 136                 pte += 8;
 137         }
 138 }
 139
 140 void
 141 nv50_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
 142 {
 143         pte <<= 3;
 144         while (cnt--) {
 145                 nv_wo32(pgt, pte + 0, 0x00000000);
 146                 nv_wo32(pgt, pte + 4, 0x00000000);
 147                 pte += 8;
 148         }
 149 }
 150
 151 void
 152 nv50_vm_flush(struct nouveau_vm *vm)
 153 {
 154         struct nouveau_engine *engine;
 155         int i;
 156
 157         for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
 158                 if (atomic_read(&vm->engref[i])) {
 159                         engine = nouveau_engine(vm->vmm, i);
 160                         if (engine && engine->tlb_flush)
 161                                 engine->tlb_flush(engine);
 162                 }
 163         }
 164 }
 165
 166 void
 167 nv50_vm_flush_engine(struct nouveau_subdev *subdev, int engine)
 168 {
 169         struct nv50_vmmgr_priv *priv = (void *)nouveau_vmmgr(subdev);
 170         unsigned long flags;
 171
 172         spin_lock_irqsave(&priv->lock, flags);
 173         nv_wr32(subdev, 0x100c80, (engine << 16) | 1);
 174         if (!nv_wait(subdev, 0x100c80, 0x00000001, 0x00000000))
 175                 nv_error(subdev, "vm flush timeout: engine %d\n", engine);
 176         spin_unlock_irqrestore(&priv->lock, flags);
 177 }
 178
 179 static int
 180 nv50_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64 length,
 181                u64 mm_offset, struct nouveau_vm **pvm)
 182 {
 183         u32 block = (1 << (vmm->pgt_bits + 12));
 184         if (block > length)
 185                 block = length;
 186
 187         return nouveau_vm_create(vmm, offset, length, mm_offset, block, pvm);
 188 }
 189
 190 static int
 191 nv50_vmmgr_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 192                 struct nouveau_oclass *oclass, void *data, u32 size,
 193                 struct nouveau_object **pobject)
 194 {
 195         struct nv50_vmmgr_priv *priv;
 196         int ret;
 197
 198         ret = nouveau_vmmgr_create(parent, engine, oclass, "VM", "vm", &priv);
 199         *pobject = nv_object(priv);
 200         if (ret)
 201                 return ret;
 202
 203         priv->base.limit = 1ULL << 40;
 204         priv->base.pgt_bits  = 29 - 12;
 205         priv->base.spg_shift = 12;
 206         priv->base.lpg_shift = 16;
 207         priv->base.create = nv50_vm_create;
 208         priv->base.map_pgt = nv50_vm_map_pgt;
 209         priv->base.map = nv50_vm_map;
 210         priv->base.map_sg = nv50_vm_map_sg;
 211         priv->base.unmap = nv50_vm_unmap;
 212         priv->base.flush = nv50_vm_flush;
 213         spin_lock_init(&priv->lock);
 214         return 0;
 215 }
 216
 217 struct nouveau_oclass
 218 nv50_vmmgr_oclass = {
 219         .handle = NV_SUBDEV(VM, 0x50),
 220         .ofuncs = &(struct nouveau_ofuncs) {
 221                 .ctor = nv50_vmmgr_ctor,
 222                 .dtor = _nouveau_vmmgr_dtor,
 223                 .init = _nouveau_vmmgr_init,
 224                 .fini = _nouveau_vmmgr_fini,
 225         },
 226 };