Skip to contentSkip to conversation

igt_dummyload: clear signal handler on the desructor

PATCH In review Abdiel Janulgue 3 participants 6 messages Dec 1, 2016 – Mar 24, 2017 Library code

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 igt_dummyload: clear signal handler on the desructor

Fixes an issue when calling igt_spin_batch_set_timeout and then tearing down the spinner right away before it has the chance to timeout, causes the associated signal handler to linger. Make sure to remove the handler on the destructor as well.

---
 lib/igt_dummyload.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)
lib/igt_dummyload.c +11 -4
index 99515ea..b201af0 100644
@@ -169,9 +169,17 @@ igt_spin_batch_new(int fd, int engine, unsigned int dep_handle)
 	return spin;
 }
 
-static void sig_handler(int sig, siginfo_t *info, void *arg)
+static void clear_sig_handler(int sig)
 {
 	struct sigaction act;
+
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = SIG_DFL;
+	igt_assert(sigaction(sig, &act, NULL) == 0);
+}
+
+static void sig_handler(int sig, siginfo_t *info, void *arg)
+{
 	struct igt_spin *iter;
 
 	igt_list_for_each(iter, &spin_list, link) {
@@ -181,9 +189,7 @@ static void sig_handler(int sig, siginfo_t *info, void *arg)
 		}
 	}
 
-	memset(&act, 0, sizeof(act));
-	act.sa_handler = SIG_DFL;
-	igt_assert(sigaction(info->si_signo, &act, NULL) == 0);
+	clear_sig_handler(info->si_signo);
 }
 
 /**
@@ -248,6 +254,7 @@ void igt_spin_batch_end(igt_spin_t *spin)
 
 	*spin->batch = MI_BATCH_BUFFER_END;
 	__sync_synchronize();
+	clear_sig_handler(spin->signo);
 }
 
 /**
Signature · 6 lines
2.7.4

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

PATCH 2/2 igt/kms_flip: Fix set_dpms called with an idle bo

---
 tests/kms_flip.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tests/kms_flip.c +2 -2
index 289335a..f744b3d 100644
@@ -757,9 +757,9 @@ static unsigned int run_test_step(struct test_output *o)
 
 	if (o->flags & TEST_DPMS) {
 		if (spin_rcs)
-			igt_spin_batch_end(spin_rcs);
+			igt_spin_batch_set_timeout(spin_rcs, NSEC_PER_SEC);
 		if (spin_bcs)
-			igt_spin_batch_end(spin_bcs);
+		        igt_spin_batch_set_timeout(spin_bcs, NSEC_PER_SEC);
 		set_dpms(o, DRM_MODE_DPMS_ON);
 	}
 
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Reviewed-by on v1 2/2
#2 lore raw

Re: [i-g-t PATCH 2/2] igt/kms_flip: Fix set_dpms called with an idle bo

Quoted 10 lines · Abdiel Janulgue
---
 tests/kms_flip.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tests/kms_flip.c +0 -0
index 289335a..f744b3d 100644

tests/kms_flip.c static unsigned int run_test_step(struct test_output *o) · line 764

Quoted 10 lines
tests/kms_flip.c +2 -2
@@ -757,9 +757,9 @@ static unsigned int run_test_step(struct test_output *o)
 
 	if (o->flags & TEST_DPMS) {
 		if (spin_rcs)
-			igt_spin_batch_end(spin_rcs);
+			igt_spin_batch_set_timeout(spin_rcs, NSEC_PER_SEC);
 		if (spin_bcs)
-			igt_spin_batch_end(spin_bcs);
+		        igt_spin_batch_set_timeout(spin_bcs, NSEC_PER_SEC);
 		set_dpms(o, DRM_MODE_DPMS_ON);
 	}

-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: [i-g-t PATCH 2/2] igt/kms_flip: Fix set_dpms called with an idle bo

Quoted 2 lines · Chris Wilson
Quoted 1 line · Abdiel Janulgue

How is the bug that this commit fixes triggered? Reverting this change seems to fix [1] which, if I understand correctly, is caused by the atomic commit of set_dpms() to be waiting for the dummy write to the frontbuffer complete, while the spin batch never ends since the test is blocked and so doesn't handle the timer signal.

[1] https://bugs.freedesktop.org/show_bug.cgi?id=100261

Thanks,
Ander

Quoted 21 lines
Quoted 9 lines
---
 tests/kms_flip.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tests/kms_flip.c +0 -0
index 289335a..f744b3d 100644

tests/kms_flip.c static unsigned int run_test_step(struct test_output *o) · line 764

Quoted 10 lines
tests/kms_flip.c +2 -2
@@ -757,9 +757,9 @@ static unsigned int run_test_step(struct test_output *o)
 
 	if (o->flags & TEST_DPMS) {
 		if (spin_rcs)
-			igt_spin_batch_end(spin_rcs);
+			igt_spin_batch_set_timeout(spin_rcs, NSEC_PER_SEC);
 		if (spin_bcs)
-			igt_spin_batch_end(spin_bcs);
+		        igt_spin_batch_set_timeout(spin_bcs, NSEC_PER_SEC);
 		set_dpms(o, DRM_MODE_DPMS_ON);
 	}
Quoted 2 lines

-Chris

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

#4 lore raw

Re: [i-g-t PATCH 2/2] igt/kms_flip: Fix set_dpms called with an idle bo

Quoted 9 lines · Ander Conselvan De Oliveira
Quoted 2 lines · Chris Wilson
Quoted 1 line · Abdiel Janulgue

How is the bug that this commit fixes triggered? Reverting this change seems to fix [1] which, if I understand correctly, is caused by the atomic commit of set_dpms() to be waiting for the dummy write to the frontbuffer complete, while the spin batch never ends since the test is blocked and so doesn't handle the timer signal.

That is actually a kernel regression (that I'm responsible for). However it raises a good point that we don't actually want to interrupt the modeset to handle the signal in this case, otherwise the interrupt modeset and when it restarts, it will find it doesn't have to wait - invalidating our test that it can wait for completion. Hmm, that affects all users. We want to delegate that task to a thread/child to avoid interrupting the syscall and perturbing the actual code under test. -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
Reviewed-by on v1 1/2
#5 lore raw
In reply to Abdiel Janulgue
Quoted 45 lines · Abdiel Janulgue

Fixes an issue when calling igt_spin_batch_set_timeout and then tearing down the spinner right away before it has the chance to timeout, causes the associated signal handler to linger. Make sure to remove the handler on the destructor as well.

---
 lib/igt_dummyload.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)
lib/igt_dummyload.c +10 -4
index 99515ea..b201af0 100644
@@ -169,9 +169,17 @@ igt_spin_batch_new(int fd, int engine, unsigned int dep_handle)
 	return spin;
 }
 
-static void sig_handler(int sig, siginfo_t *info, void *arg)
+static void clear_sig_handler(int sig)
 {
 	struct sigaction act;
+
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = SIG_DFL;
+	igt_assert(sigaction(sig, &act, NULL) == 0);
+}
+
+static void sig_handler(int sig, siginfo_t *info, void *arg)
+{
 	struct igt_spin *iter;
 
 	igt_list_for_each(iter, &spin_list, link) {
@@ -181,9 +189,7 @@ static void sig_handler(int sig, siginfo_t *info, void *arg)
 		}
 	}
 
-	memset(&act, 0, sizeof(act));
-	act.sa_handler = SIG_DFL;
-	igt_assert(sigaction(info->si_signo, &act, NULL) == 0);
+	clear_sig_handler(info->si_signo);
 }
 
 /**

lib/igt_dummyload.c void igt_spin_batch_end(igt_spin_t *spin) · line 256

Quoted 3 lines
lib/igt_dummyload.c +0 -0
@@ -248,6 +254,7 @@ void igt_spin_batch_end(igt_spin_t *spin)
 
 	*spin->batch = MI_BATCH_BUFFER_END;
 	__sync_synchronize();

Newline here since these are two different phases.

lib/igt_dummyload.c void igt_spin_batch_end(igt_spin_t *spin) · line 257

Quoted 1 line
lib/igt_dummyload.c +1 -0
+	clear_sig_handler(spin->signo);

I was also thinking about decoupling from the list of spinners here as well. Just looked like some more work to use safe iterators and list_del_init().

-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