2 * Copyright(c) 2011-2015 Intel Corporation. All rights reserved.
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:
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
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
24 #include "i915_vgpu.h"
27 * DOC: Intel GVT-g guest support
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.
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.
54 * i915_detect_vgpu - detect virtual GPU
55 * @dev_priv: i915 device private
57 * This function is called at the initialization stage, to detect whether
60 void i915_detect_vgpu(struct drm_i915_private *dev_priv)
62 struct pci_dev *pdev = dev_priv->drm.pdev;
65 void __iomem *shared_area;
67 BUILD_BUG_ON(sizeof(struct vgt_if) != VGT_PVINFO_SIZE);
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
75 if (INTEL_GEN(dev_priv) < 6)
78 shared_area = pci_iomap_range(pdev, 0, VGT_PVINFO_PAGE, VGT_PVINFO_SIZE);
80 DRM_ERROR("failed to map MMIO bar to check for VGT\n");
84 magic = readq(shared_area + vgtif_offset(magic));
85 if (magic != VGT_MAGIC)
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");
94 dev_priv->vgpu.caps = readl(shared_area + vgtif_offset(vgt_caps));
96 dev_priv->vgpu.active = true;
97 DRM_INFO("Virtual GPU for Intel GVT-g detected.\n");
100 pci_iounmap(pdev, shared_area);
103 bool intel_vgpu_has_full_ppgtt(struct drm_i915_private *dev_priv)
105 return dev_priv->vgpu.caps & VGT_CAPS_FULL_PPGTT;
108 struct _balloon_info_ {
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.
114 struct drm_mm_node space[4];
117 static struct _balloon_info_ bl_info;
119 static void vgt_deballoon_space(struct i915_ggtt *ggtt,
120 struct drm_mm_node *node)
122 DRM_DEBUG_DRIVER("deballoon space: range [0x%llx - 0x%llx] %llu KiB.\n",
124 node->start + node->size,
127 ggtt->vm.reserved -= node->size;
128 drm_mm_remove_node(node);
132 * intel_vgt_deballoon - deballoon reserved graphics address trunks
133 * @ggtt: the global GGTT from which we reserved earlier
135 * This function is called to deallocate the ballooned-out graphic memory, when
136 * driver is unloaded or when ballooning fails.
138 void intel_vgt_deballoon(struct i915_ggtt *ggtt)
142 if (!intel_vgpu_active(ggtt->vm.i915))
145 DRM_DEBUG("VGT deballoon.\n");
147 for (i = 0; i < 4; i++)
148 vgt_deballoon_space(ggtt, &bl_info.space[i]);
151 static int vgt_balloon_space(struct i915_ggtt *ggtt,
152 struct drm_mm_node *node,
153 unsigned long start, unsigned long end)
155 unsigned long size = end - start;
161 DRM_INFO("balloon space: range [ 0x%lx - 0x%lx ] %lu KiB.\n",
162 start, end, size / 1024);
163 ret = i915_gem_gtt_reserve(&ggtt->vm, node,
164 size, start, I915_COLOR_UNEVICTABLE,
167 ggtt->vm.reserved += size;
173 * intel_vgt_balloon - balloon out reserved graphics address trunks
174 * @ggtt: the global GGTT from which to reserve
176 * This function is called at the initialization stage, to balloon out the
177 * graphic address space allocated to other vGPUs, by marking these spaces as
178 * reserved. The ballooning related knowledge(starting address and size of
179 * the mappable/unmappable graphic memory) is described in the vgt_if structure
180 * in a reserved mmio range.
182 * To give an example, the drawing below depicts one typical scenario after
183 * ballooning. Here the vGPU1 has 2 pieces of graphic address spaces ballooned
184 * out each for the mappable and the non-mappable part. From the vGPU1 point of
185 * view, the total size is the same as the physical one, with the start address
186 * of its graphic space being zero. Yet there are some portions ballooned out(
187 * the shadow part, which are marked as reserved by drm allocator). From the
188 * host point of view, the graphic address space is partitioned by multiple
189 * vGPUs in different VMs. ::
191 * vGPU1 view Host view
192 * 0 ------> +-----------+ +-----------+
193 * ^ |###########| | vGPU3 |
194 * | |###########| +-----------+
195 * | |###########| | vGPU2 |
196 * | +-----------+ +-----------+
197 * mappable GM | available | ==> | vGPU1 |
198 * | +-----------+ +-----------+
199 * | |###########| | |
200 * v |###########| | Host |
201 * +=======+===========+ +===========+
202 * ^ |###########| | vGPU3 |
203 * | |###########| +-----------+
204 * | |###########| | vGPU2 |
205 * | +-----------+ +-----------+
206 * unmappable GM | available | ==> | vGPU1 |
207 * | +-----------+ +-----------+
208 * | |###########| | |
209 * | |###########| | Host |
210 * v |###########| | |
211 * total GM size ------> +-----------+ +-----------+
214 * zero on success, non-zero if configuration invalid or ballooning failed
216 int intel_vgt_balloon(struct i915_ggtt *ggtt)
218 struct intel_uncore *uncore = &ggtt->vm.i915->uncore;
219 unsigned long ggtt_end = ggtt->vm.total;
221 unsigned long mappable_base, mappable_size, mappable_end;
222 unsigned long unmappable_base, unmappable_size, unmappable_end;
225 if (!intel_vgpu_active(ggtt->vm.i915))
229 intel_uncore_read(uncore, vgtif_reg(avail_rs.mappable_gmadr.base));
231 intel_uncore_read(uncore, vgtif_reg(avail_rs.mappable_gmadr.size));
233 intel_uncore_read(uncore, vgtif_reg(avail_rs.nonmappable_gmadr.base));
235 intel_uncore_read(uncore, vgtif_reg(avail_rs.nonmappable_gmadr.size));
237 mappable_end = mappable_base + mappable_size;
238 unmappable_end = unmappable_base + unmappable_size;
240 DRM_INFO("VGT ballooning configuration:\n");
241 DRM_INFO("Mappable graphic memory: base 0x%lx size %ldKiB\n",
242 mappable_base, mappable_size / 1024);
243 DRM_INFO("Unmappable graphic memory: base 0x%lx size %ldKiB\n",
244 unmappable_base, unmappable_size / 1024);
246 if (mappable_end > ggtt->mappable_end ||
247 unmappable_base < ggtt->mappable_end ||
248 unmappable_end > ggtt_end) {
249 DRM_ERROR("Invalid ballooning configuration!\n");
253 /* Unmappable graphic memory ballooning */
254 if (unmappable_base > ggtt->mappable_end) {
255 ret = vgt_balloon_space(ggtt, &bl_info.space[2],
256 ggtt->mappable_end, unmappable_base);
262 if (unmappable_end < ggtt_end) {
263 ret = vgt_balloon_space(ggtt, &bl_info.space[3],
264 unmappable_end, ggtt_end);
266 goto err_upon_mappable;
269 /* Mappable graphic memory ballooning */
271 ret = vgt_balloon_space(ggtt, &bl_info.space[0],
275 goto err_upon_unmappable;
278 if (mappable_end < ggtt->mappable_end) {
279 ret = vgt_balloon_space(ggtt, &bl_info.space[1],
280 mappable_end, ggtt->mappable_end);
283 goto err_below_mappable;
286 DRM_INFO("VGT balloon successfully\n");
290 vgt_deballoon_space(ggtt, &bl_info.space[0]);
292 vgt_deballoon_space(ggtt, &bl_info.space[3]);
294 vgt_deballoon_space(ggtt, &bl_info.space[2]);
296 DRM_ERROR("VGT balloon fail\n");