xref: /linux/drivers/gpu/drm/i915/gt/intel_renderstate.c (revision 6fdcba32711044c35c0e1b094cbd8f3f0b4472c9)
1 /*
2  * Copyright © 2014 Intel Corporation
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
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  *
23  * Authors:
24  *    Mika Kuoppala <mika.kuoppala@intel.com>
25  *
26  */
27 
28 #include "i915_drv.h"
29 #include "intel_renderstate.h"
30 #include "intel_ring.h"
31 
32 struct intel_renderstate {
33 	const struct intel_renderstate_rodata *rodata;
34 	struct drm_i915_gem_object *obj;
35 	struct i915_vma *vma;
36 	u32 batch_offset;
37 	u32 batch_size;
38 	u32 aux_offset;
39 	u32 aux_size;
40 };
41 
42 static const struct intel_renderstate_rodata *
43 render_state_get_rodata(const struct intel_engine_cs *engine)
44 {
45 	if (engine->class != RENDER_CLASS)
46 		return NULL;
47 
48 	switch (INTEL_GEN(engine->i915)) {
49 	case 6:
50 		return &gen6_null_state;
51 	case 7:
52 		return &gen7_null_state;
53 	case 8:
54 		return &gen8_null_state;
55 	case 9:
56 		return &gen9_null_state;
57 	}
58 
59 	return NULL;
60 }
61 
62 /*
63  * Macro to add commands to auxiliary batch.
64  * This macro only checks for page overflow before inserting the commands,
65  * this is sufficient as the null state generator makes the final batch
66  * with two passes to build command and state separately. At this point
67  * the size of both are known and it compacts them by relocating the state
68  * right after the commands taking care of alignment so we should sufficient
69  * space below them for adding new commands.
70  */
71 #define OUT_BATCH(batch, i, val)				\
72 	do {							\
73 		if ((i) >= PAGE_SIZE / sizeof(u32))		\
74 			goto err;				\
75 		(batch)[(i)++] = (val);				\
76 	} while(0)
77 
78 static int render_state_setup(struct intel_renderstate *so,
79 			      struct drm_i915_private *i915)
80 {
81 	const struct intel_renderstate_rodata *rodata = so->rodata;
82 	unsigned int i = 0, reloc_index = 0;
83 	unsigned int needs_clflush;
84 	u32 *d;
85 	int ret;
86 
87 	ret = i915_gem_object_prepare_write(so->obj, &needs_clflush);
88 	if (ret)
89 		return ret;
90 
91 	d = kmap_atomic(i915_gem_object_get_dirty_page(so->obj, 0));
92 
93 	while (i < rodata->batch_items) {
94 		u32 s = rodata->batch[i];
95 
96 		if (i * 4  == rodata->reloc[reloc_index]) {
97 			u64 r = s + so->vma->node.start;
98 			s = lower_32_bits(r);
99 			if (HAS_64BIT_RELOC(i915)) {
100 				if (i + 1 >= rodata->batch_items ||
101 				    rodata->batch[i + 1] != 0)
102 					goto err;
103 
104 				d[i++] = s;
105 				s = upper_32_bits(r);
106 			}
107 
108 			reloc_index++;
109 		}
110 
111 		d[i++] = s;
112 	}
113 
114 	if (rodata->reloc[reloc_index] != -1) {
115 		DRM_ERROR("only %d relocs resolved\n", reloc_index);
116 		goto err;
117 	}
118 
119 	so->batch_offset = i915_ggtt_offset(so->vma);
120 	so->batch_size = rodata->batch_items * sizeof(u32);
121 
122 	while (i % CACHELINE_DWORDS)
123 		OUT_BATCH(d, i, MI_NOOP);
124 
125 	so->aux_offset = i * sizeof(u32);
126 
127 	if (HAS_POOLED_EU(i915)) {
128 		/*
129 		 * We always program 3x6 pool config but depending upon which
130 		 * subslice is disabled HW drops down to appropriate config
131 		 * shown below.
132 		 *
133 		 * In the below table 2x6 config always refers to
134 		 * fused-down version, native 2x6 is not available and can
135 		 * be ignored
136 		 *
137 		 * SNo  subslices config                eu pool configuration
138 		 * -----------------------------------------------------------
139 		 * 1    3 subslices enabled (3x6)  -    0x00777000  (9+9)
140 		 * 2    ss0 disabled (2x6)         -    0x00777000  (3+9)
141 		 * 3    ss1 disabled (2x6)         -    0x00770000  (6+6)
142 		 * 4    ss2 disabled (2x6)         -    0x00007000  (9+3)
143 		 */
144 		u32 eu_pool_config = 0x00777000;
145 
146 		OUT_BATCH(d, i, GEN9_MEDIA_POOL_STATE);
147 		OUT_BATCH(d, i, GEN9_MEDIA_POOL_ENABLE);
148 		OUT_BATCH(d, i, eu_pool_config);
149 		OUT_BATCH(d, i, 0);
150 		OUT_BATCH(d, i, 0);
151 		OUT_BATCH(d, i, 0);
152 	}
153 
154 	OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
155 	so->aux_size = i * sizeof(u32) - so->aux_offset;
156 	so->aux_offset += so->batch_offset;
157 	/*
158 	 * Since we are sending length, we need to strictly conform to
159 	 * all requirements. For Gen2 this must be a multiple of 8.
160 	 */
161 	so->aux_size = ALIGN(so->aux_size, 8);
162 
163 	if (needs_clflush)
164 		drm_clflush_virt_range(d, i * sizeof(u32));
165 	kunmap_atomic(d);
166 
167 	ret = 0;
168 out:
169 	i915_gem_object_finish_access(so->obj);
170 	return ret;
171 
172 err:
173 	kunmap_atomic(d);
174 	ret = -EINVAL;
175 	goto out;
176 }
177 
178 #undef OUT_BATCH
179 
180 int intel_renderstate_emit(struct i915_request *rq)
181 {
182 	struct intel_engine_cs *engine = rq->engine;
183 	struct intel_renderstate so = {}; /* keep the compiler happy */
184 	int err;
185 
186 	so.rodata = render_state_get_rodata(engine);
187 	if (!so.rodata)
188 		return 0;
189 
190 	if (so.rodata->batch_items * 4 > PAGE_SIZE)
191 		return -EINVAL;
192 
193 	so.obj = i915_gem_object_create_internal(engine->i915, PAGE_SIZE);
194 	if (IS_ERR(so.obj))
195 		return PTR_ERR(so.obj);
196 
197 	so.vma = i915_vma_instance(so.obj, &engine->gt->ggtt->vm, NULL);
198 	if (IS_ERR(so.vma)) {
199 		err = PTR_ERR(so.vma);
200 		goto err_obj;
201 	}
202 
203 	err = i915_vma_pin(so.vma, 0, 0, PIN_GLOBAL | PIN_HIGH);
204 	if (err)
205 		goto err_vma;
206 
207 	err = render_state_setup(&so, rq->i915);
208 	if (err)
209 		goto err_unpin;
210 
211 	err = engine->emit_bb_start(rq,
212 				    so.batch_offset, so.batch_size,
213 				    I915_DISPATCH_SECURE);
214 	if (err)
215 		goto err_unpin;
216 
217 	if (so.aux_size > 8) {
218 		err = engine->emit_bb_start(rq,
219 					    so.aux_offset, so.aux_size,
220 					    I915_DISPATCH_SECURE);
221 		if (err)
222 			goto err_unpin;
223 	}
224 
225 	i915_vma_lock(so.vma);
226 	err = i915_request_await_object(rq, so.vma->obj, false);
227 	if (err == 0)
228 		err = i915_vma_move_to_active(so.vma, rq, 0);
229 	i915_vma_unlock(so.vma);
230 err_unpin:
231 	i915_vma_unpin(so.vma);
232 err_vma:
233 	i915_vma_close(so.vma);
234 err_obj:
235 	i915_gem_object_put(so.obj);
236 	return err;
237 }
238