gem_exec_big.c 7.75 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34
/*
 * Copyright © 2011,2012 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 * Authors:
 *    Chris Wilson <chris@chris-wilson.co.uk>
 *    Daniel Vetter <daniel.vetter@ffwll.ch>
 *
 */

/*
 * Testcase: run a nop batch which is really big
 *
 * Mostly useful to stress-test the error-capture code
 */

35
#include "igt.h"
36 37 38 39 40 41 42 43 44 45 46 47 48
#include <unistd.h>
#include <stdlib.h>
#include <stdint.h>
#include <stdio.h>
#include <string.h>
#include <fcntl.h>
#include <inttypes.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/ioctl.h>
#include <sys/time.h>
#include "drm.h"

49 50 51
IGT_TEST_DESCRIPTION("Run a large nop batch to stress test the error capture"
		     " code.");

52 53 54 55 56
#define FORCE_PREAD_PWRITE 0

static int use_64bit_relocs;

static void exec1(int fd, uint32_t handle, uint64_t reloc_ofs, unsigned flags, char *ptr)
57 58 59
{
	struct drm_i915_gem_execbuffer2 execbuf;
	struct drm_i915_gem_exec_object2 gem_exec[1];
60
	struct drm_i915_gem_relocation_entry gem_reloc[1];
61

62 63 64 65
	gem_reloc[0].offset = reloc_ofs;
	gem_reloc[0].delta = 0;
	gem_reloc[0].target_handle = handle;
	gem_reloc[0].read_domains = I915_GEM_DOMAIN_RENDER;
66
	gem_reloc[0].write_domain = 0;
67 68
	gem_reloc[0].presumed_offset = 0;

69
	gem_exec[0].handle = handle;
70
	gem_exec[0].relocation_count = 1;
71
	gem_exec[0].relocs_ptr = to_user_pointer(gem_reloc);
72 73 74 75 76 77
	gem_exec[0].alignment = 0;
	gem_exec[0].offset = 0;
	gem_exec[0].flags = 0;
	gem_exec[0].rsvd1 = 0;
	gem_exec[0].rsvd2 = 0;

78
	execbuf.buffers_ptr = to_user_pointer(gem_exec);
79 80 81 82 83 84 85
	execbuf.buffer_count = 1;
	execbuf.batch_start_offset = 0;
	execbuf.batch_len = 8;
	execbuf.cliprects_ptr = 0;
	execbuf.num_cliprects = 0;
	execbuf.DR1 = 0;
	execbuf.DR4 = 0;
86
	execbuf.flags = flags;
87 88 89
	i915_execbuffer2_set_context_id(execbuf, 0);
	execbuf.rsvd2 = 0;

90 91 92 93 94
	/* Avoid hitting slowpaths in the reloc processing which might yield a
	 * presumed_offset of -1. Happens when the batch is still busy from the
	 * last round. */
	gem_sync(fd, handle);

95
	gem_execbuf(fd, &execbuf);
96

97 98
	igt_warn_on(gem_reloc[0].presumed_offset == -1);

99 100 101 102 103 104 105
	if (use_64bit_relocs) {
		uint64_t tmp;
		if (ptr)
			tmp = *(uint64_t *)(ptr+reloc_ofs);
		else
			gem_read(fd, handle, reloc_ofs, &tmp, sizeof(tmp));
		igt_assert_eq(tmp, gem_reloc[0].presumed_offset);
106
	} else {
107 108 109 110 111
		uint32_t tmp;
		if (ptr)
			tmp = *(uint32_t *)(ptr+reloc_ofs);
		else
			gem_read(fd, handle, reloc_ofs, &tmp, sizeof(tmp));
112 113
		igt_assert_eq(tmp, gem_reloc[0].presumed_offset);
	}
114 115
}

116 117 118 119 120 121 122 123 124 125 126 127
static void xchg_reloc(void *array, unsigned i, unsigned j)
{
	struct drm_i915_gem_relocation_entry *reloc = array;
	struct drm_i915_gem_relocation_entry *a = &reloc[i];
	struct drm_i915_gem_relocation_entry *b = &reloc[j];
	struct drm_i915_gem_relocation_entry tmp;

	tmp = *a;
	*a = *b;
	*b = tmp;
}

128
static void execN(int fd, uint32_t handle, uint64_t batch_size, unsigned flags, char *ptr)
129 130 131 132 133
{
#define reloc_ofs(N, T) ((((N)+1) << 12) - 4*(1 + ((N) == ((T)-1))))
	struct drm_i915_gem_execbuffer2 execbuf;
	struct drm_i915_gem_exec_object2 gem_exec[1];
	struct drm_i915_gem_relocation_entry *gem_reloc;
134
	uint64_t n, nreloc = batch_size >> 12;
135 136 137 138 139 140 141 142

	gem_reloc = calloc(nreloc, sizeof(*gem_reloc));
	igt_assert(gem_reloc);

	for (n = 0; n < nreloc; n++) {
		gem_reloc[n].offset = reloc_ofs(n, nreloc);
		gem_reloc[n].target_handle = handle;
		gem_reloc[n].read_domains = I915_GEM_DOMAIN_RENDER;
143 144 145 146 147 148 149 150
		gem_reloc[n].presumed_offset = n ^ 0xbeefdeaddeadbeef;
		if (ptr) {
			if (use_64bit_relocs)
				*(uint64_t *)(ptr + gem_reloc[n].offset) = gem_reloc[n].presumed_offset;
			else
				*(uint32_t *)(ptr + gem_reloc[n].offset) = gem_reloc[n].presumed_offset;
		} else
			gem_write(fd, handle, gem_reloc[n].offset, &gem_reloc[n].presumed_offset, 4*(1+use_64bit_relocs));
151 152 153 154 155
	}

	memset(gem_exec, 0, sizeof(gem_exec));
	gem_exec[0].handle = handle;
	gem_exec[0].relocation_count = nreloc;
156
	gem_exec[0].relocs_ptr = to_user_pointer(gem_reloc);
157 158

	memset(&execbuf, 0, sizeof(execbuf));
159
	execbuf.buffers_ptr = to_user_pointer(gem_exec);
160 161 162 163 164 165 166 167 168 169
	execbuf.buffer_count = 1;
	execbuf.batch_start_offset = 0;
	execbuf.batch_len = 8;
	execbuf.flags = flags;

	/* Avoid hitting slowpaths in the reloc processing which might yield a
	 * presumed_offset of -1. Happens when the batch is still busy from the
	 * last round. */
	gem_sync(fd, handle);

170 171
	igt_permute_array(gem_reloc, nreloc, xchg_reloc);

172 173 174 175
	gem_execbuf(fd, &execbuf);
	for (n = 0; n < nreloc; n++)
		igt_warn_on(gem_reloc[n].presumed_offset == -1);

176 177 178 179 180 181 182 183 184
	if (use_64bit_relocs) {
		for (n = 0; n < nreloc; n++) {
			uint64_t tmp;
			if (ptr)
				tmp = *(uint64_t *)(ptr+reloc_ofs(n, nreloc));
			else
				gem_read(fd, handle, reloc_ofs(n, nreloc), &tmp, sizeof(tmp));
			igt_assert_eq(tmp, gem_reloc[n].presumed_offset);
		}
185 186
	} else {
		for (n = 0; n < nreloc; n++) {
187 188 189 190 191
			uint32_t tmp;
			if (ptr)
				tmp = *(uint32_t *)(ptr+reloc_ofs(n, nreloc));
			else
				gem_read(fd, handle, reloc_ofs(n, nreloc), &tmp, sizeof(tmp));
192 193
			igt_assert_eq(tmp, gem_reloc[n].presumed_offset);
		}
194 195 196 197 198 199
	}

	free(gem_reloc);
#undef reloc_ofs
}

200
igt_simple_main
201 202
{
	uint32_t batch[2] = {MI_BATCH_BUFFER_END};
203
	uint64_t batch_size, max, ggtt_max, reloc_ofs;
204
	int fd;
205

206
	fd = drm_open_driver(DRIVER_INTEL);
207 208
	igt_require_gem(fd);

209
	use_64bit_relocs = intel_gen(intel_get_drm_devid(fd)) >= 8;
210

211
	max = 3 * gem_aperture_size(fd) / 4;
212
	ggtt_max = 3 * gem_global_aperture_size(fd) / 4;
213
	intel_require_memory(1, max, CHECK_RAM);
214

215
	for (batch_size = 4096; batch_size <= max; ) {
216 217
		uint32_t handle;
		void *ptr;
218 219

		handle = gem_create(fd, batch_size);
220
		gem_write(fd, handle, 0, batch, sizeof(batch));
221

222
		if (!FORCE_PREAD_PWRITE && gem_has_llc(fd))
223
			ptr = __gem_mmap__cpu(fd, handle, 0, batch_size, PROT_READ);
224
		else if (!FORCE_PREAD_PWRITE && gem_mmap__has_wc(fd))
225
			ptr = __gem_mmap__wc(fd, handle, 0, batch_size, PROT_READ);
226 227
		else
			ptr = NULL;
228

229
		igt_debug("Forwards (%lld)\n", (long long)batch_size);
230
		for (reloc_ofs = 4096; reloc_ofs < batch_size; reloc_ofs += 4096) {
231 232
			igt_debug("batch_size %llu, reloc_ofs %llu\n",
				  (long long)batch_size, (long long)reloc_ofs);
233
			exec1(fd, handle, reloc_ofs, 0, ptr);
234 235
			if (batch_size < ggtt_max)
				exec1(fd, handle, reloc_ofs, I915_EXEC_SECURE, ptr);
236
		}
237

238 239 240 241 242
		igt_debug("Backwards (%lld)\n", (long long)batch_size);
		for (reloc_ofs = batch_size - 4096; reloc_ofs; reloc_ofs -= 4096) {
			igt_debug("batch_size %llu, reloc_ofs %llu\n",
				  (long long)batch_size, (long long)reloc_ofs);
			exec1(fd, handle, reloc_ofs, 0, ptr);
243 244
			if (batch_size < ggtt_max)
				exec1(fd, handle, reloc_ofs, I915_EXEC_SECURE, ptr);
245 246 247
		}

		igt_debug("Random (%lld)\n", (long long)batch_size);
248
		execN(fd, handle, batch_size, 0, ptr);
249 250
		if (batch_size < ggtt_max)
			execN(fd, handle, batch_size, I915_EXEC_SECURE, ptr);
251

252 253
		if (ptr)
			munmap(ptr, batch_size);
254 255 256 257 258 259 260
		gem_madvise(fd, handle, I915_MADV_DONTNEED);

		if (batch_size < max && 2*batch_size > max)
			batch_size = max;
		else
			batch_size *= 2;
	}
261 262 263

	close(fd);
}