[v10,3/4] drm/i915/perf: prepare driver to receive multiple ctx handles

Commit Message

Lionel Landwerlin April 30, 2020, 1:55 p.m. UTC

Make all the internal necessary changes before we flip the switch.

v2: Use an unlimited number of intel contexts (Chris)

v3: Handle GEM context with multiple RCS0 logical contexts (Chris)

v4: Let the intel_context create its own timeline (Chris)
    Only pin configuration context when needed (Chris)

v5: Pass filtering context ID by argument (Chris)

Signed-off-by: Lionel Landwerlin <[e-mail]>
---
 drivers/gpu/drm/i915/i915_perf.c       | 584 +++++++++++++++----------
 drivers/gpu/drm/i915/i915_perf_types.h |  37 +-
 2 files changed, 379 insertions(+), 242 deletions(-)

Comments

Chris Wilson April 30, 2020, 2:55 p.m. UTC | #1

Quoting Lionel Landwerlin (2020-04-30 14:55:35)
> @@ -1382,6 +1446,12 @@ static void i915_oa_stream_destroy(struct i915_perf_stream *stream)
>  
>         BUG_ON(stream != perf->exclusive_stream);
>  
> +       err = intel_context_pin(stream->config_context);
> +       if (err) {
> +               drm_err(&perf->i915->drm,
> +                       "Error pinning i915-perf context\n");
> +       }

Reading through, this is the last thing that leapt out at me.

Taking an error here would be nasty, and we do allow the pin to be
interrupted (which would be a surprise for most people who call
close()).

The only suggestion I can make, is that the stream always keeps it
pinned.

From the robustness point of view, it's better if we break
stream->config_context than if we break stream->engine->kernel_context,
but equally pinning another context image [about 80k in total] just for
safety? Probably worth it. [If we break the kernel_context, we cannot do
hang detection or pm/idle, the driver comes grinding to a halt.]
-Chris

Lionel Landwerlin April 30, 2020, 2:59 p.m. UTC | #2

On 30/04/2020 17:55, Chris Wilson wrote:
> Quoting Lionel Landwerlin (2020-04-30 14:55:35)
>> @@ -1382,6 +1446,12 @@ static void i915_oa_stream_destroy(struct i915_perf_stream *stream)
>>   
>>          BUG_ON(stream != perf->exclusive_stream);
>>   
>> +       err = intel_context_pin(stream->config_context);
>> +       if (err) {
>> +               drm_err(&perf->i915->drm,
>> +                       "Error pinning i915-perf context\n");
>> +       }
> Reading through, this is the last thing that leapt out at me.
>
> Taking an error here would be nasty, and we do allow the pin to be
> interrupted (which would be a surprise for most people who call
> close()).
>
> The only suggestion I can make, is that the stream always keeps it
> pinned.
>
> >From the robustness point of view, it's better if we break
> stream->config_context than if we break stream->engine->kernel_context,
> but equally pinning another context image [about 80k in total] just for
> safety? Probably worth it. [If we break the kernel_context, we cannot do
> hang detection or pm/idle, the driver comes grinding to a halt.]
> -Chris

Yeah, I didn't like writing this error path there either.

-Lionel

Patch

diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
index 44148b705a5a..1d5a7d31516a 100644
--- a/drivers/gpu/drm/i915/i915_perf.c
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -192,6 +192,7 @@ 
  */
 
 #include <linux/anon_inodes.h>
+#include <linux/bsearch.h>
 #include <linux/sizes.h>
 #include <linux/uuid.h>
 
@@ -329,7 +330,8 @@  static const struct i915_oa_format gen12_oa_formats[I915_OA_FORMAT_MAX] = {
  * @single_context: Whether a single or all gpu contexts should be monitored
  * @hold_preemption: Whether the preemption is disabled for the filtered
  *                   context
- * @ctx_handle: A gem ctx handle for use with @single_context
+ * @n_ctx_handles: Length of @ctx_handles
+ * @ctx_handles: An array of gem context handles
  * @metrics_set: An ID for an OA unit metric set advertised via sysfs
  * @oa_format: An OA unit HW report format
  * @oa_periodic: Whether to enable periodic OA unit sampling
@@ -349,9 +351,10 @@  static const struct i915_oa_format gen12_oa_formats[I915_OA_FORMAT_MAX] = {
 struct perf_open_properties {
 	u32 sample_flags;
 
-	u64 single_context:1;
 	u64 hold_preemption:1;
-	u64 ctx_handle;
+
+	u32 n_ctx_handles;
+	u32 *ctx_handles;
 
 	/* OA sampling state */
 	int metrics_set;
@@ -631,6 +634,23 @@  static int append_oa_sample(struct i915_perf_stream *stream,
 	return 0;
 }
 
+static int ctx_id_equal(const void *key, const void *elem)
+{
+	const struct i915_perf_context_detail *details = elem;
+
+	return ((int)details->id) - (uintptr_t)key;
+}
+
+static inline bool ctx_id_match(struct i915_perf_stream *stream,
+				u32 masked_ctx_id)
+{
+	return bsearch((void *)(uintptr_t)masked_ctx_id,
+		       stream->pinned_ctxs,
+		       stream->n_pinned_ctxs,
+		       sizeof(*stream->pinned_ctxs),
+		       ctx_id_equal) != NULL;
+}
+
 /**
  * Copies all buffered OA reports into userspace read() buffer.
  * @stream: An i915-perf stream opened for OA metrics
@@ -742,7 +762,7 @@  static int gen8_append_oa_reports(struct i915_perf_stream *stream,
 			continue;
 		}
 
-		ctx_id = report32[2] & stream->specific_ctx_id_mask;
+		ctx_id = report32[2] & stream->ctx_id_mask;
 
 		/*
 		 * Squash whatever is in the CTX_ID field if it's marked as
@@ -787,26 +807,32 @@  static int gen8_append_oa_reports(struct i915_perf_stream *stream,
 		 * switches since it's not-uncommon for periodic samples to
 		 * identify a switch before any 'context switch' report.
 		 */
-		if (!stream->perf->exclusive_stream->ctx ||
-		    stream->specific_ctx_id == ctx_id ||
-		    stream->oa_buffer.last_ctx_id == stream->specific_ctx_id ||
-		    reason & OAREPORT_REASON_CTX_SWITCH) {
-
-			/*
-			 * While filtering for a single context we avoid
-			 * leaking the IDs of other contexts.
-			 */
-			if (stream->perf->exclusive_stream->ctx &&
-			    stream->specific_ctx_id != ctx_id) {
-				report32[2] = INVALID_CTX_ID;
-			}
-
+		if (!stream->perf->exclusive_stream->n_ctxs) {
 			ret = append_oa_sample(stream, buf, count, offset,
 					       report);
 			if (ret)
 				break;
+		} else {
+			bool ctx_match = ctx_id != INVALID_CTX_ID &&
+				ctx_id_match(stream, ctx_id);
+
+			if (ctx_match ||
+			    stream->oa_buffer.last_ctx_match ||
+			    reason & OAREPORT_REASON_CTX_SWITCH) {
+				/*
+				 * While filtering for a single context we avoid
+				 * leaking the IDs of other contexts.
+				 */
+				if (!ctx_match)
+					report32[2] = INVALID_CTX_ID;
+
+				ret = append_oa_sample(stream, buf, count, offset,
+						       report);
+				if (ret)
+					break;
+			}
 
-			stream->oa_buffer.last_ctx_id = ctx_id;
+			stream->oa_buffer.last_ctx_match = ctx_match;
 		}
 
 		/*
@@ -1197,136 +1223,174 @@  static int i915_oa_read(struct i915_perf_stream *stream,
 	return stream->perf->ops.read(stream, buf, count, offset);
 }
 
-static struct intel_context *oa_pin_context(struct i915_perf_stream *stream)
+static u32 get_ctx_id_mask(struct intel_engine_cs *engine)
 {
-	struct i915_gem_engines_iter it;
-	struct i915_gem_context *ctx = stream->ctx;
-	struct intel_context *ce;
-	int err;
+	switch (INTEL_GEN(engine->i915)) {
+	case 7:
+		/*
+		 * On Haswell we don't do any post processing of the reports
+		 * and don't need to use the mask.
+		 */
+		return 0;
 
-	for_each_gem_engine(ce, i915_gem_context_lock_engines(ctx), it) {
-		if (ce->engine != stream->engine) /* first match! */
-			continue;
+	case 8:
+	case 9:
+	case 10:
+		if (intel_engine_in_execlists_submission_mode(engine))
+			return (1U << GEN8_CTX_ID_WIDTH) - 1;
 
 		/*
-		 * As the ID is the gtt offset of the context's vma we
-		 * pin the vma to ensure the ID remains fixed.
+		 * GuC uses the top bit to signal proxy submission, so ignore
+		 * that bit.
 		 */
-		err = intel_context_pin(ce);
-		if (err == 0) {
-			stream->pinned_ctx = ce;
-			break;
-		}
-	}
-	i915_gem_context_unlock_engines(ctx);
+		return (1U << (GEN8_CTX_ID_WIDTH - 1)) - 1;
 
-	return stream->pinned_ctx;
+	case 11:
+	case 12:
+		/*
+		 * 0x7ff is used by idle context.
+		 */
+		return ((1U << GEN11_SW_CTX_ID_WIDTH) - 1) << (GEN11_SW_CTX_ID_SHIFT - 32);
+
+	default:
+		MISSING_CASE(INTEL_GEN(engine->i915));
+		return 0;
+	}
 }
 
-/**
- * oa_get_render_ctx_id - determine and hold ctx hw id
- * @stream: An i915-perf stream opened for OA metrics
- *
- * Determine the render context hw id, and ensure it remains fixed for the
- * lifetime of the stream. This ensures that we don't have to worry about
- * updating the context ID in OACONTROL on the fly.
- *
- * Returns: zero on success or a negative error code
- */
-static int oa_get_render_ctx_id(struct i915_perf_stream *stream)
+static u32 get_ctx_id(struct intel_context *ce, int idx)
 {
-	struct intel_context *ce;
-
-	ce = oa_pin_context(stream);
-	if (IS_ERR(ce))
-		return PTR_ERR(ce);
 
 	switch (INTEL_GEN(ce->engine->i915)) {
-	case 7: {
-		/*
-		 * On Haswell we don't do any post processing of the reports
-		 * and don't need to use the mask.
-		 */
-		stream->specific_ctx_id = i915_ggtt_offset(ce->state);
-		stream->specific_ctx_id_mask = 0;
-		break;
-	}
+	case 7:
+		return i915_ggtt_offset(ce->state);
 
 	case 8:
 	case 9:
 	case 10:
-		if (intel_engine_in_execlists_submission_mode(ce->engine)) {
-			stream->specific_ctx_id_mask =
-				(1U << GEN8_CTX_ID_WIDTH) - 1;
-			stream->specific_ctx_id = stream->specific_ctx_id_mask;
-		} else {
-			/*
-			 * When using GuC, the context descriptor we write in
-			 * i915 is read by GuC and rewritten before it's
-			 * actually written into the hardware. The LRCA is
-			 * what is put into the context id field of the
-			 * context descriptor by GuC. Because it's aligned to
-			 * a page, the lower 12bits are always at 0 and
-			 * dropped by GuC. They won't be part of the context
-			 * ID in the OA reports, so squash those lower bits.
-			 */
-			stream->specific_ctx_id = ce->lrc.lrca >> 12;
+		if (intel_engine_in_execlists_submission_mode(ce->engine))
+			return (1U << GEN8_CTX_ID_WIDTH) - 1 - idx;
 
-			/*
-			 * GuC uses the top bit to signal proxy submission, so
-			 * ignore that bit.
-			 */
-			stream->specific_ctx_id_mask =
-				(1U << (GEN8_CTX_ID_WIDTH - 1)) - 1;
-		}
-		break;
+		/*
+		 * When using GuC, the context descriptor we write in i915 is
+		 * read by GuC and rewritten before it's actually written into
+		 * the hardware. The LRCA is what is put into the context id
+		 * field of the context descriptor by GuC. Because it's
+		 * aligned to a page, the lower 12bits are always at 0 and
+		 * dropped by GuC. They won't be part of the context ID in the
+		 * OA reports, so squash those lower bits.
+		 */
+		return ce->lrc.lrca >> 12;
 
 	case 11:
-	case 12: {
-		stream->specific_ctx_id_mask =
-			((1U << GEN11_SW_CTX_ID_WIDTH) - 1) << (GEN11_SW_CTX_ID_SHIFT - 32);
+	case 12:
 		/*
 		 * Pick an unused context id
 		 * 0 - BITS_PER_LONG are used by other contexts
 		 * GEN12_MAX_CONTEXT_HW_ID (0x7ff) is used by idle context
 		 */
-		stream->specific_ctx_id = (GEN12_MAX_CONTEXT_HW_ID - 1) << (GEN11_SW_CTX_ID_SHIFT - 32);
-		break;
-	}
+		return (GEN12_MAX_CONTEXT_HW_ID - 1 - idx) << (GEN11_SW_CTX_ID_SHIFT - 32);
 
 	default:
 		MISSING_CASE(INTEL_GEN(ce->engine->i915));
+		return 0;
 	}
-
-	ce->tag = stream->specific_ctx_id;
-
-	drm_dbg(&stream->perf->i915->drm,
-		"filtering on ctx_id=0x%x ctx_id_mask=0x%x\n",
-		stream->specific_ctx_id,
-		stream->specific_ctx_id_mask);
-
-	return 0;
 }
 
 /**
- * oa_put_render_ctx_id - counterpart to oa_get_render_ctx_id releases hold
+ * oa_put_render_ctx_id - counterpart to oa_get_render_ctx_ids releases hold
  * @stream: An i915-perf stream opened for OA metrics
  *
  * In case anything needed doing to ensure the context HW ID would remain valid
  * for the lifetime of the stream, then that can be undone here.
  */
-static void oa_put_render_ctx_id(struct i915_perf_stream *stream)
+static void oa_put_render_ctx_ids(struct i915_perf_stream *stream)
+{
+	int i;
+
+	for (i = 0; i < stream->n_pinned_ctxs; i++) {
+		struct intel_context *ce;
+
+		ce = fetch_and_zero(&stream->pinned_ctxs[i].ce);
+		if (ce) {
+			ce->tag = 0; /* recomputed on next submission after parking */
+			intel_context_unpin(ce);
+		}
+
+		stream->pinned_ctxs[i].id = INVALID_CTX_ID;
+	}
+
+	stream->ctx_id_mask = 0;
+	stream->n_pinned_ctxs = 0;
+
+	kfree(stream->pinned_ctxs);
+}
+
+static int oa_get_render_ctx_ids(struct i915_perf_stream *stream)
 {
 	struct intel_context *ce;
+	int i, err = 0;
+	u32 n_allocated_ctxs = 0;
 
-	ce = fetch_and_zero(&stream->pinned_ctx);
-	if (ce) {
-		ce->tag = 0; /* recomputed on next submission after parking */
-		intel_context_unpin(ce);
+	stream->ctx_id_mask = get_ctx_id_mask(stream->engine);
+
+	for (i = 0; i < stream->n_ctxs; i++) {
+		struct i915_gem_context *ctx = stream->ctxs[i];
+		struct i915_gem_engines_iter it;
+
+		for_each_gem_engine(ce, i915_gem_context_lock_engines(ctx), it) {
+			if (ce->engine != stream->engine) /* first match! */
+				continue;
+
+			/*
+			 * As the ID is the gtt offset of the context's vma we
+			 * pin the vma to ensure the ID remains fixed.
+			 */
+			err = intel_context_pin(ce);
+			if (err) {
+				i915_gem_context_unlock_engines(ctx);
+				break;
+			}
+
+			if (stream->n_pinned_ctxs >= n_allocated_ctxs) {
+				u32 new_allocated_len = max(n_allocated_ctxs * 2, 2u);
+				struct i915_perf_context_detail *new_ctxs =
+					krealloc(stream->pinned_ctxs,
+						 sizeof(*stream->pinned_ctxs) *
+						 new_allocated_len,
+						 GFP_KERNEL);
+
+				if (!new_ctxs) {
+					err = -ENOMEM;
+					break;
+				}
+
+				n_allocated_ctxs = new_allocated_len;
+				stream->pinned_ctxs = new_ctxs;
+			}
+
+			stream->pinned_ctxs[stream->n_pinned_ctxs].ce = ce;
+			stream->pinned_ctxs[stream->n_pinned_ctxs].id = get_ctx_id(ce, i);
+
+			drm_dbg(&stream->perf->i915->drm,
+				"filtering on ctx_id%i=0x%x ctx_id_mask=0x%x\n",
+				i, stream->pinned_ctxs[i].id, stream->ctx_id_mask);
+
+			ce->tag = stream->pinned_ctxs[stream->n_pinned_ctxs].id;
+
+			stream->n_pinned_ctxs++;
+		}
+		i915_gem_context_unlock_engines(ctx);
+		if (err)
+			goto err;
 	}
 
-	stream->specific_ctx_id = INVALID_CTX_ID;
-	stream->specific_ctx_id_mask = 0;
+	return 0;
+
+err:
+	oa_put_render_ctx_ids(stream);
+
+	return err;
 }
 
 static void
@@ -1382,6 +1446,12 @@  static void i915_oa_stream_destroy(struct i915_perf_stream *stream)
 
 	BUG_ON(stream != perf->exclusive_stream);
 
+	err = intel_context_pin(stream->config_context);
+	if (err) {
+		drm_err(&perf->i915->drm,
+			"Error pinning i915-perf context\n");
+	}
+
 	/*
 	 * Unset exclusive_stream first, it will be checked while disabling
 	 * the metric set on gen8+.
@@ -1389,10 +1459,13 @@  static void i915_oa_stream_destroy(struct i915_perf_stream *stream)
 	 * See i915_oa_init_reg_state() and lrc_configure_all_contexts()
 	 */
 	WRITE_ONCE(perf->exclusive_stream, NULL);
-	err = i915_perf_stream_sync(stream, false /* enable */);
-	if (err) {
-		drm_err(&perf->i915->drm,
-			"Error while disabling OA stream\n");
+
+	if (!err) {
+		err = i915_perf_stream_sync(stream, false /* enable */);
+		if (err) {
+			drm_err(&perf->i915->drm,
+				"Error while disabling OA stream\n");
+		}
 	}
 
 	intel_context_unpin(stream->config_context);
@@ -1403,8 +1476,7 @@  static void i915_oa_stream_destroy(struct i915_perf_stream *stream)
 	intel_uncore_forcewake_put(stream->uncore, FORCEWAKE_ALL);
 	intel_engine_pm_put(stream->engine);
 
-	if (stream->ctx)
-		oa_put_render_ctx_id(stream);
+	oa_put_render_ctx_ids(stream);
 
 	free_oa_configs(stream);
 	free_noa_wait(stream);
@@ -1496,7 +1568,7 @@  static void gen8_init_oa_buffer(struct i915_perf_stream *stream)
 	 * reports we will forward to userspace while filtering for a single
 	 * context.
 	 */
-	stream->oa_buffer.last_ctx_id = INVALID_CTX_ID;
+	stream->oa_buffer.last_ctx_match = false;
 
 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
 
@@ -1550,7 +1622,7 @@  static void gen12_init_oa_buffer(struct i915_perf_stream *stream)
 	 * reports we will forward to userspace while filtering for a single
 	 * context.
 	 */
-	stream->oa_buffer.last_ctx_id = INVALID_CTX_ID;
+	stream->oa_buffer.last_ctx_match = false;
 
 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
 
@@ -2273,11 +2345,10 @@  static int gen8_configure_context(struct i915_perf_stream *stream,
 	return err;
 }
 
-static int gen12_configure_oar_context(struct i915_perf_stream *stream,
-				       struct i915_active *active)
+static int gen12_configure_oar_contexts(struct i915_perf_stream *stream,
+					struct i915_active *active)
 {
-	int err;
-	struct intel_context *ce = stream->pinned_ctx;
+	int i;
 	u32 format = stream->oa_buffer.format;
 	struct flex regs_context[] = {
 		{
@@ -2298,7 +2369,7 @@  static int gen12_configure_oar_context(struct i915_perf_stream *stream,
 			(active ? GEN12_OAR_OACONTROL_COUNTER_ENABLE : 0)
 		},
 		{
-			RING_CONTEXT_CONTROL(ce->engine->mmio_base),
+			RING_CONTEXT_CONTROL(stream->engine->mmio_base),
 			CTX_CONTEXT_CONTROL,
 			_MASKED_FIELD(GEN12_CTX_CTRL_OAR_CONTEXT_ENABLE,
 				      active ?
@@ -2307,18 +2378,28 @@  static int gen12_configure_oar_context(struct i915_perf_stream *stream,
 		},
 	};
 
-	/* Modify the context image of pinned context with regs_context*/
-	err = intel_context_lock_pinned(ce);
-	if (err)
-		return err;
+	for (i = 0; i < stream->n_pinned_ctxs; i++) {
+		struct intel_context *ce = stream->pinned_ctxs[i].ce;
+		int err;
 
-	err = gen8_modify_context(stream, ce, regs_context, ARRAY_SIZE(regs_context));
-	intel_context_unlock_pinned(ce);
-	if (err)
-		return err;
+		/* Modify the context image of pinned context with regs_context*/
+		err = intel_context_lock_pinned(ce);
+		if (err)
+			return err;
 
-	/* Apply regs_lri using LRI with pinned context */
-	return gen8_modify_self(ce, regs_lri, ARRAY_SIZE(regs_lri), active);
+		err = gen8_modify_context(stream, ce, regs_context, ARRAY_SIZE(regs_context));
+		intel_context_unlock_pinned(ce);
+		if (err)
+			return err;
+
+		/* Apply regs_lri using LRI with pinned context */
+		err = gen8_modify_self(ce, regs_lri, ARRAY_SIZE(regs_lri),
+				       active);
+		if (err)
+			return err;
+	}
+
+	return 0;
 }
 
 /*
@@ -2586,11 +2667,9 @@  gen12_enable_metric_set(struct i915_perf_stream *stream,
 	 * another set of performance registers. Configure the unit dealing
 	 * with those.
 	 */
-	if (stream->ctx) {
-		ret = gen12_configure_oar_context(stream, active);
-		if (ret)
-			return ret;
-	}
+	ret = gen12_configure_oar_contexts(stream, active);
+	if (ret)
+		return ret;
 
 	return emit_oa_config(stream, stream->oa_config,
 			      stream->config_context,
@@ -2631,8 +2710,7 @@  static void gen12_disable_metric_set(struct i915_perf_stream *stream,
 	gen12_configure_all_contexts(stream, NULL, active);
 
 	/* disable the context save/restore or OAR counters */
-	if (stream->ctx)
-		gen12_configure_oar_context(stream, active);
+	gen12_configure_oar_contexts(stream, active);
 
 	/* Make sure we disable noa to save power. */
 	intel_uncore_rmw(uncore, RPM_CONFIG1, GEN10_GT_NOA_ENABLE, 0);
@@ -2641,8 +2719,7 @@  static void gen12_disable_metric_set(struct i915_perf_stream *stream,
 static void gen7_oa_enable(struct i915_perf_stream *stream)
 {
 	struct intel_uncore *uncore = stream->uncore;
-	struct i915_gem_context *ctx = stream->ctx;
-	u32 ctx_id = stream->specific_ctx_id;
+	u32 ctx_id = stream->n_pinned_ctxs ? stream->pinned_ctxs[0].id : 0;
 	bool periodic = stream->periodic;
 	u32 period_exponent = stream->period_exponent;
 	u32 report_format = stream->oa_buffer.format;
@@ -2664,7 +2741,7 @@  static void gen7_oa_enable(struct i915_perf_stream *stream)
 			    GEN7_OACONTROL_TIMER_PERIOD_SHIFT) |
 			   (periodic ? GEN7_OACONTROL_TIMER_ENABLE : 0) |
 			   (report_format << GEN7_OACONTROL_FORMAT_SHIFT) |
-			   (ctx ? GEN7_OACONTROL_PER_CTX_ENABLE : 0) |
+			   (stream->n_ctxs ? GEN7_OACONTROL_PER_CTX_ENABLE : 0) |
 			   GEN7_OACONTROL_ENABLE);
 }
 
@@ -2881,7 +2958,7 @@  static int i915_oa_stream_init(struct i915_perf_stream *stream,
 	}
 
 	if (!(props->sample_flags & SAMPLE_OA_REPORT) &&
-	    (INTEL_GEN(perf->i915) < 12 || !stream->ctx)) {
+	    (INTEL_GEN(perf->i915) < 12 || !stream->n_ctxs)) {
 		DRM_DEBUG("Only OA report sampling supported\n");
 		return -EINVAL;
 	}
@@ -2929,12 +3006,10 @@  static int i915_oa_stream_init(struct i915_perf_stream *stream,
 	if (stream->periodic)
 		stream->period_exponent = props->oa_period_exponent;
 
-	if (stream->ctx) {
-		ret = oa_get_render_ctx_id(stream);
-		if (ret) {
-			DRM_DEBUG("Invalid context id to filter with\n");
-			return ret;
-		}
+	ret = oa_get_render_ctx_ids(stream);
+	if (ret) {
+		DRM_DEBUG("Invalid context id to filter with\n");
+		return ret;
 	}
 
 	ret = alloc_noa_wait(stream);
@@ -2971,21 +3046,14 @@  static int i915_oa_stream_init(struct i915_perf_stream *stream,
 
 	stream->ops = &i915_oa_stream_ops;
 
-	timeline = intel_timeline_create(stream->engine->gt, NULL);
-	if (IS_ERR(timeline)) {
-		ret = PTR_ERR(timeline);
-		goto err_timeline;
-	}
-
 	stream->config_context = intel_context_create(stream->engine);
 	if (IS_ERR(stream->config_context)) {
 		intel_timeline_put(timeline);
 		ret = PTR_ERR(stream->config_context);
-		goto err_timeline;
+		goto err_context;
 	}
 
 	stream->config_context->sseu = props->sseu;
-	stream->config_context->timeline = timeline;
 
 	ret = intel_context_pin(stream->config_context);
 	if (ret)
@@ -3000,6 +3068,8 @@  static int i915_oa_stream_init(struct i915_perf_stream *stream,
 		goto err_enable;
 	}
 
+	intel_context_unpin(stream->config_context);
+
 	DRM_DEBUG("opening stream oa config uuid=%s\n",
 		  stream->oa_config->uuid);
 
@@ -3020,7 +3090,7 @@  static int i915_oa_stream_init(struct i915_perf_stream *stream,
 err_context_pin:
 	intel_context_put(stream->config_context);
 
-err_timeline:
+err_context:
 	free_oa_buffer(stream);
 
 err_oa_buf_alloc:
@@ -3033,8 +3103,7 @@  static int i915_oa_stream_init(struct i915_perf_stream *stream,
 	free_noa_wait(stream);
 
 err_noa_wait_alloc:
-	if (stream->ctx)
-		oa_put_render_ctx_id(stream);
+	oa_put_render_ctx_ids(stream);
 
 	return ret;
 }
@@ -3227,8 +3296,12 @@  static void i915_perf_enable_locked(struct i915_perf_stream *stream)
 	if (stream->ops->enable)
 		stream->ops->enable(stream);
 
-	if (stream->hold_preemption)
-		intel_context_set_nopreempt(stream->pinned_ctx);
+	if (stream->hold_preemption) {
+		int i;
+
+		for (i = 0; i < stream->n_pinned_ctxs; i++)
+			intel_context_set_nopreempt(stream->pinned_ctxs[i].ce);
+	}
 }
 
 /**
@@ -3253,8 +3326,12 @@  static void i915_perf_disable_locked(struct i915_perf_stream *stream)
 	/* Allow stream->ops->disable() to refer to this */
 	stream->enabled = false;
 
-	if (stream->hold_preemption)
-		intel_context_clear_nopreempt(stream->pinned_ctx);
+	if (stream->hold_preemption) {
+		int i;
+
+		for (i = 0; i < stream->n_pinned_ctxs; i++)
+			intel_context_clear_nopreempt(stream->pinned_ctxs[i].ce);
+	}
 
 	if (stream->ops->disable)
 		stream->ops->disable(stream);
@@ -3272,36 +3349,46 @@  static long i915_perf_config_locked(struct i915_perf_stream *stream,
 		return -EINVAL;
 
 	if (config != stream->oa_config) {
-		struct intel_context *ce = stream->pinned_ctx ?: stream->config_context;
+		int i;
+
+		ret = intel_context_pin(stream->config_context);
+		if (ret)
+			goto err_config;
 
 		active = i915_active_create();
 		if (!active) {
+			intel_context_unpin(stream->config_context);
 			ret = -ENOMEM;
 			goto err_config;
 		}
 
-		/*
-		 * If OA is bound to a specific context, emit the
-		 * reconfiguration inline from that context. The update
-		 * will then be ordered with respect to submission on that
-		 * context.
-		 *
-		 * When set globally, we use a low priority kernel context,
-		 * so it will effectively take effect when idle.
-		 */
 		ret = emit_oa_config(stream, config,
-				     ce,
+				     stream->config_context,
 				     active,
 				     BIT(I915_OA_CONFIG_PART_GLOBAL) |
 				     BIT(I915_OA_CONFIG_PART_PER_CONTEXT));
-		if (ret)
+		if (ret) {
+			intel_context_unpin(stream->config_context);
 			goto err_active;
+		}
+
+		for (i = 0; i < stream->n_pinned_ctxs; i++) {
+			ret = emit_oa_config(stream, config,
+					     stream->pinned_ctxs[i].ce,
+					     active,
+					     BIT(I915_OA_CONFIG_PART_PER_CONTEXT));
+			if (ret) {
+				intel_context_unpin(stream->config_context);
+				goto err_active;
+			}
+		}
 
 		config = xchg(&stream->oa_config, config);
 	}
 
 err_active:
-	i915_active_put(active);
+	if (active)
+		i915_active_put(active);
 err_config:
 	i915_oa_config_put(config);
 
@@ -3382,9 +3469,10 @@  static void i915_perf_destroy_locked(struct i915_perf_stream *stream)
 	if (stream->ops->destroy)
 		stream->ops->destroy(stream);
 
-	if (stream->ctx)
-		i915_gem_context_put(stream->ctx);
+	while (stream->n_ctxs--)
+		i915_gem_context_put(stream->ctxs[stream->n_ctxs]);
 
+	kfree(stream->ctxs);
 	kfree(stream);
 }
 
@@ -3459,25 +3547,12 @@  i915_perf_open_ioctl_locked(struct i915_perf *perf,
 			    struct perf_open_properties *props,
 			    struct drm_file *file)
 {
-	struct i915_gem_context *specific_ctx = NULL;
+	struct drm_i915_file_private *file_priv = file->driver_priv;
 	struct i915_perf_stream *stream = NULL;
 	unsigned long f_flags = 0;
 	bool privileged_op = true;
 	int stream_fd;
-	int ret;
-
-	if (props->single_context) {
-		u32 ctx_handle = props->ctx_handle;
-		struct drm_i915_file_private *file_priv = file->driver_priv;
-
-		specific_ctx = i915_gem_context_lookup(file_priv, ctx_handle);
-		if (!specific_ctx) {
-			DRM_DEBUG("Failed to look up context with ID %u for opening perf stream\n",
-				  ctx_handle);
-			ret = -ENOENT;
-			goto err;
-		}
-	}
+	int i, ret;
 
 	/*
 	 * On Haswell the OA unit supports clock gating off for a specific
@@ -3498,17 +3573,16 @@  i915_perf_open_ioctl_locked(struct i915_perf *perf,
 	 * doesn't request global stream access (i.e. query based sampling
 	 * using MI_RECORD_PERF_COUNT.
 	 */
-	if (IS_HASWELL(perf->i915) && specific_ctx)
+	if (IS_HASWELL(perf->i915) && props->n_ctx_handles > 0)
 		privileged_op = false;
-	else if (IS_GEN(perf->i915, 12) && specific_ctx &&
+	else if (IS_GEN(perf->i915, 12) && (props->n_ctx_handles > 0) &&
 		 (props->sample_flags & SAMPLE_OA_REPORT) == 0)
 		privileged_op = false;
 
 	if (props->hold_preemption) {
-		if (!props->single_context) {
+		if (!props->n_ctx_handles) {
 			DRM_DEBUG("preemption disable with no context\n");
-			ret = -EINVAL;
-			goto err;
+			return -EINVAL;
 		}
 		privileged_op = true;
 	}
@@ -3529,23 +3603,43 @@  i915_perf_open_ioctl_locked(struct i915_perf *perf,
 	if (privileged_op &&
 	    i915_perf_stream_paranoid && !capable(CAP_SYS_ADMIN)) {
 		DRM_DEBUG("Insufficient privileges to open i915 perf stream\n");
-		ret = -EACCES;
-		goto err_ctx;
+		return -EACCES;
 	}
 
 	stream = kzalloc(sizeof(*stream), GFP_KERNEL);
-	if (!stream) {
-		ret = -ENOMEM;
-		goto err_ctx;
+	if (!stream)
+		return -ENOMEM;
+
+	if (props->n_ctx_handles) {
+		gfp_t alloc_flags = GFP_KERNEL | __GFP_ZERO;
+
+		stream->ctxs = kmalloc_array(props->n_ctx_handles,
+					     sizeof(*stream->ctxs),
+					     alloc_flags);
+		if (!stream->ctxs)
+			goto err_ctx;
 	}
 
 	stream->perf = perf;
-	stream->ctx = specific_ctx;
 	stream->poll_oa_period = props->poll_oa_period;
 
+	for (i = 0; i < props->n_ctx_handles; i++) {
+		stream->ctxs[i] = i915_gem_context_lookup(file_priv,
+							  props->ctx_handles[i]);
+		if (!stream->ctxs[i]) {
+			DRM_DEBUG("Failed to look up context with ID %u for opening perf stream\n",
+				  props->ctx_handles[i]);
+
+			ret = -ENOENT;
+			goto err_ctx;
+		}
+
+		stream->n_ctxs++;
+	}
+
 	ret = i915_oa_stream_init(stream, param, props);
 	if (ret)
-		goto err_alloc;
+		goto err_ctx;
 
 	/* we avoid simply assigning stream->sample_flags = props->sample_flags
 	 * to have _stream_init check the combination of sample flags more
@@ -3580,12 +3674,11 @@  i915_perf_open_ioctl_locked(struct i915_perf *perf,
 err_flags:
 	if (stream->ops->destroy)
 		stream->ops->destroy(stream);
-err_alloc:
-	kfree(stream);
 err_ctx:
-	if (specific_ctx)
-		i915_gem_context_put(specific_ctx);
-err:
+	while (stream->n_ctxs--)
+		i915_gem_context_put(stream->ctxs[stream->n_ctxs]);
+	kfree(stream->ctxs);
+	kfree(stream);
 	return ret;
 }
 
@@ -3617,7 +3710,7 @@  static int read_properties_unlocked(struct i915_perf *perf,
 {
 	u64 __user *uprop = uprops;
 	u32 i;
-	int ret;
+	int err;
 
 	memset(props, 0, sizeof(struct perf_open_properties));
 	props->poll_oa_period = DEFAULT_POLL_PERIOD_NS;
@@ -3651,23 +3744,36 @@  static int read_properties_unlocked(struct i915_perf *perf,
 		u64 oa_period, oa_freq_hz;
 		u64 id, value;
 
-		ret = get_user(id, uprop);
-		if (ret)
-			return ret;
+		err = get_user(id, uprop);
+		if (err)
+			goto error;
 
-		ret = get_user(value, uprop + 1);
-		if (ret)
-			return ret;
+		err = get_user(value, uprop + 1);
+		if (err)
+			goto error;
 
 		if (id == 0 || id >= DRM_I915_PERF_PROP_MAX) {
 			DRM_DEBUG("Unknown i915 perf property ID\n");
-			return -EINVAL;
+			err = -EINVAL;
+			goto error;
 		}
 
 		switch ((enum drm_i915_perf_property_id)id) {
 		case DRM_I915_PERF_PROP_CTX_HANDLE:
-			props->single_context = 1;
-			props->ctx_handle = value;
+			if (props->n_ctx_handles > 0) {
+				DRM_DEBUG("Context handle specified multiple times\n");
+				err = -EINVAL;
+				goto error;
+			}
+			props->ctx_handles =
+				kmalloc_array(1, sizeof(*props->ctx_handles),
+					      GFP_KERNEL);
+			if (!props->ctx_handles) {
+				err = -ENOMEM;
+				goto error;
+			}
+			props->ctx_handles[0] = value;
+			props->n_ctx_handles = 1;
 			break;
 		case DRM_I915_PERF_PROP_SAMPLE_OA:
 			if (value)
@@ -3676,7 +3782,8 @@  static int read_properties_unlocked(struct i915_perf *perf,
 		case DRM_I915_PERF_PROP_OA_METRICS_SET:
 			if (value == 0) {
 				DRM_DEBUG("Unknown OA metric set ID\n");
-				return -EINVAL;
+				err = -EINVAL;
+				goto error;
 			}
 			props->metrics_set = value;
 			break;
@@ -3684,12 +3791,14 @@  static int read_properties_unlocked(struct i915_perf *perf,
 			if (value == 0 || value >= I915_OA_FORMAT_MAX) {
 				DRM_DEBUG("Out-of-range OA report format %llu\n",
 					  value);
-				return -EINVAL;
+				err = -EINVAL;
+				goto error;
 			}
 			if (!perf->oa_formats[value].size) {
 				DRM_DEBUG("Unsupported OA report format %llu\n",
 					  value);
-				return -EINVAL;
+				err = -EINVAL;
+				goto error;
 			}
 			props->oa_format = value;
 			break;
@@ -3697,7 +3806,8 @@  static int read_properties_unlocked(struct i915_perf *perf,
 			if (value > OA_EXPONENT_MAX) {
 				DRM_DEBUG("OA timer exponent too high (> %u)\n",
 					 OA_EXPONENT_MAX);
-				return -EINVAL;
+				err = -EINVAL;
+				goto error;
 			}
 
 			/* Theoretically we can program the OA unit to sample
@@ -3726,7 +3836,8 @@  static int read_properties_unlocked(struct i915_perf *perf,
 			    !capable(CAP_SYS_ADMIN)) {
 				DRM_DEBUG("OA exponent would exceed the max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without root privileges\n",
 					  i915_oa_max_sample_rate);
-				return -EACCES;
+				err = -EACCES;
+				goto error;
 			}
 
 			props->oa_periodic = true;
@@ -3742,13 +3853,14 @@  static int read_properties_unlocked(struct i915_perf *perf,
 					   u64_to_user_ptr(value),
 					   sizeof(user_sseu))) {
 				DRM_DEBUG("Unable to copy global sseu parameter\n");
-				return -EFAULT;
+				err = -EFAULT;
+				goto error;
 			}
 
-			ret = get_sseu_config(&props->sseu, props->engine, &user_sseu);
-			if (ret) {
+			err = get_sseu_config(&props->sseu, props->engine, &user_sseu);
+			if (err) {
 				DRM_DEBUG("Invalid SSEU configuration\n");
-				return ret;
+				goto error;
 			}
 			props->has_sseu = true;
 			break;
@@ -3757,19 +3869,25 @@  static int read_properties_unlocked(struct i915_perf *perf,
 			if (value < 100000 /* 100us */) {
 				DRM_DEBUG("OA availability timer too small (%lluns < 100us)\n",
 					  value);
-				return -EINVAL;
+				err = -EINVAL;
+				goto error;
 			}
 			props->poll_oa_period = value;
 			break;
 		case DRM_I915_PERF_PROP_MAX:
 			MISSING_CASE(id);
-			return -EINVAL;
+			err = -EINVAL;
+			goto error;
 		}
 
 		uprop += 2;
 	}
 
 	return 0;
+
+error:
+	kfree(props->ctx_handles);
+	return err;
 }
 
 /**
@@ -3829,6 +3947,8 @@  int i915_perf_open_ioctl(struct drm_device *dev, void *data,
 	ret = i915_perf_open_ioctl_locked(perf, param, &props, file);
 	mutex_unlock(&perf->lock);
 
+	kfree(props.ctx_handles);
+
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/i915/i915_perf_types.h b/drivers/gpu/drm/i915/i915_perf_types.h
index a8b903592a39..278defe0b456 100644
--- a/drivers/gpu/drm/i915/i915_perf_types.h
+++ b/drivers/gpu/drm/i915/i915_perf_types.h
@@ -161,10 +161,15 @@  struct i915_perf_stream {
 	int sample_size;
 
 	/**
-	 * @ctx: %NULL if measuring system-wide across all contexts or a
-	 * specific context that is being monitored.
+	 * @n_ctxs: Number of contexts pinned for the recording.
 	 */
-	struct i915_gem_context *ctx;
+	u32 n_ctxs;
+
+	/**
+	 * @ctxs: All to %NULL if measuring system-wide across all contexts or
+	 * a list specific contexts that are being monitored.
+	 */
+	struct i915_gem_context **ctxs;
 
 	/**
 	 * @enabled: Whether the stream is currently enabled, considering
@@ -199,19 +204,31 @@  struct i915_perf_stream {
 	struct llist_head oa_config_bos;
 
 	/**
-	 * @pinned_ctx: The OA context specific information.
+	 * @pinned_ctxs: A array of logical context details needed for
+	 * filtering and their associated pinned ID.
 	 */
-	struct intel_context *pinned_ctx;
+	struct i915_perf_context_detail {
+		/**
+		 * @ce: The OA context specific information.
+		 */
+		struct intel_context *ce;
+
+		/**
+		 * @id: The ids of the specific contexts.
+		 */
+		u32 id;
+	} *pinned_ctxs;
 
 	/**
-	 * @specific_ctx_id: The id of the specific context.
+	 * @n_pinned_ctxs: Length of the @pinned_ctxs array, 0 if measuring
+	 * system-wide across all contexts.
 	 */
-	u32 specific_ctx_id;
+	u32 n_pinned_ctxs;
 
 	/**
-	 * @specific_ctx_id_mask: The mask used to masking specific_ctx_id bits.
+	 * @ctx_id_mask: The mask used to masking specific_ctx_id bits.
 	 */
-	u32 specific_ctx_id_mask;
+	u32 ctx_id_mask;
 
 	/**
 	 * @poll_check_timer: High resolution timer that will periodically
@@ -247,7 +264,7 @@  struct i915_perf_stream {
 	struct {
 		struct i915_vma *vma;
 		u8 *vaddr;
-		u32 last_ctx_id;
+		bool last_ctx_match;
 		int format;
 		int format_size;
 		int size_exponent;