[Intel-gfx] [PATCH i-g-t 1/2] i915: Inject invalid CS into hanging spinners

[Intel-gfx] [PATCH i-g-t 1/2] i915: Inject invalid CS into hanging spinners

[Chris Wilson]

Some spinners are used with the intent of never ending and being
declared hung by the kernel. In some cases, these are being used to
simulate invalid payloads and so we can use an invalid command to
trigger a GPU hang. (Other cases, they are simulating infinite workloads
that truly never end, but we still need to be able to curtail to provide
multi-tasking). This patch adds IGT_SPIN_INVALID_CS to request the
injection of 0xdeadbeef into the command stream that should trigger a
GPU hang.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 lib/igt_dummyload.c              |  2 ++
 lib/igt_dummyload.h              |  1 +
 tests/i915/gem_busy.c            |  3 ++-
 tests/i915/gem_ctx_persistence.c | 39 +++++++++++++++++++++++++++++++-
 tests/i915/gem_eio.c             |  1 +
 tests/i915/gem_exec_balancer.c   |  4 +++-
 tests/i915/gem_exec_fence.c      |  3 ++-
 7 files changed, 49 insertions(+), 4 deletions(-)

diff --git a/lib/igt_dummyload.c b/lib/igt_dummyload.c
index b7f4caca3..041122af9 100644
--- a/lib/igt_dummyload.c
+++ b/lib/igt_dummyload.c
@@ -189,6 +189,8 @@ emit_recursive_batch(igt_spin_t *spin,
 	/* Allow ourselves to be preempted */
 	if (!(opts->flags & IGT_SPIN_NO_PREEMPTION))
 		*cs++ = MI_ARB_CHK;
+	if (opts->flags & IGT_SPIN_INVALID_CS)
+		*cs++ = 0xdeadbeef;
 
 	/* Pad with a few nops so that we do not completely hog the system.
 	 *
diff --git a/lib/igt_dummyload.h b/lib/igt_dummyload.h
index 421ca183b..cb696009f 100644
--- a/lib/igt_dummyload.h
+++ b/lib/igt_dummyload.h
@@ -62,6 +62,7 @@ struct igt_spin_factory {
 #define IGT_SPIN_POLL_RUN      (1 << 2)
 #define IGT_SPIN_FAST          (1 << 3)
 #define IGT_SPIN_NO_PREEMPTION (1 << 4)
+#define IGT_SPIN_INVALID_CS    (1 << 5)
 
 igt_spin_t *
 __igt_spin_factory(int fd, const struct igt_spin_factory *opts);
diff --git a/tests/i915/gem_busy.c b/tests/i915/gem_busy.c
index 39a6714c2..2f1b04e3c 100644
--- a/tests/i915/gem_busy.c
+++ b/tests/i915/gem_busy.c
@@ -436,7 +436,8 @@ static void basic(int fd, const struct intel_execution_engine2 *e, unsigned flag
 	igt_spin_t *spin =
 		igt_spin_new(fd,
 			     .engine = e->flags,
-			     .flags = IGT_SPIN_NO_PREEMPTION);
+			     .flags = IGT_SPIN_NO_PREEMPTION |
+			     (flags & HANG ? IGT_SPIN_INVALID_CS : 0));
 	struct timespec tv;
 	int timeout;
 	bool busy;
diff --git a/tests/i915/gem_ctx_persistence.c b/tests/i915/gem_ctx_persistence.c
index d68431ae0..d48234450 100644
--- a/tests/i915/gem_ctx_persistence.c
+++ b/tests/i915/gem_ctx_persistence.c
@@ -345,6 +345,41 @@ static void test_nohangcheck_hostile(int i915)
 	close(dir);
 }
 
+static void test_nohangcheck_hang(int i915)
+{
+	int64_t timeout = reset_timeout_ms * NSEC_PER_MSEC;
+	int dir;
+
+	/*
+	 * Even if the user disables hangcheck during their context,
+	 * we forcibly terminate that context.
+	 */
+
+	dir = igt_sysfs_open_parameters(i915);
+	igt_require(dir != -1);
+
+	igt_require(__enable_hangcheck(dir, false));
+
+	for_each_physical_engine(e, i915) {
+		uint32_t ctx = gem_context_create(i915);
+		igt_spin_t *spin;
+
+		spin = igt_spin_new(i915, ctx,
+				    .engine = eb_ring(e),
+				    .flags = IGT_SPIN_INVALID_CS);
+		gem_context_destroy(i915, ctx);
+
+		igt_assert_eq(gem_wait(i915, spin->handle, &timeout), 0);
+
+		igt_spin_free(i915, spin);
+	}
+
+	igt_require(__enable_hangcheck(dir, true));
+
+	gem_quiescent_gpu(i915);
+	close(dir);
+}
+
 static void test_nonpersistent_file(int i915)
 {
 	int debugfs = i915;
@@ -724,8 +759,10 @@ igt_main
 	igt_subtest("processes")
 		test_processes(i915);
 
-	igt_subtest("hangcheck")
+	igt_subtest("hostile")
 		test_nohangcheck_hostile(i915);
+	igt_subtest("hang")
+		test_nohangcheck_hang(i915);
 
 	__for_each_static_engine(e) {
 		igt_subtest_group {
diff --git a/tests/i915/gem_eio.c b/tests/i915/gem_eio.c
index b23dfecc6..aa4accc9d 100644
--- a/tests/i915/gem_eio.c
+++ b/tests/i915/gem_eio.c
@@ -187,6 +187,7 @@ static igt_spin_t * __spin_poll(int fd, uint32_t ctx, unsigned long flags)
 		.engine = flags,
 		.flags = (IGT_SPIN_FAST |
 			  IGT_SPIN_NO_PREEMPTION |
+			  IGT_SPIN_INVALID_CS |
 			  IGT_SPIN_FENCE_OUT),
 	};
 
diff --git a/tests/i915/gem_exec_balancer.c b/tests/i915/gem_exec_balancer.c
index cebcc39c7..6b0c47f09 100644
--- a/tests/i915/gem_exec_balancer.c
+++ b/tests/i915/gem_exec_balancer.c
@@ -1654,7 +1654,9 @@ static void hangme(int i915)
 			set_unbannable(i915, ctx);
 			set_load_balancer(i915, ctx, ci, count, NULL);
 
-			flags = IGT_SPIN_FENCE_OUT | IGT_SPIN_NO_PREEMPTION;
+			flags = IGT_SPIN_FENCE_OUT |
+				IGT_SPIN_NO_PREEMPTION |
+				IGT_SPIN_INVALID_CS;
 			for (int j = 0; j < ARRAY_SIZE(c->spin); j++)  {
 				c->spin[j] = igt_spin_new(i915, ctx,
 							  .flags = flags);
diff --git a/tests/i915/gem_exec_fence.c b/tests/i915/gem_exec_fence.c
index 2f802eece..6d369f589 100644
--- a/tests/i915/gem_exec_fence.c
+++ b/tests/i915/gem_exec_fence.c
@@ -335,7 +335,8 @@ static void test_fence_await(int fd, unsigned ring, unsigned flags)
 	spin = igt_spin_new(fd,
 			    .engine = ring,
 			    .flags = (IGT_SPIN_FENCE_OUT |
-				      IGT_SPIN_NO_PREEMPTION));
+				      IGT_SPIN_NO_PREEMPTION |
+				      (flags & HANG ? IGT_SPIN_INVALID_CS : 0)));
 	igt_assert(spin->out_fence != -1);
 
 	i = 0;
-- 
2.25.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[Intel-gfx] [PATCH i-g-t 2/2] i915_pm_rps: Be wary if RP0 == RPn

[Chris Wilson]

If the HW min/max frequencies are the same, there is not much range to
deal with and a couple of our invalid tests become confused as they are
actually no-ops.

Error reporting in i915_pm_rps is rudimentary and we deserve better.

Closes: https://gitlab.freedesktop.org/drm/intel/issues/1008
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 tests/i915/i915_pm_rps.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/tests/i915/i915_pm_rps.c b/tests/i915/i915_pm_rps.c
index b65eefb03..56c745a5b 100644
--- a/tests/i915/i915_pm_rps.c
+++ b/tests/i915/i915_pm_rps.c
@@ -397,9 +397,11 @@ static void min_max_config(void (*check)(void), bool load_gpu)
 	writeval_inval(sysfs_files[MIN].filp, origfreqs[RP0] + 1000);
 	check();
 
-	igt_debug("\nDecrease max to RPn (invalid)...\n");
-	writeval_inval(sysfs_files[MAX].filp, origfreqs[RPn]);
-	check();
+	if (origfreqs[RPn] < origfreqs[RP0]) {
+		igt_debug("\nDecrease max to RPn (invalid)...\n");
+		writeval_inval(sysfs_files[MAX].filp, origfreqs[RPn]);
+		check();
+	}
 
 	igt_debug("\nDecrease min to midpoint...\n");
 	writeval(sysfs_files[MIN].filp, fmid);
@@ -429,9 +431,11 @@ static void min_max_config(void (*check)(void), bool load_gpu)
 	writeval_inval(sysfs_files[MAX].filp, 0);
 	check();
 
-	igt_debug("\nIncrease min to RP0 (invalid)...\n");
-	writeval_inval(sysfs_files[MIN].filp, origfreqs[RP0]);
-	check();
+	if (origfreqs[RP0] > origfreqs[RP0]) {
+		igt_debug("\nIncrease min to RP0 (invalid)...\n");
+		writeval_inval(sysfs_files[MIN].filp, origfreqs[RP0]);
+		check();
+	}
 
 	igt_debug("\nIncrease max to midpoint...\n");
 	writeval(sysfs_files[MAX].filp, fmid);
-- 
2.25.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[Intel-gfx] [PATCH i-g-t v2] i915_pm_rps: Be wary if RP0 == RPn

[Chris Wilson]

If the HW min/max frequencies are the same, there is not much range to
deal with and a couple of our invalid tests become confused as they are
actually no-ops.

Error reporting in i915_pm_rps is rudimentary and we deserve better.

Closes: https://gitlab.freedesktop.org/drm/intel/issues/1008
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 tests/i915/i915_pm_rps.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/tests/i915/i915_pm_rps.c b/tests/i915/i915_pm_rps.c
index b65eefb03..cb1500d2f 100644
--- a/tests/i915/i915_pm_rps.c
+++ b/tests/i915/i915_pm_rps.c
@@ -397,9 +397,11 @@ static void min_max_config(void (*check)(void), bool load_gpu)
 	writeval_inval(sysfs_files[MIN].filp, origfreqs[RP0] + 1000);
 	check();
 
-	igt_debug("\nDecrease max to RPn (invalid)...\n");
-	writeval_inval(sysfs_files[MAX].filp, origfreqs[RPn]);
-	check();
+	if (origfreqs[RPn] < origfreqs[RP0]) {
+		igt_debug("\nDecrease max to RPn (invalid)...\n");
+		writeval_inval(sysfs_files[MAX].filp, origfreqs[RPn]);
+		check();
+	}
 
 	igt_debug("\nDecrease min to midpoint...\n");
 	writeval(sysfs_files[MIN].filp, fmid);
@@ -429,9 +431,11 @@ static void min_max_config(void (*check)(void), bool load_gpu)
 	writeval_inval(sysfs_files[MAX].filp, 0);
 	check();
 
-	igt_debug("\nIncrease min to RP0 (invalid)...\n");
-	writeval_inval(sysfs_files[MIN].filp, origfreqs[RP0]);
-	check();
+	if (origfreqs[RP0] > origfreqs[RPn]) {
+		igt_debug("\nIncrease min to RP0 (invalid)...\n");
+		writeval_inval(sysfs_files[MIN].filp, origfreqs[RP0]);
+		check();
+	}
 
 	igt_debug("\nIncrease max to midpoint...\n");
 	writeval(sysfs_files[MAX].filp, fmid);
-- 
2.25.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [Intel-gfx] [PATCH i-g-t 1/2] i915: Inject invalid CS into hanging spinners

[Mika Kuoppala]

Chris Wilson <chris@chris-wilson.co.uk> writes:

> Some spinners are used with the intent of never ending and being
> declared hung by the kernel. In some cases, these are being used to
> simulate invalid payloads and so we can use an invalid command to
> trigger a GPU hang. (Other cases, they are simulating infinite workloads
> that truly never end, but we still need to be able to curtail to provide
> multi-tasking). This patch adds IGT_SPIN_INVALID_CS to request the
> injection of 0xdeadbeef into the command stream that should trigger a
> GPU hang.

Ok so you want to differentiate between a never ending
and invalid payload as a separate. And also quicken the
resolve.

>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  lib/igt_dummyload.c              |  2 ++
>  lib/igt_dummyload.h              |  1 +
>  tests/i915/gem_busy.c            |  3 ++-
>  tests/i915/gem_ctx_persistence.c | 39 +++++++++++++++++++++++++++++++-
>  tests/i915/gem_eio.c             |  1 +
>  tests/i915/gem_exec_balancer.c   |  4 +++-
>  tests/i915/gem_exec_fence.c      |  3 ++-
>  7 files changed, 49 insertions(+), 4 deletions(-)
>
> diff --git a/lib/igt_dummyload.c b/lib/igt_dummyload.c
> index b7f4caca3..041122af9 100644
> --- a/lib/igt_dummyload.c
> +++ b/lib/igt_dummyload.c
> @@ -189,6 +189,8 @@ emit_recursive_batch(igt_spin_t *spin,
>  	/* Allow ourselves to be preempted */
>  	if (!(opts->flags & IGT_SPIN_NO_PREEMPTION))
>  		*cs++ = MI_ARB_CHK;
> +	if (opts->flags & IGT_SPIN_INVALID_CS)
> +		*cs++ = 0xdeadbeef;

Some cmd streamers might just ignore this but lets
see how it goes.

Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>

>  
>  	/* Pad with a few nops so that we do not completely hog the system.
>  	 *
> diff --git a/lib/igt_dummyload.h b/lib/igt_dummyload.h
> index 421ca183b..cb696009f 100644
> --- a/lib/igt_dummyload.h
> +++ b/lib/igt_dummyload.h
> @@ -62,6 +62,7 @@ struct igt_spin_factory {
>  #define IGT_SPIN_POLL_RUN      (1 << 2)
>  #define IGT_SPIN_FAST          (1 << 3)
>  #define IGT_SPIN_NO_PREEMPTION (1 << 4)
> +#define IGT_SPIN_INVALID_CS    (1 << 5)
>  
>  igt_spin_t *
>  __igt_spin_factory(int fd, const struct igt_spin_factory *opts);
> diff --git a/tests/i915/gem_busy.c b/tests/i915/gem_busy.c
> index 39a6714c2..2f1b04e3c 100644
> --- a/tests/i915/gem_busy.c
> +++ b/tests/i915/gem_busy.c
> @@ -436,7 +436,8 @@ static void basic(int fd, const struct intel_execution_engine2 *e, unsigned flag
>  	igt_spin_t *spin =
>  		igt_spin_new(fd,
>  			     .engine = e->flags,
> -			     .flags = IGT_SPIN_NO_PREEMPTION);
> +			     .flags = IGT_SPIN_NO_PREEMPTION |
> +			     (flags & HANG ? IGT_SPIN_INVALID_CS : 0));
>  	struct timespec tv;
>  	int timeout;
>  	bool busy;
> diff --git a/tests/i915/gem_ctx_persistence.c b/tests/i915/gem_ctx_persistence.c
> index d68431ae0..d48234450 100644
> --- a/tests/i915/gem_ctx_persistence.c
> +++ b/tests/i915/gem_ctx_persistence.c
> @@ -345,6 +345,41 @@ static void test_nohangcheck_hostile(int i915)
>  	close(dir);
>  }
>  
> +static void test_nohangcheck_hang(int i915)
> +{
> +	int64_t timeout = reset_timeout_ms * NSEC_PER_MSEC;
> +	int dir;
> +
> +	/*
> +	 * Even if the user disables hangcheck during their context,
> +	 * we forcibly terminate that context.
> +	 */
> +
> +	dir = igt_sysfs_open_parameters(i915);
> +	igt_require(dir != -1);
> +
> +	igt_require(__enable_hangcheck(dir, false));
> +
> +	for_each_physical_engine(e, i915) {
> +		uint32_t ctx = gem_context_create(i915);
> +		igt_spin_t *spin;
> +
> +		spin = igt_spin_new(i915, ctx,
> +				    .engine = eb_ring(e),
> +				    .flags = IGT_SPIN_INVALID_CS);
> +		gem_context_destroy(i915, ctx);
> +
> +		igt_assert_eq(gem_wait(i915, spin->handle, &timeout), 0);
> +
> +		igt_spin_free(i915, spin);
> +	}
> +
> +	igt_require(__enable_hangcheck(dir, true));
> +
> +	gem_quiescent_gpu(i915);
> +	close(dir);
> +}
> +
>  static void test_nonpersistent_file(int i915)
>  {
>  	int debugfs = i915;
> @@ -724,8 +759,10 @@ igt_main
>  	igt_subtest("processes")
>  		test_processes(i915);
>  
> -	igt_subtest("hangcheck")
> +	igt_subtest("hostile")
>  		test_nohangcheck_hostile(i915);
> +	igt_subtest("hang")
> +		test_nohangcheck_hang(i915);
>  
>  	__for_each_static_engine(e) {
>  		igt_subtest_group {
> diff --git a/tests/i915/gem_eio.c b/tests/i915/gem_eio.c
> index b23dfecc6..aa4accc9d 100644
> --- a/tests/i915/gem_eio.c
> +++ b/tests/i915/gem_eio.c
> @@ -187,6 +187,7 @@ static igt_spin_t * __spin_poll(int fd, uint32_t ctx, unsigned long flags)
>  		.engine = flags,
>  		.flags = (IGT_SPIN_FAST |
>  			  IGT_SPIN_NO_PREEMPTION |
> +			  IGT_SPIN_INVALID_CS |
>  			  IGT_SPIN_FENCE_OUT),
>  	};
>  
> diff --git a/tests/i915/gem_exec_balancer.c b/tests/i915/gem_exec_balancer.c
> index cebcc39c7..6b0c47f09 100644
> --- a/tests/i915/gem_exec_balancer.c
> +++ b/tests/i915/gem_exec_balancer.c
> @@ -1654,7 +1654,9 @@ static void hangme(int i915)
>  			set_unbannable(i915, ctx);
>  			set_load_balancer(i915, ctx, ci, count, NULL);
>  
> -			flags = IGT_SPIN_FENCE_OUT | IGT_SPIN_NO_PREEMPTION;
> +			flags = IGT_SPIN_FENCE_OUT |
> +				IGT_SPIN_NO_PREEMPTION |
> +				IGT_SPIN_INVALID_CS;
>  			for (int j = 0; j < ARRAY_SIZE(c->spin); j++)  {
>  				c->spin[j] = igt_spin_new(i915, ctx,
>  							  .flags = flags);
> diff --git a/tests/i915/gem_exec_fence.c b/tests/i915/gem_exec_fence.c
> index 2f802eece..6d369f589 100644
> --- a/tests/i915/gem_exec_fence.c
> +++ b/tests/i915/gem_exec_fence.c
> @@ -335,7 +335,8 @@ static void test_fence_await(int fd, unsigned ring, unsigned flags)
>  	spin = igt_spin_new(fd,
>  			    .engine = ring,
>  			    .flags = (IGT_SPIN_FENCE_OUT |
> -				      IGT_SPIN_NO_PREEMPTION));
> +				      IGT_SPIN_NO_PREEMPTION |
> +				      (flags & HANG ? IGT_SPIN_INVALID_CS : 0)));
>  	igt_assert(spin->out_fence != -1);
>  
>  	i = 0;
> -- 
> 2.25.0
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [Intel-gfx] [PATCH i-g-t 1/2] i915: Inject invalid CS into hanging spinners

[Chris Wilson]

Quoting Mika Kuoppala (2020-01-28 09:59:23)
> Chris Wilson <chris@chris-wilson.co.uk> writes:
> 
> > Some spinners are used with the intent of never ending and being
> > declared hung by the kernel. In some cases, these are being used to
> > simulate invalid payloads and so we can use an invalid command to
> > trigger a GPU hang. (Other cases, they are simulating infinite workloads
> > that truly never end, but we still need to be able to curtail to provide
> > multi-tasking). This patch adds IGT_SPIN_INVALID_CS to request the
> > injection of 0xdeadbeef into the command stream that should trigger a
> > GPU hang.
> 
> Ok so you want to differentiate between a never ending
> and invalid payload as a separate. And also quicken the
> resolve.
> 
> >
> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > ---
> >  lib/igt_dummyload.c              |  2 ++
> >  lib/igt_dummyload.h              |  1 +
> >  tests/i915/gem_busy.c            |  3 ++-
> >  tests/i915/gem_ctx_persistence.c | 39 +++++++++++++++++++++++++++++++-
> >  tests/i915/gem_eio.c             |  1 +
> >  tests/i915/gem_exec_balancer.c   |  4 +++-
> >  tests/i915/gem_exec_fence.c      |  3 ++-
> >  7 files changed, 49 insertions(+), 4 deletions(-)
> >
> > diff --git a/lib/igt_dummyload.c b/lib/igt_dummyload.c
> > index b7f4caca3..041122af9 100644
> > --- a/lib/igt_dummyload.c
> > +++ b/lib/igt_dummyload.c
> > @@ -189,6 +189,8 @@ emit_recursive_batch(igt_spin_t *spin,
> >       /* Allow ourselves to be preempted */
> >       if (!(opts->flags & IGT_SPIN_NO_PREEMPTION))
> >               *cs++ = MI_ARB_CHK;
> > +     if (opts->flags & IGT_SPIN_INVALID_CS)
> > +             *cs++ = 0xdeadbeef;
> 
> Some cmd streamers might just ignore this but lets
> see how it goes.

See the selftest in
https://patchwork.freedesktop.org/patch/350890/?series=72639&rev=1
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx