[igt-dev] [PATCH 0/3] igt tests for hot-unplug feature in amdgpu

[igt-dev] [PATCH 0/3] igt tests for hot-unplug feature in amdgpu

[Andrey Grodzovsky]

Adding some code to acompany the recently added hot-unplug
feature[1]. While the code is amdgpu specific it was suggested
by Daniel to add this to IGT hot unplug test[2]. There is also
parallel RFC for this in libdrm[3].

[1] - https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.14-AMDGPU-Hot-Unplug
[2] - https://lists.freedeskt op.org/archives/amd-gfx/2021-January/058646.html
[3] - https://www.spinics.net/lists/dri-devel/msg300998.html

Andrey Grodzovsky (3):
  tests/core_hotunplug: Add command submission for amdgpu
  tests/core_hotunplug: Add hotunplug with exported dma-buf test.
  tests/core_hotunplug: Add hotunplug with exported sync obj test

 tests/Makefile.am      |   3 +
 tests/core_hotunplug.c | 364 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   2 +-
 3 files changed, 368 insertions(+), 1 deletion(-)

-- 
2.25.1

_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

[igt-dev] [PATCH 1/3] tests/core_hotunplug: Add command submission for amdgpu

[Andrey Grodzovsky]

Enables better testing of hot remove

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
---
 tests/Makefile.am      |   3 +
 tests/core_hotunplug.c | 200 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   2 +-
 3 files changed, 204 insertions(+), 1 deletion(-)

diff --git a/tests/Makefile.am b/tests/Makefile.am
index a6c807d8..0e6a3f2f 100644
--- a/tests/Makefile.am
+++ b/tests/Makefile.am
@@ -150,4 +150,7 @@ amdgpu_amd_prime_LDADD = $(LDADD) $(DRM_AMDGPU_LIBS)
 amdgpu_amd_info_CFLAGS = $(AM_CFLAGS) $(DRM_AMDGPU_CFLAGS)
 amdgpu_amd_info_LDADD = $(LDADD) $(DRM_AMDGPU_LIBS)
 
+core_hotunplug_CFLAGS = $(AM_CFLAGS) $(DRM_AMDGPU_CFLAGS)
+core_hotunplug_LDADD = $(LDADD) $(DRM_AMDGPU_LIBS)
+
 endif
diff --git a/tests/core_hotunplug.c b/tests/core_hotunplug.c
index e7d2a447..b553bc87 100644
--- a/tests/core_hotunplug.c
+++ b/tests/core_hotunplug.c
@@ -60,6 +60,188 @@ struct hotunplug {
 	igt_kmsg(KMSG_DEBUG "%s: " fmt, igt_test_name(), msg); \
 })
 
+/* amdgpu specific code */
+
+#include <amdgpu.h>
+#include <amdgpu_drm.h>
+#include <pthread.h>
+
+#define GFX_COMPUTE_NOP  0xffff1000
+
+
+static bool do_cs;
+
+static int
+amdgpu_bo_alloc_and_map(amdgpu_device_handle dev, unsigned size,
+			unsigned alignment, unsigned heap, uint64_t flags,
+			amdgpu_bo_handle *bo, void **cpu, uint64_t *mc_address,
+			amdgpu_va_handle *va_handle)
+{
+	struct amdgpu_bo_alloc_request request = {
+		.alloc_size = size,
+		.phys_alignment = alignment,
+		.preferred_heap = heap,
+		.flags = flags,
+	};
+	amdgpu_bo_handle buf_handle;
+	amdgpu_va_handle handle;
+	uint64_t vmc_addr;
+	int r;
+
+	r = amdgpu_bo_alloc(dev, &request, &buf_handle);
+	if (r)
+		return r;
+
+	r = amdgpu_va_range_alloc(dev,
+				  amdgpu_gpu_va_range_general,
+				  size, alignment, 0, &vmc_addr,
+				  &handle, 0);
+	if (r)
+		goto error_va_alloc;
+
+	r = amdgpu_bo_va_op(buf_handle, 0, size, vmc_addr, 0, AMDGPU_VA_OP_MAP);
+	if (r)
+		goto error_va_map;
+
+	r = amdgpu_bo_cpu_map(buf_handle, cpu);
+	if (r)
+		goto error_cpu_map;
+
+	*bo = buf_handle;
+	*mc_address = vmc_addr;
+	*va_handle = handle;
+
+	return 0;
+
+error_cpu_map:
+	amdgpu_bo_cpu_unmap(buf_handle);
+
+error_va_map:
+	amdgpu_bo_va_op(buf_handle, 0, size, vmc_addr, 0, AMDGPU_VA_OP_UNMAP);
+
+error_va_alloc:
+	amdgpu_bo_free(buf_handle);
+	return r;
+}
+
+static void
+amdgpu_bo_unmap_and_free(amdgpu_bo_handle bo, amdgpu_va_handle va_handle,
+			 uint64_t mc_addr, uint64_t size)
+{
+	amdgpu_bo_cpu_unmap(bo);
+	amdgpu_bo_va_op(bo, 0, size, mc_addr, 0, AMDGPU_VA_OP_UNMAP);
+	amdgpu_va_range_free(va_handle);
+	amdgpu_bo_free(bo);
+}
+
+static void amdgpu_cs_sync(amdgpu_context_handle context,
+			   unsigned int ip_type,
+			   int ring,
+			   unsigned int seqno)
+{
+	struct amdgpu_cs_fence fence = {
+		.context = context,
+		.ip_type = ip_type,
+		.ring = ring,
+		.fence = seqno,
+	};
+	uint32_t expired;
+	int err;
+
+	err = amdgpu_cs_query_fence_status(&fence,
+					   AMDGPU_TIMEOUT_INFINITE,
+					   0, &expired);
+}
+
+static void *amdgpu_nop_cs(void *p)
+{
+	int fd = *(int *)p;
+	amdgpu_bo_handle ib_result_handle;
+	void *ib_result_cpu;
+	uint64_t ib_result_mc_address;
+	uint32_t *ptr;
+	int i, r;
+	amdgpu_bo_list_handle bo_list;
+	amdgpu_va_handle va_handle;
+	uint32_t major, minor;
+	amdgpu_device_handle device;
+	amdgpu_context_handle context;
+	struct amdgpu_cs_request ibs_request;
+	struct amdgpu_cs_ib_info ib_info;
+
+	r = amdgpu_device_initialize(fd, &major, &minor, &device);
+	igt_require(r == 0);
+
+	r = amdgpu_cs_ctx_create(device, &context);
+	igt_assert_eq(r, 0);
+
+	r = amdgpu_bo_alloc_and_map(device, 4096, 4096,
+				    AMDGPU_GEM_DOMAIN_GTT, 0,
+				    &ib_result_handle, &ib_result_cpu,
+				    &ib_result_mc_address, &va_handle);
+	igt_assert_eq(r, 0);
+
+	ptr = ib_result_cpu;
+	for (i = 0; i < 16; ++i)
+		ptr[i] = GFX_COMPUTE_NOP;
+
+	r = amdgpu_bo_list_create(device, 1, &ib_result_handle, NULL, &bo_list);
+	igt_assert_eq(r, 0);
+
+	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 = AMDGPU_HW_IP_GFX;
+	ibs_request.ring = 0;
+	ibs_request.number_of_ibs = 1;
+	ibs_request.ibs = &ib_info;
+	ibs_request.resources = bo_list;
+
+	while (do_cs)
+		amdgpu_cs_submit(context, 0, &ibs_request, 1);
+
+	amdgpu_cs_sync(context, AMDGPU_HW_IP_GFX, 0, ibs_request.seq_no);
+
+	amdgpu_bo_list_destroy(bo_list);
+	amdgpu_bo_unmap_and_free(ib_result_handle, va_handle,
+				 ib_result_mc_address, 4096);
+
+	amdgpu_cs_ctx_free(context);
+
+	amdgpu_device_deinitialize(device);
+
+	return (void *)0;
+}
+
+static pthread_t* amdgpu_create_cs_thread(int *fd)
+{
+	int r;
+	pthread_t *thread = malloc(sizeof(*thread));
+
+	do_cs = true;
+
+	r = pthread_create(thread, NULL, amdgpu_nop_cs, (void *)fd);
+	igt_assert_eq(r, 0);
+
+	/* Give thread enough time to start*/
+	usleep(100000);
+	return thread;
+}
+
+static void amdgpu_destroy_cs_thread(pthread_t *thread)
+{
+	void *status;
+
+	do_cs = false;
+
+	pthread_join(*thread, &status);
+	igt_assert(status == 0);
+
+	free(thread);
+}
+
 /**
  * Subtests must be able to close examined devices completely.  Don't
  * use drm_open_driver() since in case of an i915 device it opens it
@@ -455,12 +637,21 @@ static void unplug_rescan(struct hotunplug *priv)
 
 static void hotunbind_rebind(struct hotunplug *priv)
 {
+	pthread_t *thread = NULL;
+
 	pre_check(priv);
 
 	priv->fd.drm = local_drm_open_driver(false, "", " for hot unbind");
 
+	if (is_amdgpu_device(priv->fd.drm))
+		thread = amdgpu_create_cs_thread(&priv->fd.drm);
+
 	driver_unbind(priv, "hot ", 0);
 
+	if (thread)
+		amdgpu_destroy_cs_thread(thread);
+
+
 	priv->fd.drm = close_device(priv->fd.drm, "late ", "unbound ");
 	igt_assert_eq(priv->fd.drm, -1);
 
@@ -471,12 +662,20 @@ static void hotunbind_rebind(struct hotunplug *priv)
 
 static void hotunplug_rescan(struct hotunplug *priv)
 {
+	pthread_t *thread = NULL;
+
 	pre_check(priv);
 
 	priv->fd.drm = local_drm_open_driver(false, "", " for hot unplug");
 
+	if (is_amdgpu_device(priv->fd.drm))
+		thread = amdgpu_create_cs_thread(&priv->fd.drm);
+
 	device_unplug(priv, "hot ", 0);
 
+	if (thread)
+		amdgpu_destroy_cs_thread(thread);
+
 	priv->fd.drm = close_device(priv->fd.drm, "late ", "removed ");
 	igt_assert_eq(priv->fd.drm, -1);
 
@@ -543,6 +742,7 @@ static void hotreplug_lateclose(struct hotunplug *priv)
 	igt_assert_f(healthcheck(priv, false), "%s\n", priv->failure);
 }
 
+
 /* Main */
 
 igt_main
diff --git a/tests/meson.build b/tests/meson.build
index 825e0183..1de6cc5a 100644
--- a/tests/meson.build
+++ b/tests/meson.build
@@ -243,7 +243,7 @@ i915_progs = [
 	'sysfs_timeslice_duration',
 ]
 
-test_deps = [ igt_deps ]
+test_deps = [ igt_deps + [ libdrm_amdgpu ] ]
 
 if libdrm_nouveau.found()
 	test_progs += [
-- 
2.25.1

_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

[igt-dev] [PATCH 2/3] tests/core_hotunplug: Add hotunplug with exported dma-buf test.

[Andrey Grodzovsky]

Unplug a device while an exported BO is still around.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
---
 tests/core_hotunplug.c | 62 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 62 insertions(+)

diff --git a/tests/core_hotunplug.c b/tests/core_hotunplug.c
index b553bc87..7cde0157 100644
--- a/tests/core_hotunplug.c
+++ b/tests/core_hotunplug.c
@@ -684,6 +684,59 @@ static void hotunplug_rescan(struct hotunplug *priv)
 	igt_assert_f(healthcheck(priv, false), "%s\n", priv->failure);
 }
 
+static void hotunplug_with_exported_bo_rescan(struct hotunplug *priv)
+{
+	int r;
+	uint32_t dma_buf_fd;
+	unsigned int *ptr;
+	amdgpu_bo_handle bo_handle;
+	uint32_t major, minor;
+	amdgpu_device_handle device;
+
+	struct amdgpu_bo_alloc_request request = {
+		.alloc_size = 4096,
+		.phys_alignment = 4096,
+		.preferred_heap = AMDGPU_GEM_DOMAIN_GTT,
+		.flags = 0,
+	};
+
+	pre_check(priv);
+
+	priv->fd.drm = local_drm_open_driver(false, "", " for hot unplug with export");
+
+
+	r = amdgpu_device_initialize(priv->fd.drm, &major, &minor, &device);
+	igt_require(r == 0);
+
+	amdgpu_bo_alloc(device, &request, &bo_handle);
+	igt_assert_eq(r, 0);
+
+	r = amdgpu_bo_export(bo_handle, amdgpu_bo_handle_type_dma_buf_fd, &dma_buf_fd);
+	igt_assert(r == 0);
+
+	ptr = mmap(NULL, 4096, PROT_READ | PROT_WRITE, MAP_SHARED, dma_buf_fd, 0);
+	igt_assert(ptr != MAP_FAILED);
+
+	amdgpu_bo_free(bo_handle);
+	amdgpu_device_deinitialize(device);
+	priv->fd.drm = close_device(priv->fd.drm, "late ", "removed ");
+	igt_assert_eq(priv->fd.drm, -1);
+
+	device_unplug(priv, "hot ", 0);
+
+	*ptr = 0xdeafbeef;
+
+	munmap(ptr, 4096);
+	close (dma_buf_fd);
+	return;
+
+	bus_rescan(priv, 0);
+
+	igt_assert_f(healthcheck(priv, false), "%s\n", priv->failure);
+}
+
+
+
 static void hotrebind(struct hotunplug *priv)
 {
 	pre_check(priv);
@@ -832,6 +885,15 @@ igt_main
 			recover(&priv);
 	}
 
+	igt_subtest_group {
+			igt_describe("Check if device with exported dma buf can be cleanly unplugged, then released and restored");
+			igt_subtest("hotunplug-with-exported-bo-rescan")
+			hotunplug_with_exported_bo_rescan(&priv);
+
+			igt_fixture
+				recover(&priv);
+	}
+
 	igt_fixture
 		post_healthcheck(&priv);
 
-- 
2.25.1

_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

[igt-dev] [PATCH 3/3] tests/core_hotunplug: Add hotunplug with exported sync obj test

[Andrey Grodzovsky]

Unplug a device while an exported sync_obj and it's fence
are  is still around.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
---
 tests/core_hotunplug.c | 102 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 102 insertions(+)

diff --git a/tests/core_hotunplug.c b/tests/core_hotunplug.c
index 7cde0157..16bbbf44 100644
--- a/tests/core_hotunplug.c
+++ b/tests/core_hotunplug.c
@@ -735,6 +735,99 @@ static void hotunplug_with_exported_bo_rescan(struct hotunplug *priv)
 	igt_assert_f(healthcheck(priv, false), "%s\n", priv->failure);
 }
 
+static void hotunplug_with_exported_fence_rescan(struct hotunplug *priv)
+{
+	amdgpu_bo_handle ib_result_handle;
+	void *ib_result_cpu;
+	uint64_t ib_result_mc_address;
+	uint32_t *ptr, sync_obj_handle, sync_obj_handle2;
+	int i, r;
+	amdgpu_bo_list_handle bo_list;
+	amdgpu_va_handle va_handle;
+	uint32_t major, minor, major2, minor2;
+	amdgpu_device_handle device, device2;
+	amdgpu_context_handle context;
+	struct amdgpu_cs_request ibs_request;
+	struct amdgpu_cs_ib_info ib_info;
+	struct amdgpu_cs_fence fence_status = {0};
+	int shared_fd, second_fd = -1;
+
+	pre_check(priv);
+
+	priv->fd.drm = __drm_open_driver_another(0, DRIVER_AMDGPU);
+	igt_require(priv->fd.drm >= 0);
+	second_fd = __drm_open_driver_another(1, DRIVER_AMDGPU);
+	igt_require(second_fd >= 0);
+
+	r = amdgpu_device_initialize(priv->fd.drm, &major, &minor, &device);
+	igt_require(r == 0);
+
+	r = amdgpu_device_initialize(second_fd, &major2, &minor2, &device2);
+	igt_require(r == 0);
+
+	r = amdgpu_cs_ctx_create(device, &context);
+	igt_assert_eq(r, 0);
+
+	r = amdgpu_bo_alloc_and_map(device, 4096, 4096,
+				    AMDGPU_GEM_DOMAIN_GTT, 0,
+				    &ib_result_handle, &ib_result_cpu,
+				    &ib_result_mc_address, &va_handle);
+	igt_assert_eq(r, 0);
+
+	ptr = ib_result_cpu;
+	for (i = 0; i < 16; ++i)
+		ptr[i] = GFX_COMPUTE_NOP;
+
+	r = amdgpu_bo_list_create(device, 1, &ib_result_handle, NULL, &bo_list);
+	igt_assert_eq(r, 0);
+
+	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 = AMDGPU_HW_IP_GFX;
+	ibs_request.ring = 0;
+	ibs_request.number_of_ibs = 1;
+	ibs_request.ibs = &ib_info;
+	ibs_request.resources = bo_list;
+
+	igt_assert_eq(amdgpu_cs_submit(context, 0, &ibs_request, 1), 0);
+
+	fence_status.context = context;
+	fence_status.ip_type = AMDGPU_HW_IP_GFX;
+	fence_status.ip_instance = 0;
+	fence_status.fence = ibs_request.seq_no;
+
+	igt_assert_eq(amdgpu_cs_fence_to_handle(device, &fence_status,
+						AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ,
+						&sync_obj_handle),
+						0);
+
+	igt_assert_eq(amdgpu_cs_export_syncobj(device, sync_obj_handle, &shared_fd), 0);
+
+	igt_assert_eq(amdgpu_cs_import_syncobj(device2, shared_fd, &sync_obj_handle2), 0);
+
+	igt_assert_eq(amdgpu_cs_destroy_syncobj(device, sync_obj_handle), 0);
+	amdgpu_bo_list_destroy(bo_list);
+	amdgpu_bo_unmap_and_free(ib_result_handle, va_handle,
+				 ib_result_mc_address, 4096);
+	amdgpu_cs_ctx_free(context);
+	amdgpu_device_deinitialize(device);
+	priv->fd.drm = close_device(priv->fd.drm, "late ", "removed ");
+	igt_assert_eq(priv->fd.drm, -1);
+
+	device_unplug(priv, "hot ", 0);
+
+	igt_assert_eq(amdgpu_cs_syncobj_wait(device2, &sync_obj_handle2, 1, 100000000, 0, NULL),
+		      0);
+
+	igt_assert_eq(amdgpu_cs_destroy_syncobj(device2, sync_obj_handle2), 0);
+	amdgpu_device_deinitialize(device2);
+	close_device(second_fd, "late ", "removed ");
+
+}
+
 
 
 static void hotrebind(struct hotunplug *priv)
@@ -894,6 +987,15 @@ igt_main
 				recover(&priv);
 	}
 
+	igt_subtest_group {
+			igt_describe("Check if device with exported dma fence can be cleanly unplugged, then released and restored");
+			igt_subtest("hotunplug-with-exported-fence-rescan")
+			hotunplug_with_exported_fence_rescan(&priv);
+
+			igt_fixture
+				recover(&priv);
+	}
+
 	igt_fixture
 		post_healthcheck(&priv);
 
-- 
2.25.1

_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

[igt-dev] ✗ Fi.CI.BUILD: failure for igt tests for hot-unplug feature in amdgpu

[Patchwork]

== Series Details ==

Series: igt tests for hot-unplug feature in amdgpu
URL   : https://patchwork.freedesktop.org/series/90949/
State : failure

== Summary ==

Applying: tests/core_hotunplug: Add command submission for amdgpu
Using index info to reconstruct a base tree...
A	tests/Makefile.am
M	tests/core_hotunplug.c
M	tests/meson.build
Falling back to patching base and 3-way merge...
Auto-merging tests/meson.build
Auto-merging tests/core_hotunplug.c
CONFLICT (modify/delete): tests/Makefile.am deleted in HEAD and modified in tests/core_hotunplug: Add command submission for amdgpu. Version tests/core_hotunplug: Add command submission for amdgpu of tests/Makefile.am left in tree.
Patch failed at 0001 tests/core_hotunplug: Add command submission for amdgpu
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".


_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev