Re: [PATCH V7 6/7] remoteproc: imx_rproc: request mbox channel later

On Mon, Oct 17, 2022 at 03:13:16AM +0000, Peng Fan wrote:

Hi Mathieu,

quoted

Subject: Re: [PATCH V7 6/7] remoteproc: imx_rproc: request mbox channel
later

On Fri, Oct 14, 2022 at 11:10:36AM +0800, Peng Fan (OSS) wrote:

quoted

From: Peng Fan <peng.fan@nxp.com>

It is possible that when remote processor crash, the communication
channel will be broken with garbage value in mailbox, such as when
Linux is issuing a message through mailbox, remote processor crashes,
we need free & rebuild the mailbox channels to make sure no garbage
value in mailbox channels.

So move the request/free to start/stop for managing remote procesosr
in Linux, move to attach/detach for remote processor is out of control
of Linux.

Previous, we just request mbox when attach for CM4 boot early before
Linux, but if mbox defer probe, remoteproc core will do resource
cleanup and corrupt resource table for later probe.

So move request mbox ealier and still keep mbox request when attach
for self recovery case, but keep a check when request/free mbox.

Signed-off-by: Peng Fan <peng.fan@nxp.com>
---
 drivers/remoteproc/imx_rproc.c | 39
++++++++++++++++++++++++++++++++--
 1 file changed, 37 insertions(+), 2 deletions(-)

diff --git a/drivers/remoteproc/imx_rproc.c

b/drivers/remoteproc/imx_rproc.c index 917e6db39572..1183de84a4c0
100644

--- a/drivers/remoteproc/imx_rproc.c
+++ b/drivers/remoteproc/imx_rproc.c

@@ -84,6 +84,8 @@ struct imx_rproc_mem {
 #define ATT_CORE_MASK   0xffff
 #define ATT_CORE(I)     BIT((I))

+static int imx_rproc_xtr_mbox_init(struct rproc *rproc); static void
+imx_rproc_free_mbox(struct rproc *rproc);
 static int imx_rproc_detach_pd(struct rproc *rproc);

 struct imx_rproc {

@@ -357,6 +359,10 @@ static int imx_rproc_start(struct rproc *rproc)
 	struct arm_smccc_res res;
 	int ret;

+	ret = imx_rproc_xtr_mbox_init(rproc);
+	if (ret)
+		return ret;
+
 	switch (dcfg->method) {
 	case IMX_RPROC_MMIO:
 		ret = regmap_update_bits(priv->regmap, dcfg->src_reg,

dcfg->src_mask, @@ -407,6 +413,8 @@ static int imx_rproc_stop(struct
rproc *rproc)

 	if (ret)
 		dev_err(dev, "Failed to stop remote core\n");
+	else
+		imx_rproc_free_mbox(rproc);

 	return ret;
 }

@@ -592,6 +600,22 @@ static void imx_rproc_kick(struct rproc *rproc,

int vqid)

 static int imx_rproc_attach(struct rproc *rproc)  {
+	return imx_rproc_xtr_mbox_init(rproc); }
+
+static int imx_rproc_detach(struct rproc *rproc) {
+	struct imx_rproc *priv = rproc->priv;
+	const struct imx_rproc_dcfg *dcfg = priv->dcfg;
+
+	if (dcfg->method != IMX_RPROC_SCU_API)
+		return -EOPNOTSUPP;
+
+	if (imx_sc_rm_is_resource_owned(priv->ipc_handle, priv->rsrc_id))
+		return -EOPNOTSUPP;
+
+	imx_rproc_free_mbox(rproc);
+
 	return 0;
 }

@@ -610,6 +634,7 @@ static struct resource_table

*imx_rproc_get_loaded_rsc_table(struct rproc *rproc  static const struct

rproc_ops imx_rproc_ops = {

quoted

 	.prepare	= imx_rproc_prepare,
 	.attach		= imx_rproc_attach,
+	.detach		= imx_rproc_detach,
 	.start		= imx_rproc_start,
 	.stop		= imx_rproc_stop,
 	.kick		= imx_rproc_kick,

@@ -720,6 +745,9 @@ static int imx_rproc_xtr_mbox_init(struct rproc

*rproc)

quoted

 	struct device *dev = priv->dev;
 	struct mbox_client *cl;

+	if (priv->tx_ch && priv->rx_ch)
+		return 0;
+

You did exactly the same things as in V6.  I asked you why this is needed and
all you did is point me to the code in _probe(), which I can read on my own.

Sorry for not wrote down clear.

quoted

Again - why is this needed when we know it will be done in start() and
attach()?

start() and attach() not able to handle mbox defer probe. So I add

We are finally at the heart of the problem.  I had to go look at the
implementation of imx_rproc_xtr_mbox_init() to understand that it can return
-EPROBE_DEFER.  Had there been a comment in the code to highlight _why_ the if()
condition is needed, I would have understood right away and all this waste of
time avoided.

the mbox requesting in probe to handle mbox defer probe, and add
a check when requesting mbox channel in start/attach. During first
time attach/start remote core, the imx_rproc_xtr_mbox_init just
return, because channel requested in probe flow. 

Since mbox requested in probe, why still add it in start() and attach()?
It is to support runtime stop and start(M4 is under control of Linux),
to support runtime detach(only for i.MX8QM/QXP attach recovery,
m4 out of control from linux) and attach.

Thanks,
Peng.

quoted

quoted

 	if (!of_get_property(dev->of_node, "mbox-names", NULL))
 		return 0;

@@ -749,8 +777,15 @@ static void imx_rproc_free_mbox(struct rproc

*rproc)  {
 	struct imx_rproc *priv = rproc->priv;

-	mbox_free_channel(priv->tx_ch);
-	mbox_free_channel(priv->rx_ch);
+	if (priv->tx_ch) {
+		mbox_free_channel(priv->tx_ch);
+		priv->tx_ch = NULL;
+	}
+
+	if (priv->rx_ch) {
+		mbox_free_channel(priv->rx_ch);
+		priv->rx_ch = NULL;
+	}
 }

 static void imx_rproc_put_scu(struct rproc *rproc)
--
2.37.1