• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright(c) 2011-2016 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  * Authors:
24  *    Eddie Dong <eddie.dong@intel.com>
25  *    Kevin Tian <kevin.tian@intel.com>
26  *
27  * Contributors:
28  *    Zhi Wang <zhi.a.wang@intel.com>
29  *    Changbin Du <changbin.du@intel.com>
30  *    Zhenyu Wang <zhenyuw@linux.intel.com>
31  *    Tina Zhang <tina.zhang@intel.com>
32  *    Bing Niu <bing.niu@intel.com>
33  *
34  */
35 
36 #include "i915_drv.h"
37 #include "gt/intel_context.h"
38 #include "gt/intel_ring.h"
39 #include "gvt.h"
40 #include "trace.h"
41 
42 #define GEN9_MOCS_SIZE		64
43 
44 /* Raw offset is appened to each line for convenience. */
45 static struct engine_mmio gen8_engine_mmio_list[] __cacheline_aligned = {
46 	{RCS0, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
47 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
48 	{RCS0, HWSTAM, 0x0, false}, /* 0x2098 */
49 	{RCS0, INSTPM, 0xffff, true}, /* 0x20c0 */
50 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
51 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
52 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
53 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
54 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
55 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
56 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
57 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
58 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
59 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
60 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
61 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
62 	{RCS0, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
63 	{RCS0, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
64 	{RCS0, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
65 	{RCS0, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
66 	{RCS0, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
67 	{RCS0, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
68 
69 	{BCS0, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
70 	{BCS0, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
71 	{BCS0, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
72 	{BCS0, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
73 	{BCS0, RING_EXCC(BLT_RING_BASE), 0xffff, false}, /* 0x22028 */
74 	{RCS0, INVALID_MMIO_REG, 0, false } /* Terminated */
75 };
76 
77 static struct engine_mmio gen9_engine_mmio_list[] __cacheline_aligned = {
78 	{RCS0, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
79 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
80 	{RCS0, HWSTAM, 0x0, false}, /* 0x2098 */
81 	{RCS0, INSTPM, 0xffff, true}, /* 0x20c0 */
82 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
83 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
84 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
85 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
86 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
87 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
88 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
89 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
90 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
91 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
92 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
93 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
94 	{RCS0, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
95 	{RCS0, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
96 	{RCS0, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
97 	{RCS0, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
98 	{RCS0, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
99 	{RCS0, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
100 
101 	{RCS0, GEN8_PRIVATE_PAT_LO, 0, false}, /* 0x40e0 */
102 	{RCS0, GEN8_PRIVATE_PAT_HI, 0, false}, /* 0x40e4 */
103 	{RCS0, GEN8_CS_CHICKEN1, 0xffff, true}, /* 0x2580 */
104 	{RCS0, COMMON_SLICE_CHICKEN2, 0xffff, true}, /* 0x7014 */
105 	{RCS0, GEN9_CS_DEBUG_MODE1, 0xffff, false}, /* 0x20ec */
106 	{RCS0, GEN8_L3SQCREG4, 0, false}, /* 0xb118 */
107 	{RCS0, GEN7_HALF_SLICE_CHICKEN1, 0xffff, true}, /* 0xe100 */
108 	{RCS0, HALF_SLICE_CHICKEN2, 0xffff, true}, /* 0xe180 */
109 	{RCS0, HALF_SLICE_CHICKEN3, 0xffff, true}, /* 0xe184 */
110 	{RCS0, GEN9_HALF_SLICE_CHICKEN5, 0xffff, true}, /* 0xe188 */
111 	{RCS0, GEN9_HALF_SLICE_CHICKEN7, 0xffff, true}, /* 0xe194 */
112 	{RCS0, GEN8_ROW_CHICKEN, 0xffff, true}, /* 0xe4f0 */
113 	{RCS0, TRVATTL3PTRDW(0), 0, true}, /* 0x4de0 */
114 	{RCS0, TRVATTL3PTRDW(1), 0, true}, /* 0x4de4 */
115 	{RCS0, TRNULLDETCT, 0, true}, /* 0x4de8 */
116 	{RCS0, TRINVTILEDETCT, 0, true}, /* 0x4dec */
117 	{RCS0, TRVADR, 0, true}, /* 0x4df0 */
118 	{RCS0, TRTTE, 0, true}, /* 0x4df4 */
119 	{RCS0, _MMIO(0x4dfc), 0, true},
120 
121 	{BCS0, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
122 	{BCS0, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
123 	{BCS0, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
124 	{BCS0, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
125 	{BCS0, RING_EXCC(BLT_RING_BASE), 0xffff, false}, /* 0x22028 */
126 
127 	{VCS1, RING_EXCC(GEN8_BSD2_RING_BASE), 0xffff, false}, /* 0x1c028 */
128 
129 	{VECS0, RING_EXCC(VEBOX_RING_BASE), 0xffff, false}, /* 0x1a028 */
130 
131 	{RCS0, GEN8_HDC_CHICKEN1, 0xffff, true}, /* 0x7304 */
132 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
133 	{RCS0, GEN7_UCGCTL4, 0x0, false}, /* 0x940c */
134 	{RCS0, GAMT_CHKN_BIT_REG, 0x0, false}, /* 0x4ab8 */
135 
136 	{RCS0, GEN9_GAMT_ECO_REG_RW_IA, 0x0, false}, /* 0x4ab0 */
137 	{RCS0, GEN9_CSFE_CHICKEN1_RCS, 0xffff, false}, /* 0x20d4 */
138 	{RCS0, _MMIO(0x20D8), 0xffff, true}, /* 0x20d8 */
139 
140 	{RCS0, GEN8_GARBCNTL, 0x0, false}, /* 0xb004 */
141 	{RCS0, GEN7_FF_THREAD_MODE, 0x0, false}, /* 0x20a0 */
142 	{RCS0, FF_SLICE_CS_CHICKEN2, 0xffff, false}, /* 0x20e4 */
143 	{RCS0, INVALID_MMIO_REG, 0, false } /* Terminated */
144 };
145 
146 static struct {
147 	bool initialized;
148 	u32 control_table[I915_NUM_ENGINES][GEN9_MOCS_SIZE];
149 	u32 l3cc_table[GEN9_MOCS_SIZE / 2];
150 } gen9_render_mocs;
151 
152 static u32 gen9_mocs_mmio_offset_list[] = {
153 	[RCS0]  = 0xc800,
154 	[VCS0]  = 0xc900,
155 	[VCS1]  = 0xca00,
156 	[BCS0]  = 0xcc00,
157 	[VECS0] = 0xcb00,
158 };
159 
load_render_mocs(const struct intel_engine_cs * engine)160 static void load_render_mocs(const struct intel_engine_cs *engine)
161 {
162 	struct intel_gvt *gvt = engine->i915->gvt;
163 	struct intel_uncore *uncore = engine->uncore;
164 	u32 cnt = gvt->engine_mmio_list.mocs_mmio_offset_list_cnt;
165 	u32 *regs = gvt->engine_mmio_list.mocs_mmio_offset_list;
166 	i915_reg_t offset;
167 	int ring_id, i;
168 
169 	/* Platform doesn't have mocs mmios. */
170 	if (!regs)
171 		return;
172 
173 	for (ring_id = 0; ring_id < cnt; ring_id++) {
174 		if (!HAS_ENGINE(engine->gt, ring_id))
175 			continue;
176 
177 		offset.reg = regs[ring_id];
178 		for (i = 0; i < GEN9_MOCS_SIZE; i++) {
179 			gen9_render_mocs.control_table[ring_id][i] =
180 				intel_uncore_read_fw(uncore, offset);
181 			offset.reg += 4;
182 		}
183 	}
184 
185 	offset.reg = 0xb020;
186 	for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
187 		gen9_render_mocs.l3cc_table[i] =
188 			intel_uncore_read_fw(uncore, offset);
189 		offset.reg += 4;
190 	}
191 	gen9_render_mocs.initialized = true;
192 }
193 
194 static int
restore_context_mmio_for_inhibit(struct intel_vgpu * vgpu,struct i915_request * req)195 restore_context_mmio_for_inhibit(struct intel_vgpu *vgpu,
196 				 struct i915_request *req)
197 {
198 	u32 *cs;
199 	int ret;
200 	struct engine_mmio *mmio;
201 	struct intel_gvt *gvt = vgpu->gvt;
202 	int ring_id = req->engine->id;
203 	int count = gvt->engine_mmio_list.ctx_mmio_count[ring_id];
204 
205 	if (count == 0)
206 		return 0;
207 
208 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
209 	if (ret)
210 		return ret;
211 
212 	cs = intel_ring_begin(req, count * 2 + 2);
213 	if (IS_ERR(cs))
214 		return PTR_ERR(cs);
215 
216 	*cs++ = MI_LOAD_REGISTER_IMM(count);
217 	for (mmio = gvt->engine_mmio_list.mmio;
218 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
219 		if (mmio->id != ring_id || !mmio->in_context)
220 			continue;
221 
222 		*cs++ = i915_mmio_reg_offset(mmio->reg);
223 		*cs++ = vgpu_vreg_t(vgpu, mmio->reg) | (mmio->mask << 16);
224 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
225 			      *(cs-2), *(cs-1), vgpu->id, ring_id);
226 	}
227 
228 	*cs++ = MI_NOOP;
229 	intel_ring_advance(req, cs);
230 
231 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
232 	if (ret)
233 		return ret;
234 
235 	return 0;
236 }
237 
238 static int
restore_render_mocs_control_for_inhibit(struct intel_vgpu * vgpu,struct i915_request * req)239 restore_render_mocs_control_for_inhibit(struct intel_vgpu *vgpu,
240 					struct i915_request *req)
241 {
242 	unsigned int index;
243 	u32 *cs;
244 
245 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE + 2);
246 	if (IS_ERR(cs))
247 		return PTR_ERR(cs);
248 
249 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE);
250 
251 	for (index = 0; index < GEN9_MOCS_SIZE; index++) {
252 		*cs++ = i915_mmio_reg_offset(GEN9_GFX_MOCS(index));
253 		*cs++ = vgpu_vreg_t(vgpu, GEN9_GFX_MOCS(index));
254 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
255 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
256 
257 	}
258 
259 	*cs++ = MI_NOOP;
260 	intel_ring_advance(req, cs);
261 
262 	return 0;
263 }
264 
265 static int
restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu * vgpu,struct i915_request * req)266 restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu *vgpu,
267 				     struct i915_request *req)
268 {
269 	unsigned int index;
270 	u32 *cs;
271 
272 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE / 2 + 2);
273 	if (IS_ERR(cs))
274 		return PTR_ERR(cs);
275 
276 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE / 2);
277 
278 	for (index = 0; index < GEN9_MOCS_SIZE / 2; index++) {
279 		*cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(index));
280 		*cs++ = vgpu_vreg_t(vgpu, GEN9_LNCFCMOCS(index));
281 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
282 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
283 
284 	}
285 
286 	*cs++ = MI_NOOP;
287 	intel_ring_advance(req, cs);
288 
289 	return 0;
290 }
291 
292 /*
293  * Use lri command to initialize the mmio which is in context state image for
294  * inhibit context, it contains tracked engine mmio, render_mocs and
295  * render_mocs_l3cc.
296  */
intel_vgpu_restore_inhibit_context(struct intel_vgpu * vgpu,struct i915_request * req)297 int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu,
298 				       struct i915_request *req)
299 {
300 	int ret;
301 	u32 *cs;
302 
303 	cs = intel_ring_begin(req, 2);
304 	if (IS_ERR(cs))
305 		return PTR_ERR(cs);
306 
307 	*cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
308 	*cs++ = MI_NOOP;
309 	intel_ring_advance(req, cs);
310 
311 	ret = restore_context_mmio_for_inhibit(vgpu, req);
312 	if (ret)
313 		goto out;
314 
315 	/* no MOCS register in context except render engine */
316 	if (req->engine->id != RCS0)
317 		goto out;
318 
319 	ret = restore_render_mocs_control_for_inhibit(vgpu, req);
320 	if (ret)
321 		goto out;
322 
323 	ret = restore_render_mocs_l3cc_for_inhibit(vgpu, req);
324 	if (ret)
325 		goto out;
326 
327 out:
328 	cs = intel_ring_begin(req, 2);
329 	if (IS_ERR(cs))
330 		return PTR_ERR(cs);
331 
332 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
333 	*cs++ = MI_NOOP;
334 	intel_ring_advance(req, cs);
335 
336 	return ret;
337 }
338 
339 static u32 gen8_tlb_mmio_offset_list[] = {
340 	[RCS0]  = 0x4260,
341 	[VCS0]  = 0x4264,
342 	[VCS1]  = 0x4268,
343 	[BCS0]  = 0x426c,
344 	[VECS0] = 0x4270,
345 };
346 
handle_tlb_pending_event(struct intel_vgpu * vgpu,const struct intel_engine_cs * engine)347 static void handle_tlb_pending_event(struct intel_vgpu *vgpu,
348 				     const struct intel_engine_cs *engine)
349 {
350 	struct intel_uncore *uncore = engine->uncore;
351 	struct intel_vgpu_submission *s = &vgpu->submission;
352 	u32 *regs = vgpu->gvt->engine_mmio_list.tlb_mmio_offset_list;
353 	u32 cnt = vgpu->gvt->engine_mmio_list.tlb_mmio_offset_list_cnt;
354 	enum forcewake_domains fw;
355 	i915_reg_t reg;
356 
357 	if (!regs)
358 		return;
359 
360 	if (drm_WARN_ON(&engine->i915->drm, engine->id >= cnt))
361 		return;
362 
363 	if (!test_and_clear_bit(engine->id, (void *)s->tlb_handle_pending))
364 		return;
365 
366 	reg = _MMIO(regs[engine->id]);
367 
368 	/* WaForceWakeRenderDuringMmioTLBInvalidate:skl
369 	 * we need to put a forcewake when invalidating RCS TLB caches,
370 	 * otherwise device can go to RC6 state and interrupt invalidation
371 	 * process
372 	 */
373 	fw = intel_uncore_forcewake_for_reg(uncore, reg,
374 					    FW_REG_READ | FW_REG_WRITE);
375 	if (engine->id == RCS0 && INTEL_GEN(engine->i915) >= 9)
376 		fw |= FORCEWAKE_RENDER;
377 
378 	intel_uncore_forcewake_get(uncore, fw);
379 
380 	intel_uncore_write_fw(uncore, reg, 0x1);
381 
382 	if (wait_for_atomic(intel_uncore_read_fw(uncore, reg) == 0, 50))
383 		gvt_vgpu_err("timeout in invalidate ring %s tlb\n",
384 			     engine->name);
385 	else
386 		vgpu_vreg_t(vgpu, reg) = 0;
387 
388 	intel_uncore_forcewake_put(uncore, fw);
389 
390 	gvt_dbg_core("invalidate TLB for ring %s\n", engine->name);
391 }
392 
switch_mocs(struct intel_vgpu * pre,struct intel_vgpu * next,const struct intel_engine_cs * engine)393 static void switch_mocs(struct intel_vgpu *pre, struct intel_vgpu *next,
394 			const struct intel_engine_cs *engine)
395 {
396 	u32 regs[] = {
397 		[RCS0]  = 0xc800,
398 		[VCS0]  = 0xc900,
399 		[VCS1]  = 0xca00,
400 		[BCS0]  = 0xcc00,
401 		[VECS0] = 0xcb00,
402 	};
403 	struct intel_uncore *uncore = engine->uncore;
404 	i915_reg_t offset, l3_offset;
405 	u32 old_v, new_v;
406 	int i;
407 
408 	if (drm_WARN_ON(&engine->i915->drm, engine->id >= ARRAY_SIZE(regs)))
409 		return;
410 
411 	if (engine->id == RCS0 && IS_GEN(engine->i915, 9))
412 		return;
413 
414 	if (!pre && !gen9_render_mocs.initialized)
415 		load_render_mocs(engine);
416 
417 	offset.reg = regs[engine->id];
418 	for (i = 0; i < GEN9_MOCS_SIZE; i++) {
419 		if (pre)
420 			old_v = vgpu_vreg_t(pre, offset);
421 		else
422 			old_v = gen9_render_mocs.control_table[engine->id][i];
423 		if (next)
424 			new_v = vgpu_vreg_t(next, offset);
425 		else
426 			new_v = gen9_render_mocs.control_table[engine->id][i];
427 
428 		if (old_v != new_v)
429 			intel_uncore_write_fw(uncore, offset, new_v);
430 
431 		offset.reg += 4;
432 	}
433 
434 	if (engine->id == RCS0) {
435 		l3_offset.reg = 0xb020;
436 		for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
437 			if (pre)
438 				old_v = vgpu_vreg_t(pre, l3_offset);
439 			else
440 				old_v = gen9_render_mocs.l3cc_table[i];
441 			if (next)
442 				new_v = vgpu_vreg_t(next, l3_offset);
443 			else
444 				new_v = gen9_render_mocs.l3cc_table[i];
445 
446 			if (old_v != new_v)
447 				intel_uncore_write_fw(uncore, l3_offset, new_v);
448 
449 			l3_offset.reg += 4;
450 		}
451 	}
452 }
453 
454 #define CTX_CONTEXT_CONTROL_VAL	0x03
455 
is_inhibit_context(struct intel_context * ce)456 bool is_inhibit_context(struct intel_context *ce)
457 {
458 	const u32 *reg_state = ce->lrc_reg_state;
459 	u32 inhibit_mask =
460 		_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
461 
462 	return inhibit_mask ==
463 		(reg_state[CTX_CONTEXT_CONTROL_VAL] & inhibit_mask);
464 }
465 
466 /* Switch ring mmio values (context). */
switch_mmio(struct intel_vgpu * pre,struct intel_vgpu * next,const struct intel_engine_cs * engine)467 static void switch_mmio(struct intel_vgpu *pre,
468 			struct intel_vgpu *next,
469 			const struct intel_engine_cs *engine)
470 {
471 	struct intel_uncore *uncore = engine->uncore;
472 	struct intel_vgpu_submission *s;
473 	struct engine_mmio *mmio;
474 	u32 old_v, new_v;
475 
476 	if (INTEL_GEN(engine->i915) >= 9)
477 		switch_mocs(pre, next, engine);
478 
479 	for (mmio = engine->i915->gvt->engine_mmio_list.mmio;
480 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
481 		if (mmio->id != engine->id)
482 			continue;
483 		/*
484 		 * No need to do save or restore of the mmio which is in context
485 		 * state image on gen9, it's initialized by lri command and
486 		 * save or restore with context together.
487 		 */
488 		if (IS_GEN(engine->i915, 9) && mmio->in_context)
489 			continue;
490 
491 		// save
492 		if (pre) {
493 			vgpu_vreg_t(pre, mmio->reg) =
494 				intel_uncore_read_fw(uncore, mmio->reg);
495 			if (mmio->mask)
496 				vgpu_vreg_t(pre, mmio->reg) &=
497 					~(mmio->mask << 16);
498 			old_v = vgpu_vreg_t(pre, mmio->reg);
499 		} else {
500 			old_v = mmio->value =
501 				intel_uncore_read_fw(uncore, mmio->reg);
502 		}
503 
504 		// restore
505 		if (next) {
506 			s = &next->submission;
507 			/*
508 			 * No need to restore the mmio which is in context state
509 			 * image if it's not inhibit context, it will restore
510 			 * itself.
511 			 */
512 			if (mmio->in_context &&
513 			    !is_inhibit_context(s->shadow[engine->id]))
514 				continue;
515 
516 			if (mmio->mask)
517 				new_v = vgpu_vreg_t(next, mmio->reg) |
518 					(mmio->mask << 16);
519 			else
520 				new_v = vgpu_vreg_t(next, mmio->reg);
521 		} else {
522 			if (mmio->in_context)
523 				continue;
524 			if (mmio->mask)
525 				new_v = mmio->value | (mmio->mask << 16);
526 			else
527 				new_v = mmio->value;
528 		}
529 
530 		intel_uncore_write_fw(uncore, mmio->reg, new_v);
531 
532 		trace_render_mmio(pre ? pre->id : 0,
533 				  next ? next->id : 0,
534 				  "switch",
535 				  i915_mmio_reg_offset(mmio->reg),
536 				  old_v, new_v);
537 	}
538 
539 	if (next)
540 		handle_tlb_pending_event(next, engine);
541 }
542 
543 /**
544  * intel_gvt_switch_render_mmio - switch mmio context of specific engine
545  * @pre: the last vGPU that own the engine
546  * @next: the vGPU to switch to
547  * @engine: the engine
548  *
549  * If pre is null indicates that host own the engine. If next is null
550  * indicates that we are switching to host workload.
551  */
intel_gvt_switch_mmio(struct intel_vgpu * pre,struct intel_vgpu * next,const struct intel_engine_cs * engine)552 void intel_gvt_switch_mmio(struct intel_vgpu *pre,
553 			   struct intel_vgpu *next,
554 			   const struct intel_engine_cs *engine)
555 {
556 	if (WARN(!pre && !next, "switch ring %s from host to HOST\n",
557 		 engine->name))
558 		return;
559 
560 	gvt_dbg_render("switch ring %s from %s to %s\n", engine->name,
561 		       pre ? "vGPU" : "host", next ? "vGPU" : "HOST");
562 
563 	/**
564 	 * We are using raw mmio access wrapper to improve the
565 	 * performace for batch mmio read/write, so we need
566 	 * handle forcewake mannually.
567 	 */
568 	intel_uncore_forcewake_get(engine->uncore, FORCEWAKE_ALL);
569 	switch_mmio(pre, next, engine);
570 	intel_uncore_forcewake_put(engine->uncore, FORCEWAKE_ALL);
571 }
572 
573 /**
574  * intel_gvt_init_engine_mmio_context - Initiate the engine mmio list
575  * @gvt: GVT device
576  *
577  */
intel_gvt_init_engine_mmio_context(struct intel_gvt * gvt)578 void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt)
579 {
580 	struct engine_mmio *mmio;
581 
582 	if (INTEL_GEN(gvt->gt->i915) >= 9) {
583 		gvt->engine_mmio_list.mmio = gen9_engine_mmio_list;
584 		gvt->engine_mmio_list.tlb_mmio_offset_list = gen8_tlb_mmio_offset_list;
585 		gvt->engine_mmio_list.tlb_mmio_offset_list_cnt = ARRAY_SIZE(gen8_tlb_mmio_offset_list);
586 		gvt->engine_mmio_list.mocs_mmio_offset_list = gen9_mocs_mmio_offset_list;
587 		gvt->engine_mmio_list.mocs_mmio_offset_list_cnt = ARRAY_SIZE(gen9_mocs_mmio_offset_list);
588 	} else {
589 		gvt->engine_mmio_list.mmio = gen8_engine_mmio_list;
590 		gvt->engine_mmio_list.tlb_mmio_offset_list = gen8_tlb_mmio_offset_list;
591 		gvt->engine_mmio_list.tlb_mmio_offset_list_cnt = ARRAY_SIZE(gen8_tlb_mmio_offset_list);
592 	}
593 
594 	for (mmio = gvt->engine_mmio_list.mmio;
595 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
596 		if (mmio->in_context) {
597 			gvt->engine_mmio_list.ctx_mmio_count[mmio->id]++;
598 			intel_gvt_mmio_set_sr_in_ctx(gvt, mmio->reg.reg);
599 		}
600 	}
601 }
602