gem_set_tiling_vs_blt.c 8.01 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
/*
 * Copyright © 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:
 *    Daniel Vetter <daniel.vetter@ffwll.ch>
 *
 */

28
/** @file gem_set_tiling_vs_blt.c
29
 *
30 31
 * Testcase: Check for proper synchronization of tiling changes vs. tiled gpu
 * access
32 33 34 35 36 37 38 39 40 41
 *
 * The blitter on gen3 and earlier needs properly set up fences. Which also
 * means that for untiled blits we may not set up a fence before that blt has
 * finished.
 *
 * Current kernels have a bug there, but it's pretty hard to hit because you
 * need:
 * - a blt on an untiled object which is aligned correctly for tiling.
 * - a set_tiling to switch that object to tiling
 * - another blt without any intervening cpu access that uses this object.
42 43 44
 *
 * Testcase has been extended to also check tiled->untiled and tiled->tiled
 * transitions (i.e. changing stride).
45 46
 */

47
#include "igt.h"
48 49 50 51 52 53 54 55
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <fcntl.h>
#include <inttypes.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/time.h>
56
#include <stdbool.h>
57 58 59
#include "drm.h"
#include "intel_bufmgr.h"

60 61 62
IGT_TEST_DESCRIPTION("Check for proper synchronization of tiling changes vs."
		     " tiled gpu access.");

63 64 65 66 67 68
static drm_intel_bufmgr *bufmgr;
struct intel_batchbuffer *batch;
uint32_t devid;

#define TEST_SIZE (1024*1024)
#define TEST_STRIDE (4*1024)
69 70
#define TEST_HEIGHT(stride)	(TEST_SIZE/(stride))
#define TEST_WIDTH(stride)	((stride)/4)
71

72 73 74 75
uint32_t data[TEST_SIZE/4];

static void do_test(uint32_t tiling, unsigned stride,
		    uint32_t tiling_after, unsigned stride_after)
76 77
{
	drm_intel_bo *busy_bo, *test_bo, *target_bo;
78
	int i, ret;
79 80
	uint32_t *ptr;
	uint32_t test_bo_handle;
81
	uint32_t blt_stride, blt_bits;
82
	bool tiling_changed = false;
83

84
	igt_info("filling ring .. ");
85 86 87
	busy_bo = drm_intel_bo_alloc(bufmgr, "busy bo bo", 16*1024*1024, 4096);

	for (i = 0; i < 250; i++) {
88
		BLIT_COPY_BATCH_START(0);
89 90 91 92 93 94 95 96 97 98 99
		OUT_BATCH((3 << 24) | /* 32 bits */
			  (0xcc << 16) | /* copy ROP */
			  2*1024*4);
		OUT_BATCH(0 << 16 | 1024);
		OUT_BATCH((2048) << 16 | (2048));
		OUT_RELOC_FENCED(busy_bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
		OUT_BATCH(0 << 16 | 0);
		OUT_BATCH(2*1024*4);
		OUT_RELOC_FENCED(busy_bo, I915_GEM_DOMAIN_RENDER, 0, 0);
		ADVANCE_BATCH();

100 101
		if (batch->gen >= 6) {
			BEGIN_BATCH(3, 0);
102 103 104 105 106 107 108 109
			OUT_BATCH(XY_SETUP_CLIP_BLT_CMD);
			OUT_BATCH(0);
			OUT_BATCH(0);
			ADVANCE_BATCH();
		}
	}
	intel_batchbuffer_flush(batch);

110
	igt_info("playing tricks .. ");
111 112 113 114
	/* first allocate the target so it gets out of the way of playing funky
	 * tricks */
	target_bo = drm_intel_bo_alloc(bufmgr, "target bo", TEST_SIZE, 4096);

115 116
	/* allocate buffer with parameters _after_ transition we want to check
	 * and touch it, so that it's properly aligned in the gtt. */
117 118
	test_bo = drm_intel_bo_alloc(bufmgr, "tiled busy bo", TEST_SIZE, 4096);
	test_bo_handle = test_bo->handle;
119
	ret = drm_intel_bo_set_tiling(test_bo, &tiling_after, stride_after);
120
	igt_assert_eq(ret, 0);
121 122 123
	drm_intel_gem_bo_map_gtt(test_bo);
	ptr = test_bo->virtual;
	*ptr = 0;
124
	ptr = NULL;
125 126 127 128 129 130 131 132 133
	drm_intel_gem_bo_unmap_gtt(test_bo);

	drm_intel_bo_unreference(test_bo);

	test_bo = NULL;

	/* note we need a bo bigger than batches, otherwise the buffer reuse
	 * trick will fail. */
	test_bo = drm_intel_bo_alloc(bufmgr, "busy bo", TEST_SIZE, 4096);
134 135 136
	/* double check that the reuse trick worked */
	igt_assert(test_bo_handle == test_bo->handle);

137
	test_bo_handle = test_bo->handle;
138 139
	/* ensure we have the right tiling before we start. */
	ret = drm_intel_bo_set_tiling(test_bo, &tiling, stride);
140
	igt_assert_eq(ret, 0);
141

142 143 144 145 146 147 148 149 150
	if (tiling == I915_TILING_NONE) {
		drm_intel_bo_subdata(test_bo, 0, TEST_SIZE, data);
	} else {
		drm_intel_gem_bo_map_gtt(test_bo);
		ptr = test_bo->virtual;
		memcpy(ptr, data, TEST_SIZE);
		ptr = NULL;
		drm_intel_gem_bo_unmap_gtt(test_bo);
	}
151

152 153 154 155 156 157 158
	blt_stride = stride;
	blt_bits = 0;
	if (intel_gen(devid) >= 4 && tiling != I915_TILING_NONE) {
		blt_stride /= 4;
		blt_bits = XY_SRC_COPY_BLT_SRC_TILED;
	}

159
	BLIT_COPY_BATCH_START(blt_bits);
160 161
	OUT_BATCH((3 << 24) | /* 32 bits */
		  (0xcc << 16) | /* copy ROP */
162
		  stride);
163
	OUT_BATCH(0 << 16 | 0);
164
	OUT_BATCH((TEST_HEIGHT(stride)) << 16 | (TEST_WIDTH(stride)));
165 166
	OUT_RELOC_FENCED(target_bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
	OUT_BATCH(0 << 16 | 0);
167
	OUT_BATCH(blt_stride);
168 169 170 171 172 173 174
	OUT_RELOC_FENCED(test_bo, I915_GEM_DOMAIN_RENDER, 0, 0);
	ADVANCE_BATCH();
	intel_batchbuffer_flush(batch);

	drm_intel_bo_unreference(test_bo);

	test_bo = drm_intel_bo_alloc_for_render(bufmgr, "tiled busy bo", TEST_SIZE, 4096);
175 176
	/* double check that the reuse trick worked */
	igt_assert(test_bo_handle == test_bo->handle);
177
	ret = drm_intel_bo_set_tiling(test_bo, &tiling_after, stride_after);
178
	igt_assert_eq(ret, 0);
179 180 181 182

	/* Note: We don't care about gen4+ here because the blitter doesn't use
	 * fences there. So not setting tiling flags on the tiled buffer is ok.
	 */
183
	BLIT_COPY_BATCH_START(0);
184 185
	OUT_BATCH((3 << 24) | /* 32 bits */
		  (0xcc << 16) | /* copy ROP */
186
		  stride_after);
187 188 189 190
	OUT_BATCH(0 << 16 | 0);
	OUT_BATCH((1) << 16 | (1));
	OUT_RELOC_FENCED(test_bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
	OUT_BATCH(0 << 16 | 0);
191
	OUT_BATCH(stride_after);
192 193 194 195
	OUT_RELOC_FENCED(test_bo, I915_GEM_DOMAIN_RENDER, 0, 0);
	ADVANCE_BATCH();
	intel_batchbuffer_flush(batch);

196 197
	/* Now try to trick the kernel the kernel into changing up the fencing
	 * too early. */
198

199
	igt_info("checking .. ");
200 201 202
	memset(data, 0, TEST_SIZE);
	drm_intel_bo_get_subdata(target_bo, 0, TEST_SIZE, data);
	for (i = 0; i < TEST_SIZE/4; i++)
203
		igt_assert(data[i] == i);
204

205 206 207 208 209 210 211 212
	/* check whether tiling on the test_bo actually changed. */
	drm_intel_gem_bo_map_gtt(test_bo);
	ptr = test_bo->virtual;
	for (i = 0; i < TEST_SIZE/4; i++)
		if (ptr[i] != data[i])
			tiling_changed = true;
	ptr = NULL;
	drm_intel_gem_bo_unmap_gtt(test_bo);
213
	igt_assert(tiling_changed);
214 215 216 217

	drm_intel_bo_unreference(test_bo);
	drm_intel_bo_unreference(target_bo);
	drm_intel_bo_unreference(busy_bo);
218
	igt_info("done\n");
219 220
}

221 222
int fd;

Daniel Vetter's avatar
Daniel Vetter committed
223
igt_main
224
{
225
	int i;
226 227
	uint32_t tiling, tiling_after;

Daniel Vetter's avatar
Daniel Vetter committed
228
	igt_skip_on_simulation();
229

230 231 232
	igt_fixture {
		for (i = 0; i < 1024*256; i++)
			data[i] = i;
233

234
		fd = drm_open_driver(DRIVER_INTEL);
235
		igt_require_gem(fd);
236

237 238 239 240 241
		bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
		drm_intel_bufmgr_gem_enable_reuse(bufmgr);
		devid = intel_get_drm_devid(fd);
		batch = intel_batchbuffer_alloc(bufmgr, devid);
	}
242

Daniel Vetter's avatar
Daniel Vetter committed
243
	igt_subtest("untiled-to-tiled") {
244 245 246
		tiling = I915_TILING_NONE;
		tiling_after = I915_TILING_X;
		do_test(tiling, TEST_STRIDE, tiling_after, TEST_STRIDE);
247 248
		igt_assert(tiling == I915_TILING_NONE);
		igt_assert(tiling_after == I915_TILING_X);
249
	}
250

Daniel Vetter's avatar
Daniel Vetter committed
251
	igt_subtest("tiled-to-untiled") {
252 253 254
		tiling = I915_TILING_X;
		tiling_after = I915_TILING_NONE;
		do_test(tiling, TEST_STRIDE, tiling_after, TEST_STRIDE);
255 256
		igt_assert(tiling == I915_TILING_X);
		igt_assert(tiling_after == I915_TILING_NONE);
257 258
	}

Daniel Vetter's avatar
Daniel Vetter committed
259
	igt_subtest("tiled-to-tiled") {
260 261 262
		tiling = I915_TILING_X;
		tiling_after = I915_TILING_X;
		do_test(tiling, TEST_STRIDE/2, tiling_after, TEST_STRIDE);
263 264
		igt_assert(tiling == I915_TILING_X);
		igt_assert(tiling_after == I915_TILING_X);
265
	}
266
}