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 * Eddie Dong <eddie.dong@intel.com>
25 * Kevin Tian <kevin.tian@intel.com>
28 * Ping Gao <ping.a.gao@intel.com>
29 * Zhi Wang <zhi.a.wang@intel.com>
30 * Bing Niu <bing.niu@intel.com>
36 #include "i915_pvinfo.h"
38 void populate_pvinfo_page(struct intel_vgpu
*vgpu
)
40 struct drm_i915_private
*i915
= vgpu
->gvt
->gt
->i915
;
41 /* setup the ballooning information */
42 vgpu_vreg64_t(vgpu
, vgtif_reg(magic
)) = VGT_MAGIC
;
43 vgpu_vreg_t(vgpu
, vgtif_reg(version_major
)) = 1;
44 vgpu_vreg_t(vgpu
, vgtif_reg(version_minor
)) = 0;
45 vgpu_vreg_t(vgpu
, vgtif_reg(display_ready
)) = 0;
46 vgpu_vreg_t(vgpu
, vgtif_reg(vgt_id
)) = vgpu
->id
;
48 vgpu_vreg_t(vgpu
, vgtif_reg(vgt_caps
)) = VGT_CAPS_FULL_PPGTT
;
49 vgpu_vreg_t(vgpu
, vgtif_reg(vgt_caps
)) |= VGT_CAPS_HWSP_EMULATION
;
50 vgpu_vreg_t(vgpu
, vgtif_reg(vgt_caps
)) |= VGT_CAPS_HUGE_GTT
;
52 vgpu_vreg_t(vgpu
, vgtif_reg(avail_rs
.mappable_gmadr
.base
)) =
53 vgpu_aperture_gmadr_base(vgpu
);
54 vgpu_vreg_t(vgpu
, vgtif_reg(avail_rs
.mappable_gmadr
.size
)) =
55 vgpu_aperture_sz(vgpu
);
56 vgpu_vreg_t(vgpu
, vgtif_reg(avail_rs
.nonmappable_gmadr
.base
)) =
57 vgpu_hidden_gmadr_base(vgpu
);
58 vgpu_vreg_t(vgpu
, vgtif_reg(avail_rs
.nonmappable_gmadr
.size
)) =
61 vgpu_vreg_t(vgpu
, vgtif_reg(avail_rs
.fence_num
)) = vgpu_fence_sz(vgpu
);
63 vgpu_vreg_t(vgpu
, vgtif_reg(cursor_x_hot
)) = UINT_MAX
;
64 vgpu_vreg_t(vgpu
, vgtif_reg(cursor_y_hot
)) = UINT_MAX
;
66 gvt_dbg_core("Populate PVINFO PAGE for vGPU %d\n", vgpu
->id
);
67 gvt_dbg_core("aperture base [GMADR] 0x%llx size 0x%llx\n",
68 vgpu_aperture_gmadr_base(vgpu
), vgpu_aperture_sz(vgpu
));
69 gvt_dbg_core("hidden base [GMADR] 0x%llx size=0x%llx\n",
70 vgpu_hidden_gmadr_base(vgpu
), vgpu_hidden_sz(vgpu
));
71 gvt_dbg_core("fence size %d\n", vgpu_fence_sz(vgpu
));
73 drm_WARN_ON(&i915
->drm
, sizeof(struct vgt_if
) != VGT_PVINFO_SIZE
);
76 #define VGPU_MAX_WEIGHT 16
77 #define VGPU_WEIGHT(vgpu_num) \
78 (VGPU_MAX_WEIGHT / (vgpu_num))
85 /* A vGPU with a weight of 8 will get twice as much GPU as a vGPU
86 * with a weight of 4 on a contended host, different vGPU type has
87 * different weight set. Legal weights range from 1 to 16.
90 enum intel_vgpu_edid edid
;
93 /* Fixed vGPU type table */
94 { MB_TO_BYTES(64), MB_TO_BYTES(384), 4, VGPU_WEIGHT(8), GVT_EDID_1024_768
, "8" },
95 { MB_TO_BYTES(128), MB_TO_BYTES(512), 4, VGPU_WEIGHT(4), GVT_EDID_1920_1200
, "4" },
96 { MB_TO_BYTES(256), MB_TO_BYTES(1024), 4, VGPU_WEIGHT(2), GVT_EDID_1920_1200
, "2" },
97 { MB_TO_BYTES(512), MB_TO_BYTES(2048), 4, VGPU_WEIGHT(1), GVT_EDID_1920_1200
, "1" },
101 * intel_gvt_init_vgpu_types - initialize vGPU type list
104 * Initialize vGPU type list based on available resource.
107 int intel_gvt_init_vgpu_types(struct intel_gvt
*gvt
)
109 unsigned int num_types
;
110 unsigned int i
, low_avail
, high_avail
;
111 unsigned int min_low
;
113 /* vGPU type name is defined as GVTg_Vx_y which contains
114 * physical GPU generation type (e.g V4 as BDW server, V5 as
117 * Depend on physical SKU resource, might see vGPU types like
118 * GVTg_V4_8, GVTg_V4_4, GVTg_V4_2, etc. We can create
119 * different types of vGPU on same physical GPU depending on
120 * available resource. Each vGPU type will have "avail_instance"
121 * to indicate how many vGPU instance can be created for this
125 low_avail
= gvt_aperture_sz(gvt
) - HOST_LOW_GM_SIZE
;
126 high_avail
= gvt_hidden_sz(gvt
) - HOST_HIGH_GM_SIZE
;
127 num_types
= ARRAY_SIZE(vgpu_types
);
129 gvt
->types
= kcalloc(num_types
, sizeof(struct intel_vgpu_type
),
134 min_low
= MB_TO_BYTES(32);
135 for (i
= 0; i
< num_types
; ++i
) {
136 if (low_avail
/ vgpu_types
[i
].low_mm
== 0)
139 gvt
->types
[i
].low_gm_size
= vgpu_types
[i
].low_mm
;
140 gvt
->types
[i
].high_gm_size
= vgpu_types
[i
].high_mm
;
141 gvt
->types
[i
].fence
= vgpu_types
[i
].fence
;
143 if (vgpu_types
[i
].weight
< 1 ||
144 vgpu_types
[i
].weight
> VGPU_MAX_WEIGHT
)
147 gvt
->types
[i
].weight
= vgpu_types
[i
].weight
;
148 gvt
->types
[i
].resolution
= vgpu_types
[i
].edid
;
149 gvt
->types
[i
].avail_instance
= min(low_avail
/ vgpu_types
[i
].low_mm
,
150 high_avail
/ vgpu_types
[i
].high_mm
);
152 if (IS_GEN(gvt
->gt
->i915
, 8))
153 sprintf(gvt
->types
[i
].name
, "GVTg_V4_%s",
155 else if (IS_GEN(gvt
->gt
->i915
, 9))
156 sprintf(gvt
->types
[i
].name
, "GVTg_V5_%s",
159 gvt_dbg_core("type[%d]: %s avail %u low %u high %u fence %u weight %u res %s\n",
160 i
, gvt
->types
[i
].name
,
161 gvt
->types
[i
].avail_instance
,
162 gvt
->types
[i
].low_gm_size
,
163 gvt
->types
[i
].high_gm_size
, gvt
->types
[i
].fence
,
164 gvt
->types
[i
].weight
,
165 vgpu_edid_str(gvt
->types
[i
].resolution
));
172 void intel_gvt_clean_vgpu_types(struct intel_gvt
*gvt
)
177 static void intel_gvt_update_vgpu_types(struct intel_gvt
*gvt
)
180 unsigned int low_gm_avail
, high_gm_avail
, fence_avail
;
181 unsigned int low_gm_min
, high_gm_min
, fence_min
;
183 /* Need to depend on maxium hw resource size but keep on
184 * static config for now.
186 low_gm_avail
= gvt_aperture_sz(gvt
) - HOST_LOW_GM_SIZE
-
187 gvt
->gm
.vgpu_allocated_low_gm_size
;
188 high_gm_avail
= gvt_hidden_sz(gvt
) - HOST_HIGH_GM_SIZE
-
189 gvt
->gm
.vgpu_allocated_high_gm_size
;
190 fence_avail
= gvt_fence_sz(gvt
) - HOST_FENCE
-
191 gvt
->fence
.vgpu_allocated_fence_num
;
193 for (i
= 0; i
< gvt
->num_types
; i
++) {
194 low_gm_min
= low_gm_avail
/ gvt
->types
[i
].low_gm_size
;
195 high_gm_min
= high_gm_avail
/ gvt
->types
[i
].high_gm_size
;
196 fence_min
= fence_avail
/ gvt
->types
[i
].fence
;
197 gvt
->types
[i
].avail_instance
= min(min(low_gm_min
, high_gm_min
),
200 gvt_dbg_core("update type[%d]: %s avail %u low %u high %u fence %u\n",
201 i
, gvt
->types
[i
].name
,
202 gvt
->types
[i
].avail_instance
, gvt
->types
[i
].low_gm_size
,
203 gvt
->types
[i
].high_gm_size
, gvt
->types
[i
].fence
);
208 * intel_gvt_active_vgpu - activate a virtual GPU
211 * This function is called when user wants to activate a virtual GPU.
214 void intel_gvt_activate_vgpu(struct intel_vgpu
*vgpu
)
216 mutex_lock(&vgpu
->vgpu_lock
);
218 mutex_unlock(&vgpu
->vgpu_lock
);
222 * intel_gvt_deactive_vgpu - deactivate a virtual GPU
225 * This function is called when user wants to deactivate a virtual GPU.
226 * The virtual GPU will be stopped.
229 void intel_gvt_deactivate_vgpu(struct intel_vgpu
*vgpu
)
231 mutex_lock(&vgpu
->vgpu_lock
);
233 vgpu
->active
= false;
235 if (atomic_read(&vgpu
->submission
.running_workload_num
)) {
236 mutex_unlock(&vgpu
->vgpu_lock
);
237 intel_gvt_wait_vgpu_idle(vgpu
);
238 mutex_lock(&vgpu
->vgpu_lock
);
241 intel_vgpu_stop_schedule(vgpu
);
243 mutex_unlock(&vgpu
->vgpu_lock
);
247 * intel_gvt_release_vgpu - release a virtual GPU
250 * This function is called when user wants to release a virtual GPU.
251 * The virtual GPU will be stopped and all runtime information will be
255 void intel_gvt_release_vgpu(struct intel_vgpu
*vgpu
)
257 intel_gvt_deactivate_vgpu(vgpu
);
259 mutex_lock(&vgpu
->vgpu_lock
);
260 vgpu
->d3_entered
= false;
261 intel_vgpu_clean_workloads(vgpu
, ALL_ENGINES
);
262 intel_vgpu_dmabuf_cleanup(vgpu
);
263 mutex_unlock(&vgpu
->vgpu_lock
);
267 * intel_gvt_destroy_vgpu - destroy a virtual GPU
270 * This function is called when user wants to destroy a virtual GPU.
273 void intel_gvt_destroy_vgpu(struct intel_vgpu
*vgpu
)
275 struct intel_gvt
*gvt
= vgpu
->gvt
;
276 struct drm_i915_private
*i915
= gvt
->gt
->i915
;
278 drm_WARN(&i915
->drm
, vgpu
->active
, "vGPU is still active!\n");
281 * remove idr first so later clean can judge if need to stop
282 * service if no active vgpu.
284 mutex_lock(&gvt
->lock
);
285 idr_remove(&gvt
->vgpu_idr
, vgpu
->id
);
286 mutex_unlock(&gvt
->lock
);
288 mutex_lock(&vgpu
->vgpu_lock
);
289 intel_gvt_debugfs_remove_vgpu(vgpu
);
290 intel_vgpu_clean_sched_policy(vgpu
);
291 intel_vgpu_clean_submission(vgpu
);
292 intel_vgpu_clean_display(vgpu
);
293 intel_vgpu_clean_opregion(vgpu
);
294 intel_vgpu_reset_ggtt(vgpu
, true);
295 intel_vgpu_clean_gtt(vgpu
);
296 intel_gvt_hypervisor_detach_vgpu(vgpu
);
297 intel_vgpu_free_resource(vgpu
);
298 intel_vgpu_clean_mmio(vgpu
);
299 intel_vgpu_dmabuf_cleanup(vgpu
);
300 mutex_unlock(&vgpu
->vgpu_lock
);
302 mutex_lock(&gvt
->lock
);
303 if (idr_is_empty(&gvt
->vgpu_idr
))
304 intel_gvt_clean_irq(gvt
);
305 intel_gvt_update_vgpu_types(gvt
);
306 mutex_unlock(&gvt
->lock
);
311 #define IDLE_VGPU_IDR 0
314 * intel_gvt_create_idle_vgpu - create an idle virtual GPU
317 * This function is called when user wants to create an idle virtual GPU.
320 * pointer to intel_vgpu, error pointer if failed.
322 struct intel_vgpu
*intel_gvt_create_idle_vgpu(struct intel_gvt
*gvt
)
324 struct intel_vgpu
*vgpu
;
325 enum intel_engine_id i
;
328 vgpu
= vzalloc(sizeof(*vgpu
));
330 return ERR_PTR(-ENOMEM
);
332 vgpu
->id
= IDLE_VGPU_IDR
;
334 mutex_init(&vgpu
->vgpu_lock
);
336 for (i
= 0; i
< I915_NUM_ENGINES
; i
++)
337 INIT_LIST_HEAD(&vgpu
->submission
.workload_q_head
[i
]);
339 ret
= intel_vgpu_init_sched_policy(vgpu
);
343 vgpu
->active
= false;
353 * intel_gvt_destroy_vgpu - destroy an idle virtual GPU
356 * This function is called when user wants to destroy an idle virtual GPU.
359 void intel_gvt_destroy_idle_vgpu(struct intel_vgpu
*vgpu
)
361 mutex_lock(&vgpu
->vgpu_lock
);
362 intel_vgpu_clean_sched_policy(vgpu
);
363 mutex_unlock(&vgpu
->vgpu_lock
);
368 static struct intel_vgpu
*__intel_gvt_create_vgpu(struct intel_gvt
*gvt
,
369 struct intel_vgpu_creation_params
*param
)
371 struct drm_i915_private
*dev_priv
= gvt
->gt
->i915
;
372 struct intel_vgpu
*vgpu
;
375 gvt_dbg_core("handle %llu low %llu MB high %llu MB fence %llu\n",
376 param
->handle
, param
->low_gm_sz
, param
->high_gm_sz
,
379 vgpu
= vzalloc(sizeof(*vgpu
));
381 return ERR_PTR(-ENOMEM
);
383 ret
= idr_alloc(&gvt
->vgpu_idr
, vgpu
, IDLE_VGPU_IDR
+ 1, GVT_MAX_VGPU
,
389 vgpu
->handle
= param
->handle
;
391 vgpu
->sched_ctl
.weight
= param
->weight
;
392 mutex_init(&vgpu
->vgpu_lock
);
393 mutex_init(&vgpu
->dmabuf_lock
);
394 INIT_LIST_HEAD(&vgpu
->dmabuf_obj_list_head
);
395 INIT_RADIX_TREE(&vgpu
->page_track_tree
, GFP_KERNEL
);
396 idr_init_base(&vgpu
->object_idr
, 1);
397 intel_vgpu_init_cfg_space(vgpu
, param
->primary
);
398 vgpu
->d3_entered
= false;
400 ret
= intel_vgpu_init_mmio(vgpu
);
404 ret
= intel_vgpu_alloc_resource(vgpu
, param
);
406 goto out_clean_vgpu_mmio
;
408 populate_pvinfo_page(vgpu
);
410 ret
= intel_gvt_hypervisor_attach_vgpu(vgpu
);
412 goto out_clean_vgpu_resource
;
414 ret
= intel_vgpu_init_gtt(vgpu
);
416 goto out_detach_hypervisor_vgpu
;
418 ret
= intel_vgpu_init_opregion(vgpu
);
422 ret
= intel_vgpu_init_display(vgpu
, param
->resolution
);
424 goto out_clean_opregion
;
426 ret
= intel_vgpu_setup_submission(vgpu
);
428 goto out_clean_display
;
430 ret
= intel_vgpu_init_sched_policy(vgpu
);
432 goto out_clean_submission
;
434 intel_gvt_debugfs_add_vgpu(vgpu
);
436 ret
= intel_gvt_hypervisor_set_opregion(vgpu
);
438 goto out_clean_sched_policy
;
440 if (IS_BROADWELL(dev_priv
) || IS_BROXTON(dev_priv
))
441 ret
= intel_gvt_hypervisor_set_edid(vgpu
, PORT_B
);
443 ret
= intel_gvt_hypervisor_set_edid(vgpu
, PORT_D
);
445 goto out_clean_sched_policy
;
449 out_clean_sched_policy
:
450 intel_vgpu_clean_sched_policy(vgpu
);
451 out_clean_submission
:
452 intel_vgpu_clean_submission(vgpu
);
454 intel_vgpu_clean_display(vgpu
);
456 intel_vgpu_clean_opregion(vgpu
);
458 intel_vgpu_clean_gtt(vgpu
);
459 out_detach_hypervisor_vgpu
:
460 intel_gvt_hypervisor_detach_vgpu(vgpu
);
461 out_clean_vgpu_resource
:
462 intel_vgpu_free_resource(vgpu
);
464 intel_vgpu_clean_mmio(vgpu
);
466 idr_remove(&gvt
->vgpu_idr
, vgpu
->id
);
473 * intel_gvt_create_vgpu - create a virtual GPU
475 * @type: type of the vGPU to create
477 * This function is called when user wants to create a virtual GPU.
480 * pointer to intel_vgpu, error pointer if failed.
482 struct intel_vgpu
*intel_gvt_create_vgpu(struct intel_gvt
*gvt
,
483 struct intel_vgpu_type
*type
)
485 struct intel_vgpu_creation_params param
;
486 struct intel_vgpu
*vgpu
;
490 param
.low_gm_sz
= type
->low_gm_size
;
491 param
.high_gm_sz
= type
->high_gm_size
;
492 param
.fence_sz
= type
->fence
;
493 param
.weight
= type
->weight
;
494 param
.resolution
= type
->resolution
;
496 /* XXX current param based on MB */
497 param
.low_gm_sz
= BYTES_TO_MB(param
.low_gm_sz
);
498 param
.high_gm_sz
= BYTES_TO_MB(param
.high_gm_sz
);
500 mutex_lock(&gvt
->lock
);
501 vgpu
= __intel_gvt_create_vgpu(gvt
, ¶m
);
503 /* calculate left instance change for types */
504 intel_gvt_update_vgpu_types(gvt
);
505 mutex_unlock(&gvt
->lock
);
511 * intel_gvt_reset_vgpu_locked - reset a virtual GPU by DMLR or GT reset
513 * @dmlr: vGPU Device Model Level Reset or GT Reset
514 * @engine_mask: engines to reset for GT reset
516 * This function is called when user wants to reset a virtual GPU through
517 * device model reset or GT reset. The caller should hold the vgpu lock.
519 * vGPU Device Model Level Reset (DMLR) simulates the PCI level reset to reset
520 * the whole vGPU to default state as when it is created. This vGPU function
521 * is required both for functionary and security concerns.The ultimate goal
522 * of vGPU FLR is that reuse a vGPU instance by virtual machines. When we
523 * assign a vGPU to a virtual machine we must isse such reset first.
525 * Full GT Reset and Per-Engine GT Reset are soft reset flow for GPU engines
526 * (Render, Blitter, Video, Video Enhancement). It is defined by GPU Spec.
527 * Unlike the FLR, GT reset only reset particular resource of a vGPU per
528 * the reset request. Guest driver can issue a GT reset by programming the
529 * virtual GDRST register to reset specific virtual GPU engine or all
532 * The parameter dev_level is to identify if we will do DMLR or GT reset.
533 * The parameter engine_mask is to specific the engines that need to be
534 * resetted. If value ALL_ENGINES is given for engine_mask, it means
535 * the caller requests a full GT reset that we will reset all virtual
536 * GPU engines. For FLR, engine_mask is ignored.
538 void intel_gvt_reset_vgpu_locked(struct intel_vgpu
*vgpu
, bool dmlr
,
539 intel_engine_mask_t engine_mask
)
541 struct intel_gvt
*gvt
= vgpu
->gvt
;
542 struct intel_gvt_workload_scheduler
*scheduler
= &gvt
->scheduler
;
543 intel_engine_mask_t resetting_eng
= dmlr
? ALL_ENGINES
: engine_mask
;
545 gvt_dbg_core("------------------------------------------\n");
546 gvt_dbg_core("resseting vgpu%d, dmlr %d, engine_mask %08x\n",
547 vgpu
->id
, dmlr
, engine_mask
);
549 vgpu
->resetting_eng
= resetting_eng
;
551 intel_vgpu_stop_schedule(vgpu
);
553 * The current_vgpu will set to NULL after stopping the
554 * scheduler when the reset is triggered by current vgpu.
556 if (scheduler
->current_vgpu
== NULL
) {
557 mutex_unlock(&vgpu
->vgpu_lock
);
558 intel_gvt_wait_vgpu_idle(vgpu
);
559 mutex_lock(&vgpu
->vgpu_lock
);
562 intel_vgpu_reset_submission(vgpu
, resetting_eng
);
563 /* full GPU reset or device model level reset */
564 if (engine_mask
== ALL_ENGINES
|| dmlr
) {
565 intel_vgpu_select_submission_ops(vgpu
, ALL_ENGINES
, 0);
566 if (engine_mask
== ALL_ENGINES
)
567 intel_vgpu_invalidate_ppgtt(vgpu
);
568 /*fence will not be reset during virtual reset */
570 if(!vgpu
->d3_entered
) {
571 intel_vgpu_invalidate_ppgtt(vgpu
);
572 intel_vgpu_destroy_all_ppgtt_mm(vgpu
);
574 intel_vgpu_reset_ggtt(vgpu
, true);
575 intel_vgpu_reset_resource(vgpu
);
578 intel_vgpu_reset_mmio(vgpu
, dmlr
);
579 populate_pvinfo_page(vgpu
);
582 intel_vgpu_reset_display(vgpu
);
583 intel_vgpu_reset_cfg_space(vgpu
);
584 /* only reset the failsafe mode when dmlr reset */
585 vgpu
->failsafe
= false;
587 * PCI_D0 is set before dmlr, so reset d3_entered here
591 vgpu
->d3_entered
= false;
593 vgpu
->pv_notified
= false;
597 vgpu
->resetting_eng
= 0;
598 gvt_dbg_core("reset vgpu%d done\n", vgpu
->id
);
599 gvt_dbg_core("------------------------------------------\n");
603 * intel_gvt_reset_vgpu - reset a virtual GPU (Function Level)
606 * This function is called when user wants to reset a virtual GPU.
609 void intel_gvt_reset_vgpu(struct intel_vgpu
*vgpu
)
611 mutex_lock(&vgpu
->vgpu_lock
);
612 intel_gvt_reset_vgpu_locked(vgpu
, true, 0);
613 mutex_unlock(&vgpu
->vgpu_lock
);