drm/i915/selftests: Spin on all engines simultaneously

Vanshidhar Konda asked for the simplest test "to verify that the kernel
can submit and hardware can execute batch buffers on all the command
streamers in parallel." We have a number of tests in userspace that
submit load to each engine and verify that it is present, but strictly
we have no selftest to prove that the kernel can _simultaneously_
execute on all known engines. (We have tests to demonstrate that we can
submit to HW in parallel, but we don't insist that they execute in
parallel.)

v2: Improve the igt_spinner support for older gen.
Suggested-by: NVanshidhar Konda <vanshidhar.r.konda@intel.com>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Vanshidhar Konda <vanshidhar.r.konda@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: NVanshidhar Konda <vanshidhar.r.konda@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191101101528.10553-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/i915_drv.h

@@ -891,6 +891,10 @@ struct intel_cdclk_state {
 	u8 voltage_level;
 };
 
+struct i915_selftest_stash {
+	atomic_t counter;
+};
+
 struct drm_i915_private {
 	struct drm_device drm;
 
@@ -1286,6 +1290,8 @@ struct drm_i915_private {
 	/* Mutex to protect the above hdcp component related values. */
 	struct mutex hdcp_comp_mutex;
 
+	I915_SELFTEST_DECLARE(struct i915_selftest_stash selftest;)
+
 	/*
 	 * NOTE: This is the dri1/ums dungeon, don't add stuff here. Your patch
 	 * will be rejected. Instead look for a better place.

drivers/gpu/drm/i915/selftests/i915_request.c

@@ -32,6 +32,7 @@
 #include "i915_random.h"
 #include "i915_selftest.h"
 #include "igt_live_test.h"
+#include "igt_spinner.h"
 #include "lib_sw_fence.h"
 
 #include "mock_drm.h"
@@ -1116,12 +1117,85 @@ static int __live_parallel_engineN(void *arg)
 	return 0;
 }
 
+static bool wake_all(struct drm_i915_private *i915)
+{
+	if (atomic_dec_and_test(&i915->selftest.counter)) {
+		wake_up_var(&i915->selftest.counter);
+		return true;
+	}
+
+	return false;
+}
+
+static int wait_for_all(struct drm_i915_private *i915)
+{
+	if (wake_all(i915))
+		return 0;
+
+	if (wait_var_event_timeout(&i915->selftest.counter,
+				   !atomic_read(&i915->selftest.counter),
+				   i915_selftest.timeout_jiffies))
+		return 0;
+
+	return -ETIME;
+}
+
+static int __live_parallel_spin(void *arg)
+{
+	struct intel_engine_cs *engine = arg;
+	struct igt_spinner spin;
+	struct i915_request *rq;
+	int err = 0;
+
+	/*
+	 * Create a spinner running for eternity on each engine. If a second
+	 * spinner is incorrectly placed on the same engine, it will not be
+	 * able to start in time.
+	 */
+
+	if (igt_spinner_init(&spin, engine->gt)) {
+		wake_all(engine->i915);
+		return -ENOMEM;
+	}
+
+	rq = igt_spinner_create_request(&spin,
+					engine->kernel_context,
+					MI_NOOP); /* no preemption */
+	if (IS_ERR(rq)) {
+		err = PTR_ERR(rq);
+		if (err == -ENODEV)
+			err = 0;
+		wake_all(engine->i915);
+		goto out_spin;
+	}
+
+	i915_request_get(rq);
+	i915_request_add(rq);
+	if (igt_wait_for_spinner(&spin, rq)) {
+		/* Occupy this engine for the whole test */
+		err = wait_for_all(engine->i915);
+	} else {
+		pr_err("Failed to start spinner on %s\n", engine->name);
+		err = -EINVAL;
+	}
+	igt_spinner_end(&spin);
+
+	if (err == 0 && i915_request_wait(rq, 0, HZ / 5) < 0)
+		err = -EIO;
+	i915_request_put(rq);
+
+out_spin:
+	igt_spinner_fini(&spin);
+	return err;
+}
+
 static int live_parallel_engines(void *arg)
 {
 	struct drm_i915_private *i915 = arg;
 	static int (* const func[])(void *arg) = {
 		__live_parallel_engine1,
 		__live_parallel_engineN,
+		__live_parallel_spin,
 		NULL,
 	};
 	const unsigned int nengines = num_uabi_engines(i915);
@@ -1147,6 +1221,8 @@ static int live_parallel_engines(void *arg)
 		if (err)
 			break;
 
+		atomic_set(&i915->selftest.counter, nengines);
+
 		idx = 0;
 		for_each_uabi_engine(engine, i915) {
 			tsk[idx] = kthread_run(*fn, engine,

drivers/gpu/drm/i915/selftests/igt_spinner.c

@@ -15,8 +15,6 @@ int igt_spinner_init(struct igt_spinner *spin, struct intel_gt *gt)
 	void *vaddr;
 	int err;
 
-	GEM_BUG_ON(INTEL_GEN(gt->i915) < 8);
-
 	memset(spin, 0, sizeof(*spin));
 	spin->gt = gt;
 
@@ -95,11 +93,15 @@ igt_spinner_create_request(struct igt_spinner *spin,
 	struct intel_engine_cs *engine = ce->engine;
 	struct i915_request *rq = NULL;
 	struct i915_vma *hws, *vma;
+	unsigned int flags;
 	u32 *batch;
 	int err;
 
 	GEM_BUG_ON(spin->gt != ce->vm->gt);
 
+	if (!intel_engine_can_store_dword(ce->engine))
+		return ERR_PTR(-ENODEV);
+
 	vma = i915_vma_instance(spin->obj, ce->vm, NULL);
 	if (IS_ERR(vma))
 		return ERR_CAST(vma);
@@ -132,16 +134,37 @@ igt_spinner_create_request(struct igt_spinner *spin,
 
 	batch = spin->batch;
 
-	*batch++ = MI_STORE_DWORD_IMM_GEN4;
-	*batch++ = lower_32_bits(hws_address(hws, rq));
-	*batch++ = upper_32_bits(hws_address(hws, rq));
+	if (INTEL_GEN(rq->i915) >= 8) {
+		*batch++ = MI_STORE_DWORD_IMM_GEN4;
+		*batch++ = lower_32_bits(hws_address(hws, rq));
+		*batch++ = upper_32_bits(hws_address(hws, rq));
+	} else if (INTEL_GEN(rq->i915) >= 6) {
+		*batch++ = MI_STORE_DWORD_IMM_GEN4;
+		*batch++ = 0;
+		*batch++ = hws_address(hws, rq);
+	} else if (INTEL_GEN(rq->i915) >= 4) {
+		*batch++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
+		*batch++ = 0;
+		*batch++ = hws_address(hws, rq);
+	} else {
+		*batch++ = MI_STORE_DWORD_IMM | MI_MEM_VIRTUAL;
+		*batch++ = hws_address(hws, rq);
+	}
 	*batch++ = rq->fence.seqno;
 
 	*batch++ = arbitration_command;
 
-	*batch++ = MI_BATCH_BUFFER_START | 1 << 8 | 1;
+	if (INTEL_GEN(rq->i915) >= 8)
+		*batch++ = MI_BATCH_BUFFER_START | BIT(8) | 1;
+	else if (IS_HASWELL(rq->i915))
+		*batch++ = MI_BATCH_BUFFER_START | MI_BATCH_PPGTT_HSW;
+	else if (INTEL_GEN(rq->i915) >= 6)
+		*batch++ = MI_BATCH_BUFFER_START;
+	else
+		*batch++ = MI_BATCH_BUFFER_START | MI_BATCH_GTT;
 	*batch++ = lower_32_bits(vma->node.start);
 	*batch++ = upper_32_bits(vma->node.start);
+
 	*batch++ = MI_BATCH_BUFFER_END; /* not reached */
 
 	intel_gt_chipset_flush(engine->gt);
@@ -153,7 +176,10 @@ igt_spinner_create_request(struct igt_spinner *spin,
 			goto cancel_rq;
 	}
 
-	err = engine->emit_bb_start(rq, vma->node.start, PAGE_SIZE, 0);
+	flags = 0;
+	if (INTEL_GEN(rq->i915) <= 5)
+		flags |= I915_DISPATCH_SECURE;
+	err = engine->emit_bb_start(rq, vma->node.start, PAGE_SIZE, flags);
 
 cancel_rq:
 	if (err) {