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"
52 #define GVT_MAX_VGPU 8
55 INTEL_GVT_HYPERVISOR_XEN = 0,
56 INTEL_GVT_HYPERVISOR_KVM,
59 struct intel_gvt_host {
62 struct intel_gvt_mpt *mpt;
65 extern struct intel_gvt_host intel_gvt_host;
67 /* Describe per-platform limitations. */
68 struct intel_gvt_device_info {
69 u32 max_support_vgpus;
73 unsigned long msi_cap_offset;
76 u32 gtt_entry_size_shift;
77 int gmadr_bytes_in_cmd;
81 /* GM resources owned by a vGPU */
82 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 {
102 bool disable_warn_untrack;
105 #define INTEL_GVT_MAX_BAR_NUM 4
107 struct intel_vgpu_pci_bar {
112 struct intel_vgpu_cfg_space {
113 unsigned char virtual_cfg_space[PCI_CFG_SPACE_EXP_SIZE];
114 struct intel_vgpu_pci_bar bar[INTEL_GVT_MAX_BAR_NUM];
117 #define vgpu_cfg_space(vgpu) ((vgpu)->cfg_space.virtual_cfg_space)
119 #define INTEL_GVT_MAX_PIPE 4
121 struct intel_vgpu_irq {
122 bool irq_warn_once[INTEL_GVT_EVENT_MAX];
123 DECLARE_BITMAP(flip_done_event[INTEL_GVT_MAX_PIPE],
124 INTEL_GVT_EVENT_MAX);
127 struct intel_vgpu_opregion {
131 u32 gfn[INTEL_GVT_OPREGION_PAGES];
134 #define vgpu_opregion(vgpu) (&(vgpu->opregion))
136 #define INTEL_GVT_MAX_PORT 5
138 struct intel_vgpu_display {
139 struct intel_vgpu_i2c_edid i2c_edid;
140 struct intel_vgpu_port ports[INTEL_GVT_MAX_PORT];
141 struct intel_vgpu_sbi sbi;
144 struct vgpu_sched_ctl {
149 INTEL_VGPU_EXECLIST_SUBMISSION = 1,
150 INTEL_VGPU_GUC_SUBMISSION,
153 struct intel_vgpu_submission_ops {
155 int (*init)(struct intel_vgpu *vgpu);
156 void (*clean)(struct intel_vgpu *vgpu);
157 void (*reset)(struct intel_vgpu *vgpu, unsigned long engine_mask);
160 struct intel_vgpu_submission {
161 struct intel_vgpu_execlist execlist[I915_NUM_ENGINES];
162 struct list_head workload_q_head[I915_NUM_ENGINES];
163 struct kmem_cache *workloads;
164 atomic_t running_workload_num;
165 struct i915_gem_context *shadow_ctx;
166 DECLARE_BITMAP(shadow_ctx_desc_updated, I915_NUM_ENGINES);
167 DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
168 void *ring_scan_buffer[I915_NUM_ENGINES];
169 int ring_scan_buffer_size[I915_NUM_ENGINES];
170 const struct intel_vgpu_submission_ops *ops;
171 int virtual_submission_interface;
176 struct intel_gvt *gvt;
178 unsigned long handle; /* vGPU handle used by hypervisor MPT modules */
182 unsigned int resetting_eng;
184 struct vgpu_sched_ctl sched_ctl;
186 struct intel_vgpu_fence fence;
187 struct intel_vgpu_gm gm;
188 struct intel_vgpu_cfg_space cfg_space;
189 struct intel_vgpu_mmio mmio;
190 struct intel_vgpu_irq irq;
191 struct intel_vgpu_gtt gtt;
192 struct intel_vgpu_opregion opregion;
193 struct intel_vgpu_display display;
194 struct intel_vgpu_submission submission;
195 u32 hws_pga[I915_NUM_ENGINES];
197 struct dentry *debugfs;
199 #if IS_ENABLED(CONFIG_DRM_I915_GVT_KVMGT)
201 struct mdev_device *mdev;
202 struct vfio_region *region;
204 struct eventfd_ctx *intx_trigger;
205 struct eventfd_ctx *msi_trigger;
206 struct rb_root cache;
207 struct mutex cache_lock;
208 struct notifier_block iommu_notifier;
209 struct notifier_block group_notifier;
211 struct work_struct release_work;
213 struct vfio_device *vfio_device;
217 struct list_head dmabuf_obj_list_head;
218 struct mutex dmabuf_lock;
219 struct idr object_idr;
221 struct completion vblank_done;
225 /* validating GM healthy status*/
226 #define vgpu_is_vm_unhealthy(ret_val) \
227 (((ret_val) == -EBADRQC) || ((ret_val) == -EFAULT))
229 struct intel_gvt_gm {
230 unsigned long vgpu_allocated_low_gm_size;
231 unsigned long vgpu_allocated_high_gm_size;
234 struct intel_gvt_fence {
235 unsigned long vgpu_allocated_fence_num;
238 /* Special MMIO blocks. */
239 struct gvt_mmio_block {
247 #define INTEL_GVT_MMIO_HASH_BITS 11
249 struct intel_gvt_mmio {
251 /* Register contains RO bits */
252 #define F_RO (1 << 0)
253 /* Register contains graphics address */
254 #define F_GMADR (1 << 1)
255 /* Mode mask registers with high 16 bits as the mask bits */
256 #define F_MODE_MASK (1 << 2)
257 /* This reg can be accessed by GPU commands */
258 #define F_CMD_ACCESS (1 << 3)
259 /* This reg has been accessed by a VM */
260 #define F_ACCESSED (1 << 4)
261 /* This reg has been accessed through GPU commands */
262 #define F_CMD_ACCESSED (1 << 5)
263 /* This reg could be accessed by unaligned address */
264 #define F_UNALIGN (1 << 6)
266 struct gvt_mmio_block *mmio_block;
267 unsigned int num_mmio_block;
269 DECLARE_HASHTABLE(mmio_info_table, INTEL_GVT_MMIO_HASH_BITS);
270 unsigned long num_tracked_mmio;
273 struct intel_gvt_firmware {
276 bool firmware_loaded;
279 #define NR_MAX_INTEL_VGPU_TYPES 20
280 struct intel_vgpu_type {
282 unsigned int avail_instance;
283 unsigned int low_gm_size;
284 unsigned int high_gm_size;
287 enum intel_vgpu_edid resolution;
292 struct drm_i915_private *dev_priv;
293 struct idr vgpu_idr; /* vGPU IDR pool */
295 struct intel_gvt_device_info device_info;
296 struct intel_gvt_gm gm;
297 struct intel_gvt_fence fence;
298 struct intel_gvt_mmio mmio;
299 struct intel_gvt_firmware firmware;
300 struct intel_gvt_irq irq;
301 struct intel_gvt_gtt gtt;
302 struct intel_gvt_workload_scheduler scheduler;
303 struct notifier_block shadow_ctx_notifier_block[I915_NUM_ENGINES];
304 DECLARE_HASHTABLE(cmd_table, GVT_CMD_HASH_BITS);
305 struct intel_vgpu_type *types;
306 unsigned int num_types;
307 struct intel_vgpu *idle_vgpu;
309 struct task_struct *service_thread;
310 wait_queue_head_t service_thread_wq;
311 unsigned long service_request;
313 struct engine_mmio *engine_mmio_list;
315 struct dentry *debugfs_root;
318 static inline struct intel_gvt *to_gvt(struct drm_i915_private *i915)
324 INTEL_GVT_REQUEST_EMULATE_VBLANK = 0,
326 /* Scheduling trigger by timer */
327 INTEL_GVT_REQUEST_SCHED = 1,
329 /* Scheduling trigger by event */
330 INTEL_GVT_REQUEST_EVENT_SCHED = 2,
333 static inline void intel_gvt_request_service(struct intel_gvt *gvt,
336 set_bit(service, (void *)&gvt->service_request);
337 wake_up(&gvt->service_thread_wq);
340 void intel_gvt_free_firmware(struct intel_gvt *gvt);
341 int intel_gvt_load_firmware(struct intel_gvt *gvt);
343 /* Aperture/GM space definitions for GVT device */
344 #define MB_TO_BYTES(mb) ((mb) << 20ULL)
345 #define BYTES_TO_MB(b) ((b) >> 20ULL)
347 #define HOST_LOW_GM_SIZE MB_TO_BYTES(128)
348 #define HOST_HIGH_GM_SIZE MB_TO_BYTES(384)
351 /* Aperture/GM space definitions for GVT device */
352 #define gvt_aperture_sz(gvt) (gvt->dev_priv->ggtt.mappable_end)
353 #define gvt_aperture_pa_base(gvt) (gvt->dev_priv->ggtt.gmadr.start)
355 #define gvt_ggtt_gm_sz(gvt) (gvt->dev_priv->ggtt.base.total)
356 #define gvt_ggtt_sz(gvt) \
357 ((gvt->dev_priv->ggtt.base.total >> PAGE_SHIFT) << 3)
358 #define gvt_hidden_sz(gvt) (gvt_ggtt_gm_sz(gvt) - gvt_aperture_sz(gvt))
360 #define gvt_aperture_gmadr_base(gvt) (0)
361 #define gvt_aperture_gmadr_end(gvt) (gvt_aperture_gmadr_base(gvt) \
362 + gvt_aperture_sz(gvt) - 1)
364 #define gvt_hidden_gmadr_base(gvt) (gvt_aperture_gmadr_base(gvt) \
365 + gvt_aperture_sz(gvt))
366 #define gvt_hidden_gmadr_end(gvt) (gvt_hidden_gmadr_base(gvt) \
367 + gvt_hidden_sz(gvt) - 1)
369 #define gvt_fence_sz(gvt) (gvt->dev_priv->num_fence_regs)
371 /* Aperture/GM space definitions for vGPU */
372 #define vgpu_aperture_offset(vgpu) ((vgpu)->gm.low_gm_node.start)
373 #define vgpu_hidden_offset(vgpu) ((vgpu)->gm.high_gm_node.start)
374 #define vgpu_aperture_sz(vgpu) ((vgpu)->gm.aperture_sz)
375 #define vgpu_hidden_sz(vgpu) ((vgpu)->gm.hidden_sz)
377 #define vgpu_aperture_pa_base(vgpu) \
378 (gvt_aperture_pa_base(vgpu->gvt) + vgpu_aperture_offset(vgpu))
380 #define vgpu_ggtt_gm_sz(vgpu) ((vgpu)->gm.aperture_sz + (vgpu)->gm.hidden_sz)
382 #define vgpu_aperture_pa_end(vgpu) \
383 (vgpu_aperture_pa_base(vgpu) + vgpu_aperture_sz(vgpu) - 1)
385 #define vgpu_aperture_gmadr_base(vgpu) (vgpu_aperture_offset(vgpu))
386 #define vgpu_aperture_gmadr_end(vgpu) \
387 (vgpu_aperture_gmadr_base(vgpu) + vgpu_aperture_sz(vgpu) - 1)
389 #define vgpu_hidden_gmadr_base(vgpu) (vgpu_hidden_offset(vgpu))
390 #define vgpu_hidden_gmadr_end(vgpu) \
391 (vgpu_hidden_gmadr_base(vgpu) + vgpu_hidden_sz(vgpu) - 1)
393 #define vgpu_fence_base(vgpu) (vgpu->fence.base)
394 #define vgpu_fence_sz(vgpu) (vgpu->fence.size)
396 struct intel_vgpu_creation_params {
398 __u64 low_gm_sz; /* in MB */
399 __u64 high_gm_sz; /* in MB */
408 int intel_vgpu_alloc_resource(struct intel_vgpu *vgpu,
409 struct intel_vgpu_creation_params *param);
410 void intel_vgpu_reset_resource(struct intel_vgpu *vgpu);
411 void intel_vgpu_free_resource(struct intel_vgpu *vgpu);
412 void intel_vgpu_write_fence(struct intel_vgpu *vgpu,
413 u32 fence, u64 value);
415 /* Macros for easily accessing vGPU virtual/shadow register.
416 Explicitly seperate use for typed MMIO reg or real offset.*/
417 #define vgpu_vreg_t(vgpu, reg) \
418 (*(u32 *)(vgpu->mmio.vreg + i915_mmio_reg_offset(reg)))
419 #define vgpu_vreg(vgpu, offset) \
420 (*(u32 *)(vgpu->mmio.vreg + (offset)))
421 #define vgpu_vreg64_t(vgpu, reg) \
422 (*(u64 *)(vgpu->mmio.vreg + i915_mmio_reg_offset(reg)))
423 #define vgpu_vreg64(vgpu, offset) \
424 (*(u64 *)(vgpu->mmio.vreg + (offset)))
425 #define vgpu_sreg_t(vgpu, reg) \
426 (*(u32 *)(vgpu->mmio.sreg + i915_mmio_reg_offset(reg)))
427 #define vgpu_sreg(vgpu, offset) \
428 (*(u32 *)(vgpu->mmio.sreg + (offset)))
430 #define for_each_active_vgpu(gvt, vgpu, id) \
431 idr_for_each_entry((&(gvt)->vgpu_idr), (vgpu), (id)) \
432 for_each_if(vgpu->active)
434 static inline void intel_vgpu_write_pci_bar(struct intel_vgpu *vgpu,
435 u32 offset, u32 val, bool low)
439 /* BAR offset should be 32 bits algiend */
440 offset = rounddown(offset, 4);
441 pval = (u32 *)(vgpu_cfg_space(vgpu) + offset);
445 * only update bit 31 - bit 4,
446 * leave the bit 3 - bit 0 unchanged.
448 *pval = (val & GENMASK(31, 4)) | (*pval & GENMASK(3, 0));
454 int intel_gvt_init_vgpu_types(struct intel_gvt *gvt);
455 void intel_gvt_clean_vgpu_types(struct intel_gvt *gvt);
457 struct intel_vgpu *intel_gvt_create_idle_vgpu(struct intel_gvt *gvt);
458 void intel_gvt_destroy_idle_vgpu(struct intel_vgpu *vgpu);
459 struct intel_vgpu *intel_gvt_create_vgpu(struct intel_gvt *gvt,
460 struct intel_vgpu_type *type);
461 void intel_gvt_destroy_vgpu(struct intel_vgpu *vgpu);
462 void intel_gvt_reset_vgpu_locked(struct intel_vgpu *vgpu, bool dmlr,
463 unsigned int engine_mask);
464 void intel_gvt_reset_vgpu(struct intel_vgpu *vgpu);
465 void intel_gvt_activate_vgpu(struct intel_vgpu *vgpu);
466 void intel_gvt_deactivate_vgpu(struct intel_vgpu *vgpu);
468 /* validating GM functions */
469 #define vgpu_gmadr_is_aperture(vgpu, gmadr) \
470 ((gmadr >= vgpu_aperture_gmadr_base(vgpu)) && \
471 (gmadr <= vgpu_aperture_gmadr_end(vgpu)))
473 #define vgpu_gmadr_is_hidden(vgpu, gmadr) \
474 ((gmadr >= vgpu_hidden_gmadr_base(vgpu)) && \
475 (gmadr <= vgpu_hidden_gmadr_end(vgpu)))
477 #define vgpu_gmadr_is_valid(vgpu, gmadr) \
478 ((vgpu_gmadr_is_aperture(vgpu, gmadr) || \
479 (vgpu_gmadr_is_hidden(vgpu, gmadr))))
481 #define gvt_gmadr_is_aperture(gvt, gmadr) \
482 ((gmadr >= gvt_aperture_gmadr_base(gvt)) && \
483 (gmadr <= gvt_aperture_gmadr_end(gvt)))
485 #define gvt_gmadr_is_hidden(gvt, gmadr) \
486 ((gmadr >= gvt_hidden_gmadr_base(gvt)) && \
487 (gmadr <= gvt_hidden_gmadr_end(gvt)))
489 #define gvt_gmadr_is_valid(gvt, gmadr) \
490 (gvt_gmadr_is_aperture(gvt, gmadr) || \
491 gvt_gmadr_is_hidden(gvt, gmadr))
493 bool intel_gvt_ggtt_validate_range(struct intel_vgpu *vgpu, u64 addr, u32 size);
494 int intel_gvt_ggtt_gmadr_g2h(struct intel_vgpu *vgpu, u64 g_addr, u64 *h_addr);
495 int intel_gvt_ggtt_gmadr_h2g(struct intel_vgpu *vgpu, u64 h_addr, u64 *g_addr);
496 int intel_gvt_ggtt_index_g2h(struct intel_vgpu *vgpu, unsigned long g_index,
497 unsigned long *h_index);
498 int intel_gvt_ggtt_h2g_index(struct intel_vgpu *vgpu, unsigned long h_index,
499 unsigned long *g_index);
501 void intel_vgpu_init_cfg_space(struct intel_vgpu *vgpu,
503 void intel_vgpu_reset_cfg_space(struct intel_vgpu *vgpu);
505 int intel_vgpu_emulate_cfg_read(struct intel_vgpu *vgpu, unsigned int offset,
506 void *p_data, unsigned int bytes);
508 int intel_vgpu_emulate_cfg_write(struct intel_vgpu *vgpu, unsigned int offset,
509 void *p_data, unsigned int bytes);
511 static inline u64 intel_vgpu_get_bar_gpa(struct intel_vgpu *vgpu, int bar)
513 /* We are 64bit bar. */
514 return (*(u64 *)(vgpu->cfg_space.virtual_cfg_space + bar)) &
515 PCI_BASE_ADDRESS_MEM_MASK;
518 void intel_vgpu_clean_opregion(struct intel_vgpu *vgpu);
519 int intel_vgpu_init_opregion(struct intel_vgpu *vgpu);
520 int intel_vgpu_opregion_base_write_handler(struct intel_vgpu *vgpu, u32 gpa);
522 int intel_vgpu_emulate_opregion_request(struct intel_vgpu *vgpu, u32 swsci);
523 void populate_pvinfo_page(struct intel_vgpu *vgpu);
525 int intel_gvt_scan_and_shadow_workload(struct intel_vgpu_workload *workload);
526 void enter_failsafe_mode(struct intel_vgpu *vgpu, int reason);
528 struct intel_gvt_ops {
529 int (*emulate_cfg_read)(struct intel_vgpu *, unsigned int, void *,
531 int (*emulate_cfg_write)(struct intel_vgpu *, unsigned int, void *,
533 int (*emulate_mmio_read)(struct intel_vgpu *, u64, void *,
535 int (*emulate_mmio_write)(struct intel_vgpu *, u64, void *,
537 struct intel_vgpu *(*vgpu_create)(struct intel_gvt *,
538 struct intel_vgpu_type *);
539 void (*vgpu_destroy)(struct intel_vgpu *);
540 void (*vgpu_reset)(struct intel_vgpu *);
541 void (*vgpu_activate)(struct intel_vgpu *);
542 void (*vgpu_deactivate)(struct intel_vgpu *);
543 struct intel_vgpu_type *(*gvt_find_vgpu_type)(struct intel_gvt *gvt,
545 bool (*get_gvt_attrs)(struct attribute ***type_attrs,
546 struct attribute_group ***intel_vgpu_type_groups);
547 int (*vgpu_query_plane)(struct intel_vgpu *vgpu, void *);
548 int (*vgpu_get_dmabuf)(struct intel_vgpu *vgpu, unsigned int);
549 int (*write_protect_handler)(struct intel_vgpu *, u64, void *,
555 GVT_FAILSAFE_UNSUPPORTED_GUEST,
556 GVT_FAILSAFE_INSUFFICIENT_RESOURCE,
557 GVT_FAILSAFE_GUEST_ERR,
560 static inline void mmio_hw_access_pre(struct drm_i915_private *dev_priv)
562 intel_runtime_pm_get(dev_priv);
565 static inline void mmio_hw_access_post(struct drm_i915_private *dev_priv)
567 intel_runtime_pm_put(dev_priv);
571 * intel_gvt_mmio_set_accessed - mark a MMIO has been accessed
573 * @offset: register offset
576 static inline void intel_gvt_mmio_set_accessed(
577 struct intel_gvt *gvt, unsigned int offset)
579 gvt->mmio.mmio_attribute[offset >> 2] |= F_ACCESSED;
583 * intel_gvt_mmio_is_cmd_accessed - mark a MMIO could be accessed by command
585 * @offset: register offset
588 static inline bool intel_gvt_mmio_is_cmd_access(
589 struct intel_gvt *gvt, unsigned int offset)
591 return gvt->mmio.mmio_attribute[offset >> 2] & F_CMD_ACCESS;
595 * intel_gvt_mmio_is_unalign - mark a MMIO could be accessed unaligned
597 * @offset: register offset
600 static inline bool intel_gvt_mmio_is_unalign(
601 struct intel_gvt *gvt, unsigned int offset)
603 return gvt->mmio.mmio_attribute[offset >> 2] & F_UNALIGN;
607 * intel_gvt_mmio_set_cmd_accessed - mark a MMIO has been accessed by command
609 * @offset: register offset
612 static inline void intel_gvt_mmio_set_cmd_accessed(
613 struct intel_gvt *gvt, unsigned int offset)
615 gvt->mmio.mmio_attribute[offset >> 2] |= F_CMD_ACCESSED;
619 * intel_gvt_mmio_has_mode_mask - if a MMIO has a mode mask
621 * @offset: register offset
624 * True if a MMIO has a mode mask in its higher 16 bits, false if it isn't.
627 static inline bool intel_gvt_mmio_has_mode_mask(
628 struct intel_gvt *gvt, unsigned int offset)
630 return gvt->mmio.mmio_attribute[offset >> 2] & F_MODE_MASK;
633 int intel_gvt_debugfs_add_vgpu(struct intel_vgpu *vgpu);
634 void intel_gvt_debugfs_remove_vgpu(struct intel_vgpu *vgpu);
635 int intel_gvt_debugfs_init(struct intel_gvt *gvt);
636 void intel_gvt_debugfs_clean(struct intel_gvt *gvt);