| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Thu, 6 Jul 2023 22:37:51 +0300
From: Jarkko Sonninen <kasper@....fi>
To: Johan Hovold <johan@...nel.org>
Cc: Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v4] USB: serial: xr: Add TIOCGRS485 and TIOCSRS485 ioctls
On 6/20/23 15:38, Johan Hovold wrote:
> Hi Jarkko,
>
> I've been waiting for you to send a v5 that addresses the issue that the
> kernel test robot found, but I guess I can review the other changes in
> v4 first.
Thank you. Sorry for slow responses. My mind has been elsewhere.
> On Sun, Apr 23, 2023 at 09:59:28PM +0300, Jarkko Sonninen wrote:
>> Exar devices like XR21B1411 can control an RS485 transceiver by
>> automatically asserting the RTS#/RS485 pin before sending data
>> and deasserting it when the last stop bit has been transmitted.
>> The polarity of the RST#/RS485 signal is configurable and the
>> hardware also supports half-duplex turn-around delay and
>> address matching mode.
>>
>> Add support for enabling and disabling RS-485 mode and
>> configuring the RST#/RS485 signal polarity using the TIOCGRS485
>> and TIOCSRS485 ioctls. Support for half-duplex turn-around delay
>> and address matching mode are left unimplemented for now.
>>
>> User enables RS-485 mode by setting SER_RS485_ENABLED flag in
>> struct serial_rs485 flags. User should also set either
>> SER_RS485_RTS_ON_SEND or SER_RS485_RTS_AFTER_SEND to select the
>> behaviour of the RTS#/RS485 pin. Setting SER_RS485_RTS_ON_SEND
>> will drive RTS#/RS485 high during transmission. As this is the
>> typical application described by Exar, it is selected when
>> user sets neither or both flags.
> Thanks for updating the commit message here.
>
> Since RTS# is active low, shouldn't SER_RS485_RTS_ON_SEND drive RTS# low
> rather than high during transmission as I also pointed out earlier?
I guess you are right. I'll change that.
I use an exar usb adapter to control a solar charging controller. I
haven't found any other type of exar adapters in ebay.
This adapter uses high level (RTS off) on TX. So I really would like it
to work with the default configuration.
I hope it is ok to use SER_RS485_RTS_AFTER_SEND as the default
>> Signed-off-by: Jarkko Sonninen <kasper@....fi>
>> ---
>> Changes in v3:
>> - In this version only rs485.flags are stored to state.
>> - There is no locking as only one bit of the flags is used.
>> - ioctl returns -ENOIOCTLCMD as the actual error handling is in tty code.
>> Changes in v4:
>> - Store struct rs485 to data
>> - Add mutex to protect data->rs485.
>> - Implement SER_RS485_RTS_ON_SEND or SER_RS485_RTS_AFTER_SEND flags
>> - SER_RS485_RTS_ON_SEND is the default like in serial_core.c
>>
>>
>> drivers/usb/serial/xr_serial.c | 93 +++++++++++++++++++++++++++++++++-
>> 1 file changed, 92 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/usb/serial/xr_serial.c b/drivers/usb/serial/xr_serial.c
>> index fdb0aae546c3..a2157619b63c 100644
>> --- a/drivers/usb/serial/xr_serial.c
>> +++ b/drivers/usb/serial/xr_serial.c
>> @@ -19,6 +19,7 @@
>> #include <linux/usb.h>
>> #include <linux/usb/cdc.h>
>> #include <linux/usb/serial.h>
>> +#include <linux/mutex.h>
> Nit: keep the includes sorted alphabetically.
>
>> struct xr_txrx_clk_mask {
>> u16 tx;
>> @@ -93,6 +94,7 @@ struct xr_txrx_clk_mask {
>> #define XR_GPIO_MODE_SEL_DTR_DSR 0x2
>> #define XR_GPIO_MODE_SEL_RS485 0x3
>> #define XR_GPIO_MODE_SEL_RS485_ADDR 0x4
>> +#define XR_GPIO_MODE_RS485_TX_H 0x8
>> #define XR_GPIO_MODE_TX_TOGGLE 0x100
>> #define XR_GPIO_MODE_RX_TOGGLE 0x200
>>
>> @@ -237,6 +239,8 @@ static const struct xr_type xr_types[] = {
>> struct xr_data {
>> const struct xr_type *type;
>> u8 channel; /* zero-based index or interface number */
>> + struct serial_rs485 rs485;
>> + struct mutex lock;
>> };
>>
>> static int xr_set_reg(struct usb_serial_port *port, u8 channel, u16 reg, u16 val)
>> @@ -630,6 +634,7 @@ static void xr_set_flow_mode(struct tty_struct *tty,
>> const struct xr_type *type = data->type;
>> u16 flow, gpio_mode;
>> int ret;
>> + bool rs485_enabled;
> Nit: move above ret to maintain some approximation of reverse xmas style
> declaration.
>
>>
>> ret = xr_get_reg_uart(port, type->gpio_mode, &gpio_mode);
>> if (ret)
>> @@ -645,7 +650,17 @@ static void xr_set_flow_mode(struct tty_struct *tty,
>> /* Set GPIO mode for controlling the pins manually by default. */
>> gpio_mode &= ~XR_GPIO_MODE_SEL_MASK;
>>
>> - if (C_CRTSCTS(tty) && C_BAUD(tty) != B0) {
>> + rs485_enabled = !!(data->rs485.flags & SER_RS485_ENABLED);
>> + if (rs485_enabled) {
>> + dev_dbg(&port->dev, "Enabling RS-485\n");
>> + gpio_mode |= XR_GPIO_MODE_SEL_RS485;
>> + if (data->rs485.flags & SER_RS485_RTS_ON_SEND)
>> + gpio_mode |= XR_GPIO_MODE_RS485_TX_H;
>> + else
>> + gpio_mode &= ~XR_GPIO_MODE_RS485_TX_H;
> As mentioned above, should this not be inverted?
>
>> + }
>> +
>> + if (C_CRTSCTS(tty) && C_BAUD(tty) != B0 && !rs485_enabled) {
>> dev_dbg(&port->dev, "Enabling hardware flow ctrl\n");
>> gpio_mode |= XR_GPIO_MODE_SEL_RTS_CTS;
>> flow = XR_UART_FLOW_MODE_HW;
>> @@ -809,6 +824,80 @@ static void xr_cdc_set_line_coding(struct tty_struct *tty,
>> kfree(lc);
>> }
>>
>> +static void xr_sanitize_serial_rs485(struct serial_rs485 *rs485)
>> +{
>> + if (!(rs485->flags & SER_RS485_ENABLED)) {
>> + memset(rs485, 0, sizeof(*rs485));
>> + return;
>> + }
> This looks odd to me, but I see that this is what serial core is
> currently doing...
>
>> +
>> + /* Select RTS on send if the user hasn't selected the mode properly */
>> + if (!!(rs485->flags & SER_RS485_RTS_ON_SEND) ==
>> + !!(rs485->flags & SER_RS485_RTS_AFTER_SEND)) {
>> + rs485->flags |= SER_RS485_RTS_ON_SEND;
>> + rs485->flags &= ~SER_RS485_RTS_AFTER_SEND;
>> + }
>> +
>> + /* Only the flags are implemented at the moment */
>> + rs485->flags &= SER_RS485_ENABLED | SER_RS485_RTS_ON_SEND |
>> + SER_RS485_RTS_AFTER_SEND;
>> + rs485->delay_rts_before_send = 0;
>> + rs485->delay_rts_after_send = 0;
>> + memset(rs485->padding, 0, sizeof(rs485->padding));
>> +}
>> +
>> +static int xr_get_rs485_config(struct tty_struct *tty,
>> + struct serial_rs485 *argp)
> You still need __user annotation here as the build robot reported.
>
>> +{
>> + struct usb_serial_port *port = tty->driver_data;
>> + struct xr_data *data = usb_get_serial_port_data(port);
>> +
>> + mutex_lock(&data->lock);
> This does not prevent the termios structure from changing underneath
> you. You need to take a write lock on the termios rw sem here instead.
I'll change these to "down_read(&tty->termios_rwsem);" and
"up_read(&tty->termios_rwsem);"
I would feel safer by using a local variable "struct serial_rs485" to
minimize the holding of the semaphore.
>
>> + if (copy_to_user(argp, &data->rs485, sizeof(data->rs485))) {
>> + mutex_unlock(&data->lock);
>> + return -EFAULT;
>> + }
>> + mutex_unlock(&data->lock);
>> +
>> + return 0;
>> +}
>> +
>> +static int xr_set_rs485_config(struct tty_struct *tty,
>> + struct serial_rs485 *argp)
> missing __user
>
>> +{
>> + struct usb_serial_port *port = tty->driver_data;
>> + struct xr_data *data = usb_get_serial_port_data(port);
>> + struct serial_rs485 rs485;
>> +
>> + if (copy_from_user(&rs485, argp, sizeof(rs485)))
>> + return -EFAULT;
>> + xr_sanitize_serial_rs485(&rs485);
>> +
>> + mutex_lock(&data->lock);
> Same here as set_termios() (and xr_set_flow_mode()) can otherwise be
> called in parallel.
I'll change these to "down_write(&tty->termios_rwsem);" and
"up_write(&tty->termios_rwsem);"
>> + data->rs485 = rs485;
>> + xr_set_flow_mode(tty, port, NULL);
>> + mutex_unlock(&data->lock);
>> +
>> + if (copy_to_user(argp, &rs485, sizeof(rs485)))
>> + return -EFAULT;
>> +
>> + return 0;
>> +}
> Looks good otherwise.
>
> Johan
- Jarkko
Powered by blists - more mailing lists