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 "intel_drv.h" 25 #include "i915_vgpu.h" 26 27 /** 28 * DOC: Intel GVT-g guest support 29 * 30 * Intel GVT-g is a graphics virtualization technology which shares the 31 * GPU among multiple virtual machines on a time-sharing basis. Each 32 * virtual machine is presented a virtual GPU (vGPU), which has equivalent 33 * features as the underlying physical GPU (pGPU), so i915 driver can run 34 * seamlessly in a virtual machine. This file provides vGPU specific 35 * optimizations when running in a virtual machine, to reduce the complexity 36 * of vGPU emulation and to improve the overall performance. 37 * 38 * A primary function introduced here is so-called "address space ballooning" 39 * technique. Intel GVT-g partitions global graphics memory among multiple VMs, 40 * so each VM can directly access a portion of the memory without hypervisor's 41 * intervention, e.g. filling textures or queuing commands. However with the 42 * partitioning an unmodified i915 driver would assume a smaller graphics 43 * memory starting from address ZERO, then requires vGPU emulation module to 44 * translate the graphics address between 'guest view' and 'host view', for 45 * all registers and command opcodes which contain a graphics memory address. 46 * To reduce the complexity, Intel GVT-g introduces "address space ballooning", 47 * by telling the exact partitioning knowledge to each guest i915 driver, which 48 * then reserves and prevents non-allocated portions from allocation. Thus vGPU 49 * emulation module only needs to scan and validate graphics addresses without 50 * complexity of address translation. 51 * 52 */ 53 54 /** 55 * i915_detect_vgpu - detect virtual GPU 56 * @dev_priv: i915 device private 57 * 58 * This function is called at the initialization stage, to detect whether 59 * running on a vGPU. 60 */ 61 void i915_detect_vgpu(struct drm_i915_private *dev_priv) 62 { 63 struct pci_dev *pdev = dev_priv->drm.pdev; 64 u64 magic; 65 u16 version_major; 66 void __iomem *shared_area; 67 68 BUILD_BUG_ON(sizeof(struct vgt_if) != VGT_PVINFO_SIZE); 69 70 /* 71 * This is called before we setup the main MMIO BAR mappings used via 72 * the uncore structure, so we need to access the BAR directly. Since 73 * we do not support VGT on older gens, return early so we don't have 74 * to consider differently numbered or sized MMIO bars 75 */ 76 if (INTEL_GEN(dev_priv) < 6) 77 return; 78 79 shared_area = pci_iomap_range(pdev, 0, VGT_PVINFO_PAGE, VGT_PVINFO_SIZE); 80 if (!shared_area) { 81 DRM_ERROR("failed to map MMIO bar to check for VGT\n"); 82 return; 83 } 84 85 magic = readq(shared_area + vgtif_offset(magic)); 86 if (magic != VGT_MAGIC) 87 goto out; 88 89 version_major = readw(shared_area + vgtif_offset(version_major)); 90 if (version_major < VGT_VERSION_MAJOR) { 91 DRM_INFO("VGT interface version mismatch!\n"); 92 goto out; 93 } 94 95 dev_priv->vgpu.caps = readl(shared_area + vgtif_offset(vgt_caps)); 96 97 dev_priv->vgpu.active = true; 98 DRM_INFO("Virtual GPU for Intel GVT-g detected.\n"); 99 100 out: 101 pci_iounmap(pdev, shared_area); 102 } 103 104 bool intel_vgpu_has_full_ppgtt(struct drm_i915_private *dev_priv) 105 { 106 return dev_priv->vgpu.caps & VGT_CAPS_FULL_PPGTT; 107 } 108 109 struct _balloon_info_ { 110 /* 111 * There are up to 2 regions per mappable/unmappable graphic 112 * memory that might be ballooned. Here, index 0/1 is for mappable 113 * graphic memory, 2/3 for unmappable graphic memory. 114 */ 115 struct drm_mm_node space[4]; 116 }; 117 118 static struct _balloon_info_ bl_info; 119 120 static void vgt_deballoon_space(struct i915_ggtt *ggtt, 121 struct drm_mm_node *node) 122 { 123 DRM_DEBUG_DRIVER("deballoon space: range [0x%llx - 0x%llx] %llu KiB.\n", 124 node->start, 125 node->start + node->size, 126 node->size / 1024); 127 128 ggtt->vm.reserved -= node->size; 129 drm_mm_remove_node(node); 130 } 131 132 /** 133 * intel_vgt_deballoon - deballoon reserved graphics address trunks 134 * @ggtt: the global GGTT from which we reserved earlier 135 * 136 * This function is called to deallocate the ballooned-out graphic memory, when 137 * driver is unloaded or when ballooning fails. 138 */ 139 void intel_vgt_deballoon(struct i915_ggtt *ggtt) 140 { 141 int i; 142 143 if (!intel_vgpu_active(ggtt->vm.i915)) 144 return; 145 146 DRM_DEBUG("VGT deballoon.\n"); 147 148 for (i = 0; i < 4; i++) 149 vgt_deballoon_space(ggtt, &bl_info.space[i]); 150 } 151 152 static int vgt_balloon_space(struct i915_ggtt *ggtt, 153 struct drm_mm_node *node, 154 unsigned long start, unsigned long end) 155 { 156 unsigned long size = end - start; 157 int ret; 158 159 if (start >= end) 160 return -EINVAL; 161 162 DRM_INFO("balloon space: range [ 0x%lx - 0x%lx ] %lu KiB.\n", 163 start, end, size / 1024); 164 ret = i915_gem_gtt_reserve(&ggtt->vm, node, 165 size, start, I915_COLOR_UNEVICTABLE, 166 0); 167 if (!ret) 168 ggtt->vm.reserved += size; 169 170 return ret; 171 } 172 173 /** 174 * intel_vgt_balloon - balloon out reserved graphics address trunks 175 * @ggtt: the global GGTT from which to reserve 176 * 177 * This function is called at the initialization stage, to balloon out the 178 * graphic address space allocated to other vGPUs, by marking these spaces as 179 * reserved. The ballooning related knowledge(starting address and size of 180 * the mappable/unmappable graphic memory) is described in the vgt_if structure 181 * in a reserved mmio range. 182 * 183 * To give an example, the drawing below depicts one typical scenario after 184 * ballooning. Here the vGPU1 has 2 pieces of graphic address spaces ballooned 185 * out each for the mappable and the non-mappable part. From the vGPU1 point of 186 * view, the total size is the same as the physical one, with the start address 187 * of its graphic space being zero. Yet there are some portions ballooned out( 188 * the shadow part, which are marked as reserved by drm allocator). From the 189 * host point of view, the graphic address space is partitioned by multiple 190 * vGPUs in different VMs. :: 191 * 192 * vGPU1 view Host view 193 * 0 ------> +-----------+ +-----------+ 194 * ^ |###########| | vGPU3 | 195 * | |###########| +-----------+ 196 * | |###########| | vGPU2 | 197 * | +-----------+ +-----------+ 198 * mappable GM | available | ==> | vGPU1 | 199 * | +-----------+ +-----------+ 200 * | |###########| | | 201 * v |###########| | Host | 202 * +=======+===========+ +===========+ 203 * ^ |###########| | vGPU3 | 204 * | |###########| +-----------+ 205 * | |###########| | vGPU2 | 206 * | +-----------+ +-----------+ 207 * unmappable GM | available | ==> | vGPU1 | 208 * | +-----------+ +-----------+ 209 * | |###########| | | 210 * | |###########| | Host | 211 * v |###########| | | 212 * total GM size ------> +-----------+ +-----------+ 213 * 214 * Returns: 215 * zero on success, non-zero if configuration invalid or ballooning failed 216 */ 217 int intel_vgt_balloon(struct i915_ggtt *ggtt) 218 { 219 struct intel_uncore *uncore = &ggtt->vm.i915->uncore; 220 unsigned long ggtt_end = ggtt->vm.total; 221 222 unsigned long mappable_base, mappable_size, mappable_end; 223 unsigned long unmappable_base, unmappable_size, unmappable_end; 224 int ret; 225 226 if (!intel_vgpu_active(ggtt->vm.i915)) 227 return 0; 228 229 mappable_base = 230 intel_uncore_read(uncore, vgtif_reg(avail_rs.mappable_gmadr.base)); 231 mappable_size = 232 intel_uncore_read(uncore, vgtif_reg(avail_rs.mappable_gmadr.size)); 233 unmappable_base = 234 intel_uncore_read(uncore, vgtif_reg(avail_rs.nonmappable_gmadr.base)); 235 unmappable_size = 236 intel_uncore_read(uncore, vgtif_reg(avail_rs.nonmappable_gmadr.size)); 237 238 mappable_end = mappable_base + mappable_size; 239 unmappable_end = unmappable_base + unmappable_size; 240 241 DRM_INFO("VGT ballooning configuration:\n"); 242 DRM_INFO("Mappable graphic memory: base 0x%lx size %ldKiB\n", 243 mappable_base, mappable_size / 1024); 244 DRM_INFO("Unmappable graphic memory: base 0x%lx size %ldKiB\n", 245 unmappable_base, unmappable_size / 1024); 246 247 if (mappable_end > ggtt->mappable_end || 248 unmappable_base < ggtt->mappable_end || 249 unmappable_end > ggtt_end) { 250 DRM_ERROR("Invalid ballooning configuration!\n"); 251 return -EINVAL; 252 } 253 254 /* Unmappable graphic memory ballooning */ 255 if (unmappable_base > ggtt->mappable_end) { 256 ret = vgt_balloon_space(ggtt, &bl_info.space[2], 257 ggtt->mappable_end, unmappable_base); 258 259 if (ret) 260 goto err; 261 } 262 263 if (unmappable_end < ggtt_end) { 264 ret = vgt_balloon_space(ggtt, &bl_info.space[3], 265 unmappable_end, ggtt_end); 266 if (ret) 267 goto err_upon_mappable; 268 } 269 270 /* Mappable graphic memory ballooning */ 271 if (mappable_base) { 272 ret = vgt_balloon_space(ggtt, &bl_info.space[0], 273 0, mappable_base); 274 275 if (ret) 276 goto err_upon_unmappable; 277 } 278 279 if (mappable_end < ggtt->mappable_end) { 280 ret = vgt_balloon_space(ggtt, &bl_info.space[1], 281 mappable_end, ggtt->mappable_end); 282 283 if (ret) 284 goto err_below_mappable; 285 } 286 287 DRM_INFO("VGT balloon successfully\n"); 288 return 0; 289 290 err_below_mappable: 291 vgt_deballoon_space(ggtt, &bl_info.space[0]); 292 err_upon_unmappable: 293 vgt_deballoon_space(ggtt, &bl_info.space[3]); 294 err_upon_mappable: 295 vgt_deballoon_space(ggtt, &bl_info.space[2]); 296 err: 297 DRM_ERROR("VGT balloon fail\n"); 298 return ret; 299 } 300