Skip to contentSkip to conversation

Implement sw_sync test

PATCH Superseded [email protected] 3 participants 18 messages Sep 14, 2016 – Sep 15, 2016 Library code

v4 of 11 Superseded Latest is v11 →

lore
Share & tools
lore
#0 lore raw

v4 cover Implement sw_sync test

From: Robert Foss <[email protected]>

s series implements the sw_sync test and the lib/sw_sync helper functions for said test.

Gustavo Padovans sw_sync series was just de-staged in gregkh-staging/staging-next [1], and this test is targeted at verifying the functionality implemented in that series.

The sw_sync subtests range from very basic tests of the sw_sync functionality, to stress testing and randomized tests.

[1] http://git.kernel.org/cgit/linux/kernel/git/gregkh/staging.git/

Changes since v1:
Added "Reviewed-by: Eric Engestrom <[email protected]>" tag

lib/sw_sync:
- Fixed fd value checking
- Fixed fd validity check in sw_sync_fd_close()
- Moved sw_sync related paths to define
- Switched from malloc+memset to calloc in sync_file_info()
- Switched sizeof to dereferenced ptr

tests/sw_sync:
- Moved lib/sw_sync related code to lib/sw_sync commit
- Replaced memset with assignment in loop

Changes since v2:

lib/sw_sync:
- Replaced fd validity check in sw_sync_timeline_create()
- Replace sw_sync_XXX_destroy() functions with close()
- Simplified sw_sync_timeline_inc() comparison
- Changed sw_sync_merge() return value to -errno
- Changed name of sw_sync_fence_size() to sw_sync_fence_count()
- Reworked implementation of sw_sync_fence_count()
- Reworked implementation of sw_sync_fence_count_status()

tests/sw_sync:
- Replace sw_sync_XXX_destroy() functions with close()

Changes since v3:

lib/sw_sync:
- Changed sw_sync_fence_create() to take uint32_t seqno
- Added raw __sw_sync_fence_create() and failure check sw_sync_fence_create()

tests/sw_sync:
- Switch to using __sw_sync_fence_create() for failure cases

Robert Foss (13):
lib/sw_sync: Add helper functions for managing synchronization
primitives
tests/sw_sync: Add sw_sync test
tests/sw_sync: Add subtest test_alloc_fence tests/sw_sync: Add subtest test_alloc_fence_invalid_timeline tests/sw_sync: Add subtest test_alloc_merge_fence tests/sw_sync: Add subtest test_sync_wait
tests/sw_sync: Add subtest test_sync_merge tests/sw_sync: Add subtest test_sync_merge_same tests/sw_sync: Add subtest test_sync_multi_consumer tests/sw_sync: Add subtest test_sync_multi_consumer_producer tests/sw_sync: Add subtest test_sync_random_merge tests/sw_sync: Add subtest test_sync_multi_timeline_wait tests/sw_sync: Add subtest test_sync_multi_producer_single_consumer

 lib/Makefile.sources   |   2 +
 lib/sw_sync.c          | 190 ++++++++++++++
 lib/sw_sync.h          |  48 ++++
 tests/Makefile.sources |   1 +
 tests/sw_sync.c        | 694 +++++++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 935 insertions(+)
 create mode 100644 lib/sw_sync.c
 create mode 100644 lib/sw_sync.h
 create mode 100644 tests/sw_sync.c
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw

PATCH v4 1/13 lib/sw_sync: Add helper functions for managing synchronization primitives

From: Robert Foss <[email protected]>

Base functions to help testing the Sync File Framework (explicit fencing mechanism ported from Android).
These functions allow you to create, use and destroy timelines and fences.

---
 lib/Makefile.sources |   2 +
 lib/sw_sync.c        | 190 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/sw_sync.h        |  48 +++++++++++++
 3 files changed, 240 insertions(+)
 create mode 100644 lib/sw_sync.c
 create mode 100644 lib/sw_sync.h
lib/Makefile.sources +2 -0
index bac9a7f..3dc7c3c 100644
@@ -61,6 +61,8 @@ lib_source_list =	 	\
 	rendercopy_gen8.c	\
 	rendercopy_gen9.c	\
 	rendercopy.h		\
+	sw_sync.c		\
+	sw_sync.h		\
 	intel_reg_map.c		\
 	intel_iosf.c		\
 	igt_kms.c		\
lib/sw_sync.c new file +190 -0

190 lines not on this page · raw message

lib/sw_sync.h new file +48 -0

48 lines not on this page · raw message

Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw

Re: [PATCH i-g-t v4 01/13] lib/sw_sync: Add helper functions for managing synchronization primitives

lib/sw_sync.c line 111

Quoted 6 lines · On Wed, Sep 14, 2016 at 11:04:30AM -0400, [email protected] wrote:
lib/sw_sync.c new file +6 -0
+void sw_sync_timeline_inc(int fd, uint32_t count)
+{
+	uint32_t arg = count;
+
+	if (fd == 0)
+		return;

But fd = 0 is a valid fd, and might be a timeline somewhere.

Did you mean count == 0 ?

And even then (unless it is defined as an error condition in the kernel ABI, and it should not be...) we should pass it through to the kernel.

lib/sw_sync.c line 138

Quoted 11 lines
lib/sw_sync.c new file +10 -0
+	do_ioctl(fd, SW_SYNC_IOC_INC, &arg);
+}
+
 
+int sw_sync_wait(int fence, int timeout)
+{
+	struct pollfd fds;
+	int ret;
+
+	fds.fd = fence;
+	fds.events = POLLIN | POLLERR;

POLLERR is always implied and doesn't need to be specified (it is meaningless in .events).

int sw_sync_wait(int fence, int timeout)
{

#if BEING_FANCY
	return poll(&(struct pollfd){fd, POLLIN}, 1, timeout);
#else
	struct pollfd pfd = { fd, POLLIN };
	return poll(&pfd, 1, timeout);

#endif
}

Indentation has gone wrong, double check the whitespace.

How do fences operate after their timeline is closed? (Are they automatically signaled, or do they persist and are signaled normally?) Is there a test for using fences from a closed timeline (I was looking but didn't notice one).
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#3 lore raw

Re: [PATCH i-g-t v4 01/13] lib/sw_sync: Add helper functions for managing synchronization primitives

Quoted 10 lines · Chris Wilson

lib/sw_sync.c line 111

Quoted 6 lines · On Wed, Sep 14, 2016 at 11:04:30AM -0400, [email protected] wrote:
lib/sw_sync.c new file +6 -0
+void sw_sync_timeline_inc(int fd, uint32_t count)
+{
+	uint32_t arg = count;
+
+	if (fd == 0)
+		return;

But fd = 0 is a valid fd, and might be a timeline somewhere.

Quoted 4 lines

Did you mean count == 0 ?

And even then (unless it is defined as an error condition in the kernel ABI, and it should not be...) we should pass it through to the kernel.

You're right, I'll change it in v5.

Quoted 23 lines

lib/sw_sync.c line 138

Quoted 12 lines
lib/sw_sync.c new file +10 -0
+	do_ioctl(fd, SW_SYNC_IOC_INC, &arg);
+}
+
 
+int sw_sync_wait(int fence, int timeout)
+{
+	struct pollfd fds;
+	int ret;
+
+	fds.fd = fence;
+	fds.events = POLLIN | POLLERR;

POLLERR is always implied and doesn't need to be specified (it is meaningless in .events).

int sw_sync_wait(int fence, int timeout)
{

#if BEING_FANCY
	return poll(&(struct pollfd){fd, POLLIN}, 1, timeout);
#else
	struct pollfd pfd = { fd, POLLIN };
	return poll(&pfd, 1, timeout);
Quoted 4 lines

#endif
}

Indentation has gone wrong, double check the whitespace.

That is definitely nicer looking. I'll drop it in for v5.

Quoted 6 lines

How do fences operate after their timeline is closed? (Are they automatically signaled, or do they persist and are signaled normally?) Is there a test for using fences from a closed timeline (I was looking but didn't notice one).

I did some quick tests just to confirm, closing the timeline signals all of its fences.
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#4 lore raw

Re: [PATCH i-g-t v4 01/13] lib/sw_sync: Add helper functions for managing synchronization primitives

Quoted 53 lines · Robert Foss
Quoted 11 lines · Chris Wilson

lib/sw_sync.c line 111

Quoted 6 lines · On Wed, Sep 14, 2016 at 11:04:30AM -0400, [email protected] wrote:
lib/sw_sync.c new file +6 -0
+void sw_sync_timeline_inc(int fd, uint32_t count)
+{
+    uint32_t arg = count;
+
+    if (fd == 0)
+        return;

But fd = 0 is a valid fd, and might be a timeline somewhere.

Quoted 4 lines

Did you mean count == 0 ?

And even then (unless it is defined as an error condition in the kernel ABI, and it should not be...) we should pass it through to the kernel.

You're right, I'll change it in v5.

Quoted 23 lines

lib/sw_sync.c line 138

Quoted 12 lines
lib/sw_sync.c new file +10 -0
+    do_ioctl(fd, SW_SYNC_IOC_INC, &arg);
+}
+
 
+int sw_sync_wait(int fence, int timeout)
+{
+    struct pollfd fds;
+    int ret;
+
+    fds.fd = fence;
+    fds.events = POLLIN | POLLERR;

POLLERR is always implied and doesn't need to be specified (it is meaningless in .events).

int sw_sync_wait(int fence, int timeout)
{

#if BEING_FANCY
    return poll(&(struct pollfd){fd, POLLIN}, 1, timeout);
#else
    struct pollfd pfd = { fd, POLLIN };
    return poll(&pfd, 1, timeout);
Quoted 4 lines

#endif
}

Indentation has gone wrong, double check the whitespace.

That is definitely nicer looking. I'll drop it in for v5.

Quoted 2 lines
Quoted 7 lines
Quoted 4 lines

How do fences operate after their timeline is closed? (Are they automatically signaled, or do they persist and are signaled normally?) Is there a test for using fences from a closed timeline (I was looking but didn't notice one).

I did some quick tests just to confirm, closing the timeline signals all of its fences.

Actually, my quick test was wrong. A fence is _not_ signaled on when its timeline has been closed.

So you would like to see a test that confirms that a fence on closed timeline is not signaled?

Rob.
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#5 lore raw

Re: [PATCH i-g-t v4 01/13] lib/sw_sync: Add helper functions for managing synchronization primitives

2016-09-15 Robert Foss <[email protected]>:

Quoted 61 lines · Robert Foss
Quoted 58 lines · Chris Wilson
Quoted 11 lines

lib/sw_sync.c line 111

Quoted 6 lines · On Wed, Sep 14, 2016 at 11:04:30AM -0400, [email protected] wrote:
lib/sw_sync.c new file +6 -0
+void sw_sync_timeline_inc(int fd, uint32_t count)
+{
+    uint32_t arg = count;
+
+    if (fd == 0)
+        return;

But fd = 0 is a valid fd, and might be a timeline somewhere.

Quoted 4 lines

Did you mean count == 0 ?

And even then (unless it is defined as an error condition in the kernel ABI, and it should not be...) we should pass it through to the kernel.

You're right, I'll change it in v5.

Quoted 23 lines

lib/sw_sync.c line 138

Quoted 12 lines
lib/sw_sync.c new file +10 -0
+    do_ioctl(fd, SW_SYNC_IOC_INC, &arg);
+}
+
 
+int sw_sync_wait(int fence, int timeout)
+{
+    struct pollfd fds;
+    int ret;
+
+    fds.fd = fence;
+    fds.events = POLLIN | POLLERR;

POLLERR is always implied and doesn't need to be specified (it is meaningless in .events).

int sw_sync_wait(int fence, int timeout)
{

#if BEING_FANCY
    return poll(&(struct pollfd){fd, POLLIN}, 1, timeout);
#else
    struct pollfd pfd = { fd, POLLIN };
    return poll(&pfd, 1, timeout);
Quoted 4 lines

#endif
}

Indentation has gone wrong, double check the whitespace.

That is definitely nicer looking. I'll drop it in for v5.

Quoted 6 lines

How do fences operate after their timeline is closed? (Are they automatically signaled, or do they persist and are signaled normally?) Is there a test for using fences from a closed timeline (I was looking but didn't notice one).

Quoted 8 lines
Quoted 2 lines

I did some quick tests just to confirm, closing the timeline signals all of its fences.

Actually, my quick test was wrong. A fence is _not_ signaled on when its timeline has been closed.

So you would like to see a test that confirms that a fence on closed timeline is not signaled?

Yes, please. Add this test and maybe another the signals the fence before closing the timeline and the check if the fence is indeed signaled.

Gustavo

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#6 lore raw

PATCH v4 2/13 tests/sw_sync: Add sw_sync test

In reply to [email protected]

From: Robert Foss <[email protected]>

Add initial tests for sw_sync.

---
 tests/Makefile.sources |  1 +
 tests/sw_sync.c        | 52 ++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 53 insertions(+)
 create mode 100644 tests/sw_sync.c
tests/Makefile.sources +1 -0
index 72a58ad..0ba769f 100644
@@ -125,6 +125,7 @@ TESTS_progs_M = \
 	prime_mmap_kms \
 	prime_self_import \
 	prime_vgem \
+	sw_sync \
 	template \
 	vgem_basic \
 	vgem_slow \
tests/sw_sync.c new file +52 -0

52 lines not on this page · raw message

Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#7 lore raw

PATCH v4 3/13 tests/sw_sync: Add subtest test_alloc_fence

In reply to [email protected]

From: Robert Foss <[email protected]>

Add subtest alloc_fence that verifies that it's possible to allocate a fence on a timeline.

---
 tests/sw_sync.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
tests/sw_sync.c +16 -0
index 47e2dfa..8f6208b 100644
@@ -44,9 +44,25 @@ static void test_alloc_timeline(void)
 	close(timeline);
 }
 
+static void test_alloc_fence(void)
+{
+	int in_fence;
+	int timeline;
+
+	timeline = sw_sync_timeline_create();
+	in_fence = sw_sync_fence_create(timeline, 0);
+
+	close(in_fence);
+	close(timeline);
+}
+
+
 igt_main
 {
 	igt_subtest("alloc_timeline")
 		test_alloc_timeline();
+
+	igt_subtest("alloc_fence")
+		test_alloc_fence();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#8 lore raw

PATCH v4 4/13 tests/sw_sync: Add subtest test_alloc_fence_invalid_timeline

In reply to [email protected]

From: Robert Foss <[email protected]>

This subtests tests that creating fences on negative timelines fail.

---
 tests/sw_sync.c | 7 +++++++
 1 file changed, 7 insertions(+)
tests/sw_sync.c +7 -0
index 8f6208b..75bd471 100644
@@ -56,6 +56,10 @@ static void test_alloc_fence(void)
 	close(timeline);
 }
 
+static void test_alloc_fence_invalid_timeline(void)
+{
+	igt_assert(__sw_sync_fence_create(-1, 0) < 0);
+}
 
 igt_main
 {
@@ -64,5 +68,8 @@ igt_main
 
 	igt_subtest("alloc_fence")
 		test_alloc_fence();
+
+	igt_subtest("alloc_fence_invalid_timeline")
+		test_alloc_fence_invalid_timeline();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#9 lore raw

PATCH v4 5/13 tests/sw_sync: Add subtest test_alloc_merge_fence

In reply to [email protected]

From: Robert Foss <[email protected]>

This subtest verifies that merging two fences works in the simples possible case.

---
 tests/sw_sync.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
tests/sw_sync.c +23 -0
index 75bd471..fcb2f57 100644
@@ -61,6 +61,26 @@ static void test_alloc_fence_invalid_timeline(void)
 	igt_assert(__sw_sync_fence_create(-1, 0) < 0);
 }
 
+static void test_alloc_merge_fence(void)
+{
+	int in_fence[2];
+	int fence_merge;
+	int timeline[2];
+
+	timeline[0] = sw_sync_timeline_create();
+	timeline[1] = sw_sync_timeline_create();
+
+	in_fence[0] = sw_sync_fence_create(timeline[0], 1);
+	in_fence[1] = sw_sync_fence_create(timeline[1], 1);
+	fence_merge = sw_sync_merge(in_fence[1], in_fence[0]);
+
+	close(in_fence[0]);
+	close(in_fence[1]);
+	close(fence_merge);
+	close(timeline[0]);
+	close(timeline[1]);
+}
+
 igt_main
 {
 	igt_subtest("alloc_timeline")
@@ -71,5 +91,8 @@ igt_main
 
 	igt_subtest("alloc_fence_invalid_timeline")
 		test_alloc_fence_invalid_timeline();
+
+	igt_subtest("alloc_merge_fence")
+		test_alloc_merge_fence();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#10 lore raw

PATCH v4 6/13 tests/sw_sync: Add subtest test_sync_wait

In reply to [email protected]

From: Robert Foss <[email protected]>

This subtest verifies that waiting on fences works properly.

---
 tests/sw_sync.c | 38 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 38 insertions(+)
tests/sw_sync.c +38 -0
index fcb2f57..3061279 100644
@@ -81,6 +81,41 @@ static void test_alloc_merge_fence(void)
 	close(timeline[1]);
 }
 
+static void test_sync_wait(void)
+{
+	int fence, ret;
+	int timeline;
+
+	timeline = sw_sync_timeline_create();
+	fence = sw_sync_fence_create(timeline, 5);
+
+	/* Wait on fence until timeout */
+	ret = sw_sync_wait(fence, 0);
+	igt_assert_f(ret == 0, "Failure waiting on fence until timeout\n");
+
+	/* Advance timeline from 0 -> 1 */
+	sw_sync_timeline_inc(timeline, 1);
+
+	/* Wait on fence until timeout */
+	ret = sw_sync_wait(fence, 0);
+	igt_assert_f(ret == 0, "Failure waiting on fence until timeout\n");
+
+	/* Signal the fence */
+	sw_sync_timeline_inc(timeline, 4);
+
+	/* Wait successfully */
+	ret = sw_sync_wait(fence, 0);
+	igt_assert_f(ret > 0, "Failure waiting on fence\n");
+
+	/* Go even further, and confirm wait still succeeds */
+	sw_sync_timeline_inc(timeline, 10);
+	ret = sw_sync_wait(fence, 0);
+	igt_assert_f(ret > 0, "Failure waiting ahead\n");
+
+	close(fence);
+	close(timeline);
+}
+
 igt_main
 {
 	igt_subtest("alloc_timeline")
@@ -94,5 +129,8 @@ igt_main
 
 	igt_subtest("alloc_merge_fence")
 		test_alloc_merge_fence();
+
+	igt_subtest("sync_wait")
+		test_sync_wait();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#11 lore raw

PATCH v4 7/13 tests/sw_sync: Add subtest test_sync_merge

In reply to [email protected]

From: Robert Foss <[email protected]>

Add subtest test_sync_merge that tests merging fences and the validity of the resulting merged fence.

---
 tests/sw_sync.c | 67 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 67 insertions(+)
tests/sw_sync.c +67 -0
index 3061279..26226bd 100644
@@ -116,6 +116,70 @@ static void test_sync_wait(void)
 	close(timeline);
 }
 
+static void test_sync_merge(void)
+{
+	int in_fence[3];
+	int fence_merge;
+	int timeline;
+	int active, signaled;
+
+	timeline = sw_sync_timeline_create();
+	in_fence[0] = sw_sync_fence_create(timeline, 1);
+	in_fence[1] = sw_sync_fence_create(timeline, 2);
+	in_fence[2] = sw_sync_fence_create(timeline, 3);
+
+	fence_merge = sw_sync_merge(in_fence[0], in_fence[1]);
+	fence_merge = sw_sync_merge(in_fence[2], fence_merge);
+
+	/* confirm all fences have one active point (even d) */
+	active = sw_sync_fence_count_status(in_fence[0],
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	igt_assert_f(active == 1, "in_fence[0] has too many active fences\n");
+	active = sw_sync_fence_count_status(in_fence[1],
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	igt_assert_f(active == 1, "in_fence[1] has too many active fences\n");
+	active = sw_sync_fence_count_status(in_fence[2],
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	igt_assert_f(active == 1, "in_fence[2] has too many active fences\n");
+	active = sw_sync_fence_count_status(fence_merge,
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	igt_assert_f(active == 1, "fence_merge has too many active fences\n");
+
+	/* confirm that fence_merge is not signaled until the max of fence 0,1,2 */
+	sw_sync_timeline_inc(timeline, 1);
+	signaled = sw_sync_fence_count_status(in_fence[0],
+					      SW_SYNC_FENCE_STATUS_SIGNALED);
+	active = sw_sync_fence_count_status(fence_merge,
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	igt_assert_f(signaled == 1, "in_fence[0] did not signal\n");
+	igt_assert_f(active == 1, "fence_merge signaled too early\n");
+
+	sw_sync_timeline_inc(timeline, 1);
+	signaled = sw_sync_fence_count_status(in_fence[1],
+					      SW_SYNC_FENCE_STATUS_SIGNALED);
+	active = sw_sync_fence_count_status(fence_merge,
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	igt_assert_f(signaled == 1, "in_fence[1] did not signal\n");
+	igt_assert_f(active == 1, "fence_merge signaled too early\n");
+
+	sw_sync_timeline_inc(timeline, 1);
+	signaled = sw_sync_fence_count_status(in_fence[2],
+					      SW_SYNC_FENCE_STATUS_SIGNALED);
+	igt_assert_f(signaled == 1, "in_fence[2] did not signal\n");
+	signaled = sw_sync_fence_count_status(fence_merge,
+					       SW_SYNC_FENCE_STATUS_SIGNALED);
+	active = sw_sync_fence_count_status(fence_merge,
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	igt_assert_f(active == 0 && signaled == 1,
+		     "fence_merge did not signal\n");
+
+	sw_sync_fence_destroy(in_fence[0]);
+	sw_sync_fence_destroy(in_fence[1]);
+	sw_sync_fence_destroy(in_fence[2]);
+	sw_sync_fence_destroy(fence_merge);
+	sw_sync_timeline_destroy(timeline);
+}
+
 igt_main
 {
 	igt_subtest("alloc_timeline")
@@ -132,5 +196,8 @@ igt_main
 
 	igt_subtest("sync_wait")
 		test_sync_wait();
+
+	igt_subtest("sync_merge")
+		test_sync_merge();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#12 lore raw

PATCH v4 8/13 tests/sw_sync: Add subtest test_sync_merge_same

In reply to [email protected]

From: Robert Foss <[email protected]>

This subtest verifies merging a fence with itself does not fail.

---
 tests/sw_sync.c | 37 ++++++++++++++++++++++++++++++++-----
 1 file changed, 32 insertions(+), 5 deletions(-)
tests/sw_sync.c +32 -5
index 26226bd..ff87cfe 100644
@@ -173,11 +173,35 @@ static void test_sync_merge(void)
 	igt_assert_f(active == 0 && signaled == 1,
 		     "fence_merge did not signal\n");
 
-	sw_sync_fence_destroy(in_fence[0]);
-	sw_sync_fence_destroy(in_fence[1]);
-	sw_sync_fence_destroy(in_fence[2]);
-	sw_sync_fence_destroy(fence_merge);
-	sw_sync_timeline_destroy(timeline);
+	close(in_fence[0]);
+	close(in_fence[1]);
+	close(in_fence[2]);
+	close(fence_merge);
+	close(timeline);
+}
+
+static void test_sync_merge_same(void)
+{
+	int in_fence[2];
+	int timeline;
+	int signaled;
+
+	timeline = sw_sync_timeline_create();
+	in_fence[0] = sw_sync_fence_create(timeline, 1);
+	in_fence[1] = sw_sync_merge(in_fence[0], in_fence[0]);
+
+	signaled = sw_sync_fence_count_status(in_fence[0],
+					      SW_SYNC_FENCE_STATUS_SIGNALED);
+	igt_assert_f(signaled == 0, "fence signaled too early\n");
+
+	sw_sync_timeline_inc(timeline, 1);
+	signaled = sw_sync_fence_count_status(in_fence[0],
+					      SW_SYNC_FENCE_STATUS_SIGNALED);
+	igt_assert_f(signaled == 1, "fence did not signal\n");
+
+	close(in_fence[0]);
+	close(in_fence[1]);
+	close(timeline);
 }
 
 igt_main
@@ -199,5 +223,8 @@ igt_main
 
 	igt_subtest("sync_merge")
 		test_sync_merge();
+
+	igt_subtest("sync_merge_same")
+		test_sync_merge_same();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#13 lore raw

PATCH v4 9/13 tests/sw_sync: Add subtest test_sync_multi_consumer

In reply to [email protected]

From: Robert Foss <[email protected]>

This subtest verifies the access ordering of multiple consumer threads.

---
 tests/sw_sync.c | 103 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 103 insertions(+)
tests/sw_sync.c +103 -0
index ff87cfe..e607b75 100644
@@ -27,6 +27,8 @@
  *    Robert Foss <[email protected]>
  */
 
+#include <pthread.h>
+#include <semaphore.h>
 #include <stdint.h>
 #include <unistd.h>
 
@@ -36,6 +38,15 @@
 
 IGT_TEST_DESCRIPTION("Test SW Sync Framework");
 
+typedef struct {
+	int timeline;
+	uint32_t thread_id;
+	uint32_t nbr_threads;
+	uint32_t nbr_iterations;
+	volatile uint32_t * volatile counter;
+	sem_t *sem;
+} data_t;
+
 static void test_alloc_timeline(void)
 {
 	int timeline;
@@ -204,6 +215,95 @@ static void test_sync_merge_same(void)
 	close(timeline);
 }
 
+static void * test_sync_multi_consumer_thread(void *arg)
+{
+	data_t *data = arg;
+	int thread_id = data->thread_id;
+	int nbr_threads = data->nbr_threads;
+	int timeline = data->timeline;
+	int iterations = data->nbr_iterations;
+	int ret, i;
+
+	for (i = 0; i < iterations; i++) {
+		int next_point = i * nbr_threads + thread_id;
+		int fence = sw_sync_fence_create(timeline, next_point);
+
+		ret = sw_sync_wait(fence, 1000);
+		if (ret <= 0)
+		{
+			return (void *) 1;
+		}
+
+		if (*(data->counter) != next_point)
+		{
+			return (void *) 1;
+		}
+
+		sem_post(data->sem);
+		close(fence);
+	}
+	return NULL;
+}
+
+static void test_sync_multi_consumer(void)
+{
+	const uint32_t nbr_threads = 8;
+	const uint32_t nbr_iterations = 1 << 14;
+	data_t data_arr[nbr_threads];
+	pthread_t thread_arr[nbr_threads];
+	sem_t sem;
+	int timeline;
+	volatile uint32_t counter = 0;
+	uintptr_t thread_ret = 0;
+	data_t data;
+	int i, ret;
+
+	sem_init(&sem, 0, 0);
+	timeline = sw_sync_timeline_create();
+
+	data.nbr_iterations = nbr_iterations;
+	data.nbr_threads = nbr_threads;
+	data.counter = &counter;
+	data.timeline = timeline;
+	data.sem = &sem;
+
+	/* Start sync threads. */
+	for (i = 0; i < nbr_threads; i++)
+	{
+		data_arr[i] = data;
+		data_arr[i].thread_id = i;
+		ret = pthread_create(&thread_arr[i], NULL,
+				     test_sync_multi_consumer_thread,
+				     (void *) &(data_arr[i]));
+		igt_assert_eq(ret, 0);
+	}
+
+	/* Produce 'content'. */
+	for (i = 0; i < nbr_threads * nbr_iterations; i++)
+	{
+		sem_wait(&sem);
+
+		counter++;
+		sw_sync_timeline_inc(timeline, 1);
+	}
+
+	/* Wait for threads to complete. */
+	for (i = 0; i < nbr_threads; i++)
+	{
+		uintptr_t local_thread_ret;
+		pthread_join(thread_arr[i], (void **)&local_thread_ret);
+		thread_ret |= local_thread_ret;
+	}
+
+	close(timeline);
+	sem_destroy(&sem);
+
+	igt_assert_f(counter == nbr_threads * nbr_iterations,
+		     "Counter has unexpected value.\n");
+
+	igt_assert_f(thread_ret == 0, "A sync thread reported failure.\n");
+}
+
 igt_main
 {
 	igt_subtest("alloc_timeline")
@@ -226,5 +326,8 @@ igt_main
 
 	igt_subtest("sync_merge_same")
 		test_sync_merge_same();
+
+	igt_subtest("sync_multi_consumer")
+		test_sync_multi_consumer();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#14 lore raw

PATCH v4 10/13 tests/sw_sync: Add subtest test_sync_multi_consumer_producer

In reply to [email protected]

From: Robert Foss <[email protected]>

This test verifies that stressing the kernel by creating multiple consumer/producer threads that wait on a single timeline to be incremented by another conumer/producer thread does not fail.
And that the order amongst the threads is maintained.

---
 tests/sw_sync.c | 83 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 83 insertions(+)
tests/sw_sync.c +83 -0
index e607b75..13a5643 100644
@@ -304,6 +304,86 @@ static void test_sync_multi_consumer(void)
 	igt_assert_f(thread_ret == 0, "A sync thread reported failure.\n");
 }
 
+static void * test_sync_multi_consumer_producer_thread(void *arg)
+{
+	data_t *data = arg;
+	int thread_id = data->thread_id;
+	int nbr_threads = data->nbr_threads;
+	int timeline = data->timeline;
+	int iterations = data->nbr_iterations;
+	int ret, i;
+
+	for (i = 0; i < iterations; i++) {
+		int next_point = i * nbr_threads + thread_id;
+		int fence = sw_sync_fence_create(timeline, next_point);
+
+		ret = sw_sync_wait(fence, 1000);
+		if (ret <= 0)
+		{
+			return (void *) 1;
+		}
+
+		if (*(data->counter) != next_point)
+		{
+			return (void *) 1;
+		}
+
+		(*data->counter)++;
+
+		/* Kick off the next thread. */
+		sw_sync_timeline_inc(timeline, 1);
+
+		close(fence);
+	}
+	return NULL;
+}
+
+static void test_sync_multi_consumer_producer(void)
+{
+	const uint32_t nbr_threads = 8;
+	const uint32_t nbr_iterations = 1 << 14;
+	data_t data_arr[nbr_threads];
+	pthread_t thread_arr[nbr_threads];
+	int timeline;
+	volatile uint32_t counter = 0;
+	uintptr_t thread_ret = 0;
+	data_t data;
+	int i, ret;
+
+	timeline = sw_sync_timeline_create();
+
+	data.nbr_iterations = nbr_iterations;
+	data.nbr_threads = nbr_threads;
+	data.counter = &counter;
+	data.timeline = timeline;
+
+	/* Start consumer threads. */
+	for (i = 0; i < nbr_threads; i++)
+	{
+		data_arr[i] = data;
+		data_arr[i].thread_id = i;
+		ret = pthread_create(&thread_arr[i], NULL,
+				     test_sync_multi_consumer_producer_thread,
+				     (void *) &(data_arr[i]));
+		igt_assert_eq(ret, 0);
+	}
+
+	/* Wait for threads to complete. */
+	for (i = 0; i < nbr_threads; i++)
+	{
+		uintptr_t local_thread_ret;
+		pthread_join(thread_arr[i], (void **)&local_thread_ret);
+		thread_ret |= local_thread_ret;
+	}
+
+	close(timeline);
+
+	igt_assert_f(counter == nbr_threads * nbr_iterations,
+		     "Counter has unexpected value.\n");
+
+	igt_assert_f(thread_ret == 0, "A sync thread reported failure.\n");
+}
+
 igt_main
 {
 	igt_subtest("alloc_timeline")
@@ -329,5 +409,8 @@ igt_main
 
 	igt_subtest("sync_multi_consumer")
 		test_sync_multi_consumer();
+
+	igt_subtest("sync_multi_consumer_producer")
+		test_sync_multi_consumer_producer();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#15 lore raw

PATCH v4 11/13 tests/sw_sync: Add subtest test_sync_random_merge

In reply to [email protected]

From: Robert Foss <[email protected]>

This subtest verifies that creating many timelines and merging random fences from each timeline with eachother results in merged fences that are fully functional.

---
 tests/sw_sync.c | 73 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 73 insertions(+)
tests/sw_sync.c +73 -0
index 13a5643..bd86111 100644
@@ -384,6 +384,76 @@ static void test_sync_multi_consumer_producer(void)
 	igt_assert_f(thread_ret == 0, "A sync thread reported failure.\n");
 }
 
+static void test_sync_random_merge(void)
+{
+	int i, size, ret;
+	const int nbr_timeline = 32;
+	const int nbr_merge = 1024;
+	int fence_map[nbr_timeline];
+	int timeline_arr[nbr_timeline];
+	int fence, tmpfence, merged;
+	int timeline, timeline_offset, sync_pt;
+
+	srand(time(NULL));
+
+	for (i = 0; i < nbr_timeline; i++) {
+		timeline_arr[i] = sw_sync_timeline_create();
+		fence_map[i] = -1;
+	}
+
+	sync_pt = rand();
+	fence = sw_sync_fence_create(timeline_arr[0], sync_pt);
+
+	fence_map[0] = sync_pt;
+
+	/* Randomly create syncpoints out of a fixed set of timelines,
+	 * and merge them together.
+	 */
+	for (i = 0; i < nbr_merge; i++) {
+		/* Generate syncpoint. */
+		timeline_offset = rand() % nbr_timeline;
+		timeline = timeline_arr[timeline_offset];
+		sync_pt = rand();
+
+		/* Keep track of the latest sync_pt in each timeline. */
+		if (fence_map[timeline_offset] == -1)
+			fence_map[timeline_offset] = sync_pt;
+		else if (fence_map[timeline_offset] < sync_pt)
+			fence_map[timeline_offset] = sync_pt;
+
+		/* Merge. */
+		tmpfence = sw_sync_fence_create(timeline, sync_pt);
+		merged = sw_sync_merge(tmpfence, fence);
+		close(tmpfence);
+		close(fence);
+		fence = merged;
+	}
+
+	size = 0;
+	for (i = 0; i < nbr_timeline; i++)
+		if (fence_map[i] != -1)
+			size++;
+
+	/* Trigger the merged fence. */
+	for (i = 0; i < nbr_timeline; i++) {
+		if (fence_map[i] != -1) {
+			ret = sw_sync_wait(fence, 0);
+			igt_assert_f(ret == 0,
+				    "Failure waiting on fence until timeout\n");
+			/* Increment the timeline to the last sync_pt */
+			sw_sync_timeline_inc(timeline_arr[i], fence_map[i]);
+		}
+	}
+
+	/* Check that the fence is triggered. */
+	ret = sw_sync_wait(fence, 0);
+	igt_assert_f(ret > 0, "Failure triggering fence\n");
+
+	close(fence);
+	for (i = 0; i < nbr_timeline; i++)
+		close(timeline_arr[i]);
+}
+
 igt_main
 {
 	igt_subtest("alloc_timeline")
@@ -412,5 +482,8 @@ igt_main
 
 	igt_subtest("sync_multi_consumer_producer")
 		test_sync_multi_consumer_producer();
+
+	igt_subtest("sync_random_merge")
+		test_sync_random_merge();
 }
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#16 lore raw

PATCH v4 12/13 tests/sw_sync: Add subtest test_sync_multi_timeline_wait

In reply to [email protected]

From: Robert Foss <[email protected]>

This subtest verifies that waiting, timing out on a wait and that counting fences in various states works.

---
 tests/sw_sync.c | 66 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 66 insertions(+)
tests/sw_sync.c +66 -0
index bd86111..9ec2dc2 100644
@@ -215,6 +215,69 @@ static void test_sync_merge_same(void)
 	close(timeline);
 }
 
+static void test_sync_multi_timeline_wait(void)
+{
+	int timeline[3];
+	int in_fence[3];
+	int fence_merge;
+	int active, signaled, ret;
+
+	timeline[0] = sw_sync_timeline_create();
+	timeline[1] = sw_sync_timeline_create();
+	timeline[2] = sw_sync_timeline_create();
+
+	in_fence[0] = sw_sync_fence_create(timeline[0], 5);
+	in_fence[1] = sw_sync_fence_create(timeline[1], 5);
+	in_fence[2] = sw_sync_fence_create(timeline[2], 5);
+
+	fence_merge = sw_sync_merge(in_fence[0], in_fence[1]);
+	fence_merge = sw_sync_merge(in_fence[2], fence_merge);
+
+	/* Confirm fence isn't signaled */
+	active = sw_sync_fence_count_status(fence_merge,
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	igt_assert_f(active == 3, "Fence signaled too early\n");
+
+	ret = sw_sync_wait(fence_merge, 0);
+	igt_assert_f(ret == 0, "Failure waiting on fence until timeout\n");
+
+	sw_sync_timeline_inc(timeline[0], 5);
+	active = sw_sync_fence_count_status(fence_merge,
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	signaled = sw_sync_fence_count_status(fence_merge,
+					      SW_SYNC_FENCE_STATUS_SIGNALED);
+	igt_assert_f(active == 2 && signaled == 1,
+		    "Fence did not signal properly\n");
+
+	sw_sync_timeline_inc(timeline[1], 5);
+	active = sw_sync_fence_count_status(fence_merge,
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	signaled = sw_sync_fence_count_status(fence_merge,
+					      SW_SYNC_FENCE_STATUS_SIGNALED);
+	igt_assert_f(active == 1 && signaled == 2,
+		    "Fence did not signal properly\n");
+
+	sw_sync_timeline_inc(timeline[2], 5);
+	active = sw_sync_fence_count_status(fence_merge,
+					    SW_SYNC_FENCE_STATUS_ACTIVE);
+	signaled = sw_sync_fence_count_status(fence_merge,
+					      SW_SYNC_FENCE_STATUS_SIGNALED);
+	igt_assert_f(active == 0 && signaled == 3,
+		     "Fence did not signal properly\n");
+
+	/* confirm you can successfully wait */
+	ret = sw_sync_wait(fence_merge, 100);
+	igt_assert_f(ret > 0, "Failure waiting on signaled fence\n");
+
+	close(in_fence[0]);
+	close(in_fence[1]);
+	close(in_fence[2]);
+	close(fence_merge);
+	close(timeline[0]);
+	close(timeline[1]);
+	close(timeline[2]);
+}
+
 static void * test_sync_multi_consumer_thread(void *arg)
 {
 	data_t *data = arg;
@@ -477,6 +540,9 @@ igt_main
 	igt_subtest("sync_merge_same")
 		test_sync_merge_same();
 
+	igt_subtest("sync_multi_timeline_wait")
+		test_sync_multi_timeline_wait();
+
 	igt_subtest("sync_multi_consumer")
 		test_sync_multi_consumer();
 
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#17 lore raw

PATCH v4 13/13 tests/sw_sync: Add subtest test_sync_multi_producer_single_consumer

In reply to [email protected]

From: Robert Foss <[email protected]>

This subtest runs a single consumer thread and multiple producer thread that are synchronized using multiple timelines.

---
 tests/sw_sync.c | 139 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 139 insertions(+)
tests/sw_sync.c +139 -0
index 9ec2dc2..1af9a6b 100644
@@ -447,6 +447,142 @@ static void test_sync_multi_consumer_producer(void)
 	igt_assert_f(thread_ret == 0, "A sync thread reported failure.\n");
 }
 
+static int test_mspc_wait_on_fence(int fence)
+{
+	int error, active;
+
+	do {
+		error = sw_sync_fence_count_status(fence,
+						   SW_SYNC_FENCE_STATUS_ERROR);
+		igt_assert_f(error == 0, "Error occurred on fence\n");
+		active = sw_sync_fence_count_status(fence,
+						    SW_SYNC_FENCE_STATUS_ACTIVE);
+	} while (active);
+
+	return 0;
+}
+
+static struct {
+	int iterations;
+	int threads;
+	int counter;
+	int cons_timeline;
+	int *prod_timeline;
+	pthread_mutex_t lock;
+} test_mpsc_data;
+
+static int mpsc_producer_thread(void *d)
+{
+	int id = (long)d;
+	int fence, i;
+	int *prod_timeline = test_mpsc_data.prod_timeline;
+	int cons_timeline = test_mpsc_data.cons_timeline;
+	int iterations = test_mpsc_data.iterations;
+
+	for (i = 0; i < iterations; i++) {
+		fence = sw_sync_fence_create(cons_timeline, i);
+
+		/* Wait for the consumer to finish. Use alternate
+		 * means of waiting on the fence
+		 */
+		if ((iterations + id) % 8 != 0) {
+			igt_assert_f(sw_sync_wait(fence, -1) > 0,
+				     "Failure waiting on fence\n");
+		} else {
+			igt_assert_f(test_mspc_wait_on_fence(fence) == 0,
+				     "Failure waiting on fence\n");
+		}
+
+		/* Every producer increments the counter, the consumer
+		 * checks and erases it
+		 */
+		pthread_mutex_lock(&test_mpsc_data.lock);
+		test_mpsc_data.counter++;
+		pthread_mutex_unlock(&test_mpsc_data.lock);
+
+		sw_sync_timeline_inc(prod_timeline[id], 1);
+		close(fence);
+	}
+
+	return 0;
+}
+
+static int mpsc_consumer_thread(void)
+{
+	int fence, merged, tmp, it, i;
+	int *prod_timeline = test_mpsc_data.prod_timeline;
+	int cons_timeline = test_mpsc_data.cons_timeline;
+	int iterations = test_mpsc_data.iterations;
+	int n = test_mpsc_data.threads;
+
+	for (it = 1; it <= iterations; it++) {
+		fence = sw_sync_fence_create(prod_timeline[0], it);
+		for (i = 1; i < n; i++) {
+			tmp = sw_sync_fence_create(prod_timeline[i], it);
+			merged = sw_sync_merge(tmp, fence);
+			close(tmp);
+			close(fence);
+			fence = merged;
+		}
+
+		/* Make sure we see an increment from every producer thread.
+		 * Vary the means by which we wait.
+		 */
+		if (iterations % 8 != 0) {
+			igt_assert_f(sw_sync_wait(fence, -1) == 0,
+				    "Producers did not increment as expected\n");
+		} else {
+			igt_assert_f(test_mspc_wait_on_fence(fence) == 0,
+				     "Failure waiting on fence\n");
+		}
+
+		igt_assert_f(test_mpsc_data.counter == n * it,
+			     "Counter value mismatch\n");
+
+		/* Release the producer threads */
+		sw_sync_timeline_inc(cons_timeline, 1);
+		close(fence);
+	}
+
+	return 0;
+}
+
+/* IMPORTANT NOTE: if you see this test failing on your system, it may be
+ * due to a shortage of file descriptors. Please ensure your system has
+ * a sensible limit for this test to finish correctly.
+ */
+static void test_sync_multi_producer_single_consumer(void)
+{
+	int iterations = 1 << 12;
+	int n = 5;
+	int prod_timeline[n];
+	int cons_timeline;
+	pthread_t threads[n];
+	long i;
+
+	cons_timeline = sw_sync_timeline_create();
+	for (i = 0; i < n; i++)
+		prod_timeline[i] = sw_sync_timeline_create();
+
+	test_mpsc_data.prod_timeline = prod_timeline;
+	test_mpsc_data.cons_timeline = cons_timeline;
+	test_mpsc_data.iterations = iterations;
+	test_mpsc_data.threads = n;
+	test_mpsc_data.counter = 0;
+	pthread_mutex_init(&test_mpsc_data.lock, NULL);
+
+	for (i = 0; i < n; i++) {
+		pthread_create(&threads[i], NULL, (void * (*)(void *))
+			       mpsc_producer_thread,
+			       (void *)i);
+	}
+
+	mpsc_consumer_thread();
+
+	for (i = 0; i < n; i++)
+		pthread_join(threads[i], NULL);
+}
+
 static void test_sync_random_merge(void)
 {
 	int i, size, ret;
@@ -549,6 +685,9 @@ igt_main
 	igt_subtest("sync_multi_consumer_producer")
 		test_sync_multi_consumer_producer();
 
+	igt_subtest("sync_multi_producer_single_consumer")
+		test_sync_multi_producer_single_consumer();
+
 	igt_subtest("sync_random_merge")
 		test_sync_random_merge();
 }
Signature · 6 lines
2.9.3

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx