Skip to contentSkip to conversation

drm/xe/oa: Changes to OA_TAKEN

PATCH In review Ashutosh Dixit 2 participants 6 messages Mar 12, 2024 – Mar 14, 2024

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 16/17 drm/xe/oa: Changes to OA_TAKEN

Rename OA_TAKEN to xe_oa_circ_diff, since xe_oa_circ_diff better describes what the macro actually does. Also convert to function and add xe_oa_stream arg. These will be used in the following patch.

---
 drivers/gpu/drm/xe/xe_oa.c | 22 +++++++++++++---------
 1 file changed, 13 insertions(+), 9 deletions(-)
drivers/gpu/drm/xe/xe_oa.c +13 -9
index b33976db1da5..6f5bbb0787d9 100644
@@ -28,7 +28,6 @@
 #include "xe_sched_job.h"
 #include "xe_perf.h"
 
-#define OA_TAKEN(tail, head)	(((tail) - (head)) & (XE_OA_BUFFER_SIZE - 1))
 #define DEFAULT_POLL_FREQUENCY_HZ 200
 #define DEFAULT_POLL_PERIOD_NS (NSEC_PER_SEC / DEFAULT_POLL_FREQUENCY_HZ)
 #define XE_OA_UNIT_INVALID U32_MAX
@@ -105,6 +104,11 @@ static const struct xe_oa_format oa_formats[] = {
 	[XE_OA_FORMAT_PEC36u64_G1_4_G2_32]	= { 4, 320, DRM_FMT(PEC), HDR_64_BIT, 1, 0 },
 };
 
+static u32 xe_oa_circ_diff(struct xe_oa_stream *stream, u32 tail, u32 head)
+{
+	return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+}
+
 static void xe_oa_config_release(struct kref *ref)
 {
 	struct xe_oa_config *oa_config =
@@ -209,11 +213,11 @@ static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 	 * increments. Also report size may not be a power of 2. Compute potential
 	 * partially landed report in OA buffer.
 	 */
-	partial_report_size = OA_TAKEN(hw_tail, stream->oa_buffer.tail);
+	partial_report_size = xe_oa_circ_diff(stream, hw_tail, stream->oa_buffer.tail);
 	partial_report_size %= report_size;
 
 	/* Subtract partial amount off the tail */
-	hw_tail = OA_TAKEN(hw_tail, partial_report_size);
+	hw_tail = xe_oa_circ_diff(stream, hw_tail, partial_report_size);
 
 	tail = hw_tail;
 
@@ -225,24 +229,24 @@ static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 	 * This is assuming that the writes of the OA unit land in memory in the order
 	 * they were written.  If not : (╯°□°)╯︵ ┻━┻
 	 */
-	while (OA_TAKEN(tail, stream->oa_buffer.tail) >= report_size) {
+	while (xe_oa_circ_diff(stream, tail, stream->oa_buffer.tail) >= report_size) {
 		void *report = stream->oa_buffer.vaddr + tail;
 
 		if (oa_report_id(stream, report) || oa_timestamp(stream, report))
 			break;
 
-		tail = OA_TAKEN(tail, report_size);
+		tail = xe_oa_circ_diff(stream, tail, report_size);
 	}
 
-	if (OA_TAKEN(hw_tail, tail) > report_size)
+	if (xe_oa_circ_diff(stream, hw_tail, tail) > report_size)
 		drm_dbg(&stream->oa->xe->drm,
 			"unlanded report(s) head=0x%x tail=0x%x hw_tail=0x%x\n",
 			stream->oa_buffer.head, tail, hw_tail);
 
 	stream->oa_buffer.tail = tail;
 
-	pollin = OA_TAKEN(stream->oa_buffer.tail,
-			  stream->oa_buffer.head) >= report_size;
+	pollin = xe_oa_circ_diff(stream, stream->oa_buffer.tail,
+				 stream->oa_buffer.head) >= report_size;
 
 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
 
@@ -315,7 +319,7 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 
 	xe_assert(stream->oa->xe, head < XE_OA_BUFFER_SIZE && tail < XE_OA_BUFFER_SIZE);
 
-	for (; OA_TAKEN(tail, head); head = (head + report_size) & mask) {
+	for (; xe_oa_circ_diff(stream, tail, head); head = (head + report_size) & mask) {
 		u8 *report = oa_buf_base + head;
 
 		ret = xe_oa_append_report(stream, buf, count, offset, report);
Signature · 1 line
2.41.0
#1 lore raw

PATCH 17/17 drm/xe/oa: Enable Xe2+ overrun mode

Enable Xe2+ overrun mode. For Xe2+, when overrun mode is enabled, there are no partial reports at the end of buffer, making the OA buffer effectively a non-power-of-2 size circular buffer whose size, circ_size, is a multiple of the report size.

---
 drivers/gpu/drm/xe/xe_oa.c       | 36 +++++++++++++++++++++++++-------
 drivers/gpu/drm/xe/xe_oa_types.h |  3 +++
 2 files changed, 31 insertions(+), 8 deletions(-)
drivers/gpu/drm/xe/xe_oa.c +28 -8
index 6f5bbb0787d9..6a0d2e229254 100644
@@ -106,7 +106,15 @@ static const struct xe_oa_format oa_formats[] = {
 
 static u32 xe_oa_circ_diff(struct xe_oa_stream *stream, u32 tail, u32 head)
 {
-	return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+	if (stream->oa_buffer.circ_size == XE_OA_BUFFER_SIZE)
+		return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+	else
+		return (tail - head) % stream->oa_buffer.circ_size;
+}
+
+static u32 xe_oa_circ_incr(struct xe_oa_stream *stream, u32 ptr, u32 n)
+{
+	return xe_oa_circ_diff(stream, ptr, -n);
 }
 
 static void xe_oa_config_release(struct kref *ref)
@@ -280,7 +288,7 @@ static int xe_oa_append_report(struct xe_oa_stream *stream, char __user *buf,
 
 	buf += *offset;
 
-	oa_buf_end = stream->oa_buffer.vaddr + XE_OA_BUFFER_SIZE;
+	oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
 	report_size_partial = oa_buf_end - report;
 
 	if (report_size_partial < report_size) {
@@ -306,7 +314,6 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 	int report_size = stream->oa_buffer.format->size;
 	u8 *oa_buf_base = stream->oa_buffer.vaddr;
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
-	u32 mask = (XE_OA_BUFFER_SIZE - 1);
 	size_t start_offset = *offset;
 	unsigned long flags;
 	u32 head, tail;
@@ -317,21 +324,23 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 	tail = stream->oa_buffer.tail;
 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
 
-	xe_assert(stream->oa->xe, head < XE_OA_BUFFER_SIZE && tail < XE_OA_BUFFER_SIZE);
+	xe_assert(stream->oa->xe,
+		  head < stream->oa_buffer.circ_size && tail < stream->oa_buffer.circ_size);
 
-	for (; xe_oa_circ_diff(stream, tail, head); head = (head + report_size) & mask) {
+	for (; xe_oa_circ_diff(stream, tail, head);
+	     head = xe_oa_circ_incr(stream, head, report_size)) {
 		u8 *report = oa_buf_base + head;
 
 		ret = xe_oa_append_report(stream, buf, count, offset, report);
 		if (ret)
 			break;
 
-		if (is_power_of_2(report_size)) {
+		if (!(stream->oa_buffer.circ_size % report_size)) {
 			/* Clear out report id and timestamp to detect unlanded reports */
 			oa_report_id_clear(stream, (void *)report);
 			oa_timestamp_clear(stream, (void *)report);
 		} else {
-			u8 *oa_buf_end = stream->oa_buffer.vaddr + XE_OA_BUFFER_SIZE;
+			u8 *oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
 			u32 part = oa_buf_end - report;
 
 			/* Zero out the entire report */
@@ -369,7 +378,6 @@ static void xe_oa_init_oa_buffer(struct xe_oa_stream *stream)
 	xe_mmio_write32(stream->gt, __oa_regs(stream)->oa_head_ptr,
 			gtt_offset & OAG_OAHEADPTR_MASK);
 	stream->oa_buffer.head = 0;
-
 	/*
 	 * PRM says: "This MMIO must be set before the OATAILPTR register and after the
 	 * OAHEADPTR register. This is to enable proper functionality of the overflow bit".
@@ -1256,6 +1264,18 @@ static int xe_oa_stream_init(struct xe_oa_stream *stream,
 	stream->periodic = param->period_exponent > 0;
 	stream->period_exponent = param->period_exponent;
 
+	/*
+	 * For Xe2+, when overrun mode is enabled, there are no partial reports at the end
+	 * of buffer, making the OA buffer effectively a non-power-of-2 size circular
+	 * buffer whose size, circ_size, is a multiple of the report size
+	 */
+	if (GRAPHICS_VER(stream->oa->xe) >= 20 &&
+	    stream->hwe->oa_unit->type == DRM_XE_OA_UNIT_TYPE_OAG && stream->sample)
+		stream->oa_buffer.circ_size =
+			XE_OA_BUFFER_SIZE - XE_OA_BUFFER_SIZE % stream->oa_buffer.format->size;
+	else
+		stream->oa_buffer.circ_size = XE_OA_BUFFER_SIZE;
+
 	if (stream->exec_q && engine_supports_mi_query(stream->hwe)) {
 		/* If we don't find the context offset, just return error */
 		ret = xe_oa_set_ctx_ctrl_offset(stream);
drivers/gpu/drm/xe/xe_oa_types.h +3 -0
index 6984e7d04be5..d8d5c9d8c22e 100644
@@ -163,6 +163,9 @@ struct xe_oa_buffer {
 
 	/** @tail: The last verified cached tail where HW has completed writing */
 	u32 tail;
+
+	/** @circ_size: The effective circular buffer size, for Xe2+ */
+	u32 circ_size;
 };
 
 /**
Signature · 1 line
2.41.0
#2 lore raw

Re: [PATCH 17/17] drm/xe/oa: Enable Xe2+ overrun mode

Quoted 16 lines · Ashutosh Dixit

Enable Xe2+ overrun mode. For Xe2+, when overrun mode is enabled, there are no partial reports at the end of buffer, making the OA buffer effectively a non-power-of-2 size circular buffer whose size, circ_size, is a multiple of the report size.

---
drivers/gpu/drm/xe/xe_oa.c | 36 +++++++++++++++++++++++++-------
drivers/gpu/drm/xe/xe_oa_types.h | 3 +++

2 files changed, 31 insertions(+), 8 deletions(-)
drivers/gpu/drm/xe/xe_oa.c +0 -0
index 6f5bbb0787d9..6a0d2e229254 100644

drivers/gpu/drm/xe/xe_oa.c static const struct xe_oa_format oa_formats[] = { · line 113

Quoted 9 lines
drivers/gpu/drm/xe/xe_oa.c +5 -1
@@ -106,7 +106,15 @@ static const struct xe_oa_format oa_formats[] = {
 
 static u32 xe_oa_circ_diff(struct xe_oa_stream *stream, u32 tail, u32 head)
 {
-	return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+	if (stream->oa_buffer.circ_size == XE_OA_BUFFER_SIZE)
+		return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+	else
+		return (tail - head) % stream->oa_buffer.circ_size;
+}

For ex: consider a 16 MB buffer with a report size of 384 bytes. At the end of the buffer, you would have an empty space of 256 bytes (16 MB % 384)

(For ref: 16 MB = 0x1000000, 384 = 0x180)
In this case circ_size = 0xFFFF00

Let's say your head is pointing to 0xFFFD80 and tail is pointing to 0x180 (essentially there is one unread report at the end of the buffer and one unread report at the beginning of the buffer).

In this case, (tail - head) % stream->oa_buffer.circ_size, is not calculating the correct size. Should be 0x300, but I am not getting that. Can you please check/verify?

I am thinking we need something like this (roughly). We don't need the mod operation.

static u32 xe_oa_circ_diff(struct xe_oa_stream *stream, u32 tail, u32 > head)

{
	u32 available = (tail - head) & (XE_OA_BUFFER_SIZE - 1);

	/* head is always aligned to report size, but tail may not be */
	available = available % format_size;

	return tail >= head ? available : available - empty_space;
}
Quoted 84 lines
drivers/gpu/drm/xe/xe_oa.c +23 -7
+
+static u32 xe_oa_circ_incr(struct xe_oa_stream *stream, u32 ptr, u32 n)
+{
+	return xe_oa_circ_diff(stream, ptr, -n);
 }
 
 static void xe_oa_config_release(struct kref *ref)
@@ -280,7 +288,7 @@ static int xe_oa_append_report(struct xe_oa_stream *stream, char __user *buf,
 
 	buf += *offset;
 
-	oa_buf_end = stream->oa_buffer.vaddr + XE_OA_BUFFER_SIZE;
+	oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
 	report_size_partial = oa_buf_end - report;
 
 	if (report_size_partial < report_size) {
@@ -306,7 +314,6 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 	int report_size = stream->oa_buffer.format->size;
 	u8 *oa_buf_base = stream->oa_buffer.vaddr;
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
-	u32 mask = (XE_OA_BUFFER_SIZE - 1);
 	size_t start_offset = *offset;
 	unsigned long flags;
 	u32 head, tail;
@@ -317,21 +324,23 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 	tail = stream->oa_buffer.tail;
 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
 
-	xe_assert(stream->oa->xe, head < XE_OA_BUFFER_SIZE && tail < XE_OA_BUFFER_SIZE);
+	xe_assert(stream->oa->xe,
+		  head < stream->oa_buffer.circ_size && tail < stream->oa_buffer.circ_size);
 
-	for (; xe_oa_circ_diff(stream, tail, head); head = (head + report_size) & mask) {
+	for (; xe_oa_circ_diff(stream, tail, head);
+	     head = xe_oa_circ_incr(stream, head, report_size)) {
 		u8 *report = oa_buf_base + head;
 
 		ret = xe_oa_append_report(stream, buf, count, offset, report);
 		if (ret)
 			break;
 
-		if (is_power_of_2(report_size)) {
+		if (!(stream->oa_buffer.circ_size % report_size)) {
 			/* Clear out report id and timestamp to detect unlanded reports */
 			oa_report_id_clear(stream, (void *)report);
 			oa_timestamp_clear(stream, (void *)report);
 		} else {
-			u8 *oa_buf_end = stream->oa_buffer.vaddr + XE_OA_BUFFER_SIZE;
+			u8 *oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
 			u32 part = oa_buf_end - report;
 
 			/* Zero out the entire report */
@@ -369,7 +378,6 @@ static void xe_oa_init_oa_buffer(struct xe_oa_stream *stream)
 	xe_mmio_write32(stream->gt, __oa_regs(stream)->oa_head_ptr,
 			gtt_offset & OAG_OAHEADPTR_MASK);
 	stream->oa_buffer.head = 0;
-
 	/*
 	 * PRM says: "This MMIO must be set before the OATAILPTR register and after the
 	 * OAHEADPTR register. This is to enable proper functionality of the overflow bit".
@@ -1256,6 +1264,18 @@ static int xe_oa_stream_init(struct xe_oa_stream *stream,
 	stream->periodic = param->period_exponent > 0;
 	stream->period_exponent = param->period_exponent;
 
+	/*
+	 * For Xe2+, when overrun mode is enabled, there are no partial reports at the end
+	 * of buffer, making the OA buffer effectively a non-power-of-2 size circular
+	 * buffer whose size, circ_size, is a multiple of the report size
+	 */
+	if (GRAPHICS_VER(stream->oa->xe) >= 20 &&
+	    stream->hwe->oa_unit->type == DRM_XE_OA_UNIT_TYPE_OAG && stream->sample)
+		stream->oa_buffer.circ_size =
+			XE_OA_BUFFER_SIZE - XE_OA_BUFFER_SIZE % stream->oa_buffer.format->size;
+	else
+		stream->oa_buffer.circ_size = XE_OA_BUFFER_SIZE;
+
 	if (stream->exec_q && engine_supports_mi_query(stream->hwe)) {
 		/* If we don't find the context offset, just return error */
 		ret = xe_oa_set_ctx_ctrl_offset(stream);
drivers/gpu/drm/xe/xe_oa_types.h +0 -0
index 6984e7d04be5..d8d5c9d8c22e 100644

drivers/gpu/drm/xe/xe_oa_types.h struct xe_oa_buffer { · line 168

Quoted 6 lines
drivers/gpu/drm/xe/xe_oa_types.h +3 -0
@@ -163,6 +163,9 @@ struct xe_oa_buffer {
 
 	/** @tail: The last verified cached tail where HW has completed writing */
 	u32 tail;
+
+	/** @circ_size: The effective circular buffer size, for Xe2+ */
+	u32 circ_size;

You could store the difference here instead.

	/** @empty_space: empty space at tend of buffer */
	u32 empty_space;

Regards,
Umesh

Quoted 6 lines
drivers/gpu/drm/xe/xe_oa_types.h +0 -0
 };
 
 /**
Signature · 1 line
2.41.0
#3 lore raw

Re: [PATCH 17/17] drm/xe/oa: Enable Xe2+ overrun mode

Quoted 36 lines · Umesh Nerlige Ramappa
Quoted 16 lines · Ashutosh Dixit

Enable Xe2+ overrun mode. For Xe2+, when overrun mode is enabled, there are no partial reports at the end of buffer, making the OA buffer effectively a non-power-of-2 size circular buffer whose size, circ_size, is a multiple of the report size.

---
drivers/gpu/drm/xe/xe_oa.c | 36 +++++++++++++++++++++++++-------
drivers/gpu/drm/xe/xe_oa_types.h | 3 +++

2 files changed, 31 insertions(+), 8 deletions(-)
drivers/gpu/drm/xe/xe_oa.c +0 -0
index 6f5bbb0787d9..6a0d2e229254 100644

drivers/gpu/drm/xe/xe_oa.c static const struct xe_oa_format oa_formats[] = { · line 113

Quoted 9 lines
drivers/gpu/drm/xe/xe_oa.c +5 -1
@@ -106,7 +106,15 @@ static const struct xe_oa_format oa_formats[] = {
 
 static u32 xe_oa_circ_diff(struct xe_oa_stream *stream, u32 tail, u32 head)
 {
-	return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+	if (stream->oa_buffer.circ_size == XE_OA_BUFFER_SIZE)
+		return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+	else
+		return (tail - head) % stream->oa_buffer.circ_size;
+}

For ex: consider a 16 MB buffer with a report size of 384 bytes. At the end of the buffer, you would have an empty space of 256 bytes (16 MB % 384)

(For ref: 16 MB = 0x1000000, 384 = 0x180)
In this case circ_size = 0xFFFF00

Let's say your head is pointing to 0xFFFD80 and tail is pointing to 0x180 (essentially there is one unread report at the end of the buffer and one unread report at the beginning of the buffer).

In this case, (tail - head) % stream->oa_buffer.circ_size, is not calculating the correct size. Should be 0x300, but I am not getting that. Can you please check/verify?

Quoted 8 lines

I am thinking we need something like this (roughly). We don't need the mod operation.

static u32 xe_oa_circ_diff(struct xe_oa_stream *stream, u32 tail, u32 > head)

{
	u32 available = (tail - head) & (XE_OA_BUFFER_SIZE - 1);

	/* head is always aligned to report size, but tail may not be */
	available = available % format_size;

should be

available -= available % format_size;

Quoted 109 lines
	return tail >= head ? available : available - empty_space;
}
Quoted 84 lines
drivers/gpu/drm/xe/xe_oa.c +23 -7
+
+static u32 xe_oa_circ_incr(struct xe_oa_stream *stream, u32 ptr, u32 n)
+{
+	return xe_oa_circ_diff(stream, ptr, -n);
 }
 
 static void xe_oa_config_release(struct kref *ref)
@@ -280,7 +288,7 @@ static int xe_oa_append_report(struct xe_oa_stream *stream, char __user *buf,
 
 	buf += *offset;
 
-	oa_buf_end = stream->oa_buffer.vaddr + XE_OA_BUFFER_SIZE;
+	oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
 	report_size_partial = oa_buf_end - report;
 
 	if (report_size_partial < report_size) {
@@ -306,7 +314,6 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 	int report_size = stream->oa_buffer.format->size;
 	u8 *oa_buf_base = stream->oa_buffer.vaddr;
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
-	u32 mask = (XE_OA_BUFFER_SIZE - 1);
 	size_t start_offset = *offset;
 	unsigned long flags;
 	u32 head, tail;
@@ -317,21 +324,23 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 	tail = stream->oa_buffer.tail;
 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
 
-	xe_assert(stream->oa->xe, head < XE_OA_BUFFER_SIZE && tail < XE_OA_BUFFER_SIZE);
+	xe_assert(stream->oa->xe,
+		  head < stream->oa_buffer.circ_size && tail < stream->oa_buffer.circ_size);
 
-	for (; xe_oa_circ_diff(stream, tail, head); head = (head + report_size) & mask) {
+	for (; xe_oa_circ_diff(stream, tail, head);
+	     head = xe_oa_circ_incr(stream, head, report_size)) {
 		u8 *report = oa_buf_base + head;
 
 		ret = xe_oa_append_report(stream, buf, count, offset, report);
 		if (ret)
 			break;
 
-		if (is_power_of_2(report_size)) {
+		if (!(stream->oa_buffer.circ_size % report_size)) {
 			/* Clear out report id and timestamp to detect unlanded reports */
 			oa_report_id_clear(stream, (void *)report);
 			oa_timestamp_clear(stream, (void *)report);
 		} else {
-			u8 *oa_buf_end = stream->oa_buffer.vaddr + XE_OA_BUFFER_SIZE;
+			u8 *oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
 			u32 part = oa_buf_end - report;
 
 			/* Zero out the entire report */
@@ -369,7 +378,6 @@ static void xe_oa_init_oa_buffer(struct xe_oa_stream *stream)
 	xe_mmio_write32(stream->gt, __oa_regs(stream)->oa_head_ptr,
 			gtt_offset & OAG_OAHEADPTR_MASK);
 	stream->oa_buffer.head = 0;
-
 	/*
 	 * PRM says: "This MMIO must be set before the OATAILPTR register and after the
 	 * OAHEADPTR register. This is to enable proper functionality of the overflow bit".
@@ -1256,6 +1264,18 @@ static int xe_oa_stream_init(struct xe_oa_stream *stream,
 	stream->periodic = param->period_exponent > 0;
 	stream->period_exponent = param->period_exponent;
 
+	/*
+	 * For Xe2+, when overrun mode is enabled, there are no partial reports at the end
+	 * of buffer, making the OA buffer effectively a non-power-of-2 size circular
+	 * buffer whose size, circ_size, is a multiple of the report size
+	 */
+	if (GRAPHICS_VER(stream->oa->xe) >= 20 &&
+	    stream->hwe->oa_unit->type == DRM_XE_OA_UNIT_TYPE_OAG && stream->sample)
+		stream->oa_buffer.circ_size =
+			XE_OA_BUFFER_SIZE - XE_OA_BUFFER_SIZE % stream->oa_buffer.format->size;
+	else
+		stream->oa_buffer.circ_size = XE_OA_BUFFER_SIZE;
+
 	if (stream->exec_q && engine_supports_mi_query(stream->hwe)) {
 		/* If we don't find the context offset, just return error */
 		ret = xe_oa_set_ctx_ctrl_offset(stream);
drivers/gpu/drm/xe/xe_oa_types.h +0 -0
index 6984e7d04be5..d8d5c9d8c22e 100644

drivers/gpu/drm/xe/xe_oa_types.h struct xe_oa_buffer { · line 168

Quoted 6 lines
drivers/gpu/drm/xe/xe_oa_types.h +3 -0
@@ -163,6 +163,9 @@ struct xe_oa_buffer {
 
 	/** @tail: The last verified cached tail where HW has completed writing */
 	u32 tail;
+
+	/** @circ_size: The effective circular buffer size, for Xe2+ */
+	u32 circ_size;

You could store the difference here instead.

	/** @empty_space: empty space at tend of buffer */
	u32 empty_space;

Regards,
Umesh

Quoted 6 lines
drivers/gpu/drm/xe/xe_oa_types.h +0 -0
 };
 
 /**
Signature · 1 line
2.41.0
#4 lore raw

Re: [PATCH 17/17] drm/xe/oa: Enable Xe2+ overrun mode

In reply to Umesh Nerlige Ramappa
Quoted 1 line · Umesh Nerlige Ramappa

Hi Umesh,

Quoted 37 lines · Ashutosh Dixit
Quoted 16 lines

Enable Xe2+ overrun mode. For Xe2+, when overrun mode is enabled, there are no partial reports at the end of buffer, making the OA buffer effectively a non-power-of-2 size circular buffer whose size, circ_size, is a multiple of the report size.

---
drivers/gpu/drm/xe/xe_oa.c | 36 +++++++++++++++++++++++++-------
drivers/gpu/drm/xe/xe_oa_types.h | 3 +++

2 files changed, 31 insertions(+), 8 deletions(-)
drivers/gpu/drm/xe/xe_oa.c +0 -0
index 6f5bbb0787d9..6a0d2e229254 100644

drivers/gpu/drm/xe/xe_oa.c static const struct xe_oa_format oa_formats[] = { · line 113

Quoted 9 lines
drivers/gpu/drm/xe/xe_oa.c +5 -1
@@ -106,7 +106,15 @@ static const struct xe_oa_format oa_formats[] = {
 
 static u32 xe_oa_circ_diff(struct xe_oa_stream *stream, u32 tail, u32 head)
 {
-	return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+	if (stream->oa_buffer.circ_size == XE_OA_BUFFER_SIZE)
+		return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+	else
+		return (tail - head) % stream->oa_buffer.circ_size;
+}

For ex: consider a 16 MB buffer with a report size of 384 bytes. At the end of the buffer, you would have an empty space of 256 bytes (16 MB % 384)

(For ref: 16 MB = 0x1000000, 384 = 0x180)
In this case circ_size = 0xFFFF00

Let's say your head is pointing to 0xFFFD80 and tail is pointing to 0x180 (essentially there is one unread report at the end of the buffer and one unread report at the beginning of the buffer).

Quoted 6 lines

In this case, (tail - head) % stream->oa_buffer.circ_size, is not calculating the correct size. Should be 0x300, but I am not getting that. Can you please check/verify?

I am thinking we need something like this (roughly). We don't need the mod operation.

First of all, thank you so much for checking, and catching, this huge bug. I did some digging into it and it is to do with how the % operator behaves with -ve numbers, as well as with u32 vs. s32 data types. And yes we should not use the % operation.

If this bug had got merged, I can only imagine what a nightmare it would be to find and fix it, so it's great it is caught in the code review.

I missed this completely because IGT tests with Xe2 overrun mode enabled were passing in spite of this bug! Specially igt@non-zero-reason would hang previously, because we wouldn't find any reports because we would be misaligned, due to the empty space at the end of the OA buffer. This test was also passing in spite of this bug, I am still trying to figure out how it could still pass.

In any case, I have sent out the latest version of just this patch (rather than the entire series) with this issue hopefully fixed. Please take a look.

Quoted 14 lines

drivers/gpu/drm/xe/xe_oa_types.h struct xe_oa_buffer { · line 168

Quoted 11 lines
drivers/gpu/drm/xe/xe_oa_types.h +3 -0
index 6984e7d04be5..d8d5c9d8c22e 100644
@@ -163,6 +163,9 @@ struct xe_oa_buffer {
 
 	/** @tail: The last verified cached tail where HW has completed writing */
 	u32 tail;
+
+	/** @circ_size: The effective circular buffer size, for Xe2+ */
+	u32 circ_size;

You could store the difference here instead.

Quoted 2 lines
	/** @empty_space: empty space at tend of buffer */
	u32 empty_space;

I have left circ_size here as it was in the previous version, since I think it better indicates that we still have a non-power-of-2 sized circular buffer even in this Xe2+ overrun case.

Thanks.
--
Ashutosh

Reviewed-by on v1 16/17
#5 lore raw
In reply to Ashutosh Dixit
Quoted 5 lines · Ashutosh Dixit

Rename OA_TAKEN to xe_oa_circ_diff, since xe_oa_circ_diff better describes what the macro actually does. Also convert to function and add xe_oa_stream arg. These will be used in the following patch.

Umesh

Quoted 84 lines

---
drivers/gpu/drm/xe/xe_oa.c | 22 +++++++++++++---------

1 file changed, 13 insertions(+), 9 deletions(-)
drivers/gpu/drm/xe/xe_oa.c +13 -9
index b33976db1da5..6f5bbb0787d9 100644
@@ -28,7 +28,6 @@
 #include "xe_sched_job.h"
 #include "xe_perf.h"
 
-#define OA_TAKEN(tail, head)	(((tail) - (head)) & (XE_OA_BUFFER_SIZE - 1))
 #define DEFAULT_POLL_FREQUENCY_HZ 200
 #define DEFAULT_POLL_PERIOD_NS (NSEC_PER_SEC / DEFAULT_POLL_FREQUENCY_HZ)
 #define XE_OA_UNIT_INVALID U32_MAX
@@ -105,6 +104,11 @@ static const struct xe_oa_format oa_formats[] = {
 	[XE_OA_FORMAT_PEC36u64_G1_4_G2_32]	= { 4, 320, DRM_FMT(PEC), HDR_64_BIT, 1, 0 },
 };
 
+static u32 xe_oa_circ_diff(struct xe_oa_stream *stream, u32 tail, u32 head)
+{
+	return (tail - head) & (XE_OA_BUFFER_SIZE - 1);
+}
+
 static void xe_oa_config_release(struct kref *ref)
 {
 	struct xe_oa_config *oa_config =
@@ -209,11 +213,11 @@ static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 	 * increments. Also report size may not be a power of 2. Compute potential
 	 * partially landed report in OA buffer.
 	 */
-	partial_report_size = OA_TAKEN(hw_tail, stream->oa_buffer.tail);
+	partial_report_size = xe_oa_circ_diff(stream, hw_tail, stream->oa_buffer.tail);
 	partial_report_size %= report_size;
 
 	/* Subtract partial amount off the tail */
-	hw_tail = OA_TAKEN(hw_tail, partial_report_size);
+	hw_tail = xe_oa_circ_diff(stream, hw_tail, partial_report_size);
 
 	tail = hw_tail;
 
@@ -225,24 +229,24 @@ static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 	 * This is assuming that the writes of the OA unit land in memory in the order
 	 * they were written.  If not : (╯°□°)╯︵ ┻━┻
 	 */
-	while (OA_TAKEN(tail, stream->oa_buffer.tail) >= report_size) {
+	while (xe_oa_circ_diff(stream, tail, stream->oa_buffer.tail) >= report_size) {
 		void *report = stream->oa_buffer.vaddr + tail;
 
 		if (oa_report_id(stream, report) || oa_timestamp(stream, report))
 			break;
 
-		tail = OA_TAKEN(tail, report_size);
+		tail = xe_oa_circ_diff(stream, tail, report_size);
 	}
 
-	if (OA_TAKEN(hw_tail, tail) > report_size)
+	if (xe_oa_circ_diff(stream, hw_tail, tail) > report_size)
 		drm_dbg(&stream->oa->xe->drm,
 			"unlanded report(s) head=0x%x tail=0x%x hw_tail=0x%x\n",
 			stream->oa_buffer.head, tail, hw_tail);
 
 	stream->oa_buffer.tail = tail;
 
-	pollin = OA_TAKEN(stream->oa_buffer.tail,
-			  stream->oa_buffer.head) >= report_size;
+	pollin = xe_oa_circ_diff(stream, stream->oa_buffer.tail,
+				 stream->oa_buffer.head) >= report_size;
 
 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
 
@@ -315,7 +319,7 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 
 	xe_assert(stream->oa->xe, head < XE_OA_BUFFER_SIZE && tail < XE_OA_BUFFER_SIZE);
 
-	for (; OA_TAKEN(tail, head); head = (head + report_size) & mask) {
+	for (; xe_oa_circ_diff(stream, tail, head); head = (head + report_size) & mask) {
 		u8 *report = oa_buf_base + head;
 
 		ret = xe_oa_append_report(stream, buf, count, offset, report);
Signature · 1 line
2.41.0