From: Andrey Grodzovsky <Andrey.Grodzovsky@xxxxxxx>
Adding initial tests for locks detection when SW
scheduler FIFO is full.
The test works by submitting a batch of identical commands which make the CP
stall waiting for condition to become true. The condition is later satisfied
form a helper thread. Other events that happen during this time
might create deadlock situations. One such example is GPU reset
triggered by this stall when amdgpu_lockup_timeout != 0.
v2:
Increase the delay from 2 to 100 ms.
Comment out the compute test until it's working.
Typos fix.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky at amd.com>
Reviewed-by: Christian König <christian.koenig at amd.com>
---
tests/amdgpu/Makefile.am | 6 +-
tests/amdgpu/amdgpu_test.c | 6 ++
tests/amdgpu/amdgpu_test.h | 15 +++
tests/amdgpu/deadlock_tests.c | 245 ++++++++++++++++++++++++++++++++++++++++++
4 files changed, 270 insertions(+), 2 deletions(-)
create mode 100644 tests/amdgpu/deadlock_tests.c
diff --git a/tests/amdgpu/Makefile.am b/tests/amdgpu/Makefile.am
index 9c02fd6..8700c4d 100644
--- a/tests/amdgpu/Makefile.am
+++ b/tests/amdgpu/Makefile.am
@@ -1,7 +1,8 @@
AM_CFLAGS = \
-I $(top_srcdir)/include/drm \
-I $(top_srcdir)/amdgpu \
- -I $(top_srcdir)
+ -I $(top_srcdir) \
+ -pthread
LDADD = $(top_builddir)/libdrm.la \
$(top_builddir)/amdgpu/libdrm_amdgpu.la \
@@ -29,4 +30,5 @@ amdgpu_test_SOURCES = \
frame.h \
uvd_enc_tests.c \
vcn_tests.c \
- uve_ib.h
+ uve_ib.h \
+ deadlock_tests.c
diff --git a/tests/amdgpu/amdgpu_test.c b/tests/amdgpu/amdgpu_test.c
index cd6b826..9925503 100644
--- a/tests/amdgpu/amdgpu_test.c
+++ b/tests/amdgpu/amdgpu_test.c
@@ -97,6 +97,12 @@ static CU_SuiteInfo suites[] = {
.pCleanupFunc = suite_uvd_enc_tests_clean,
.pTests = uvd_enc_tests,
},
+ {
+ .pName = "Deadlock Tests",
+ .pInitFunc = suite_deadlock_tests_init,
+ .pCleanupFunc = suite_deadlock_tests_clean,
+ .pTests = deadlock_tests,
+ },
CU_SUITE_INFO_NULL,
};
diff --git a/tests/amdgpu/amdgpu_test.h b/tests/amdgpu/amdgpu_test.h
index d0b61ba..ece93f4 100644
--- a/tests/amdgpu/amdgpu_test.h
+++ b/tests/amdgpu/amdgpu_test.h
@@ -135,6 +135,21 @@ int suite_uvd_enc_tests_clean();
extern CU_TestInfo uvd_enc_tests[];
/**
+ * Initialize deadlock test suite
+ */
+int suite_deadlock_tests_init();
+
+/**
+ * Deinitialize deadlock test suite
+ */
+int suite_deadlock_tests_clean();
+
+/**
+ * Tests in uvd enc test suite
+ */
+extern CU_TestInfo deadlock_tests[];
+
+/**
* Helper functions
*/
static inline amdgpu_bo_handle gpu_mem_alloc(
diff --git a/tests/amdgpu/deadlock_tests.c b/tests/amdgpu/deadlock_tests.c
new file mode 100644
index 0000000..e23d903
--- /dev/null
+++ b/tests/amdgpu/deadlock_tests.c
@@ -0,0 +1,245 @@
+/*
+ * Copyright 2017 Advanced Micro Devices, Inc.
+ *
+ * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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.
+ *
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#ifdef HAVE_ALLOCA_H
+# include <alloca.h>
+#endif
+
+#include "CUnit/Basic.h"
+
+#include "amdgpu_test.h"
+#include "amdgpu_drm.h"
+
+#include <pthread.h>
+
+
+/*
+ * This defines the delay in MS after which memory location designated for
+ * compression against reference value is written to, unblocking command
+ * processor
+ */
+#define WRITE_MEM_ADDRESS_DELAY_MS 100
+
+#define PACKET_TYPE3 3
+
+#define PACKET3(op, n) ((PACKET_TYPE3 << 30) | \
+ (((op) & 0xFF) << 8) | \
+ ((n) & 0x3FFF) << 16)
+
+#define PACKET3_WAIT_REG_MEM 0x3C
+#define WAIT_REG_MEM_FUNCTION(x) ((x) << 0)
+ /* 0 - always
+ * 1 - <
+ * 2 - <=
+ * 3 - ==
+ * 4 - !=
+ * 5 - >=
+ * 6 - >
+ */
+#define WAIT_REG_MEM_MEM_SPACE(x) ((x) << 4)
+ /* 0 - reg
+ * 1 - mem
+ */
+#define WAIT_REG_MEM_OPERATION(x) ((x) << 6)
+ /* 0 - wait_reg_mem
+ * 1 - wr_wait_wr_reg
+ */
+#define WAIT_REG_MEM_ENGINE(x) ((x) << 8)
+ /* 0 - me
+ * 1 - pfp
+ */
+
+static amdgpu_device_handle device_handle;
+static uint32_t major_version;
+static uint32_t minor_version;
+
+static pthread_t stress_thread;
+static uint32_t *ptr;
+
+static void amdgpu_deadlock_helper(unsigned ip_type);
+static void amdgpu_deadlock_gfx(void);
+static void amdgpu_deadlock_compute(void);
+
+int suite_deadlock_tests_init(void)
+{
+ struct amdgpu_gpu_info gpu_info = {0};
+ int r;
+
+ r = amdgpu_device_initialize(drm_amdgpu[0], &major_version,
+ &minor_version, &device_handle);
+
+ if (r) {
+ if ((r == -EACCES) && (errno == EACCES))
+ printf("\n\nError:%s. "
+ "Hint:Try to run this test program as root.",
+ strerror(errno));
+ return CUE_SINIT_FAILED;
+ }
+
+ return CUE_SUCCESS;
+}
+
+int suite_deadlock_tests_clean(void)
+{
+ int r = amdgpu_device_deinitialize(device_handle);
+
+ if (r == 0)
+ return CUE_SUCCESS;
+ else
+ return CUE_SCLEAN_FAILED;
+}
+
+
+CU_TestInfo deadlock_tests[] = {
+ { "gfx ring block test", amdgpu_deadlock_gfx },
+
+ /*
+ * BUG: Compute ring stalls and never recovers when the address is
+ * written after the command already submitted
+ */
+ /* { "compute ring block test", amdgpu_deadlock_compute }, */
+
+ CU_TEST_INFO_NULL,
+};
+
+static void *write_mem_address(void *data)
+{
+ int i;
+
+ /* useconds_t range is [0, 1,000,000] so use loop for waits > 1s */
+ for (i = 0; i < WRITE_MEM_ADDRESS_DELAY_MS; i++)
+ usleep(1000);
+
+ ptr[256] = 0x1;
+
+ return 0;
+}
+
+static void amdgpu_deadlock_gfx(void)
+{
+ amdgpu_deadlock_helper(AMDGPU_HW_IP_GFX);
+}
+
+static void amdgpu_deadlock_compute(void)
+{
+ amdgpu_deadlock_helper(AMDGPU_HW_IP_COMPUTE);
+}
+
+static void amdgpu_deadlock_helper(unsigned ip_type)
+{
+ amdgpu_context_handle context_handle;
+ amdgpu_bo_handle ib_result_handle;
+ void *ib_result_cpu;
+ uint64_t ib_result_mc_address;
+ struct amdgpu_cs_request ibs_request;
+ struct amdgpu_cs_ib_info ib_info;
+ struct amdgpu_cs_fence fence_status;
+ uint32_t expired;
+ int i, r, instance;
+ amdgpu_bo_list_handle bo_list;
+ amdgpu_va_handle va_handle;
+
+ r = pthread_create(&stress_thread, NULL, write_mem_address, NULL);
+ CU_ASSERT_EQUAL(r, 0);
+
+ r = amdgpu_cs_ctx_create(device_handle, &context_handle);
+ CU_ASSERT_EQUAL(r, 0);
+
+ r = amdgpu_bo_alloc_and_map(device_handle, 4096, 4096,
+ AMDGPU_GEM_DOMAIN_GTT, 0,
+ &ib_result_handle, &ib_result_cpu,
+ &ib_result_mc_address, &va_handle);
+ CU_ASSERT_EQUAL(r, 0);
+
+ r = amdgpu_get_bo_list(device_handle, ib_result_handle, NULL,
+ &bo_list);
+ CU_ASSERT_EQUAL(r, 0);
+
+ ptr = ib_result_cpu;
+
+ ptr[0] = PACKET3(PACKET3_WAIT_REG_MEM, 5);
+ ptr[1] = (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
+ WAIT_REG_MEM_FUNCTION(4) | /* != */
+ WAIT_REG_MEM_ENGINE(0)); /* me */
+ ptr[2] = (ib_result_mc_address + 256*4) & 0xfffffffc;
+ ptr[3] = ((ib_result_mc_address + 256*4) >> 32) & 0xffffffff;
+ ptr[4] = 0x00000000; /* reference value */
+ ptr[5] = 0xffffffff; /* and mask */
+ ptr[6] = 0x00000004; /* poll interval */
+
+ for (i = 7; i < 16; ++i)
+ ptr[i] = 0xffff1000;
+
+
+ ptr[256] = 0x0; /* the memory we wait on to change */
+
+
+
+ memset(&ib_info, 0, sizeof(struct amdgpu_cs_ib_info));
+ ib_info.ib_mc_address = ib_result_mc_address;
+ ib_info.size = 16;
+
+ memset(&ibs_request, 0, sizeof(struct amdgpu_cs_request));
+ ibs_request.ip_type = ip_type;
+ ibs_request.ring = 0;
+ ibs_request.number_of_ibs = 1;
+ ibs_request.ibs = &ib_info;
+ ibs_request.resources = bo_list;
+ ibs_request.fence_info.handle = NULL;
+
+ for (i = 0; i < 200; i++) {
+ r = amdgpu_cs_submit(context_handle, 0,&ibs_request, 1);
+ CU_ASSERT_EQUAL(r, 0);
+
+ }
+
+ memset(&fence_status, 0, sizeof(struct amdgpu_cs_fence));
+ fence_status.context = context_handle;
+ fence_status.ip_type = ip_type;
+ fence_status.ip_instance = 0;
+ fence_status.ring = 0;
+ fence_status.fence = ibs_request.seq_no;
+
+ r = amdgpu_cs_query_fence_status(&fence_status,
+ AMDGPU_TIMEOUT_INFINITE,0, &expired);
+ CU_ASSERT_EQUAL(r, 0);
+
+ r = amdgpu_bo_list_destroy(bo_list);
+ CU_ASSERT_EQUAL(r, 0);
+
+ r = amdgpu_bo_unmap_and_free(ib_result_handle, va_handle,
+ ib_result_mc_address, 4096);
+ CU_ASSERT_EQUAL(r, 0);
+
+ r = amdgpu_cs_ctx_free(context_handle);
+ CU_ASSERT_EQUAL(r, 0);
+
+ pthread_join(stress_thread, NULL);
+}
--
2.7.4
