On 8/4/25 2:31 PM, Hiago De Franco wrote:
Hi Andrew,
On Tue, Jul 29, 2025 at 03:04:20PM -0300, Hiago De Franco wrote:
On Sat, Jul 26, 2025 at 12:48:14PM -0500, Andrew Davis wrote:
On 7/26/25 9:39 AM, Hiago De Franco wrote:
Hi Andrew, Beleswar,
On Fri, Jul 25, 2025 at 02:29:22PM -0500, Andrew Davis wrote:
So the issue then looks to be this message we send here when we setup
the mailbox[0]. This mailbox setup is done during probe() for the K3
rproc drivers now (mailbox setup used to be done during
rproc_{start,attach}() before [1]). Moving mailbox setup to probe
is correct, but we should have factored out the test message sending
code out of mailbox setup so it could have been left in
rproc_{start,attach}(). That way we only send this message if the
core is going to be started, no sense in sending that message if
we are not even going to run the core..
Fix might be as simple as [2] (not tested, if this works feel free
to send as a fix)
I tested the patch and it works, thanks!
Andrew
[0]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/remoteproc/ti_k3_common.c#n176
[1]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f3f11cfe890733373ddbb1ce8991ccd4ee5e79e1
[2]
diff --git a/drivers/remoteproc/ti_k3_common.c
b/drivers/remoteproc/ti_k3_common.c
index a70d4879a8bea..657a200fa9040 100644
--- a/drivers/remoteproc/ti_k3_common.c
+++ b/drivers/remoteproc/ti_k3_common.c
@@ -198,6 +198,22 @@ int k3_rproc_reset(struct k3_rproc *kproc)
}
EXPORT_SYMBOL_GPL(k3_rproc_reset);
+static int k3_rproc_ping(struct k3_rproc *kproc)
+{
+ /*
+ * Ping the remote processor, this is only for sanity-sake for now;
+ * there is no functional effect whatsoever.
+ *
+ * Note that the reply will _not_ arrive immediately: this message
+ * will wait in the mailbox fifo until the remote processor is booted.
+ */
+ int ret = mbox_send_message(kproc->mbox, (void *)RP_MBOX_ECHO_REQUEST);
+ if (ret < 0)
+ dev_err(kproc->dev, "mbox_send_message failed (%pe)\n",
ERR_PTR(ret));
+
+ return ret;
+}
+
/* Release the remote processor from reset */
int k3_rproc_release(struct k3_rproc *kproc)
{
@@ -221,6 +237,8 @@ int k3_rproc_release(struct k3_rproc *kproc)
if (ret)
dev_err(dev, "module-reset deassert failed (%pe)\n",
ERR_PTR(ret));
+ k3_rproc_ping(kproc);
+
return ret;
}
EXPORT_SYMBOL_GPL(k3_rproc_release);
@@ -243,20 +261,6 @@ int k3_rproc_request_mbox(struct rproc *rproc)
return dev_err_probe(dev, PTR_ERR(kproc->mbox),
"mbox_request_channel failed\n");
- /*
- * Ping the remote processor, this is only for sanity-sake for now;
- * there is no functional effect whatsoever.
- *
- * Note that the reply will _not_ arrive immediately: this message
- * will wait in the mailbox fifo until the remote processor is booted.
- */
- ret = mbox_send_message(kproc->mbox, (void *)RP_MBOX_ECHO_REQUEST);
- if (ret < 0) {
- dev_err(dev, "mbox_send_message failed (%pe)\n", ERR_PTR(ret));
- mbox_free_channel(kproc->mbox);
- return ret;
- }
-
return 0;
}
EXPORT_SYMBOL_GPL(k3_rproc_request_mbox);
@@ -397,7 +401,12 @@ EXPORT_SYMBOL_GPL(k3_rproc_stop);
* remote core. This callback is invoked only in IPC-only mode and exists
* because rproc_validate() checks for its existence.
*/
-int k3_rproc_attach(struct rproc *rproc) { return 0; }
+int k3_rproc_attach(struct rproc *rproc)
+{
+ k3_rproc_ping(rproc->priv);
+
+ return 0;
+}
EXPORT_SYMBOL_GPL(k3_rproc_attach);
/*
On Sat, Jul 26, 2025 at 07:47:34PM +0530, Beleswar Prasad Padhi wrote:
So the issue then looks to be this message we send here when we setup
the mailbox[0]. This mailbox setup is done during probe() for the K3
rproc drivers now (mailbox setup used to be done during
rproc_{start,attach}() before [1]). Moving mailbox setup to probe
is correct, but we should have factored out the test message sending
code out of mailbox setup so it could have been left in
rproc_{start,attach}().
Or, how about we don't send that test mbox message at all. It does not
actually check if the remoteproc was able to receive and respond to the
message. It only verifies if the write to the mbox queue was successful. And
most firmwares anyways don't reply to that mailbox-level echo message.
I was thinking about the same.
I tested the patch and it indeed works, however when I boot the remote
core with a hello world firwmare from the TI MCU SDK (with IPC enabled
with the sysconfig), the ping is sent but M4 never replies to it, which
at the end causes an unread message to stay there. Later, if I stop the
remote processor, I can not got into suspend mode again because of this
message.
So I believe we should never send the message or clear the mailbox when
the remote processor is stopped, but I was not able to find a way to
clear the mailbox. So, is it ok if we never send the ping?
So right now it is okay to not send that ping, and in the past I've
thought about removing it (it is a bit of a legacy hold-over from
the OMAP RProc driver. Back then we would send other messages like
suspend and shutdown requests, you can see the different messages
here[0]. Actually using those messages never got upstream, only the
ping message part did.
For K3 we want to start making use of all these other messages and
upstream the support for the same. So removing the ping test message
felt like a step backwards as it is a good placeholder for the more
important messages we want to send later. But as said, removing it
is probably fine for now.
The second thing on the roadmap is to better deal with messages left
in the mailbox queue when we try to suspend. Basically on suspend the
mailbox IP is powered down and all messages waiting will be lost, this
can cause issues. Instead of blocking suspend, one other option would
be to attempt to read out these messages and restore them on resume.
This state saving would match what most other IP drivers. This would
fix issues like the above were the firmware doesn't consume a message
for whatever reason. But until we implement that, either we throw out
messages on suspend, or we block suspend.
Got it, thanks for you explanation Andrew. I am out this week, so next
week I will propose a patch so we can turn around this issue.
Thanks!
Hiago.
Sorry for the delay, I am back this week, I was testing the patches and
removing the ping was not enough, there is one extra message being sent,
which is the k3_rproc_kick() from ti_k3_common.c. This one is a callback
from remoteproc_virtio.c.
So tracing back, looks like this message will be added to the mailbox
when Linux tries to start communication with the remote core, and that
happens if the firmware advertises vrings in its resource table.
I belive this one is necessary to make the firmware works, but with the
hello world demo, I still have the issue where I can not go into suspend
Which "hello world demo" is this? In Zephyr, we do not add the VDEV to
the resource table if the firmware does not intend to communicate[0].
But MCU+SDK firmware might add these unconditionally, I'm not sure. You
could check what is in the table with:
$ readelf -x .resource_table <your_firmware.elf>
and empty one might look like:
Hex dump of section '.resource_table':
0xa3100000 01000000 00000000 00000000 00000000 ................
one with VDEV will be much longer.
Andrew
[0]
https://github.com/zephyrproject-rtos/zephyr/blob/main/lib/open-amp/resource_table.h#L34
mode. Removing both mbox_send_message() calls makes the suspend work
again:
root@verdin-am62-15479173:~# dmesg | grep -i -E
"remoteproc|rproc|omap-mailbox|hfranco"
[ 0.000000] Kernel command line: root=PARTUUID=096221e5-02 ro rootwait
console=tty1 console=ttyS2,115200 dyndb
g="file ti_k3_common.c +p; file remotecore_proc.c +p; file remoteproc_virtio.c
+p"
[ 10.520920] omap-mailbox 29000000.mailbox: omap mailbox rev 0x66fc9100
[ 10.711357] k3-m4-rproc 5000000.m4fss: assigned reserved memory node
m4f-dma-memory@9cb00000
[ 10.753040] k3-m4-rproc 5000000.m4fss: configured M4F for remoteproc mode
[ 10.793640] remoteproc remoteproc0: 5000000.m4fss is available
[ 10.856735] remoteproc remoteproc0: powering up 5000000.m4fss
[ 10.895961] remoteproc remoteproc0: Booting fw image am62-mcu-m4f0_0-fw,
size 451080
[ 11.000752] rproc-virtio rproc-virtio.4.auto: assigned reserved memory node
m4f-dma-memory@9cb00000
[ 11.101614] rproc-virtio rproc-virtio.4.auto: registered virtio0 (type 7)
[ 11.151665] remoteproc remoteproc0: remote processor 5000000.m4fss is now up
[ 12.123724] remoteproc remoteproc1: 30074000.pru is available
[ 12.171118] remoteproc remoteproc2: 30078000.pru is available
[ 12.337287] remoteproc remoteproc0: vring0: va 00000000cabe42be qsz 256
notifyid 0
[ 12.337337] remoteproc remoteproc0: vring1: va 00000000a651968a qsz 256
notifyid 1
[ 12.348543] remoteproc remoteproc0: kicking vq index: 0
[ 12.348559] hfranco: sending msg 0x0, name mbox-m4-0
[ 2514.508396] remoteproc remoteproc0: stopped remote processor 5000000.m4fss
[ 2518.010399] omap-mailbox 29000000.mailbox: fifo 1 has unexpected unread
messages
[ 2518.010433] omap-mailbox 29000000.mailbox: PM: dpm_run_callback():
platform_pm_suspend returns -16
[ 2518.010461] omap-mailbox 29000000.mailbox: PM: failed to suspend: error -16
In this case, I was wondering if we should drop the messages for now,
until we have the routine to save the messages first. Any suggestion you
might have?
Thanks for the help,
Hiago.
Andrew
[0] drivers/remoteproc/omap_remoteproc.h
Best regards,
Hiago.
Thanks,
Beleswar
That way we only send this message if the
core is going to be started, no sense in sending that message if
we are not even going to run the core..
Fix might be as simple as [2] (not tested, if this works feel free
to send as a fix)
Andrew
[0]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/remoteproc/ti_k3_common.c#n176
[1]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f3f11cfe890733373ddbb1ce8991ccd4ee5e79e1
[2]
diff --git a/drivers/remoteproc/ti_k3_common.c
b/drivers/remoteproc/ti_k3_common.c
index a70d4879a8bea..657a200fa9040 100644
--- a/drivers/remoteproc/ti_k3_common.c
+++ b/drivers/remoteproc/ti_k3_common.c
@@ -198,6 +198,22 @@ int k3_rproc_reset(struct k3_rproc *kproc)
}
EXPORT_SYMBOL_GPL(k3_rproc_reset);
+static int k3_rproc_ping(struct k3_rproc *kproc)
+{
+ /*
+ * Ping the remote processor, this is only for sanity-sake for
now;
+ * there is no functional effect whatsoever.
+ *
+ * Note that the reply will _not_ arrive immediately: this
message
+ * will wait in the mailbox fifo until the remote processor is
booted.
+ */
+ int ret = mbox_send_message(kproc->mbox, (void
*)RP_MBOX_ECHO_REQUEST);
+ if (ret < 0)
+ dev_err(kproc->dev, "mbox_send_message failed (%pe)\n",
ERR_PTR(ret));
+
+ return ret;
+}
+
/* Release the remote processor from reset */
int k3_rproc_release(struct k3_rproc *kproc)
{
@@ -221,6 +237,8 @@ int k3_rproc_release(struct k3_rproc *kproc)
if (ret)
dev_err(dev, "module-reset deassert failed (%pe)\n",
ERR_PTR(ret));
+ k3_rproc_ping(kproc);
+
return ret;
}
EXPORT_SYMBOL_GPL(k3_rproc_release);
@@ -243,20 +261,6 @@ int k3_rproc_request_mbox(struct rproc *rproc)
return dev_err_probe(dev, PTR_ERR(kproc->mbox),
"mbox_request_channel failed\n");
- /*
- * Ping the remote processor, this is only for sanity-sake for
now;
- * there is no functional effect whatsoever.
- *
- * Note that the reply will _not_ arrive immediately: this
message
- * will wait in the mailbox fifo until the remote processor is
booted.
- */
- ret = mbox_send_message(kproc->mbox, (void
*)RP_MBOX_ECHO_REQUEST);
- if (ret < 0) {
- dev_err(dev, "mbox_send_message failed (%pe)\n",
ERR_PTR(ret));
- mbox_free_channel(kproc->mbox);
- return ret;
- }
-
return 0;
}
EXPORT_SYMBOL_GPL(k3_rproc_request_mbox);
@@ -397,7 +401,12 @@ EXPORT_SYMBOL_GPL(k3_rproc_stop);
* remote core. This callback is invoked only in IPC-only mode and
exists
* because rproc_validate() checks for its existence.
*/
-int k3_rproc_attach(struct rproc *rproc) { return 0; }
+int k3_rproc_attach(struct rproc *rproc)
+{
+ k3_rproc_ping(rproc->priv);
+
+ return 0;
+}
EXPORT_SYMBOL_GPL(k3_rproc_attach);
/*
