New subject: [Sound-open-firmware] [PATCH] Add support to replace stale stream/trace position updates.

23 Nov 2017

Hi, Keyon,
	Previou replying may be wrong about multiply trace messages.
	Please see the following:

On Thu, 2017-11-23 at 18:10 +0800, yanwang wrote:
> On Thu, 2017-11-23 at 17:06 +0800, Keyon Jie wrote:
> > 
> > 
> > On 2017年11月23日 13:27, yanwang wrote:
> > > 
> > > 
> > > On Thu, 2017-11-23 at 03:07 +0000, Jie, Yang wrote:
> > > > 
> > > > 
> > > > > 
> > > > > 
> > > > > 
> > > > > -----Original Message-----
> > > > > From: sound-open-firmware-bounces@alsa-project.org
> > > > > [mailto:sound-
> > > > > open-
> > > > > firmware-bounces@alsa-project.org] On Behalf Of yan.wang@linu
> > > > > x.
> > > > > inte
> > > > > l.com
> > > > > Sent: Thursday, November 23, 2017 10:48 AM
> > > > > To: sound-open-firmware@alsa-project.org
> > > > > Cc: Yan Wang <yan.wang@linux.intel.com>
> > > > > Subject: [Sound-open-firmware] [PATCH] Add support to replace
> > > > > stale
> > > > > stream/trace position updates.
> > > > > 
> > > > > From: Yan Wang <yan.wang@linux.intel.com>
> > > > > 
> > > > > 1. Add "replace" parameter for ipc_queue_host_message() to
> > > > > indicate
> > > > > whether
> > > > > check duplicate message in host message queue.
> > > > > 2. Add msg_find() to search duplicate message.
> > > > > 3. If replace flag is true, search and replace duplicate
> > > > > message.
> > > > > 4. For the message of host offset of DMA trace, enable
> > > > > replace
> > > > > logic.
> > > > > 
> > > > > It will avoid that the host message queue is full because
> > > > > there
> > > > > is
> > > > > too many IPC
> > > > > messages of host offset of DMA trace.
> > > > 
> > > > Does the "duplicate" message are true duplicated? IMO, we
> > > > should
> > > > not
> > > > send
> > > > duplicate message except that it is critical/urgent and no
> > > > respond
> > > > for it from
> > > > host side.
> > > > 
> > > > If it is meaning that the DMA trace sending too much
> > > > messages(with
> > > > too high
> > > > frequency), we should reduce the frequency, instead of adding
> > > > message
> > > > replacing logic which makes it more complicate and error-prone.
> > > > 
> > > > Thanks,
> > > > ~Keyon
> > > 
> > > Host offset of DMA trace need be updated after DMA transfer
> > > completes.
> > > This patch wants solve:
> > > If previous update message of host offset hasn't been sent
> > > because
> > > of
> > > many the other messages, set the latest host offset for it.
> > > "duplicate" may cause some misunderstanding, I could add more
> > > description into commit message.
> > > Thanks.
> > 
> > OK, that make sense.
> > 
> > So if there are several host offset update messages in the queue,
> > we 
> > will only replace one(the most nearly to the head one) of it, is
> > this 
> > the intention?
> 
> Yes. The first found message should be sent at first. It will notify
> degbufs the latest offset. It will output available trace data as
> possible. In fact, it may be less possible that there are multiply
> trace messages because of current time intervals (500ms) and buffer
> size. I have not meeted this situation in my BYT MB platform. But it
> should be possible in some extreme situation e.g. IPC timeout. I
> could
> change it and make host offset of all trace mesage are updated to the
> latest value. 

Because trace message is always searched at first before try to add
into queue, it is impossible that there are more one trace message in
queue now. So it is unnecessary to search the queue list until its end.
Is it right?

Thanks.
Yan Wang

> 
> > 
> > 
> > Another question is, for some kind of IPC message, e.g. struct 
> > sof_ipc_stream_posn, we should merge only the messages with same
> > header 
> > plus same other member(e.g. posn->comp_id), which means they are
> > same 
> > kind of messages from same compononents ---- e.g. we should not
> > merge 
> > host position update to dai position update messages.
> > 
> > For this case, the msg_find() function may need to be modified.
> 
> Yes. Current impelemntation doesn't know the meaning of the message
> and
> just comparing with header. So if need more complex logic, need
> modify
> it.
> 
> > 
> > 
> > 
> > Thanks,
> > ~Keyon
> > 
> > > 
> > > 
> > > 
> > > Yan Wang
> > > 
> > > > 
> > > > 
> > > > 
> > > > > 
> > > > > 
> > > > > 
> > > > > 
> > > > > Signed-off-by: Yan Wang <yan.wang@linux.intel.com>
> > > > > ---
> > > > > src/include/reef/ipc.h |  2 +-
> > > > > src/ipc/intel-ipc.c    | 45
> > > > > ++++++++++++++++++++++++++++++++++-
> > > > > ----
> > > > > ------
> > > > > 2 files changed, 35 insertions(+), 12 deletions(-)
> > > > > 
> > > > > diff --git a/src/include/reef/ipc.h b/src/include/reef/ipc.h
> > > > > index
> > > > > fcab45b..bb814be 100644
> > > > > --- a/src/include/reef/ipc.h
> > > > > +++ b/src/include/reef/ipc.h
> > > > > @@ -125,7 +125,7 @@ int ipc_stream_send_xrun(struct comp_dev
> > > > > *cdev,
> > > > > 
> > > > > int ipc_queue_host_message(struct ipc *ipc, uint32_t header,
> > > > > 	void *tx_data, size_t tx_bytes, void *rx_data,
> > > > > -	size_t rx_bytes, void (*cb)(void*, void*), void
> > > > > *cb_data);
> > > > > +	size_t rx_bytes, void (*cb)(void*, void*), void
> > > > > *cb_data,
> > > > > uint32_t
> > > > > +replace);
> > > > > int ipc_send_short_msg(uint32_t msg);
> > > > > 
> > > > > void ipc_platform_do_cmd(struct ipc *ipc); diff --git
> > > > > a/src/ipc/intel-ipc.c
> > > > > b/src/ipc/intel-ipc.c index e13b541..391907e 100644
> > > > > --- a/src/ipc/intel-ipc.c
> > > > > +++ b/src/ipc/intel-ipc.c
> > > > > @@ -367,7 +367,7 @@ int ipc_stream_send_position(struct
> > > > > comp_dev
> > > > > *cdev,
> > > > > 	posn->comp_id = cdev->comp.id;
> > > > > 
> > > > > 	return ipc_queue_host_message(_ipc, posn-
> > > > > >rhdr.hdr.cmd, 
> > > > > posn,
> > > > > -		sizeof(*posn), NULL, 0, NULL, NULL);
> > > > > +		sizeof(*posn), NULL, 0, NULL, NULL, 0);
> > > > > }
> > > > > 
> > > > > /* send stream position TODO: send compound message  */ @@
> > > > > -379,7
> > > > > +379,7
> > > > > @@ int ipc_stream_send_xrun(struct comp_dev *cdev,
> > > > > 	posn->comp_id = cdev->comp.id;
> > > > > 
> > > > > 	return ipc_queue_host_message(_ipc, posn-
> > > > > >rhdr.hdr.cmd, 
> > > > > posn,
> > > > > -		sizeof(*posn), NULL, 0, NULL, NULL);
> > > > > +		sizeof(*posn), NULL, 0, NULL, NULL, 0);
> > > > > }
> > > > > 
> > > > > static int ipc_stream_trigger(uint32_t header) @@ -644,7
> > > > > +644,7
> > > > > @@
> > > > > int
> > > > > ipc_dma_trace_send_position(void)
> > > > > 	posn.rhdr.hdr.size = sizeof(posn);
> > > > > 
> > > > > 	return ipc_queue_host_message(_ipc, posn.rhdr.hdr.cmd,
> > > > > &posn,
> > > > > -		sizeof(posn), NULL, 0, NULL, NULL);
> > > > > +		sizeof(posn), NULL, 0, NULL, NULL, 1);
> > > > > }
> > > > > 
> > > > > static int ipc_glb_debug_message(uint32_t header) @@ -899,19
> > > > > +899,40 @@
> > > > > static inline struct ipc_msg *msg_get_empty(struct ipc *ipc)
> > > > > 	return msg;
> > > > > }
> > > > > 
> > > > > +static inline struct ipc_msg *msg_find(struct ipc *ipc,
> > > > > uint32_t
> > > > > +header) {
> > > > > +	struct list_item *plist;
> > > > > +	struct ipc_msg *msg = NULL;
> > > > > +
> > > > > +	list_for_item(plist, &ipc->msg_list) {
> > > > > +		msg = container_of(plist, struct ipc_msg,
> > > > > list);
> > > > > +		if (msg->header == header)
> > > > > +			return msg;
> > > > > +	}
> > > > > +
> > > > > +	return NULL;
> > > > > +}
> > > > > 
> > > > > int ipc_queue_host_message(struct ipc *ipc, uint32_t header,
> > > > > 	void *tx_data, size_t tx_bytes, void *rx_data,
> > > > > -	size_t rx_bytes, void (*cb)(void*, void*), void
> > > > > *cb_data)
> > > > > +	size_t rx_bytes, void (*cb)(void*, void*), void
> > > > > *cb_data,
> > > > > uint32_t
> > > > > +replace)
> > > > > {
> > > > > -	struct ipc_msg *msg;
> > > > > -	uint32_t flags;
> > > > > +	struct ipc_msg *msg = NULL;
> > > > > +	uint32_t flags, found = 0;
> > > > > 	int ret = 0;
> > > > > 
> > > > > 	spin_lock_irq(&ipc->lock, flags);
> > > > > 
> > > > > -	/* get a free message */
> > > > > -	msg = msg_get_empty(ipc);
> > > > > +	/* do we need to replace an existing message? */
> > > > > +	if (replace)
> > > > > +		msg = msg_find(ipc, header);
> > > > > +
> > > > > +	/* do we need to use a new empty message? */
> > > > > +	if (msg)
> > > > > +		found = 1;
> > > > > +	else
> > > > > +		msg = msg_get_empty(ipc);
> > > > > +
> > > > > 	if (msg == NULL) {
> > > > > 		trace_ipc_error("eQb");
> > > > > 		ret = -EBUSY;
> > > > > @@ -929,9 +950,11 @@ int ipc_queue_host_message(struct ipc
> > > > > *ipc,
> > > > > uint32_t
> > > > > header,
> > > > > 	if (tx_bytes > 0 && tx_bytes < SOF_IPC_MSG_MAX_SIZE)
> > > > > 		rmemcpy(msg->tx_data, tx_data, tx_bytes);
> > > > > 
> > > > > -	/* now queue the message */
> > > > > -	ipc->dsp_pending = 1;
> > > > > -	list_item_append(&msg->list, &ipc->msg_list);
> > > > > +	if (!found) {
> > > > > +		/* now queue the message */
> > > > > +		ipc->dsp_pending = 1;
> > > > > +		list_item_append(&msg->list, &ipc-
> > > > > >msg_list);
> > > > > +	}
> > > > > 
> > > > > out:
> > > > > 	spin_unlock_irq(&ipc->lock, flags);
> > > > > --
> > > > > 2.7.4
> > > > > 
> > > > > _______________________________________________
> > > > > Sound-open-firmware mailing list
> > > > > Sound-open-firmware@alsa-project.org
> > > > > http://mailman.alsa-project.org/mailman/listinfo/sound-open-f
> > > > > ir
> > > > > mwar
> > > > > e
> > > > _______________________________________________
> > > > Sound-open-firmware mailing list
> > > > Sound-open-firmware@alsa-project.org
> > > > http://mailman.alsa-project.org/mailman/listinfo/sound-open-fir
> > > > mw
> > > > are
> > > _______________________________________________
> > > Sound-open-firmware mailing list
> > > Sound-open-firmware@alsa-project.org
> > > http://mailman.alsa-project.org/mailman/listinfo/sound-open-firmw
> > > ar
> > > e
> > > 
> _______________________________________________
> Sound-open-firmware mailing list
> Sound-open-firmware@alsa-project.org
> http://mailman.alsa-project.org/mailman/listinfo/sound-open-firmware

[Sound-open-firmware] [PATCH] Add support to replace stale stream/trace position updates.

yan.wang＠linux.intel.com

Jie, Yang

yanwang

Keyon Jie

yanwang

yanwang

tags

participants (4)