Message-ID: <760819dc-4c26-b492-a0ba-8b27c189d5b1@st.com>
Date:   Fri, 5 Apr 2019 18:08:52 +0200
From:   Arnaud Pouliquen <arnaud.pouliquen@...com>
To:     xiang xiao <xiaoxiang781216@...il.com>
CC:     Fabien Dessenne <fabien.dessenne@...com>,
        Ohad Ben-Cohen <ohad@...ery.com>,
        Bjorn Andersson <bjorn.andersson@...aro.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Jiri Slaby <jslaby@...e.com>, <linux-kernel@...r.kernel.org>,
        <linux-remoteproc@...r.kernel.org>,
        Benjamin Gaignard <benjamin.gaignard@...com>
Subject: Re: [PATCH 2/2] tty: add rpmsg driver



On 4/5/19 4:03 PM, xiang xiao wrote:
> On Fri, Apr 5, 2019 at 8:33 PM Arnaud Pouliquen <arnaud.pouliquen@...com> wrote:
>>
>>
>>
>> On 4/5/19 12:12 PM, xiang xiao wrote:
>>> On Fri, Apr 5, 2019 at 12:14 AM Arnaud Pouliquen
>>> <arnaud.pouliquen@...com> wrote:
>>>>
>>>> Hello Xiang,
>>>>
>>>>
>>>> On 4/3/19 2:47 PM, xiang xiao wrote:
>>>>> On Thu, Mar 21, 2019 at 11:48 PM Fabien Dessenne <fabien.dessenne@...com> wrote:
>>>>>>
>>>>>> This driver exposes a standard tty interface on top of the rpmsg
>>>>>> framework through the "rpmsg-tty-channel" rpmsg service.
>>>>>>
>>>>>> This driver supports multi-instances, offering a /dev/ttyRPMSGx entry
>>>>>> per rpmsg endpoint.
>>>>>>
>>>>>
>>>>> How to support multi-instances from the same remoteproc instance? I
>>>>> saw that the channel name is fixed to "rpmsg-tty-channel" which mean
>>>>> only one channel can be created for each remote side.
>>>> The driver is multi-instance based on muti-endpoints on top of the
>>>> "rpmsg-tty-channel" service.
>>>> On remote side you just have to call rpmsg_create_ept with destination
>>>> address set to ANY. The result is a NS service announcement that trigs a
>>>>  probe with a new endpoint.
>>>> You can find code example for the remote side here:
>>>> https://github.com/STMicroelectronics/STM32CubeMP1/blob/master/Projects/STM32MP157C-DK2/Applications/OpenAMP/OpenAMP_TTY_echo/Src/main.c
>>>
>>> Demo code create two uarts(huart0 and huart1), and both use the same
>>> channel name( "rpmsg-tty-channel").
>>> But rpmsg_create_channel in kernel will complain the duplication:
>>> /* make sure a similar channel doesn't already exist */
>>> tmp = rpmsg_find_device(dev, chinfo);
>>> if (tmp) {
>>>         /* decrement the matched device's refcount back */
>>>         put_device(tmp);
>>>         dev_err(dev, "channel %s:%x:%x already exist\n",
>>>                    chinfo->name, chinfo->src, chinfo->dst);
>>>         return NULL;
>>> }
>>> Do you have some local change not upstream yet?
>> Nothing is missing. There is no complain as the function
>> rpmsg_device_match returns 0, because the chinfo->dst (that corresponds
>> to the remote ept address) is different.
>>
> 
> Yes, you are right.
> 
>> If i well remember you have also a similar implementation of the service
>> on your side. Do you see any incompatibility with your implementation?
>>
> 
> Our implementation is similar to yours, but has two major difference:
> 1.Each instance has a different channel name but share the same prefix
> "rpmsg-tty*", the benefit is that:
>    a.Device name(/dev/tty*) is derived from rpmsg-tty*, instead the
> random /dev/ttyRPMSGx
>    b.Don't need tty_idr to allocate the unique device id
I understand the need but in your implementation it look like you hack
the rpmsg service to instantiate your tty... you introduce a kind of
meta rpmsg tty service with sub-service related to the serial content.
Not sure that this could be upstreamed...
proposal to integrate your need in the ST driver: it seems possible to
have /dev/ttyRPMSGx with x corresponding to the remote endpoint address?
So if you want to have a fixed tty name you can fix the remote endpoint.
Is it something reasonable for you?


> 2.Each transfer need get response from peer to avoid the buffer
> overflow. This is very important if the peer use pull
> model(read/write) instead of push model(callback).
I not sure to understand your point... You mean that you assume that the
driver should be blocked until a response from the remote side?
This seems not compatible with a "generic" tty and with Johan remarks:
"Just a drive-by comment; it looks like rpmsg_send() may block, but
the tty-driver write() callback must never sleep."

Why no just let rpmsg should manage the case with no Tx buffer free,
with a rpmsg_trysend...

> 
> Here is the patch for kernel side:
> https://github.com/xiaoxiang781216/linux/commit/f04d2386eb11e0781f0eb47d99fae838abf7eb53
> https://github.com/xiaoxiang781216/linux/commit/1a41be362d916eb9104bf21065cb38a0a63d2e91
> 
> And RTOS side:
> https://github.com/PineconeCode/nuttx/blob/song-u1/include/nuttx/serial/uart_rpmsg.h
> https://github.com/PineconeCode/nuttx/blob/song-u1/drivers/serial/uart_rpmsg.c
> 
> Maybe, we can consider how to unify these two implementation into one.
Yes sure.

> 
>>>
>>>> using some wrapper functions available here:
>>>> https://github.com/STMicroelectronics/STM32CubeMP1/tree/master/Middlewares/Third_Party/OpenAMP
>>>>
>>>>
>>>> Best Regards
>>>> Arnaud
>>>>
>>>>>
>>>>>> Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@...com>
>>>>>> Signed-off-by: Fabien Dessenne <fabien.dessenne@...com>
>>>>>> ---
>>>>>>  drivers/tty/Kconfig     |   9 ++
>>>>>>  drivers/tty/Makefile    |   1 +
>>>>>>  drivers/tty/rpmsg_tty.c | 309 ++++++++++++++++++++++++++++++++++++++++++++++++
>>>>>>  3 files changed, 319 insertions(+)
>>>>>>  create mode 100644 drivers/tty/rpmsg_tty.c
>>>>>>
>>>>>> diff --git a/drivers/tty/Kconfig b/drivers/tty/Kconfig
>>>>>> index 0840d27..42696e6 100644
>>>>>> --- a/drivers/tty/Kconfig
>>>>>> +++ b/drivers/tty/Kconfig
>>>>>> @@ -441,4 +441,13 @@ config VCC
>>>>>>         depends on SUN_LDOMS
>>>>>>         help
>>>>>>           Support for Sun logical domain consoles.
>>>>>> +
>>>>>> +config RPMSG_TTY
>>>>>> +       tristate "RPMSG tty driver"
>>>>>> +       depends on RPMSG
>>>>>> +       help
>>>>>> +         Say y here to export rpmsg endpoints as tty devices, usually found
>>>>>> +         in /dev/ttyRPMSGx.
>>>>>> +         This makes it possible for user-space programs to send and receive
>>>>>> +         rpmsg messages as a standard tty protocol.
>>>>>>  endif # TTY
>>>>>> diff --git a/drivers/tty/Makefile b/drivers/tty/Makefile
>>>>>> index c72cafd..90a98a2 100644
>>>>>> --- a/drivers/tty/Makefile
>>>>>> +++ b/drivers/tty/Makefile
>>>>>> @@ -33,5 +33,6 @@ obj-$(CONFIG_PPC_EPAPR_HV_BYTECHAN) += ehv_bytechan.o
>>>>>>  obj-$(CONFIG_GOLDFISH_TTY)     += goldfish.o
>>>>>>  obj-$(CONFIG_MIPS_EJTAG_FDC_TTY) += mips_ejtag_fdc.o
>>>>>>  obj-$(CONFIG_VCC)              += vcc.o
>>>>>> +obj-$(CONFIG_RPMSG_TTY)                += rpmsg_tty.o
>>>>>>
>>>>>>  obj-y += ipwireless/
>>>>>> diff --git a/drivers/tty/rpmsg_tty.c b/drivers/tty/rpmsg_tty.c
>>>>>> new file mode 100644
>>>>>> index 0000000..9c2a629
>>>>>> --- /dev/null
>>>>>> +++ b/drivers/tty/rpmsg_tty.c
>>>>>> @@ -0,0 +1,309 @@
>>>>>> +// SPDX-License-Identifier: GPL-2.0
>>>>>> +/*
>>>>>> + * Copyright (C) STMicroelectronics 2019 - All Rights Reserved
>>>>>> + * Authors: Arnaud Pouliquen <arnaud.pouliquen@...com> for STMicroelectronics.
>>>>>> + *          Fabien Dessenne <fabien.dessenne@...com> for STMicroelectronics.
>>>>>> + * Derived from the imx-rpmsg and omap-rpmsg implementations.
>>>>>> + */
>>>>>> +
>>>>>> +#include <linux/module.h>
>>>>>> +#include <linux/rpmsg.h>
>>>>>> +#include <linux/slab.h>
>>>>>> +#include <linux/tty.h>
>>>>>> +#include <linux/tty_flip.h>
>>>>>> +
>>>>>> +#define MAX_TTY_RPMSG          32
>>>>>> +
>>>>>> +static DEFINE_IDR(tty_idr);            /* tty instance id */
>>>>>> +static DEFINE_MUTEX(idr_lock);         /* protects tty_idr */
>>>>>> +
>>>>>> +static struct tty_driver *rpmsg_tty_driver;
>>>>>> +
>>>>>> +struct rpmsg_tty_port {
>>>>>> +       struct tty_port         port;   /* TTY port data */
>>>>>> +       unsigned int            id;     /* TTY rpmsg index */
>>>>>> +       spinlock_t              rx_lock; /* message reception lock */
>>>>>> +       struct rpmsg_device     *rpdev; /* rpmsg device */
>>>>>> +};
>>>>>> +
>>>>>> +static int rpmsg_tty_cb(struct rpmsg_device *rpdev, void *data, int len,
>>>>>> +                       void *priv, u32 src)
>>>>>> +{
>>>>>> +       int space;
>>>>>> +       unsigned char *cbuf;
>>>>>> +       struct rpmsg_tty_port *cport = dev_get_drvdata(&rpdev->dev);
>>>>>> +
>>>>>> +       dev_dbg(&rpdev->dev, "msg(<- src 0x%x) len %d\n", src, len);
>>>>>> +
>>>>>> +       print_hex_dump_debug(__func__, DUMP_PREFIX_NONE, 16, 1, data, len,
>>>>>> +                            true);
>>>>>> +
>>>>>> +       /* Flush the recv-ed none-zero data to tty node */
>>>>>> +       if (!len)
>>>>>> +               return -EINVAL;
>>>>>> +
>>>>>> +       spin_lock(&cport->rx_lock);
>>>>>> +       space = tty_prepare_flip_string(&cport->port, &cbuf, len);
>>>>>> +       if (space <= 0) {
>>>>>> +               dev_err(&rpdev->dev, "No memory for tty_prepare_flip_string\n");
>>>>>> +               spin_unlock(&cport->rx_lock);
>>>>>> +               return -ENOMEM;
>>>>>> +       }
>>>>>> +
>>>>>> +       if (space != len)
>>>>>> +               dev_warn(&rpdev->dev, "Trunc buffer from %d to %d\n",
>>>>>> +                        len, space);
>>>>>> +
>>>>>> +       memcpy(cbuf, data, space);
>>>>>> +       tty_flip_buffer_push(&cport->port);
>>>>>> +       spin_unlock(&cport->rx_lock);
>>>>>> +
>>>>>> +       return 0;
>>>>>> +}
>>>>>> +
>>>>>> +static int rpmsg_tty_install(struct tty_driver *driver, struct tty_struct *tty)
>>>>>> +{
>>>>>> +       struct rpmsg_tty_port *cport = idr_find(&tty_idr, tty->index);
>>>>>> +
>>>>>> +       if (!cport) {
>>>>>> +               dev_err(tty->dev, "cannot get cport\n");
>>>>>> +               return -ENODEV;
>>>>>> +       }
>>>>>> +
>>>>>> +       return tty_port_install(&cport->port, driver, tty);
>>>>>> +}
>>>>>> +
>>>>>> +static int rpmsg_tty_open(struct tty_struct *tty, struct file *filp)
>>>>>> +{
>>>>>> +       return tty_port_open(tty->port, tty, filp);
>>>>>> +}
>>>>>> +
>>>>>> +static void rpmsg_tty_close(struct tty_struct *tty, struct file *filp)
>>>>>> +{
>>>>>> +       return tty_port_close(tty->port, tty, filp);
>>>>>> +}
>>>>>> +
>>>>>> +static int rpmsg_tty_write(struct tty_struct *tty, const unsigned char *buf,
>>>>>> +                          int total)
>>>>>> +{
>>>>>> +       int count, ret = 0;
>>>>>> +       const unsigned char *tbuf;
>>>>>> +       struct rpmsg_tty_port *cport = idr_find(&tty_idr, tty->index);
>>>>>> +       struct rpmsg_device *rpdev;
>>>>>> +       int msg_size;
>>>>>> +
>>>>>> +       if (!cport) {
>>>>>> +               dev_err(tty->dev, "cannot get cport\n");
>>>>>> +               return -ENODEV;
>>>>>> +       }
>>>>>> +
>>>>>> +       rpdev = cport->rpdev;
>>>>>> +
>>>>>> +       dev_dbg(&rpdev->dev, "%s: send message from tty->index = %d\n",
>>>>>> +               __func__, tty->index);
>>>>>> +
>>>>>> +       if (!buf) {
>>>>>> +               dev_err(&rpdev->dev, "buf shouldn't be null.\n");
>>>>>> +               return -ENOMEM;
>>>>>> +       }
>>>>>> +
>>>>>> +       msg_size = rpmsg_get_buf_payload_size(rpdev->ept);
>>>>>> +       if (msg_size < 0)
>>>>>> +               return msg_size;
>>>>>> +
>>>>>> +       count = total;
>>>>>> +       tbuf = buf;
>>>>>> +       do {
>>>>>> +               /* send a message to our remote processor */
>>>>>> +               ret = rpmsg_send(rpdev->ept, (void *)tbuf,
>>>>>> +                                min(count, msg_size));
>>>>>> +               if (ret) {
>>>>>> +                       dev_err(&rpdev->dev, "rpmsg_send failed: %d\n", ret);
>>>>>> +                       return ret;
>>>>>> +               }
>>>>>> +
>>>>>> +               if (count > msg_size) {
>>>>>> +                       count -= msg_size;
>>>>>> +                       tbuf += msg_size;
>>>>>> +               } else {
>>>>>> +                       count = 0;
>>>>>> +               }
>>>>>> +       } while (count > 0);
>>>>>> +
>>>>>
>>>>> We need the flow control(or handshake) here, otherwise it's very easy
>>>>> to lose the data if kernel keep send the data as fast as possible.
>>>>>
>>>>>> +       return total;
>>>>>> +}
>>>>>> +
>>>>>> +static int rpmsg_tty_write_room(struct tty_struct *tty)
>>>>>> +{
>>>>>> +       struct rpmsg_tty_port *cport = idr_find(&tty_idr, tty->index);
>>>>>> +
>>>>>> +       if (!cport) {
>>>>>> +               dev_err(tty->dev, "cannot get cport\n");
>>>>>> +               return -ENODEV;
>>>>>> +       }
>>>>>> +
>>>>>> +       /* report the space in the rpmsg buffer */
>>>>>> +       return rpmsg_get_buf_payload_size(cport->rpdev->ept);
>>>>>> +}
>>>>>> +
>>>>>> +static const struct tty_operations rpmsg_tty_ops = {
>>>>>> +       .install        = rpmsg_tty_install,
>>>>>> +       .open           = rpmsg_tty_open,
>>>>>> +       .close          = rpmsg_tty_close,
>>>>>> +       .write          = rpmsg_tty_write,
>>>>>> +       .write_room     = rpmsg_tty_write_room,
>>>>>> +};
>>>>>> +
>>>>>> +struct rpmsg_tty_port *rpmsg_tty_alloc_cport(void)
>>>>>> +{
>>>>>> +       struct rpmsg_tty_port *cport;
>>>>>> +
>>>>>> +       cport = kzalloc(sizeof(*cport), GFP_KERNEL);
>>>>>> +       if (!cport)
>>>>>> +               return ERR_PTR(-ENOMEM);
>>>>>> +
>>>>>> +       mutex_lock(&idr_lock);
>>>>>> +       cport->id = idr_alloc(&tty_idr, cport, 0, MAX_TTY_RPMSG, GFP_KERNEL);
>>>>>> +       mutex_unlock(&idr_lock);
>>>>>> +
>>>>>> +       if (cport->id < 0) {
>>>>>> +               kfree(cport);
>>>>>> +               return ERR_PTR(-ENOSPC);
>>>>>> +       }
>>>>>> +
>>>>>> +       return cport;
>>>>>> +}
>>>>>> +
>>>>>> +static void rpmsg_tty_release_cport(struct rpmsg_tty_port *cport)
>>>>>> +{
>>>>>> +       mutex_lock(&idr_lock);
>>>>>> +       idr_remove(&tty_idr, cport->id);
>>>>>> +       mutex_unlock(&idr_lock);
>>>>>> +
>>>>>> +       kfree(cport);
>>>>>> +}
>>>>>> +
>>>>>> +static int rpmsg_tty_probe(struct rpmsg_device *rpdev)
>>>>>> +{
>>>>>> +       struct rpmsg_tty_port *cport;
>>>>>> +       struct device *tty_dev;
>>>>>> +       int ret;
>>>>>> +
>>>>>> +       cport = rpmsg_tty_alloc_cport();
>>>>>> +       if (IS_ERR(cport)) {
>>>>>> +               dev_err(&rpdev->dev, "failed to alloc tty port\n");
>>>>>> +               return PTR_ERR(cport);
>>>>>> +       }
>>>>>> +
>>>>>> +       tty_port_init(&cport->port);
>>>>>> +       cport->port.low_latency = cport->port.flags | ASYNC_LOW_LATENCY;
>>>>>> +
>>>>>> +       tty_dev = tty_port_register_device(&cport->port, rpmsg_tty_driver,
>>>>>> +                                          cport->id, &rpdev->dev);
>>>>>> +       if (IS_ERR(tty_dev)) {
>>>>>> +               dev_err(&rpdev->dev, "failed to register tty port\n");
>>>>>> +               ret = PTR_ERR(tty_dev);
>>>>>> +               goto err_destroy;
>>>>>> +       }
>>>>>> +
>>>>>> +       spin_lock_init(&cport->rx_lock);
>>>>>> +       cport->rpdev = rpdev;
>>>>>> +
>>>>>> +       dev_set_drvdata(&rpdev->dev, cport);
>>>>>> +
>>>>>> +       dev_info(&rpdev->dev, "new channel: 0x%x -> 0x%x : ttyRPMSG%d\n",
>>>>>> +                rpdev->src, rpdev->dst, cport->id);
>>>>>> +
>>>>>> +       return 0;
>>>>>> +
>>>>>> +err_destroy:
>>>>>> +       tty_port_destroy(&cport->port);
>>>>>> +       rpmsg_tty_release_cport(cport);
>>>>>> +       return ret;
>>>>>> +}
>>>>>> +
>>>>>> +static void rpmsg_tty_remove(struct rpmsg_device *rpdev)
>>>>>> +{
>>>>>> +       struct rpmsg_tty_port *cport = dev_get_drvdata(&rpdev->dev);
>>>>>> +
>>>>>> +       dev_info(&rpdev->dev, "removing rpmsg tty device %d\n", cport->id);
>>>>>> +
>>>>>> +       /* User hang up to release the tty */
>>>>>> +       if (tty_port_initialized(&cport->port))
>>>>>> +               tty_port_tty_hangup(&cport->port, false);
>>>>>> +
>>>>>> +       tty_unregister_device(rpmsg_tty_driver, cport->id);
>>>>>> +       tty_port_destroy(&cport->port);
>>>>>> +
>>>>>> +       rpmsg_tty_release_cport(cport);
>>>>>> +}
>>>>>> +
>>>>>> +static struct rpmsg_device_id rpmsg_driver_tty_id_table[] = {
>>>>>> +       { .name = "rpmsg-tty-channel" },
>>>>>> +       { },
>>>>>> +};
>>>>>> +MODULE_DEVICE_TABLE(rpmsg, rpmsg_driver_tty_id_table);
>>>>>> +
>>>>>> +static struct rpmsg_driver rpmsg_tty_rpmsg_drv = {
>>>>>> +       .drv.name       = KBUILD_MODNAME,
>>>>>> +       .id_table       = rpmsg_driver_tty_id_table,
>>>>>> +       .probe          = rpmsg_tty_probe,
>>>>>> +       .callback       = rpmsg_tty_cb,
>>>>>> +       .remove         = rpmsg_tty_remove,
>>>>>> +};
>>>>>> +
>>>>>> +static int __init rpmsg_tty_init(void)
>>>>>> +{
>>>>>> +       int err;
>>>>>> +
>>>>>> +       rpmsg_tty_driver = tty_alloc_driver(MAX_TTY_RPMSG, TTY_DRIVER_REAL_RAW |
>>>>>> +                                           TTY_DRIVER_DYNAMIC_DEV);
>>>>>> +       if (IS_ERR(rpmsg_tty_driver))
>>>>>> +               return PTR_ERR(rpmsg_tty_driver);
>>>>>> +
>>>>>> +       rpmsg_tty_driver->driver_name = "rpmsg_tty";
>>>>>> +       rpmsg_tty_driver->name = "ttyRPMSG";
>>>>>> +       rpmsg_tty_driver->major = 0;
>>>>>> +       rpmsg_tty_driver->type = TTY_DRIVER_TYPE_CONSOLE;
>>>>>> +       rpmsg_tty_driver->init_termios = tty_std_termios;
>>>>>> +
>>>>>> +       tty_set_operations(rpmsg_tty_driver, &rpmsg_tty_ops);
>>>>>> +
>>>>>> +       err = tty_register_driver(rpmsg_tty_driver);
>>>>>> +       if (err < 0) {
>>>>>> +               pr_err("Couldn't install rpmsg tty driver: err %d\n", err);
>>>>>> +               goto error_put;
>>>>>> +       }
>>>>>> +
>>>>>> +       err = register_rpmsg_driver(&rpmsg_tty_rpmsg_drv);
>>>>>> +       if (err < 0) {
>>>>>> +               pr_err("Couldn't register rpmsg tty driver: err %d\n", err);
>>>>>> +               goto error_unregister;
>>>>>> +       }
>>>>>> +
>>>>>> +       return 0;
>>>>>> +
>>>>>> +error_unregister:
>>>>>> +       tty_unregister_driver(rpmsg_tty_driver);
>>>>>> +
>>>>>> +error_put:
>>>>>> +       put_tty_driver(rpmsg_tty_driver);
>>>>>> +
>>>>>> +       return err;
>>>>>> +}
>>>>>> +
>>>>>> +static void __exit rpmsg_tty_exit(void)
>>>>>> +{
>>>>>> +       unregister_rpmsg_driver(&rpmsg_tty_rpmsg_drv);
>>>>>> +       tty_unregister_driver(rpmsg_tty_driver);
>>>>>> +       put_tty_driver(rpmsg_tty_driver);
>>>>>> +       idr_destroy(&tty_idr);
>>>>>> +}
>>>>>> +
>>>>>> +module_init(rpmsg_tty_init);
>>>>>> +module_exit(rpmsg_tty_exit);
>>>>>> +
>>>>>> +MODULE_AUTHOR("Arnaud Pouliquen <arnaud.pouliquen@...com>");
>>>>>> +MODULE_AUTHOR("Fabien Dessenne <fabien.dessenne@...com>");
>>>>>> +MODULE_DESCRIPTION("virtio remote processor messaging tty driver");
>>>>>> +MODULE_LICENSE("GPL v2");
>>>>>> --
>>>>>> 2.7.4
>>>>>>

Powered by blists - more mailing lists