deadlock_tests.c revision 6532f28e
1/*
2 * Copyright 2017 Advanced Micro Devices, Inc.
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 shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22*/
23
24#include <stdio.h>
25#include <stdlib.h>
26#include <unistd.h>
27#ifdef HAVE_ALLOCA_H
28# include <alloca.h>
29#endif
30
31#include "CUnit/Basic.h"
32
33#include "amdgpu_test.h"
34#include "amdgpu_drm.h"
35#include "amdgpu_internal.h"
36
37#include <pthread.h>
38
39
40/*
41 * This defines the delay in MS after which memory location designated for
42 * compression against reference value is written to, unblocking command
43 * processor
44 */
45#define WRITE_MEM_ADDRESS_DELAY_MS 100
46
47#define	PACKET_TYPE3	3
48
49#define PACKET3(op, n)	((PACKET_TYPE3 << 30) |				\
50			 (((op) & 0xFF) << 8) |				\
51			 ((n) & 0x3FFF) << 16)
52
53#define	PACKET3_WAIT_REG_MEM				0x3C
54#define		WAIT_REG_MEM_FUNCTION(x)                ((x) << 0)
55		/* 0 - always
56		 * 1 - <
57		 * 2 - <=
58		 * 3 - ==
59		 * 4 - !=
60		 * 5 - >=
61		 * 6 - >
62		 */
63#define		WAIT_REG_MEM_MEM_SPACE(x)               ((x) << 4)
64		/* 0 - reg
65		 * 1 - mem
66		 */
67#define		WAIT_REG_MEM_OPERATION(x)               ((x) << 6)
68		/* 0 - wait_reg_mem
69		 * 1 - wr_wait_wr_reg
70		 */
71#define		WAIT_REG_MEM_ENGINE(x)                  ((x) << 8)
72		/* 0 - me
73		 * 1 - pfp
74		 */
75
76#define	PACKET3_WRITE_DATA				0x37
77#define		WRITE_DATA_DST_SEL(x)                   ((x) << 8)
78		/* 0 - register
79		 * 1 - memory (sync - via GRBM)
80		 * 2 - gl2
81		 * 3 - gds
82		 * 4 - reserved
83		 * 5 - memory (async - direct)
84		 */
85#define		WR_ONE_ADDR                             (1 << 16)
86#define		WR_CONFIRM                              (1 << 20)
87#define		WRITE_DATA_CACHE_POLICY(x)              ((x) << 25)
88		/* 0 - LRU
89		 * 1 - Stream
90		 */
91#define		WRITE_DATA_ENGINE_SEL(x)                ((x) << 30)
92		/* 0 - me
93		 * 1 - pfp
94		 * 2 - ce
95		 */
96
97#define mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR                                      0x54f
98
99static  amdgpu_device_handle device_handle;
100static  uint32_t  major_version;
101static  uint32_t  minor_version;
102
103static pthread_t stress_thread;
104static uint32_t *ptr;
105
106int use_uc_mtype = 0;
107
108static void amdgpu_deadlock_helper(unsigned ip_type);
109static void amdgpu_deadlock_gfx(void);
110static void amdgpu_deadlock_compute(void);
111static void amdgpu_illegal_reg_access();
112static void amdgpu_illegal_mem_access();
113
114CU_BOOL suite_deadlock_tests_enable(void)
115{
116	CU_BOOL enable = CU_TRUE;
117
118	if (amdgpu_device_initialize(drm_amdgpu[0], &major_version,
119					     &minor_version, &device_handle))
120		return CU_FALSE;
121
122	/*
123	 * Only enable for ASICs supporting GPU reset and for which it's enabled
124	 * by default (currently GFX8/9 dGPUS)
125	 */
126	if (device_handle->info.family_id != AMDGPU_FAMILY_VI &&
127	    device_handle->info.family_id != AMDGPU_FAMILY_AI &&
128	    device_handle->info.family_id != AMDGPU_FAMILY_CI) {
129		printf("\n\nGPU reset is not enabled for the ASIC, deadlock suite disabled\n");
130		enable = CU_FALSE;
131	}
132
133	if (device_handle->info.family_id >= AMDGPU_FAMILY_AI)
134		use_uc_mtype = 1;
135
136	if (amdgpu_device_deinitialize(device_handle))
137		return CU_FALSE;
138
139	return enable;
140}
141
142int suite_deadlock_tests_init(void)
143{
144	int r;
145
146	r = amdgpu_device_initialize(drm_amdgpu[0], &major_version,
147				   &minor_version, &device_handle);
148
149	if (r) {
150		if ((r == -EACCES) && (errno == EACCES))
151			printf("\n\nError:%s. "
152				"Hint:Try to run this test program as root.",
153				strerror(errno));
154		return CUE_SINIT_FAILED;
155	}
156
157	return CUE_SUCCESS;
158}
159
160int suite_deadlock_tests_clean(void)
161{
162	int r = amdgpu_device_deinitialize(device_handle);
163
164	if (r == 0)
165		return CUE_SUCCESS;
166	else
167		return CUE_SCLEAN_FAILED;
168}
169
170
171CU_TestInfo deadlock_tests[] = {
172	{ "gfx ring block test (set amdgpu.lockup_timeout=50)", amdgpu_deadlock_gfx },
173	{ "compute ring block test (set amdgpu.lockup_timeout=50)", amdgpu_deadlock_compute },
174	{ "illegal reg access test", amdgpu_illegal_reg_access },
175	{ "illegal mem access test (set amdgpu.vm_fault_stop=2)", amdgpu_illegal_mem_access },
176	CU_TEST_INFO_NULL,
177};
178
179static void *write_mem_address(void *data)
180{
181	int i;
182
183	/* useconds_t range is [0, 1,000,000] so use loop for waits > 1s */
184	for (i = 0; i < WRITE_MEM_ADDRESS_DELAY_MS; i++)
185		usleep(1000);
186
187	ptr[256] = 0x1;
188
189	return 0;
190}
191
192static void amdgpu_deadlock_gfx(void)
193{
194	amdgpu_deadlock_helper(AMDGPU_HW_IP_GFX);
195}
196
197static void amdgpu_deadlock_compute(void)
198{
199	amdgpu_deadlock_helper(AMDGPU_HW_IP_COMPUTE);
200}
201
202static void amdgpu_deadlock_helper(unsigned ip_type)
203{
204	amdgpu_context_handle context_handle;
205	amdgpu_bo_handle ib_result_handle;
206	void *ib_result_cpu;
207	uint64_t ib_result_mc_address;
208	struct amdgpu_cs_request ibs_request;
209	struct amdgpu_cs_ib_info ib_info;
210	struct amdgpu_cs_fence fence_status;
211	uint32_t expired;
212	int i, r;
213	amdgpu_bo_list_handle bo_list;
214	amdgpu_va_handle va_handle;
215
216	r = pthread_create(&stress_thread, NULL, write_mem_address, NULL);
217	CU_ASSERT_EQUAL(r, 0);
218
219	r = amdgpu_cs_ctx_create(device_handle, &context_handle);
220	CU_ASSERT_EQUAL(r, 0);
221
222	r = amdgpu_bo_alloc_and_map_raw(device_handle, 4096, 4096,
223			AMDGPU_GEM_DOMAIN_GTT, 0, use_uc_mtype ? AMDGPU_VM_MTYPE_UC : 0,
224						    &ib_result_handle, &ib_result_cpu,
225						    &ib_result_mc_address, &va_handle);
226	CU_ASSERT_EQUAL(r, 0);
227
228	r = amdgpu_get_bo_list(device_handle, ib_result_handle, NULL,
229			       &bo_list);
230	CU_ASSERT_EQUAL(r, 0);
231
232	ptr = ib_result_cpu;
233
234	ptr[0] = PACKET3(PACKET3_WAIT_REG_MEM, 5);
235	ptr[1] = (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
236			 WAIT_REG_MEM_FUNCTION(4) | /* != */
237			 WAIT_REG_MEM_ENGINE(0));  /* me */
238	ptr[2] = (ib_result_mc_address + 256*4) & 0xfffffffc;
239	ptr[3] = ((ib_result_mc_address + 256*4) >> 32) & 0xffffffff;
240	ptr[4] = 0x00000000; /* reference value */
241	ptr[5] = 0xffffffff; /* and mask */
242	ptr[6] = 0x00000004; /* poll interval */
243
244	for (i = 7; i < 16; ++i)
245		ptr[i] = 0xffff1000;
246
247
248	ptr[256] = 0x0; /* the memory we wait on to change */
249
250
251
252	memset(&ib_info, 0, sizeof(struct amdgpu_cs_ib_info));
253	ib_info.ib_mc_address = ib_result_mc_address;
254	ib_info.size = 16;
255
256	memset(&ibs_request, 0, sizeof(struct amdgpu_cs_request));
257	ibs_request.ip_type = ip_type;
258	ibs_request.ring = 0;
259	ibs_request.number_of_ibs = 1;
260	ibs_request.ibs = &ib_info;
261	ibs_request.resources = bo_list;
262	ibs_request.fence_info.handle = NULL;
263
264	for (i = 0; i < 200; i++) {
265		r = amdgpu_cs_submit(context_handle, 0,&ibs_request, 1);
266		CU_ASSERT_EQUAL((r == 0 || r == -ECANCELED), 1);
267
268	}
269
270	memset(&fence_status, 0, sizeof(struct amdgpu_cs_fence));
271	fence_status.context = context_handle;
272	fence_status.ip_type = ip_type;
273	fence_status.ip_instance = 0;
274	fence_status.ring = 0;
275	fence_status.fence = ibs_request.seq_no;
276
277	r = amdgpu_cs_query_fence_status(&fence_status,
278			AMDGPU_TIMEOUT_INFINITE,0, &expired);
279	CU_ASSERT_EQUAL((r == 0 || r == -ECANCELED), 1);
280
281	pthread_join(stress_thread, NULL);
282
283	r = amdgpu_bo_list_destroy(bo_list);
284	CU_ASSERT_EQUAL(r, 0);
285
286	r = amdgpu_bo_unmap_and_free(ib_result_handle, va_handle,
287				     ib_result_mc_address, 4096);
288	CU_ASSERT_EQUAL(r, 0);
289
290	r = amdgpu_cs_ctx_free(context_handle);
291	CU_ASSERT_EQUAL(r, 0);
292}
293
294static void bad_access_helper(int reg_access)
295{
296	amdgpu_context_handle context_handle;
297	amdgpu_bo_handle ib_result_handle;
298	void *ib_result_cpu;
299	uint64_t ib_result_mc_address;
300	struct amdgpu_cs_request ibs_request;
301	struct amdgpu_cs_ib_info ib_info;
302	struct amdgpu_cs_fence fence_status;
303	uint32_t expired;
304	int i, r;
305	amdgpu_bo_list_handle bo_list;
306	amdgpu_va_handle va_handle;
307
308	r = amdgpu_cs_ctx_create(device_handle, &context_handle);
309	CU_ASSERT_EQUAL(r, 0);
310
311	r = amdgpu_bo_alloc_and_map_raw(device_handle, 4096, 4096,
312			AMDGPU_GEM_DOMAIN_GTT, 0, 0,
313							&ib_result_handle, &ib_result_cpu,
314							&ib_result_mc_address, &va_handle);
315	CU_ASSERT_EQUAL(r, 0);
316
317	r = amdgpu_get_bo_list(device_handle, ib_result_handle, NULL,
318				   &bo_list);
319	CU_ASSERT_EQUAL(r, 0);
320
321	ptr = ib_result_cpu;
322	i = 0;
323
324	ptr[i++] = PACKET3(PACKET3_WRITE_DATA, 3);
325	ptr[i++] = (reg_access ? WRITE_DATA_DST_SEL(0) : WRITE_DATA_DST_SEL(5))| WR_CONFIRM;
326	ptr[i++] = reg_access ? mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR : 0xdeadbee0;
327	ptr[i++] = 0;
328	ptr[i++] = 0xdeadbeef;
329
330	for (; i < 16; ++i)
331		ptr[i] = 0xffff1000;
332
333	memset(&ib_info, 0, sizeof(struct amdgpu_cs_ib_info));
334	ib_info.ib_mc_address = ib_result_mc_address;
335	ib_info.size = 16;
336
337	memset(&ibs_request, 0, sizeof(struct amdgpu_cs_request));
338	ibs_request.ip_type = AMDGPU_HW_IP_GFX;
339	ibs_request.ring = 0;
340	ibs_request.number_of_ibs = 1;
341	ibs_request.ibs = &ib_info;
342	ibs_request.resources = bo_list;
343	ibs_request.fence_info.handle = NULL;
344
345	r = amdgpu_cs_submit(context_handle, 0,&ibs_request, 1);
346	CU_ASSERT_EQUAL((r == 0 || r == -ECANCELED), 1);
347
348
349	memset(&fence_status, 0, sizeof(struct amdgpu_cs_fence));
350	fence_status.context = context_handle;
351	fence_status.ip_type = AMDGPU_HW_IP_GFX;
352	fence_status.ip_instance = 0;
353	fence_status.ring = 0;
354	fence_status.fence = ibs_request.seq_no;
355
356	r = amdgpu_cs_query_fence_status(&fence_status,
357			AMDGPU_TIMEOUT_INFINITE,0, &expired);
358	CU_ASSERT_EQUAL((r == 0 || r == -ECANCELED), 1);
359
360	r = amdgpu_bo_list_destroy(bo_list);
361	CU_ASSERT_EQUAL(r, 0);
362
363	r = amdgpu_bo_unmap_and_free(ib_result_handle, va_handle,
364					 ib_result_mc_address, 4096);
365	CU_ASSERT_EQUAL(r, 0);
366
367	r = amdgpu_cs_ctx_free(context_handle);
368	CU_ASSERT_EQUAL(r, 0);
369}
370
371static void amdgpu_illegal_reg_access()
372{
373	bad_access_helper(1);
374}
375
376static void amdgpu_illegal_mem_access()
377{
378	bad_access_helper(0);
379}
380