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 * Chris Wilson <[email protected]>
25 *
26 */
27
28 #include "igt.h"
29 #include <unistd.h>
30 #include <stdlib.h>
31 #include <stdint.h>
32 #include <stdio.h>
33 #include <string.h>
34 #include <fcntl.h>
35 #include <inttypes.h>
36 #include <errno.h>
37 #include <sys/stat.h>
38 #include <sys/ioctl.h>
39 #include <sys/time.h>
40 #include "drm.h"
41
42 IGT_TEST_DESCRIPTION("Simulates SNA behaviour using negative self-relocations"
43 " for STATE_BASE_ADDRESS command packets.");
44
45 #define USE_LUT (1 << 12)
46 #define BIAS (256*1024)
47
48 /* Simulates SNA behaviour using negative self-relocations for
49 * STATE_BASE_ADDRESS command packets. If they wrap around (to values greater
50 * than the total size of the GTT), the GPU will hang.
51 * See https://bugs.freedesktop.org/show_bug.cgi?id=78533
52 */
negative_reloc(int fd,unsigned engine,unsigned flags)53 static void negative_reloc(int fd, unsigned engine, unsigned flags)
54 {
55 struct drm_i915_gem_execbuffer2 execbuf;
56 struct drm_i915_gem_exec_object2 obj;
57 struct drm_i915_gem_relocation_entry reloc[1000];
58 uint64_t gtt_max = gem_aperture_size(fd);
59 uint32_t bbe = MI_BATCH_BUFFER_END;
60 uint64_t *offsets;
61 int i;
62
63 gem_require_ring(fd, engine);
64 igt_require(intel_gen(intel_get_drm_devid(fd)) >= 7);
65
66 memset(&obj, 0, sizeof(obj));
67 obj.handle = gem_create(fd, 8192);
68 gem_write(fd, obj.handle, 0, &bbe, sizeof(bbe));
69
70 memset(&execbuf, 0, sizeof(execbuf));
71 execbuf.buffers_ptr = (uintptr_t)&obj;
72 execbuf.buffer_count = 1;
73 execbuf.flags = engine | (flags & USE_LUT);
74 igt_require(__gem_execbuf(fd, &execbuf) == 0);
75
76 igt_info("Found offset %lld for 4k batch\n", (long long)obj.offset);
77 /*
78 * Ideally we'd like to be able to control where the kernel is going to
79 * place the buffer. We don't SKIP here because it causes the test
80 * to "randomly" flip-flop between the SKIP and PASS states.
81 */
82 if (obj.offset < BIAS) {
83 igt_info("Offset is below BIAS, not testing anything\n");
84 return;
85 }
86
87 memset(reloc, 0, sizeof(reloc));
88 for (i = 0; i < ARRAY_SIZE(reloc); i++) {
89 reloc[i].offset = 8 + 8*i;
90 reloc[i].delta = -BIAS*i/1024;
91 reloc[i].presumed_offset = -1;
92 reloc[i].target_handle = flags & USE_LUT ? 0 : obj.handle;
93 reloc[i].read_domains = I915_GEM_DOMAIN_COMMAND;
94 }
95 obj.relocation_count = i;
96 obj.relocs_ptr = (uintptr_t)reloc;
97 gem_execbuf(fd, &execbuf);
98
99 igt_info("Batch is now at offset %#llx, max GTT %#llx\n",
100 (long long)obj.offset, (long long)gtt_max);
101
102 offsets = gem_mmap__cpu(fd, obj.handle, 0, 8192, PROT_READ);
103 gem_set_domain(fd, obj.handle, I915_GEM_DOMAIN_CPU, 0);
104 gem_close(fd, obj.handle);
105
106 for (i = 0; i < ARRAY_SIZE(reloc); i++)
107 igt_assert_f(offsets[1 + i] < gtt_max,
108 "Offset[%d]=%#llx, expected less than %#llx\n",
109 i, (long long)offsets[i+i], (long long)gtt_max);
110 munmap(offsets, 8192);
111 }
112
negative_reloc_blt(int fd)113 static void negative_reloc_blt(int fd)
114 {
115 const int gen = intel_gen(intel_get_drm_devid(fd));
116 struct drm_i915_gem_execbuffer2 execbuf;
117 struct drm_i915_gem_exec_object2 obj[1024][2];
118 struct drm_i915_gem_relocation_entry reloc;
119 uint32_t buf[1024], *b;
120 int i;
121
122 memset(&reloc, 0, sizeof(reloc));
123 reloc.offset = 4 * sizeof(uint32_t);
124 reloc.presumed_offset = ~0ULL;
125 reloc.delta = -4096;
126 reloc.target_handle = 0;
127 reloc.read_domains = I915_GEM_DOMAIN_RENDER;
128 reloc.write_domain = I915_GEM_DOMAIN_RENDER;
129
130 for (i = 0; i < 1024; i++) {
131 memset(obj[i], 0, sizeof(obj[i]));
132
133 obj[i][0].handle = gem_create(fd, 4096);
134 obj[i][0].flags = EXEC_OBJECT_NEEDS_FENCE;
135
136 b = buf;
137 *b++ = XY_COLOR_BLT_CMD_NOLEN |
138 ((gen >= 8) ? 5 : 4) |
139 COLOR_BLT_WRITE_ALPHA | XY_COLOR_BLT_WRITE_RGB;
140 *b++ = 0xf0 << 16 | 1 << 25 | 1 << 24 | 4096;
141 *b++ = 1 << 16 | 0;
142 *b++ = 2 << 16 | 1024;
143 *b++ = ~0;
144 if (gen >= 8)
145 *b++ = ~0;
146 *b++ = 0xc0ffee ^ i;
147 *b++ = MI_BATCH_BUFFER_END;
148 if ((b - buf) & 1)
149 *b++ = 0;
150
151 obj[i][1].handle = gem_create(fd, 4096);
152 gem_write(fd, obj[i][1].handle, 0, buf, (b - buf) * sizeof(uint32_t));
153 obj[i][1].relocation_count = 1;
154 obj[i][1].relocs_ptr = (uintptr_t)&reloc;
155 }
156
157 memset(&execbuf, 0, sizeof(execbuf));
158 execbuf.buffer_count = 2;
159 execbuf.batch_len = (b - buf) * sizeof(uint32_t);
160 execbuf.flags = USE_LUT;
161 if (gen >= 6)
162 execbuf.flags |= I915_EXEC_BLT;
163
164 for (i = 0; i < 1024; i++) {
165 execbuf.buffers_ptr = (uintptr_t)obj[i];
166 gem_execbuf(fd, &execbuf);
167 }
168
169 for (i = 1024; i--;) {
170 gem_read(fd, obj[i][0].handle,
171 i*sizeof(uint32_t), buf + i, sizeof(uint32_t));
172 gem_close(fd, obj[i][0].handle);
173 gem_close(fd, obj[i][1].handle);
174 }
175
176 if (0) {
177 for (i = 0; i < 1024; i += 8)
178 igt_info("%08x %08x %08x %08x %08x %08x %08x %08x\n",
179 buf[i + 0], buf[i + 1], buf[i + 2], buf[i + 3],
180 buf[i + 4], buf[i + 5], buf[i + 6], buf[i + 7]);
181 }
182 for (i = 0; i < 1024; i++)
183 igt_assert_eq(buf[i], 0xc0ffee ^ i);
184 }
185
186 igt_main
187 {
188 const struct intel_execution_engine *e;
189 int fd = -1;
190
191 igt_fixture {
192 fd = drm_open_driver(DRIVER_INTEL);
193 igt_require_gem(fd);
194 }
195
196 for (e = intel_execution_engines; e->name; e++) {
197 igt_subtest_f("negative-reloc-%s", e->name)
198 negative_reloc(fd, e->exec_id | e->flags, 0);
199
200 igt_subtest_f("negative-reloc-lut-%s", e->name)
201 negative_reloc(fd, e->exec_id | e->flags, USE_LUT);
202 }
203
204 igt_subtest("negative-reloc-bltcopy")
205 negative_reloc_blt(fd);
206
207 igt_fixture
208 close(fd);
209 }
210