1 /* 2 * Copyright(c) 2011-2015 Intel Corporation. All rights reserved. 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 (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 * SOFTWARE. 22 */ 23 24 #include "i915_vgpu.h" 25 26 /** 27 * DOC: Intel GVT-g guest support 28 * 29 * Intel GVT-g is a graphics virtualization technology which shares the 30 * GPU among multiple virtual machines on a time-sharing basis. Each 31 * virtual machine is presented a virtual GPU (vGPU), which has equivalent 32 * features as the underlying physical GPU (pGPU), so i915 driver can run 33 * seamlessly in a virtual machine. This file provides vGPU specific 34 * optimizations when running in a virtual machine, to reduce the complexity 35 * of vGPU emulation and to improve the overall performance. 36 * 37 * A primary function introduced here is so-called "address space ballooning" 38 * technique. Intel GVT-g partitions global graphics memory among multiple VMs, 39 * so each VM can directly access a portion of the memory without hypervisor's 40 * intervention, e.g. filling textures or queuing commands. However with the 41 * partitioning an unmodified i915 driver would assume a smaller graphics 42 * memory starting from address ZERO, then requires vGPU emulation module to 43 * translate the graphics address between 'guest view' and 'host view', for 44 * all registers and command opcodes which contain a graphics memory address. 45 * To reduce the complexity, Intel GVT-g introduces "address space ballooning", 46 * by telling the exact partitioning knowledge to each guest i915 driver, which 47 * then reserves and prevents non-allocated portions from allocation. Thus vGPU 48 * emulation module only needs to scan and validate graphics addresses without 49 * complexity of address translation. 50 * 51 */ 52 53 /** 54 * i915_detect_vgpu - detect virtual GPU 55 * @dev_priv: i915 device private 56 * 57 * This function is called at the initialization stage, to detect whether 58 * running on a vGPU. 59 */ 60 void i915_detect_vgpu(struct drm_i915_private *dev_priv) 61 { 62 struct pci_dev *pdev = dev_priv->drm.pdev; 63 u64 magic; 64 u16 version_major; 65 void __iomem *shared_area; 66 67 BUILD_BUG_ON(sizeof(struct vgt_if) != VGT_PVINFO_SIZE); 68 69 /* 70 * This is called before we setup the main MMIO BAR mappings used via 71 * the uncore structure, so we need to access the BAR directly. Since 72 * we do not support VGT on older gens, return early so we don't have 73 * to consider differently numbered or sized MMIO bars 74 */ 75 if (INTEL_GEN(dev_priv) < 6) 76 return; 77 78 shared_area = pci_iomap_range(pdev, 0, VGT_PVINFO_PAGE, VGT_PVINFO_SIZE); 79 if (!shared_area) { 80 DRM_ERROR("failed to map MMIO bar to check for VGT\n"); 81 return; 82 } 83 84 magic = readq(shared_area + vgtif_offset(magic)); 85 if (magic != VGT_MAGIC) 86 goto out; 87 88 version_major = readw(shared_area + vgtif_offset(version_major)); 89 if (version_major < VGT_VERSION_MAJOR) { 90 DRM_INFO("VGT interface version mismatch!\n"); 91 goto out; 92 } 93 94 dev_priv->vgpu.caps = readl(shared_area + vgtif_offset(vgt_caps)); 95 96 dev_priv->vgpu.active = true; 97 DRM_INFO("Virtual GPU for Intel GVT-g detected.\n"); 98 99 out: 100 pci_iounmap(pdev, shared_area); 101 } 102 103 bool intel_vgpu_has_full_ppgtt(struct drm_i915_private *dev_priv) 104 { 105 return dev_priv->vgpu.caps & VGT_CAPS_FULL_PPGTT; 106 } 107 108 struct _balloon_info_ { 109 /* 110 * There are up to 2 regions per mappable/unmappable graphic 111 * memory that might be ballooned. Here, index 0/1 is for mappable 112 * graphic memory, 2/3 for unmappable graphic memory. 113 */ 114 struct drm_mm_node space[4]; 115 }; 116 117 static struct _balloon_info_ bl_info; 118 119 static void vgt_deballoon_space(struct i915_ggtt *ggtt, 120 struct drm_mm_node *node) 121 { 122 if (!drm_mm_node_allocated(node)) 123 return; 124 125 DRM_DEBUG_DRIVER("deballoon space: range [0x%llx - 0x%llx] %llu KiB.\n", 126 node->start, 127 node->start + node->size, 128 node->size / 1024); 129 130 ggtt->vm.reserved -= node->size; 131 drm_mm_remove_node(node); 132 } 133 134 /** 135 * intel_vgt_deballoon - deballoon reserved graphics address trunks 136 * @ggtt: the global GGTT from which we reserved earlier 137 * 138 * This function is called to deallocate the ballooned-out graphic memory, when 139 * driver is unloaded or when ballooning fails. 140 */ 141 void intel_vgt_deballoon(struct i915_ggtt *ggtt) 142 { 143 int i; 144 145 if (!intel_vgpu_active(ggtt->vm.i915)) 146 return; 147 148 DRM_DEBUG("VGT deballoon.\n"); 149 150 for (i = 0; i < 4; i++) 151 vgt_deballoon_space(ggtt, &bl_info.space[i]); 152 } 153 154 static int vgt_balloon_space(struct i915_ggtt *ggtt, 155 struct drm_mm_node *node, 156 unsigned long start, unsigned long end) 157 { 158 unsigned long size = end - start; 159 int ret; 160 161 if (start >= end) 162 return -EINVAL; 163 164 DRM_INFO("balloon space: range [ 0x%lx - 0x%lx ] %lu KiB.\n", 165 start, end, size / 1024); 166 ret = i915_gem_gtt_reserve(&ggtt->vm, node, 167 size, start, I915_COLOR_UNEVICTABLE, 168 0); 169 if (!ret) 170 ggtt->vm.reserved += size; 171 172 return ret; 173 } 174 175 /** 176 * intel_vgt_balloon - balloon out reserved graphics address trunks 177 * @ggtt: the global GGTT from which to reserve 178 * 179 * This function is called at the initialization stage, to balloon out the 180 * graphic address space allocated to other vGPUs, by marking these spaces as 181 * reserved. The ballooning related knowledge(starting address and size of 182 * the mappable/unmappable graphic memory) is described in the vgt_if structure 183 * in a reserved mmio range. 184 * 185 * To give an example, the drawing below depicts one typical scenario after 186 * ballooning. Here the vGPU1 has 2 pieces of graphic address spaces ballooned 187 * out each for the mappable and the non-mappable part. From the vGPU1 point of 188 * view, the total size is the same as the physical one, with the start address 189 * of its graphic space being zero. Yet there are some portions ballooned out( 190 * the shadow part, which are marked as reserved by drm allocator). From the 191 * host point of view, the graphic address space is partitioned by multiple 192 * vGPUs in different VMs. :: 193 * 194 * vGPU1 view Host view 195 * 0 ------> +-----------+ +-----------+ 196 * ^ |###########| | vGPU3 | 197 * | |###########| +-----------+ 198 * | |###########| | vGPU2 | 199 * | +-----------+ +-----------+ 200 * mappable GM | available | ==> | vGPU1 | 201 * | +-----------+ +-----------+ 202 * | |###########| | | 203 * v |###########| | Host | 204 * +=======+===========+ +===========+ 205 * ^ |###########| | vGPU3 | 206 * | |###########| +-----------+ 207 * | |###########| | vGPU2 | 208 * | +-----------+ +-----------+ 209 * unmappable GM | available | ==> | vGPU1 | 210 * | +-----------+ +-----------+ 211 * | |###########| | | 212 * | |###########| | Host | 213 * v |###########| | | 214 * total GM size ------> +-----------+ +-----------+ 215 * 216 * Returns: 217 * zero on success, non-zero if configuration invalid or ballooning failed 218 */ 219 int intel_vgt_balloon(struct i915_ggtt *ggtt) 220 { 221 struct intel_uncore *uncore = &ggtt->vm.i915->uncore; 222 unsigned long ggtt_end = ggtt->vm.total; 223 224 unsigned long mappable_base, mappable_size, mappable_end; 225 unsigned long unmappable_base, unmappable_size, unmappable_end; 226 int ret; 227 228 if (!intel_vgpu_active(ggtt->vm.i915)) 229 return 0; 230 231 mappable_base = 232 intel_uncore_read(uncore, vgtif_reg(avail_rs.mappable_gmadr.base)); 233 mappable_size = 234 intel_uncore_read(uncore, vgtif_reg(avail_rs.mappable_gmadr.size)); 235 unmappable_base = 236 intel_uncore_read(uncore, vgtif_reg(avail_rs.nonmappable_gmadr.base)); 237 unmappable_size = 238 intel_uncore_read(uncore, vgtif_reg(avail_rs.nonmappable_gmadr.size)); 239 240 mappable_end = mappable_base + mappable_size; 241 unmappable_end = unmappable_base + unmappable_size; 242 243 DRM_INFO("VGT ballooning configuration:\n"); 244 DRM_INFO("Mappable graphic memory: base 0x%lx size %ldKiB\n", 245 mappable_base, mappable_size / 1024); 246 DRM_INFO("Unmappable graphic memory: base 0x%lx size %ldKiB\n", 247 unmappable_base, unmappable_size / 1024); 248 249 if (mappable_end > ggtt->mappable_end || 250 unmappable_base < ggtt->mappable_end || 251 unmappable_end > ggtt_end) { 252 DRM_ERROR("Invalid ballooning configuration!\n"); 253 return -EINVAL; 254 } 255 256 /* Unmappable graphic memory ballooning */ 257 if (unmappable_base > ggtt->mappable_end) { 258 ret = vgt_balloon_space(ggtt, &bl_info.space[2], 259 ggtt->mappable_end, unmappable_base); 260 261 if (ret) 262 goto err; 263 } 264 265 if (unmappable_end < ggtt_end) { 266 ret = vgt_balloon_space(ggtt, &bl_info.space[3], 267 unmappable_end, ggtt_end); 268 if (ret) 269 goto err_upon_mappable; 270 } 271 272 /* Mappable graphic memory ballooning */ 273 if (mappable_base) { 274 ret = vgt_balloon_space(ggtt, &bl_info.space[0], 275 0, mappable_base); 276 277 if (ret) 278 goto err_upon_unmappable; 279 } 280 281 if (mappable_end < ggtt->mappable_end) { 282 ret = vgt_balloon_space(ggtt, &bl_info.space[1], 283 mappable_end, ggtt->mappable_end); 284 285 if (ret) 286 goto err_below_mappable; 287 } 288 289 DRM_INFO("VGT balloon successfully\n"); 290 return 0; 291 292 err_below_mappable: 293 vgt_deballoon_space(ggtt, &bl_info.space[0]); 294 err_upon_unmappable: 295 vgt_deballoon_space(ggtt, &bl_info.space[3]); 296 err_upon_mappable: 297 vgt_deballoon_space(ggtt, &bl_info.space[2]); 298 err: 299 DRM_ERROR("VGT balloon fail\n"); 300 return ret; 301 } 302