2 * Copyright(c) 2011-2016 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 * Kevin Tian <kevin.tian@intel.com>
25 * Eddie Dong <eddie.dong@intel.com>
28 * Niu Bing <bing.niu@intel.com>
29 * Zhi Wang <zhi.a.wang@intel.com>
37 #include "hypercall.h"
40 #include "interrupt.h"
45 #include "scheduler.h"
46 #include "sched_policy.h"
47 #include "mmio_context.h"
48 #include "cmd_parser.h"
49 #include "fb_decoder.h"
51 #include "page_track.h"
53 #define GVT_MAX_VGPU 8
56 INTEL_GVT_HYPERVISOR_XEN = 0,
57 INTEL_GVT_HYPERVISOR_KVM,
60 struct intel_gvt_host {
63 struct intel_gvt_mpt *mpt;
66 extern struct intel_gvt_host intel_gvt_host;
68 /* Describe per-platform limitations. */
69 struct intel_gvt_device_info {
70 u32 max_support_vgpus;
74 unsigned long msi_cap_offset;
77 u32 gtt_entry_size_shift;
78 int gmadr_bytes_in_cmd;
82 /* GM resources owned by a vGPU */
83 struct intel_vgpu_gm {
86 struct drm_mm_node low_gm_node;
87 struct drm_mm_node high_gm_node;
90 #define INTEL_GVT_MAX_NUM_FENCES 32
92 /* Fences owned by a vGPU */
93 struct intel_vgpu_fence {
94 struct drm_i915_fence_reg *regs[INTEL_GVT_MAX_NUM_FENCES];
99 struct intel_vgpu_mmio {
104 #define INTEL_GVT_MAX_BAR_NUM 4
106 struct intel_vgpu_pci_bar {
111 struct intel_vgpu_cfg_space {
112 unsigned char virtual_cfg_space[PCI_CFG_SPACE_EXP_SIZE];
113 struct intel_vgpu_pci_bar bar[INTEL_GVT_MAX_BAR_NUM];
116 #define vgpu_cfg_space(vgpu) ((vgpu)->cfg_space.virtual_cfg_space)
118 #define INTEL_GVT_MAX_PIPE 4
120 struct intel_vgpu_irq {
121 bool irq_warn_once[INTEL_GVT_EVENT_MAX];
122 DECLARE_BITMAP(flip_done_event[INTEL_GVT_MAX_PIPE],
123 INTEL_GVT_EVENT_MAX);
126 struct intel_vgpu_opregion {
129 u32 gfn[INTEL_GVT_OPREGION_PAGES];
132 #define vgpu_opregion(vgpu) (&(vgpu->opregion))
134 struct intel_vgpu_display {
135 struct intel_vgpu_i2c_edid i2c_edid;
136 struct intel_vgpu_port ports[I915_MAX_PORTS];
137 struct intel_vgpu_sbi sbi;
140 struct vgpu_sched_ctl {
145 INTEL_VGPU_EXECLIST_SUBMISSION = 1,
146 INTEL_VGPU_GUC_SUBMISSION,
149 struct intel_vgpu_submission_ops {
151 int (*init)(struct intel_vgpu *vgpu, unsigned long engine_mask);
152 void (*clean)(struct intel_vgpu *vgpu, unsigned long engine_mask);
153 void (*reset)(struct intel_vgpu *vgpu, unsigned long engine_mask);
156 struct intel_vgpu_submission {
157 struct intel_vgpu_execlist execlist[I915_NUM_ENGINES];
158 struct list_head workload_q_head[I915_NUM_ENGINES];
159 struct kmem_cache *workloads;
160 atomic_t running_workload_num;
161 struct i915_gem_context *shadow_ctx;
162 DECLARE_BITMAP(shadow_ctx_desc_updated, I915_NUM_ENGINES);
163 DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
164 void *ring_scan_buffer[I915_NUM_ENGINES];
165 int ring_scan_buffer_size[I915_NUM_ENGINES];
166 const struct intel_vgpu_submission_ops *ops;
167 int virtual_submission_interface;
172 struct intel_gvt *gvt;
174 unsigned long handle; /* vGPU handle used by hypervisor MPT modules */
178 unsigned int resetting_eng;
180 struct vgpu_sched_ctl sched_ctl;
182 struct intel_vgpu_fence fence;
183 struct intel_vgpu_gm gm;
184 struct intel_vgpu_cfg_space cfg_space;
185 struct intel_vgpu_mmio mmio;
186 struct intel_vgpu_irq irq;
187 struct intel_vgpu_gtt gtt;
188 struct intel_vgpu_opregion opregion;
189 struct intel_vgpu_display display;
190 struct intel_vgpu_submission submission;
191 struct radix_tree_root page_track_tree;
192 u32 hws_pga[I915_NUM_ENGINES];
194 struct dentry *debugfs;
196 #if IS_ENABLED(CONFIG_DRM_I915_GVT_KVMGT)
198 struct mdev_device *mdev;
199 struct vfio_region *region;
201 struct eventfd_ctx *intx_trigger;
202 struct eventfd_ctx *msi_trigger;
205 * Two caches are used to avoid mapping duplicated pages (eg.
206 * scratch pages). This help to reduce dma setup overhead.
208 struct rb_root gfn_cache;
209 struct rb_root dma_addr_cache;
210 unsigned long nr_cache_entries;
211 struct mutex cache_lock;
213 struct notifier_block iommu_notifier;
214 struct notifier_block group_notifier;
216 struct work_struct release_work;
218 struct vfio_device *vfio_device;
222 struct list_head dmabuf_obj_list_head;
223 struct mutex dmabuf_lock;
224 struct idr object_idr;
226 struct completion vblank_done;
231 /* validating GM healthy status*/
232 #define vgpu_is_vm_unhealthy(ret_val) \
233 (((ret_val) == -EBADRQC) || ((ret_val) == -EFAULT))
235 struct intel_gvt_gm {
236 unsigned long vgpu_allocated_low_gm_size;
237 unsigned long vgpu_allocated_high_gm_size;
240 struct intel_gvt_fence {
241 unsigned long vgpu_allocated_fence_num;
244 /* Special MMIO blocks. */
245 struct gvt_mmio_block {
253 #define INTEL_GVT_MMIO_HASH_BITS 11
255 struct intel_gvt_mmio {
257 /* Register contains RO bits */
258 #define F_RO (1 << 0)
259 /* Register contains graphics address */
260 #define F_GMADR (1 << 1)
261 /* Mode mask registers with high 16 bits as the mask bits */
262 #define F_MODE_MASK (1 << 2)
263 /* This reg can be accessed by GPU commands */
264 #define F_CMD_ACCESS (1 << 3)
265 /* This reg has been accessed by a VM */
266 #define F_ACCESSED (1 << 4)
267 /* This reg has been accessed through GPU commands */
268 #define F_CMD_ACCESSED (1 << 5)
269 /* This reg could be accessed by unaligned address */
270 #define F_UNALIGN (1 << 6)
272 struct gvt_mmio_block *mmio_block;
273 unsigned int num_mmio_block;
275 DECLARE_HASHTABLE(mmio_info_table, INTEL_GVT_MMIO_HASH_BITS);
276 unsigned long num_tracked_mmio;
279 struct intel_gvt_firmware {
282 bool firmware_loaded;
285 #define NR_MAX_INTEL_VGPU_TYPES 20
286 struct intel_vgpu_type {
288 unsigned int avail_instance;
289 unsigned int low_gm_size;
290 unsigned int high_gm_size;
293 enum intel_vgpu_edid resolution;
298 struct drm_i915_private *dev_priv;
299 struct idr vgpu_idr; /* vGPU IDR pool */
301 struct intel_gvt_device_info device_info;
302 struct intel_gvt_gm gm;
303 struct intel_gvt_fence fence;
304 struct intel_gvt_mmio mmio;
305 struct intel_gvt_firmware firmware;
306 struct intel_gvt_irq irq;
307 struct intel_gvt_gtt gtt;
308 struct intel_gvt_workload_scheduler scheduler;
309 struct notifier_block shadow_ctx_notifier_block[I915_NUM_ENGINES];
310 DECLARE_HASHTABLE(cmd_table, GVT_CMD_HASH_BITS);
311 struct intel_vgpu_type *types;
312 unsigned int num_types;
313 struct intel_vgpu *idle_vgpu;
315 struct task_struct *service_thread;
316 wait_queue_head_t service_thread_wq;
317 unsigned long service_request;
320 struct engine_mmio *mmio;
321 int ctx_mmio_count[I915_NUM_ENGINES];
324 struct dentry *debugfs_root;
327 static inline struct intel_gvt *to_gvt(struct drm_i915_private *i915)
333 INTEL_GVT_REQUEST_EMULATE_VBLANK = 0,
335 /* Scheduling trigger by timer */
336 INTEL_GVT_REQUEST_SCHED = 1,
338 /* Scheduling trigger by event */
339 INTEL_GVT_REQUEST_EVENT_SCHED = 2,
342 static inline void intel_gvt_request_service(struct intel_gvt *gvt,
345 set_bit(service, (void *)&gvt->service_request);
346 wake_up(&gvt->service_thread_wq);
349 void intel_gvt_free_firmware(struct intel_gvt *gvt);
350 int intel_gvt_load_firmware(struct intel_gvt *gvt);
352 /* Aperture/GM space definitions for GVT device */
353 #define MB_TO_BYTES(mb) ((mb) << 20ULL)
354 #define BYTES_TO_MB(b) ((b) >> 20ULL)
356 #define HOST_LOW_GM_SIZE MB_TO_BYTES(128)
357 #define HOST_HIGH_GM_SIZE MB_TO_BYTES(384)
360 /* Aperture/GM space definitions for GVT device */
361 #define gvt_aperture_sz(gvt) (gvt->dev_priv->ggtt.mappable_end)
362 #define gvt_aperture_pa_base(gvt) (gvt->dev_priv->ggtt.gmadr.start)
364 #define gvt_ggtt_gm_sz(gvt) (gvt->dev_priv->ggtt.base.total)
365 #define gvt_ggtt_sz(gvt) \
366 ((gvt->dev_priv->ggtt.base.total >> PAGE_SHIFT) << 3)
367 #define gvt_hidden_sz(gvt) (gvt_ggtt_gm_sz(gvt) - gvt_aperture_sz(gvt))
369 #define gvt_aperture_gmadr_base(gvt) (0)
370 #define gvt_aperture_gmadr_end(gvt) (gvt_aperture_gmadr_base(gvt) \
371 + gvt_aperture_sz(gvt) - 1)
373 #define gvt_hidden_gmadr_base(gvt) (gvt_aperture_gmadr_base(gvt) \
374 + gvt_aperture_sz(gvt))
375 #define gvt_hidden_gmadr_end(gvt) (gvt_hidden_gmadr_base(gvt) \
376 + gvt_hidden_sz(gvt) - 1)
378 #define gvt_fence_sz(gvt) (gvt->dev_priv->num_fence_regs)
380 /* Aperture/GM space definitions for vGPU */
381 #define vgpu_aperture_offset(vgpu) ((vgpu)->gm.low_gm_node.start)
382 #define vgpu_hidden_offset(vgpu) ((vgpu)->gm.high_gm_node.start)
383 #define vgpu_aperture_sz(vgpu) ((vgpu)->gm.aperture_sz)
384 #define vgpu_hidden_sz(vgpu) ((vgpu)->gm.hidden_sz)
386 #define vgpu_aperture_pa_base(vgpu) \
387 (gvt_aperture_pa_base(vgpu->gvt) + vgpu_aperture_offset(vgpu))
389 #define vgpu_ggtt_gm_sz(vgpu) ((vgpu)->gm.aperture_sz + (vgpu)->gm.hidden_sz)
391 #define vgpu_aperture_pa_end(vgpu) \
392 (vgpu_aperture_pa_base(vgpu) + vgpu_aperture_sz(vgpu) - 1)
394 #define vgpu_aperture_gmadr_base(vgpu) (vgpu_aperture_offset(vgpu))
395 #define vgpu_aperture_gmadr_end(vgpu) \
396 (vgpu_aperture_gmadr_base(vgpu) + vgpu_aperture_sz(vgpu) - 1)
398 #define vgpu_hidden_gmadr_base(vgpu) (vgpu_hidden_offset(vgpu))
399 #define vgpu_hidden_gmadr_end(vgpu) \
400 (vgpu_hidden_gmadr_base(vgpu) + vgpu_hidden_sz(vgpu) - 1)
402 #define vgpu_fence_base(vgpu) (vgpu->fence.base)
403 #define vgpu_fence_sz(vgpu) (vgpu->fence.size)
405 struct intel_vgpu_creation_params {
407 __u64 low_gm_sz; /* in MB */
408 __u64 high_gm_sz; /* in MB */
417 int intel_vgpu_alloc_resource(struct intel_vgpu *vgpu,
418 struct intel_vgpu_creation_params *param);
419 void intel_vgpu_reset_resource(struct intel_vgpu *vgpu);
420 void intel_vgpu_free_resource(struct intel_vgpu *vgpu);
421 void intel_vgpu_write_fence(struct intel_vgpu *vgpu,
422 u32 fence, u64 value);
424 /* Macros for easily accessing vGPU virtual/shadow register.
425 Explicitly seperate use for typed MMIO reg or real offset.*/
426 #define vgpu_vreg_t(vgpu, reg) \
427 (*(u32 *)(vgpu->mmio.vreg + i915_mmio_reg_offset(reg)))
428 #define vgpu_vreg(vgpu, offset) \
429 (*(u32 *)(vgpu->mmio.vreg + (offset)))
430 #define vgpu_vreg64_t(vgpu, reg) \
431 (*(u64 *)(vgpu->mmio.vreg + i915_mmio_reg_offset(reg)))
432 #define vgpu_vreg64(vgpu, offset) \
433 (*(u64 *)(vgpu->mmio.vreg + (offset)))
434 #define vgpu_sreg_t(vgpu, reg) \
435 (*(u32 *)(vgpu->mmio.sreg + i915_mmio_reg_offset(reg)))
436 #define vgpu_sreg(vgpu, offset) \
437 (*(u32 *)(vgpu->mmio.sreg + (offset)))
439 #define for_each_active_vgpu(gvt, vgpu, id) \
440 idr_for_each_entry((&(gvt)->vgpu_idr), (vgpu), (id)) \
441 for_each_if(vgpu->active)
443 static inline void intel_vgpu_write_pci_bar(struct intel_vgpu *vgpu,
444 u32 offset, u32 val, bool low)
448 /* BAR offset should be 32 bits algiend */
449 offset = rounddown(offset, 4);
450 pval = (u32 *)(vgpu_cfg_space(vgpu) + offset);
454 * only update bit 31 - bit 4,
455 * leave the bit 3 - bit 0 unchanged.
457 *pval = (val & GENMASK(31, 4)) | (*pval & GENMASK(3, 0));
463 int intel_gvt_init_vgpu_types(struct intel_gvt *gvt);
464 void intel_gvt_clean_vgpu_types(struct intel_gvt *gvt);
466 struct intel_vgpu *intel_gvt_create_idle_vgpu(struct intel_gvt *gvt);
467 void intel_gvt_destroy_idle_vgpu(struct intel_vgpu *vgpu);
468 struct intel_vgpu *intel_gvt_create_vgpu(struct intel_gvt *gvt,
469 struct intel_vgpu_type *type);
470 void intel_gvt_destroy_vgpu(struct intel_vgpu *vgpu);
471 void intel_gvt_reset_vgpu_locked(struct intel_vgpu *vgpu, bool dmlr,
472 unsigned int engine_mask);
473 void intel_gvt_reset_vgpu(struct intel_vgpu *vgpu);
474 void intel_gvt_activate_vgpu(struct intel_vgpu *vgpu);
475 void intel_gvt_deactivate_vgpu(struct intel_vgpu *vgpu);
477 /* validating GM functions */
478 #define vgpu_gmadr_is_aperture(vgpu, gmadr) \
479 ((gmadr >= vgpu_aperture_gmadr_base(vgpu)) && \
480 (gmadr <= vgpu_aperture_gmadr_end(vgpu)))
482 #define vgpu_gmadr_is_hidden(vgpu, gmadr) \
483 ((gmadr >= vgpu_hidden_gmadr_base(vgpu)) && \
484 (gmadr <= vgpu_hidden_gmadr_end(vgpu)))
486 #define vgpu_gmadr_is_valid(vgpu, gmadr) \
487 ((vgpu_gmadr_is_aperture(vgpu, gmadr) || \
488 (vgpu_gmadr_is_hidden(vgpu, gmadr))))
490 #define gvt_gmadr_is_aperture(gvt, gmadr) \
491 ((gmadr >= gvt_aperture_gmadr_base(gvt)) && \
492 (gmadr <= gvt_aperture_gmadr_end(gvt)))
494 #define gvt_gmadr_is_hidden(gvt, gmadr) \
495 ((gmadr >= gvt_hidden_gmadr_base(gvt)) && \
496 (gmadr <= gvt_hidden_gmadr_end(gvt)))
498 #define gvt_gmadr_is_valid(gvt, gmadr) \
499 (gvt_gmadr_is_aperture(gvt, gmadr) || \
500 gvt_gmadr_is_hidden(gvt, gmadr))
502 bool intel_gvt_ggtt_validate_range(struct intel_vgpu *vgpu, u64 addr, u32 size);
503 int intel_gvt_ggtt_gmadr_g2h(struct intel_vgpu *vgpu, u64 g_addr, u64 *h_addr);
504 int intel_gvt_ggtt_gmadr_h2g(struct intel_vgpu *vgpu, u64 h_addr, u64 *g_addr);
505 int intel_gvt_ggtt_index_g2h(struct intel_vgpu *vgpu, unsigned long g_index,
506 unsigned long *h_index);
507 int intel_gvt_ggtt_h2g_index(struct intel_vgpu *vgpu, unsigned long h_index,
508 unsigned long *g_index);
510 void intel_vgpu_init_cfg_space(struct intel_vgpu *vgpu,
512 void intel_vgpu_reset_cfg_space(struct intel_vgpu *vgpu);
514 int intel_vgpu_emulate_cfg_read(struct intel_vgpu *vgpu, unsigned int offset,
515 void *p_data, unsigned int bytes);
517 int intel_vgpu_emulate_cfg_write(struct intel_vgpu *vgpu, unsigned int offset,
518 void *p_data, unsigned int bytes);
520 static inline u64 intel_vgpu_get_bar_gpa(struct intel_vgpu *vgpu, int bar)
522 /* We are 64bit bar. */
523 return (*(u64 *)(vgpu->cfg_space.virtual_cfg_space + bar)) &
524 PCI_BASE_ADDRESS_MEM_MASK;
527 void intel_vgpu_clean_opregion(struct intel_vgpu *vgpu);
528 int intel_vgpu_init_opregion(struct intel_vgpu *vgpu);
529 int intel_vgpu_opregion_base_write_handler(struct intel_vgpu *vgpu, u32 gpa);
531 int intel_vgpu_emulate_opregion_request(struct intel_vgpu *vgpu, u32 swsci);
532 void populate_pvinfo_page(struct intel_vgpu *vgpu);
534 int intel_gvt_scan_and_shadow_workload(struct intel_vgpu_workload *workload);
535 void enter_failsafe_mode(struct intel_vgpu *vgpu, int reason);
537 struct intel_gvt_ops {
538 int (*emulate_cfg_read)(struct intel_vgpu *, unsigned int, void *,
540 int (*emulate_cfg_write)(struct intel_vgpu *, unsigned int, void *,
542 int (*emulate_mmio_read)(struct intel_vgpu *, u64, void *,
544 int (*emulate_mmio_write)(struct intel_vgpu *, u64, void *,
546 struct intel_vgpu *(*vgpu_create)(struct intel_gvt *,
547 struct intel_vgpu_type *);
548 void (*vgpu_destroy)(struct intel_vgpu *);
549 void (*vgpu_reset)(struct intel_vgpu *);
550 void (*vgpu_activate)(struct intel_vgpu *);
551 void (*vgpu_deactivate)(struct intel_vgpu *);
552 struct intel_vgpu_type *(*gvt_find_vgpu_type)(struct intel_gvt *gvt,
554 bool (*get_gvt_attrs)(struct attribute ***type_attrs,
555 struct attribute_group ***intel_vgpu_type_groups);
556 int (*vgpu_query_plane)(struct intel_vgpu *vgpu, void *);
557 int (*vgpu_get_dmabuf)(struct intel_vgpu *vgpu, unsigned int);
558 int (*write_protect_handler)(struct intel_vgpu *, u64, void *,
564 GVT_FAILSAFE_UNSUPPORTED_GUEST,
565 GVT_FAILSAFE_INSUFFICIENT_RESOURCE,
566 GVT_FAILSAFE_GUEST_ERR,
569 static inline void mmio_hw_access_pre(struct drm_i915_private *dev_priv)
571 intel_runtime_pm_get(dev_priv);
574 static inline void mmio_hw_access_post(struct drm_i915_private *dev_priv)
576 intel_runtime_pm_put(dev_priv);
580 * intel_gvt_mmio_set_accessed - mark a MMIO has been accessed
582 * @offset: register offset
585 static inline void intel_gvt_mmio_set_accessed(
586 struct intel_gvt *gvt, unsigned int offset)
588 gvt->mmio.mmio_attribute[offset >> 2] |= F_ACCESSED;
592 * intel_gvt_mmio_is_cmd_accessed - mark a MMIO could be accessed by command
594 * @offset: register offset
597 static inline bool intel_gvt_mmio_is_cmd_access(
598 struct intel_gvt *gvt, unsigned int offset)
600 return gvt->mmio.mmio_attribute[offset >> 2] & F_CMD_ACCESS;
604 * intel_gvt_mmio_is_unalign - mark a MMIO could be accessed unaligned
606 * @offset: register offset
609 static inline bool intel_gvt_mmio_is_unalign(
610 struct intel_gvt *gvt, unsigned int offset)
612 return gvt->mmio.mmio_attribute[offset >> 2] & F_UNALIGN;
616 * intel_gvt_mmio_set_cmd_accessed - mark a MMIO has been accessed by command
618 * @offset: register offset
621 static inline void intel_gvt_mmio_set_cmd_accessed(
622 struct intel_gvt *gvt, unsigned int offset)
624 gvt->mmio.mmio_attribute[offset >> 2] |= F_CMD_ACCESSED;
628 * intel_gvt_mmio_has_mode_mask - if a MMIO has a mode mask
630 * @offset: register offset
633 * True if a MMIO has a mode mask in its higher 16 bits, false if it isn't.
636 static inline bool intel_gvt_mmio_has_mode_mask(
637 struct intel_gvt *gvt, unsigned int offset)
639 return gvt->mmio.mmio_attribute[offset >> 2] & F_MODE_MASK;
642 int intel_gvt_debugfs_add_vgpu(struct intel_vgpu *vgpu);
643 void intel_gvt_debugfs_remove_vgpu(struct intel_vgpu *vgpu);
644 int intel_gvt_debugfs_init(struct intel_gvt *gvt);
645 void intel_gvt_debugfs_clean(struct intel_gvt *gvt);