debug.h is a NuttX-specific, non-POSIX header. Placing it in the
top-level include/ directory creates naming conflicts with external
projects that define their own debug.h.
This commit moves the canonical header to include/nuttx/debug.h,
following the NuttX convention for non-POSIX/non-standard headers,
and updates all in-tree references.
A backward-compatibility shim is left at include/debug.h that
emits a deprecation #warning and re-includes <nuttx/debug.h>,
allowing out-of-tree code to continue building while migrating.
Signed-off-by: Piyush Patle <piyushpatle228@gmail.com>
support specify the address by h2r_buf_addr and r2h_buf_addr
so is can split shared memory pool
Signed-off-by: yintao <yintao@xiaomi.com>
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
When the VIRTIO_RPMSG_F_CPUNAME feature is not negotiated, the driver
previously would crash due to DEBUGASSERT. This commit adds support
for a fallback mechanism:
1. Add rpmsg_virtio_probe_cpuname() API that accepts an optional cpuname
parameter to be used when VIRTIO_RPMSG_F_CPUNAME is not available.
2. Modify rptun driver to pass the cpuname from rptun device config
to rpmsg_virtio_probe_cpuname().
3. Refactor rptun device register/unregister to prioritize RPMSG device
handling before other virtio device types.
This ensures the system works correctly in scenarios where the remote
side does not support the VIRTIO_RPMSG_F_CPUNAME feature.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Add TX/RX timestamp tracking for rpmsg_port to enable latency
measurement and debugging:
- Add rpmsg_port_get_timestamp() to retrieve buffer timestamps
- Add rpmsg_port_update_timestamp() to record timestamps at TX/RX time
- Integrate timestamp recording in rpmsg_port_spi and rpmsg_port_spi_slave
- Reserve space for rpmsg_timestamp_s in buffer size calculations
Signed-off-by: liaoao <liaoao@xiaomi.com>
Add rpmsg_get_timestamp() API to get the TX/RX timestamps of an rpmsg
buffer. This is useful for latency measurement and debugging purposes.
The new rpmsg_timestamp_s structure contains:
- tx_nsec: timestamp when the buffer was transmitted
- rx_nsec: timestamp when the buffer was received
Signed-off-by: liaoao <liaoao@xiaomi.com>
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
1. Use virtio_read_config_bytes() to read the string types in
rpmsg-virtio and virtio-net;
2. Use virtio_read_config_member() to read the elements in
virtio-snd configuration space;
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Now we can release the rx buffer in ept->cb, ept->cb should copy
the rx buffer it's private buffer and then return
RPMSG_SUCCESS_BUFFER_RETURNED.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
When recv peer poweroff command, should clear the connected status
first to avoid there is data to send when call the rpmsg_unregister(),
then tx thread hang at the data send process but peer has power
off.
When clear the connected status first, rpmsg_port_uart_tx_ready()
will drop the tx packet to avoid tx thread is waked up to send useless
data.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
rpmsg_port_unreigster() should be called after rpmsg_port_reigster(),
so add the connected condition when receive poweroff command.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
To debug the tx buffer get failed issue, the rpmsg port uart/spi
can use this mechanism to dump more debug informations.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Fix the case that tx thread want to send data to peer and try to
wakeup peer, but rx thread blocked on the tx buffer got procees and
so rx thread can't process the commands includes the wakeup ack
command, then the tx thread will hang at the wakeup process forever.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
When no free buffers, notify the lower transport, so the lower
transports can know this thing and do some work, such as continue
to process the rx data to avoid the rx thread is blocked that
even can't process the commands.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Signed-off-by: liaoao <liaoao@xiaomi.com>
Orignal issue: the new wakeup process may be waked by the last Ack
command from peer, and lead the data is lost because peer may have
enter into the sleep mode.
W = Wake, A = WakeAck, R = Relax, D = Data
TX: [0]W-[1]R--------[4]W----[6]DDDDDDDDDDDD-------
RX: ------------[3]A----[5]Enter Sleep-------------
In this figure, Ack command [3] wakeup the wakeup process [4] because
[3] may be received with a delay.
Now use the ping-pong wake/ack command to fix this issue because
the different wake/ack command can block the last Ack command
wake the wrong wakeup process.
W1 = Wake1, A1 = WakeAck1, W2 = Wake2, A2 = WakeAck2, R = Relax, D = Data
TX: [0]W1-[1]R-------[4]W2----------[6]DDDDDDDDDDD--
RX: ------------[3]A1---------[5]A2-----------------
With this patch, the Ack comamnd [3] can't wakeup the wakeup process [4]
because [3] is Ack1 but [4] is Wake2, and local will not send data until
peer response a [5]Ack2.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Try to dump more error log to catch the uart trasmission issue.
1. dump the received error char when received the non start char
at the wait start state;
2. dump the dbg buffer and received buffer when received the
duplicate end char at the wait start state;
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Peer will receive the power off command and unregister the rpmsg
connections to avoid the rpmsg services stuck at the rpmsg operations.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Support wakeup by peer core (Linux)
1. Add 3 commands: STAYWAKE/RELAXWAKE/STAYWAKEACK
Local will stay the wakeup source when received the STAYWAKE command
and relax the pm wake soucre when received the RELAXWAKE.
And local should response the STAYWAKEACK to peer to notify local has
been waked up.
Local core must send the RELAXWAKE command to peer when there is nothing
to do (send/receive data) otherwise peer can't enter into sleep mode.
So we wakeup the tx thread to send the RELAXWAKE command at the time
that may blocked in the file_read() in rx thread.
NOTE:
Local core should relax the pm wakelock when received POWEROFF
command to allow enter into sleep mode when peer core SHUTDOWN.
2. Change all the flags to event, use event is more convinent
to reprsent all the flags. But the tx sem must be kept because we need
support notify the tx is ready event the tx thread is running to avoid
miss any tx ready event.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Use the TIOCVHANGUP to notify the driver to switch the uart io
to gpio mode to avoid the current leak caused by the uart io because
peer core may already poweroff so the io can be in high voltage
state.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
To maintain compatibility in future update, the data transfer is
still valid even can't process the new added command.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
1. Dump the error log and continue process the next packet when
received error happened instead direclty panic to allow use adb
to update peer core's firmeware;
2. Add more error log to debug the receive error;
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Move rpmsg internal APIs from public header include/nuttx/rpmsg/rpmsg.h
to private header drivers/rpmsg/rpmsg.h. These APIs are only used within
the rpmsg driver implementation and should not be exposed to external
modules.
Moved APIs:
- rpmsg_modify_signals()
- rpmsg_ns_bind() / rpmsg_ns_unbind()
- rpmsg_device_created() / rpmsg_device_destory()
- rpmsg_register() / rpmsg_unregister()
Signed-off-by: Yongrong Wang <wangyongrong@xiaomi.com>
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Add a new getrecvbuf callback to the SPI slave device operations
that allows the slave device to provide a pre-allocated receive
buffer directly to the controller. This enables zero-copy data
transfer by allowing the controller to receive data directly into
the device's buffer instead of copying.
Changes:
- Add SPIS_DEV_GETRECVBUF macro and getrecvbuf callback to
spi_slave_devops_s structure in include/nuttx/spi/slave.h
- Implement getrecvbuf in rpmsg_port_spi_slave driver
- Update getdata to return 0 as it's no longer used for this path
Signed-off-by: liaoao <liaoao@xiaomi.com>
The cmdhdr->len was uninitialized before being used in SPI exchange
operations. This caused incorrect transfer lengths.
Changes:
- Initialize cmdhdr->len to sizeof(struct rpmsg_port_header_s) during
initialization for both SPI master and slave drivers
- Use port.txq.len instead of cmdhdr->len for TX exchange length
- Use port.rxq.len instead of cmdhdr->len for RX getdata length
Signed-off-by: liaoao <liaoao@xiaomi.com>
CC: vfs/fs_unlink.c rpmsg/rpmsg_port_spi.c: In function ‘rpmsg_port_spi_complete_handler’:
rpmsg/rpmsg_port_spi.c:415:11: warning: implicit declaration of function ‘rpmsg_port_spi_drop_packets’; did you mean ‘rpmsg_port_drop_packets’? [-Wimplicit-function-declaration]
415 | rpmsg_port_spi_drop_packets(rpspi, RPMSG_PORT_SPI_DROP_ALL);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~
| rpmsg_port_drop_packets
rpmsg/rpmsg_port_spi.c:415:46: error: ‘RPMSG_PORT_SPI_DROP_ALL’ undeclared (first use in this function); did you mean ‘RPMSG_PORT_DROP_ALL’?
415 | rpmsg_port_spi_drop_packets(rpspi, RPMSG_PORT_SPI_DROP_ALL);
| ^~~~~~~~~~~~~~~~~~~~~~~
| RPMSG_PORT_DROP_ALL
rpmsg/rpmsg_port_spi_slave.c:292:5: warning: implicit declaration of function ‘container_of’; did you mean ‘list_container_of’? [-Wimplicit-function-declaration]
292 | container_of(port, struct rpmsg_port_spi_s, port);
| ^~~~~~~~~~~~
| list_container_of
Signed-off-by: liaoao <liaoao@xiaomi.com>
The default stack size should be large enough to avoid the stack
overflow:
0x18551ad8 1968 1968 100.0%! 24.2% rpmsg-spi
Signed-off-by: liaoao <liaoao@xiaomi.com>
Use pm wakelock is enough to make sure system will not suspend
when there is a transfer in process, and it can actually sleep when
there is buffer to send but peer side not response in 1ms which is
the timeout we set.
Signed-off-by: liaoao <liaoao@xiaomi.com>
Unbind the spictrl to close the spi controller, and bind it again
when peer boots.
This can help to reduce the leak current caused by the spi gpio.
Signed-off-by: liaoao <liaoao@xiaomi.com>
Only call rpmsg_port_spi_pm_action(false) when no more data to
transfer to avoid unnecessary pm unlock/lock operations.
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Now rpmsg_port_spi/slave support send shutdown command to remote
core to notify then local core will reboot, so remote core can
notify all the rpmsg services to do the unregister.
Signed-off-by: liaoao <liaoao@xiaomi.com>
Add low power support for the rpmsg port spi/slave transport, now
the rpmsg port spi/slave transport will not allow enter into the
low power state when there are still tx data to sent or rx data
to processed.
Signed-off-by: liaoao <liaoao@xiaomi.com>
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
This commit fixes a memory alignment issue in the rpmsg subsystem.
The previous implementation used a flexible array member (rdev[0])
in struct rpmsg_s, but due to automatic 8-byte padding, rpmsg->rdev
could differ from the transport layer's rdev address (e.g.,
rpmsg_virtio->rvdev.rdev), causing potential memory corruption.
Changes:
- Remove rdev[0] flexible array member from struct rpmsg_s
- Add rpmsg_get_rdev_by_rpmsg() inline helper to calculate rdev
address as (rpmsg + 1), ensuring correct pointer arithmetic
- Update all rpmsg->rdev references to use the new helper function
- Fix rpmsg_virtio_notify_wait() to use correct container_of macro
Signed-off-by: Yongrong Wang <wangyongrong@xiaomi.com>
Unify all the cpuname/local_cpuname to struct rpmsg_s, so we can
remove the ops->get_cpuname and ops->get_local_cpuname in
struct rpmsg_ops_s.
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
Fix the issue that pm feature not picked before
rename the rpmsg_virtio to rpmsg_virtio_lite
Signed-off-by: buxiasen <buxiasen@xiaomi.com>
Signed-off-by: Bowen Wang <wangbowen6@xiaomi.com>
If ns_msg->flags == RPMSG_NS_CREATE_ACK, means already know peer's address
so direclty use usr_ept->dest_addr
Signed-off-by: yintao <yintao@xiaomi.com>
fist free when rpmsg_port_unregister at rpmsg spi thread
<free+12>
<rpmsg_router_hub_ept_release+6>
<rpmsg_ept_decref+28>
<rpmsg_unregister_endpoint+120>
<rpmsg_destroy_ept+40>
<rpmsg_router_hub_unbind+22>
<rpmsg_device_destory+172>
<rpmsg_port_unregister+26>
this thread will free r:droid and r:audio at last;
But in this process,
rpmsg_destroy_ept "r:droid" will send NS_destroy to audio;
audio will response NS_destroy to ap; if r:droid has not been
removed from the ept list yet, the rptun_audio thread on ap will
occur as follows:
3 0x103016fa in kasan_check_report
4 0x103018b6 in __asan_store4_noabort
5 0x106ca352 in metal_list_del
6 in rpmsg_unregister_endpoint
7 0x106ca77a in rpmsg_destroy_ept
8 0x102db232 in rpmsg_router_hub_unbind
9 0x106cabb2 in rpmsg_virtio_ns_callback
10 0x106cad5c in rpmsg_virtio_rx_callback
11 0x106cc190 in virtqueue_notification
12 0x106ca064 in rproc_virtio_notified
13 0x106c9ad0 in remoteproc_get_notification
14 0x102dd4ba in rptun_worker at rptun/rptun.c:334
15 rptun_worker (arg=<optimized out>) at rptun/rptun.c:328
16 0x102dd974 in rptun_thread rptun/rptun.c:353
17 0x102cd558 in nxtask_start () at task/task_start.c:1
this will lead to r:audio be freed again.
Signed-off-by: yintao <yintao@xiaomi.com>
Piyush Patle
yintao
Bowen Wang
liaoao
zhangbo56
Xiang Xiao
Yongrong Wang
buxiasen