openwrt/target/linux/bcm27xx/patches-5.4/950-0500-media-v4l2-mem2mem-allow-request-job-buffer-processi.patch
Álvaro Fernández Rojas f07e572f64 bcm27xx: import latest patches from the RPi foundation
bcm2708: boot tested on RPi B+ v1.2
bcm2709: boot tested on RPi 3B v1.2 and RPi 4B v1.1 4G
bcm2710: boot tested on RPi 3B v1.2
bcm2711: boot tested on RPi 4B v1.1 4G

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
2021-02-18 23:42:32 +01:00

275 lines
9.1 KiB
Diff

From b8ae9d55d468a9f55524296247dba93531c29c99 Mon Sep 17 00:00:00 2001
From: John Cox <jc@kynesim.co.uk>
Date: Thu, 5 Mar 2020 14:46:54 +0000
Subject: [PATCH] media: v4l2-mem2mem: allow request job buffer
processing after job finish
Allow the capture buffer to be detached from a v4l2 request job such
that another job can start before the capture buffer is returned. This
allows h/w codecs that can process multiple requests at the same time
to operate more efficiently.
Signed-off-by: John Cox <jc@kynesim.co.uk>
---
drivers/media/v4l2-core/v4l2-mem2mem.c | 105 +++++++++++++++++++++++--
include/media/v4l2-mem2mem.h | 47 +++++++++++
include/media/videobuf2-v4l2.h | 3 +
3 files changed, 149 insertions(+), 6 deletions(-)
--- a/drivers/media/v4l2-core/v4l2-mem2mem.c
+++ b/drivers/media/v4l2-core/v4l2-mem2mem.c
@@ -399,15 +399,18 @@ static void v4l2_m2m_cancel_job(struct v
{
struct v4l2_m2m_dev *m2m_dev;
unsigned long flags;
+ bool det_abort_req;
m2m_dev = m2m_ctx->m2m_dev;
spin_lock_irqsave(&m2m_dev->job_spinlock, flags);
+ det_abort_req = !list_empty(&m2m_ctx->det_list);
m2m_ctx->job_flags |= TRANS_ABORT;
if (m2m_ctx->job_flags & TRANS_RUNNING) {
spin_unlock_irqrestore(&m2m_dev->job_spinlock, flags);
if (m2m_dev->m2m_ops->job_abort)
m2m_dev->m2m_ops->job_abort(m2m_ctx->priv);
+ det_abort_req = false;
dprintk("m2m_ctx %p running, will wait to complete\n", m2m_ctx);
wait_event(m2m_ctx->finished,
!(m2m_ctx->job_flags & TRANS_RUNNING));
@@ -421,6 +424,11 @@ static void v4l2_m2m_cancel_job(struct v
/* Do nothing, was not on queue/running */
spin_unlock_irqrestore(&m2m_dev->job_spinlock, flags);
}
+
+ /* Wait for detached buffers to come back too */
+ if (det_abort_req && m2m_dev->m2m_ops->job_abort)
+ m2m_dev->m2m_ops->job_abort(m2m_ctx->priv);
+ wait_event(m2m_ctx->det_empty, list_empty(&m2m_ctx->det_list));
}
/*
@@ -458,6 +466,7 @@ static bool _v4l2_m2m_job_finish(struct
list_del(&m2m_dev->curr_ctx->queue);
m2m_dev->curr_ctx->job_flags &= ~(TRANS_QUEUED | TRANS_RUNNING);
+ m2m_ctx->cap_detached = false;
wake_up(&m2m_dev->curr_ctx->finished);
m2m_dev->curr_ctx = NULL;
return true;
@@ -485,6 +494,80 @@ void v4l2_m2m_job_finish(struct v4l2_m2m
}
EXPORT_SYMBOL(v4l2_m2m_job_finish);
+struct vb2_v4l2_buffer *_v4l2_m2m_cap_buf_detach(struct v4l2_m2m_ctx *m2m_ctx)
+{
+ struct vb2_v4l2_buffer *buf;
+
+ buf = v4l2_m2m_dst_buf_remove(m2m_ctx);
+ list_add_tail(&container_of(buf, struct v4l2_m2m_buffer, vb)->list,
+ &m2m_ctx->det_list);
+ m2m_ctx->cap_detached = true;
+ buf->is_held = true;
+ buf->det_state = VB2_BUF_STATE_ACTIVE;
+
+ return buf;
+}
+
+struct vb2_v4l2_buffer *v4l2_m2m_cap_buf_detach(struct v4l2_m2m_dev *m2m_dev,
+ struct v4l2_m2m_ctx *m2m_ctx)
+{
+ unsigned long flags;
+ struct vb2_v4l2_buffer *src_buf, *dst_buf;
+
+ spin_lock_irqsave(&m2m_dev->job_spinlock, flags);
+
+ dst_buf = NULL;
+ src_buf = v4l2_m2m_next_src_buf(m2m_ctx);
+
+ if (!(src_buf->flags & V4L2_BUF_FLAG_M2M_HOLD_CAPTURE_BUF) &&
+ !m2m_ctx->cap_detached)
+ dst_buf = _v4l2_m2m_cap_buf_detach(m2m_ctx);
+
+ spin_unlock_irqrestore(&m2m_dev->job_spinlock, flags);
+ return dst_buf;
+}
+EXPORT_SYMBOL(v4l2_m2m_cap_buf_detach);
+
+static void _v4l2_m2m_cap_buf_return(struct v4l2_m2m_ctx *m2m_ctx,
+ struct vb2_v4l2_buffer *buf,
+ enum vb2_buffer_state state)
+{
+ buf->det_state = state;
+
+ /*
+ * Always signal done in the order we got stuff
+ * Stop if we find a buf that is still in use
+ */
+ while (!list_empty(&m2m_ctx->det_list)) {
+ buf = &list_first_entry(&m2m_ctx->det_list,
+ struct v4l2_m2m_buffer, list)->vb;
+ state = buf->det_state;
+ if (state != VB2_BUF_STATE_DONE &&
+ state != VB2_BUF_STATE_ERROR)
+ return;
+ list_del(&container_of(buf, struct v4l2_m2m_buffer, vb)->list);
+ buf->det_state = VB2_BUF_STATE_DEQUEUED;
+ v4l2_m2m_buf_done(buf, state);
+ }
+ wake_up(&m2m_ctx->det_empty);
+}
+
+void v4l2_m2m_cap_buf_return(struct v4l2_m2m_dev *m2m_dev,
+ struct v4l2_m2m_ctx *m2m_ctx,
+ struct vb2_v4l2_buffer *buf,
+ enum vb2_buffer_state state)
+{
+ unsigned long flags;
+
+ if (!buf)
+ return;
+
+ spin_lock_irqsave(&m2m_dev->job_spinlock, flags);
+ _v4l2_m2m_cap_buf_return(m2m_ctx, buf, state);
+ spin_unlock_irqrestore(&m2m_dev->job_spinlock, flags);
+}
+EXPORT_SYMBOL(v4l2_m2m_cap_buf_return);
+
void v4l2_m2m_buf_done_and_job_finish(struct v4l2_m2m_dev *m2m_dev,
struct v4l2_m2m_ctx *m2m_ctx,
enum vb2_buffer_state state)
@@ -495,15 +578,23 @@ void v4l2_m2m_buf_done_and_job_finish(st
spin_lock_irqsave(&m2m_dev->job_spinlock, flags);
src_buf = v4l2_m2m_src_buf_remove(m2m_ctx);
- dst_buf = v4l2_m2m_next_dst_buf(m2m_ctx);
- if (WARN_ON(!src_buf || !dst_buf))
+ if (WARN_ON(!src_buf))
goto unlock;
v4l2_m2m_buf_done(src_buf, state);
- dst_buf->is_held = src_buf->flags & V4L2_BUF_FLAG_M2M_HOLD_CAPTURE_BUF;
- if (!dst_buf->is_held) {
- v4l2_m2m_dst_buf_remove(m2m_ctx);
- v4l2_m2m_buf_done(dst_buf, state);
+
+ if (!m2m_ctx->cap_detached) {
+ dst_buf = v4l2_m2m_next_dst_buf(m2m_ctx);
+ if (WARN_ON(!dst_buf))
+ goto unlock;
+
+ dst_buf->is_held = src_buf->flags
+ & V4L2_BUF_FLAG_M2M_HOLD_CAPTURE_BUF;
+
+ if (!dst_buf->is_held) {
+ dst_buf = _v4l2_m2m_cap_buf_detach(m2m_ctx);
+ _v4l2_m2m_cap_buf_return(m2m_ctx, dst_buf, state);
+ }
}
schedule_next = _v4l2_m2m_job_finish(m2m_dev, m2m_ctx);
unlock:
@@ -983,12 +1074,14 @@ struct v4l2_m2m_ctx *v4l2_m2m_ctx_init(s
m2m_ctx->priv = drv_priv;
m2m_ctx->m2m_dev = m2m_dev;
init_waitqueue_head(&m2m_ctx->finished);
+ init_waitqueue_head(&m2m_ctx->det_empty);
out_q_ctx = &m2m_ctx->out_q_ctx;
cap_q_ctx = &m2m_ctx->cap_q_ctx;
INIT_LIST_HEAD(&out_q_ctx->rdy_queue);
INIT_LIST_HEAD(&cap_q_ctx->rdy_queue);
+ INIT_LIST_HEAD(&m2m_ctx->det_list);
spin_lock_init(&out_q_ctx->rdy_spinlock);
spin_lock_init(&cap_q_ctx->rdy_spinlock);
--- a/include/media/v4l2-mem2mem.h
+++ b/include/media/v4l2-mem2mem.h
@@ -88,6 +88,9 @@ struct v4l2_m2m_queue_ctx {
* %TRANS_QUEUED, %TRANS_RUNNING and %TRANS_ABORT.
* @finished: Wait queue used to signalize when a job queue finished.
* @priv: Instance private data
+ * @cap_detached: Current job's capture buffer has been detached
+ * @det_list: List of detached (post-job but still in flight) capture buffers
+ * @det_empty: Wait queue signalled when det_list goes empty
*
* The memory to memory context is specific to a file handle, NOT to e.g.
* a device.
@@ -111,6 +114,11 @@ struct v4l2_m2m_ctx {
wait_queue_head_t finished;
void *priv;
+
+ /* Detached buffer handling */
+ bool cap_detached;
+ struct list_head det_list;
+ wait_queue_head_t det_empty;
};
/**
@@ -216,6 +224,45 @@ v4l2_m2m_buf_done(struct vb2_v4l2_buffer
}
/**
+ * v4l2_m2m_cap_buf_detach() - detach the capture buffer from the job and
+ * return it.
+ *
+ * @m2m_dev: opaque pointer to the internal data to handle M2M context
+ * @m2m_ctx: m2m context assigned to the instance given by struct &v4l2_m2m_ctx
+ *
+ * This function is designed to be used in conjunction with
+ * v4l2_m2m_buf_done_and_job_finish(). It allows the next job to start
+ * execution before the capture buffer is returned to the user which can be
+ * important if the underlying processing has multiple phases that are more
+ * efficiently executed in parallel.
+ *
+ * If used then it must be called before v4l2_m2m_buf_done_and_job_finish()
+ * as otherwise the buffer will have already gone.
+ *
+ * It is the callers reponsibilty to ensure that all detached buffers are
+ * returned.
+ */
+struct vb2_v4l2_buffer *v4l2_m2m_cap_buf_detach(struct v4l2_m2m_dev *m2m_dev,
+ struct v4l2_m2m_ctx *m2m_ctx);
+
+/**
+ * v4l2_m2m_cap_buf_return() - return a capture buffer, previously detached
+ * with v4l2_m2m_cap_buf_detach() to the user.
+ *
+ * @m2m_dev: opaque pointer to the internal data to handle M2M context
+ * @m2m_ctx: m2m context assigned to the instance given by struct &v4l2_m2m_ctx
+ * @buf: the buffer to return
+ * @state: vb2 buffer state passed to v4l2_m2m_buf_done().
+ *
+ * Buffers returned by this function will be returned to the user in the order
+ * of the original jobs rather than the order in which this function is called.
+ */
+void v4l2_m2m_cap_buf_return(struct v4l2_m2m_dev *m2m_dev,
+ struct v4l2_m2m_ctx *m2m_ctx,
+ struct vb2_v4l2_buffer *buf,
+ enum vb2_buffer_state state);
+
+/**
* v4l2_m2m_reqbufs() - multi-queue-aware REQBUFS multiplexer
*
* @file: pointer to struct &file
--- a/include/media/videobuf2-v4l2.h
+++ b/include/media/videobuf2-v4l2.h
@@ -35,6 +35,8 @@
* @request_fd: the request_fd associated with this buffer
* @is_held: if true, then this capture buffer was held
* @planes: plane information (userptr/fd, length, bytesused, data_offset).
+ * @det_state: if a detached request capture buffer then this contains its
+ * current state
*
* Should contain enough information to be able to cover all the fields
* of &struct v4l2_buffer at ``videodev2.h``.
@@ -49,6 +51,7 @@ struct vb2_v4l2_buffer {
__s32 request_fd;
bool is_held;
struct vb2_plane planes[VB2_MAX_PLANES];
+ enum vb2_buffer_state det_state;
};
/* VB2 V4L2 flags as set in vb2_queue.subsystem_flags */