serial: core: fix broken console after suspend
Subsystems (2)
v2 cover serial: core: fix broken console after suspend
This is v2 of a patch[1].v1->v2:
* Fixed typos in commit message
* Tested[2] patch "serial: core: preserve cflags, ispeed and ospeed" on all current LTS kernels (4.14, 4.19, 5.4, 5.10, 5.15, 6.1) and on top of 6.3-rc5 with positive results - console was working after resume
* During tests another issue was observed on 6.1+ - FIFO not enabled after resume - and an additional change was needed ("serial: core: enable FIFO after resume")
* Test procedure:
1) ensure the console output is ok
2) suspend device with "echo freeze > /sys/power/state"
(/sys/module/printk/parameters/console_suspend == N)
3) resume device and check the console output
4) suspend device with "echo freeze > /sys/power/state"
(/sys/module/printk/parameters/console_suspend == Y)5) resume device and check the console output
[1] https://lore.kernel.org/lkml/[email protected]/
[2] Test platforms: PC with i5-8400 + GB H370M D3H
HP Elite c1030 Chromebook Enterprise i5-10310U
Lukasz Majczak (2):
serial: core: preserve cflags, ispeed and ospeed
serial: core: enable FIFO after resume
drivers/tty/serial/serial_core.c | 57 +++++++++++++++----------------- 1 file changed, 27 insertions(+), 30 deletions(-)
Signature · 1 line
2.40.0.577.gac1e443424-goog
PATCH v2 1/2 serial: core: preserve cflags, ispeed and ospeed
Re-enable the console device after suspending, causes its cflags, ispeed and ospeed to be set anew, basing on the values stored in uport->cons. These values are set only once,
when parsing console parameters after boot (see uart_set_options()), next after configuring a port in uart_port_startup() these parameters (cflags, ispeed and ospeed) are copied to termios structure and the original one (stored in uport->cons) are cleared, but there is no place in code where those fields are checked against 0.
When kernel calls uart_resume_port() and setups console, it copies cflags, ispeed and ospeed values from uart->cons, but those are already cleared. The effect is that the console is broken.
This patch address the issue by preserving the cflags, ispeed and ospeed fields in uart->cons during uart_port_startup().
--- drivers/tty/serial/serial_core.c | 3 --- 1 file changed, 3 deletions(-)
drivers/tty/serial/serial_core.c +0 -3
@@ -225,9 +225,6 @@ static int uart_port_startup(struct tty_struct *tty, struct uart_state *state, tty->termios.c_cflag = uport->cons->cflag; tty->termios.c_ispeed = uport->cons->ispeed; tty->termios.c_ospeed = uport->cons->ospeed; - uport->cons->cflag =0 ; - uport->cons->ispeed =0 ; - uport->cons->ospeed =0 ; }/* * Initialise the hardware port settings.
Signature · 1 line
2.40.0.577.gac1e443424-goog
PATCH v2 2/2 serial: core: enable FIFO after resume
In reply to Lukasz MajczakThe "serial/8250: Use fifo in 8250 console driver" commit has revealed an issue of not re-enabling FIFO after resume. The problematic path is inside uart_resume_port() function. First, when the console device is re-enabled,
a call to uport->ops->set_termios() internally initializes FIFO (in serial8250_do_set_termios()), although further code disables it by issuing ops->startup() (pointer to serial8250_do_startup, internally calling serial8250_clear_fifos()).
There is even a comment saying "Clear the FIFO buffers and disable them. (they will be reenabled in set_termios())", but in this scenario, set_termios() has been called already and FIFO remains disabled.
This patch address the issue by reversing the order - first checks if tty port is suspended and performs actions accordingly (e.g. call to ops->startup()), then tries to re-enable
the console device after suspend (and call to uport->ops->set_termios()).
--- drivers/tty/serial/serial_core.c | 54 ++++++++++++++++---------------- 1 file changed, 27 insertions(+), 27 deletions(-)
drivers/tty/serial/serial_core.c +27 -27
@@ -2406,33 +2406,6 @@ int uart_resume_port(struct uart_driver *drv, struct uart_port *uport)put_device (tty_dev); uport->suspended =0 ; -/* -* Re-enable the console device after suspending. -*/ -if (uart_console (uport)) { -/* -* First try to use the console cflag setting. -*/ -memset (&termios,0 ,sizeof (struct ktermios )); - termios.c_cflag = uport->cons->cflag; - termios.c_ispeed = uport->cons->ispeed; - termios.c_ospeed = uport->cons->ospeed; - -/* -* If that's unset, use the tty termios setting. -*/ -if (port->tty && termios.c_cflag ==0 ) - termios = port->tty->termios; - -if (console_suspend_enabled) -uart_change_pm (state,UART_PM_STATE_ON ); - uport->ops->set_termios (uport, &termios,NULL ); -if (!console_suspend_enabled && uport->ops->start_rx) - uport->ops->start_rx (uport); -if (console_suspend_enabled) -console_start (uport->cons); - } -if (tty_port_suspended (port)) {const struct uart_ops *ops = uport->ops;int ret;
@@ -2471,6 +2444,33 @@ int uart_resume_port(struct uart_driver *drv, struct uart_port *uport)tty_port_set_suspended (port, false); } +/* +* Re-enable the console device after suspending. +*/ +if (uart_console (uport)) { +/* +* First try to use the console cflag setting. +*/ +memset (&termios,0 ,sizeof (struct ktermios )); + termios.c_cflag = uport->cons->cflag; + termios.c_ispeed = uport->cons->ispeed; + termios.c_ospeed = uport->cons->ospeed; + +/* +* If that's unset, use the tty termios setting. +*/ +if (port->tty && termios.c_cflag ==0 ) + termios = port->tty->termios; + +if (console_suspend_enabled) +uart_change_pm (state,UART_PM_STATE_ON ); + uport->ops->set_termios (uport, &termios,NULL ); +if (!console_suspend_enabled && uport->ops->start_rx) + uport->ops->start_rx (uport); +if (console_suspend_enabled) +console_start (uport->cons); + } +mutex_unlock (&port->mutex);return 0 ;
Signature · 1 line
2.40.0.577.gac1e443424-goog
Re: [PATCH v2 2/2] serial: core: enable FIFO after resume
Quoted 1 line · Lukasz Majczak
The "serial/8250: Use fifo in 8250 console driver" commit
Use canonical formatting when referring to commit:
commit 12char_SHA1 ("shortlog")
Quoted 5 lines
has revealed an issue of not re-enabling FIFO after resume. The problematic path is inside
uart_resume_port()function. First, when the console device is re-enabled,
a call touport->ops->set_termios()internally initializes FIFO (inserial8250_do_set_termios()), although further code
I'd drop "First," and start with "When" and change "although" to "then"
Quoted 5 lines
disables it by issuing
ops->startup()(pointer toserial8250_do_startup, internally callingserial8250_clear_fifos()).
There is even a comment saying "Clear the FIFO buffers and disable them. (they will be reenabled inset_termios())", but in this scenario,set_termios()has been called already and FIFO remains disabled.
Also, you should reflow the text to 72 chars per line.
Quoted 74 lines
This patch address the issue by reversing the order - first checks if tty port is suspended and performs actions accordingly (e.g. call to
ops->startup()), then tries to re-enable
the console device after suspend (and call touport->ops->set_termios()).Signed-off-by Lukasz Majczak <[email protected]>Cc <[email protected]> # 6.1+--- drivers/tty/serial/serial_core.c | 54 ++++++++++++++++---------------- 1 file changed, 27 insertions(+), 27 deletions(-)drivers/tty/serial/serial_core.c +20
-27@@ -2406,33 +2406,6 @@ int uart_resume_port(struct uart_driver *drv, struct uart_port *uport)put_device (tty_dev); uport->suspended =0 ; -/* -* Re-enable the console device after suspending. -*/ -if (uart_console (uport)) { -/* -* First try to use the console cflag setting. -*/ -memset (&termios,0 ,sizeof (struct ktermios )); - termios.c_cflag = uport->cons->cflag; - termios.c_ispeed = uport->cons->ispeed; - termios.c_ospeed = uport->cons->ospeed; - -/* -* If that's unset, use the tty termios setting. -*/ -if (port->tty && termios.c_cflag ==0 ) - termios = port->tty->termios; - -if (console_suspend_enabled) -uart_change_pm (state,UART_PM_STATE_ON ); - uport->ops->set_termios (uport, &termios,NULL ); -if (!console_suspend_enabled && uport->ops->start_rx) - uport->ops->start_rx (uport); -if (console_suspend_enabled) -console_start (uport->cons); - } -if (tty_port_suspended (port)) {const struct uart_ops *ops = uport->ops;int ret;@@ -2471,6 +2444,33 @@ int uart_resume_port(struct uart_driver *drv, struct uart_port *uport)tty_port_set_suspended (port, false); } +/* +* Re-enable the console device after suspending. +*/ +if (uart_console (uport)) { +/* +* First try to use the console cflag setting. +*/ +memset (&termios,0 ,sizeof (struct ktermios )); + termios.c_cflag = uport->cons->cflag; + termios.c_ispeed = uport->cons->ispeed; + termios.c_ospeed = uport->cons->ospeed; + +/* +* If that's unset, use the tty termios setting. +*/ +if (port->tty && termios.c_cflag ==0 ) + termios = port->tty->termios; + +if (console_suspend_enabled) +uart_change_pm (state,UART_PM_STATE_ON );
drivers/tty/serial/serial_core.c int uart_resume_port(struct uart_driver *drv, struct uart_port *uport) · line 2476
Quoted 10 lines
drivers/tty/serial/serial_core.c +7
-0+ uport->ops->set_termios (uport, &termios,NULL ); +if (!console_suspend_enabled && uport->ops->start_rx) + uport->ops->start_rx (uport); +if (console_suspend_enabled) +console_start (uport->cons); + } +mutex_unlock (&port->mutex);return 0 ;
To me it looks the whole function is too messed up to fix anything this easily. I'd start with splitting the two large ifs block so that the ordering makes sense:
- set_termios / uart_change_line_settings is called only once
- rx and tx is started only after set_termios
- failure path (the one doing uart_shutdown) logic is reverse + gotoed
Signature · 1 line
i.
Hi,
Is there anything more I should do regarding these changes ?
Best regards,
Lukasz
śr., 5 kwi 2023 o 13:19 Lukasz Majczak <[email protected]> napisał(a):
Quoted 33 lines
This is v2 of a patch[1].
v1->v2:
* Fixed typos in commit message
* Tested[2] patch "serial: core: preserve cflags, ispeed and ospeed" on all current LTS kernels (4.14, 4.19, 5.4, 5.10, 5.15, 6.1) and on top of 6.3-rc5 with positive results - console was working after resume
* During tests another issue was observed on 6.1+ - FIFO not enabled after resume - and an additional change was needed ("serial: core: enable FIFO after resume")
* Test procedure:
1) ensure the console output is ok
2) suspend device with "echo freeze >/sys/power/state"(/sys/module/printk/parameters/console_suspend == N) 3) resume device and check the console output 4) suspend device with "echo freeze > /sys/power/state" (/sys/module/printk/parameters/console_suspend == Y)5) resume device and check the console output
[1] https://lore.kernel.org/lkml/[email protected]/
[2] Test platforms: PC with i5-8400 + GB H370M D3H
HP Elite c1030 Chromebook Enterprise i5-10310ULukasz Majczak (2):
serial: core: preserve cflags, ispeed and ospeed
serial: core: enable FIFO after resumedrivers/tty/serial/serial_core.c | 57 +++++++++++++++----------------- 1 file changed, 27 insertions(+), 30 deletions(-)--
2.40.0.577.gac1e443424-goog
Greg Kroah-Hartman
maintainerQuoted 3 lines · Lukasz Majczak
Hi,
Is there anything more I should do regarding these changes ?
What changes?
The merge window just was finished, so I now have a bunch of patches to review. Hopefully someone else can help review this one in the meantime, which would help out a lot in verifying that this change really is correct or not.
thanks,
greg k-h