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

[Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>]

On Wed, Mar 13, 2024 at 08:25:32PM -0700, Ashutosh Dixit wrote:
>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.
>
>v2: Fix implementation of xe_oa_circ_diff/xe_oa_circ_incr (Umesh)
>
>Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>

Thanks,
Umesh

>---
> drivers/gpu/drm/xe/xe_oa.c       | 35 ++++++++++++++++++++++++--------
> drivers/gpu/drm/xe/xe_oa_types.h |  3 +++
> 2 files changed, 30 insertions(+), 8 deletions(-)
>
>diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
>index 6f5bbb0787d9..1ad17cc14532 100644
>--- a/drivers/gpu/drm/xe/xe_oa.c
>+++ b/drivers/gpu/drm/xe/xe_oa.c
>@@ -106,7 +106,14 @@ 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);
>+	return tail >= head ? tail - head :
>+		tail + stream->oa_buffer.circ_size - head;
>+}
>+
>+static u32 xe_oa_circ_incr(struct xe_oa_stream *stream, u32 ptr, u32 n)
>+{
>+	return ptr + n >= stream->oa_buffer.circ_size ?
>+		ptr + n - stream->oa_buffer.circ_size : ptr + n;
> }
>
> static void xe_oa_config_release(struct kref *ref)
>@@ -280,7 +287,7 @@ 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 +313,6 @@ 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 +323,23 @@ 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 +377,6 @@ 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 +1263,18 @@ 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);
>diff --git a/drivers/gpu/drm/xe/xe_oa_types.h b/drivers/gpu/drm/xe/xe_oa_types.h
>index 6984e7d04be5..d8d5c9d8c22e 100644
>--- a/drivers/gpu/drm/xe/xe_oa_types.h
>+++ b/drivers/gpu/drm/xe/xe_oa_types.h
>@@ -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;
> };
>
> /**
>-- 
>2.41.0
>