[PATCH printk v1 00/10] printk: introduce atomic consoles and sync mode

Hi,

This is the next part of our printk-rework effort (points 3 and
4 of the LPC 2019 summary [0]).

Here the concept of "atomic consoles" is introduced through  a
new (optional) write_atomic() callback for console drivers. This
callback must be implemented as an NMI-safe variant of the
write() callback, meaning that it can function from any context
without relying on questionable tactics such as ignoring locking
and also without relying on the synchronization of console
semaphore.

As an example of how such an atomic console can look like, this
series implements write_atomic() for the 8250 UART driver.

This series also introduces a new console printing mode called
"sync mode" that is only activated when the kernel is about to
end (such as panic, oops, shutdown, reboot). Sync mode can only
be activated if atomic consoles are available. A system without
registered atomic consoles will be unaffected by this series.

When in sync mode, the console printing behavior becomes:

- only consoles implementing write_atomic() will be called

- printing occurs within vprintk_store() instead of
  console_unlock(), since the console semaphore is irrelevant
  for atomic consoles

For systems that have registered atomic consoles, this series
improves the reliability of seeing crash messages by using new
locking techniques rather than "ignoring locks and hoping for
the best". In particular, atomic consoles rely on the
CPU-reentrant spinlock (i.e. the printk cpulock) for
synchronizing console output.

John Ogness

[0] https://lore.kernel.org/lkml/87k1acz5rx.fsf@linutronix.de/

John Ogness (10):
  printk: relocate printk cpulock functions
  printk: rename printk cpulock API and always disable interrupts
  kgdb: delay roundup if holding printk cpulock
  printk: relocate printk_delay()
  printk: call boot_delay_msec() in printk_delay()
  printk: use seqcount_latch for console_seq
  console: add write_atomic interface
  printk: introduce kernel sync mode
  kdb: if available, only use atomic consoles for output mirroring
  serial: 8250: implement write_atomic

 arch/powerpc/include/asm/smp.h         |   1 +
 arch/powerpc/kernel/kgdb.c             |  10 +-
 arch/powerpc/kernel/smp.c              |   5 +
 arch/x86/kernel/kgdb.c                 |   9 +-
 drivers/tty/serial/8250/8250.h         |  47 ++-
 drivers/tty/serial/8250/8250_core.c    |  17 +-
 drivers/tty/serial/8250/8250_fsl.c     |   9 +
 drivers/tty/serial/8250/8250_ingenic.c |   7 +
 drivers/tty/serial/8250/8250_mtk.c     |  29 +-
 drivers/tty/serial/8250/8250_port.c    |  92 ++--
 drivers/tty/serial/8250/Kconfig        |   1 +
 include/linux/console.h                |  32 ++
 include/linux/kgdb.h                   |   3 +
 include/linux/printk.h                 |  57 +--
 include/linux/serial_8250.h            |   5 +
 kernel/debug/debug_core.c              |  45 +-
 kernel/debug/kdb/kdb_io.c              |  16 +
 kernel/printk/printk.c                 | 554 +++++++++++++++++--------
 lib/Kconfig.debug                      |   3 +
 lib/dump_stack.c                       |   4 +-
 lib/nmi_backtrace.c                    |   4 +-
 21 files changed, 684 insertions(+), 266 deletions(-)


base-commit: 23d8adcf8022b9483605531d8985f5b77533cb3a
-- 
2.20.1

Implement an NMI-safe write_atomic() console function in order to
support synchronous console printing.

Since interrupts need to be disabled during transmit, all usage of
the IER register is wrapped with access functions that use the
printk cpulock to synchronize register access while tracking the
state of the interrupts. This is necessary because write_atomic()
can be called from an NMI context that has preempted write_atomic().

Signed-off-by: John Ogness <john.ogness@linutronix.de>
---
 drivers/tty/serial/8250/8250.h         | 47 ++++++++++++-
 drivers/tty/serial/8250/8250_core.c    | 17 +++--
 drivers/tty/serial/8250/8250_fsl.c     |  9 +++
 drivers/tty/serial/8250/8250_ingenic.c |  7 ++
 drivers/tty/serial/8250/8250_mtk.c     | 29 +++++++-
 drivers/tty/serial/8250/8250_port.c    | 92 ++++++++++++++++----------
 drivers/tty/serial/8250/Kconfig        |  1 +
 include/linux/serial_8250.h            |  5 ++
 8 files changed, 163 insertions(+), 44 deletions(-)

diff --git a/drivers/tty/serial/8250/8250.h b/drivers/tty/serial/8250/8250.h
index 6473361525d1..9e8785ab9886 100644
--- a/drivers/tty/serial/8250/8250.h
+++ b/drivers/tty/serial/8250/8250.h

@@ -132,12 +132,55 @@ static inline void serial_dl_write(struct uart_8250_port *up, int value)
 	up->dl_write(up, value);
 }
 
+static inline void serial8250_set_IER(struct uart_8250_port *up,
+				      unsigned char ier)
+{
+	struct uart_port *port = &up->port;
+	unsigned long flags;
+	bool is_console;
+
+	is_console = uart_console(port);
+
+	if (is_console)
+		console_atomic_cpu_lock(flags);
+
+	serial_out(up, UART_IER, ier);
+
+	if (is_console)
+		console_atomic_cpu_unlock(flags);
+}
+
+static inline unsigned char serial8250_clear_IER(struct uart_8250_port *up)
+{
+	struct uart_port *port = &up->port;
+	unsigned int clearval = 0;
+	unsigned long flags;
+	unsigned int prior;
+	bool is_console;
+
+	is_console = uart_console(port);
+
+	if (up->capabilities & UART_CAP_UUE)
+		clearval = UART_IER_UUE;
+
+	if (is_console)
+		console_atomic_cpu_lock(flags);
+
+	prior = serial_port_in(port, UART_IER);
+	serial_port_out(port, UART_IER, clearval);
+
+	if (is_console)
+		console_atomic_cpu_unlock(flags);
+
+	return prior;
+}
+
 static inline bool serial8250_set_THRI(struct uart_8250_port *up)
 {
 	if (up->ier & UART_IER_THRI)
 		return false;
 	up->ier |= UART_IER_THRI;
-	serial_out(up, UART_IER, up->ier);
+	serial8250_set_IER(up, up->ier);
 	return true;
 }
 

@@ -146,7 +189,7 @@ static inline bool serial8250_clear_THRI(struct uart_8250_port *up)
 	if (!(up->ier & UART_IER_THRI))
 		return false;
 	up->ier &= ~UART_IER_THRI;
-	serial_out(up, UART_IER, up->ier);
+	serial8250_set_IER(up, up->ier);
 	return true;
 }
 

diff --git a/drivers/tty/serial/8250/8250_core.c b/drivers/tty/serial/8250/8250_core.c
index 1ce193daea7f..fad00c0414e3 100644
--- a/drivers/tty/serial/8250/8250_core.c
+++ b/drivers/tty/serial/8250/8250_core.c

@@ -264,10 +264,8 @@ static void serial8250_backup_timeout(struct timer_list *t)
 	 * Must disable interrupts or else we risk racing with the interrupt
 	 * based handler.
 	 */
-	if (up->port.irq) {
-		ier = serial_in(up, UART_IER);
-		serial_out(up, UART_IER, 0);
-	}
+	if (up->port.irq)
+		ier = serial8250_clear_IER(up);
 
 	iir = serial_in(up, UART_IIR);
 

@@ -290,7 +288,7 @@ static void serial8250_backup_timeout(struct timer_list *t)
 		serial8250_tx_chars(up);
 
 	if (up->port.irq)
-		serial_out(up, UART_IER, ier);
+		serial8250_set_IER(up, ier);
 
 	spin_unlock_irqrestore(&up->port.lock, flags);
 

@@ -568,6 +566,14 @@ serial8250_register_ports(struct uart_driver *drv, struct device *dev)
 
 #ifdef CONFIG_SERIAL_8250_CONSOLE
 
+static void univ8250_console_write_atomic(struct console *co, const char *s,
+					  unsigned int count)
+{
+	struct uart_8250_port *up = &serial8250_ports[co->index];
+
+	serial8250_console_write_atomic(up, s, count);
+}
+
 static void univ8250_console_write(struct console *co, const char *s,
 				   unsigned int count)
 {

@@ -661,6 +667,7 @@ static int univ8250_console_match(struct console *co, char *name, int idx,
 
 static struct console univ8250_console = {
 	.name		= "ttyS",
+	.write_atomic	= univ8250_console_write_atomic,
 	.write		= univ8250_console_write,
 	.device		= uart_console_device,
 	.setup		= univ8250_console_setup,

diff --git a/drivers/tty/serial/8250/8250_fsl.c b/drivers/tty/serial/8250/8250_fsl.c
index 4e75d2e4f87c..00e0f8f6607b 100644
--- a/drivers/tty/serial/8250/8250_fsl.c
+++ b/drivers/tty/serial/8250/8250_fsl.c

@@ -59,9 +59,18 @@ int fsl8250_handle_irq(struct uart_port *port)
 
 	/* Stop processing interrupts on input overrun */
 	if ((orig_lsr & UART_LSR_OE) && (up->overrun_backoff_time_ms > 0)) {
+		unsigned long flags;
 		unsigned long delay;
+		bool is_console;
 
+		is_console = uart_console(port);
+
+		if (is_console)
+			console_atomic_cpu_lock(flags);
 		up->ier = port->serial_in(port, UART_IER);
+		if (is_console)
+			console_atomic_cpu_unlock(flags);
+
 		if (up->ier & (UART_IER_RLSI | UART_IER_RDI)) {
 			port->ops->stop_rx(port);
 		} else {

diff --git a/drivers/tty/serial/8250/8250_ingenic.c b/drivers/tty/serial/8250/8250_ingenic.c
index 988bf6bcce42..ff0e1d84ec7e 100644
--- a/drivers/tty/serial/8250/8250_ingenic.c
+++ b/drivers/tty/serial/8250/8250_ingenic.c

@@ -146,6 +146,8 @@ OF_EARLYCON_DECLARE(x1000_uart, "ingenic,x1000-uart",
 
 static void ingenic_uart_serial_out(struct uart_port *p, int offset, int value)
 {
+	unsigned long flags;
+	bool is_console;
 	int ier;
 
 	switch (offset) {

@@ -167,7 +169,12 @@ static void ingenic_uart_serial_out(struct uart_port *p, int offset, int value)
 		 * If we have enabled modem status IRQs we should enable
 		 * modem mode.
 		 */
+		is_console = uart_console(p);
+		if (is_console)
+			console_atomic_cpu_lock(flags);
 		ier = p->serial_in(p, UART_IER);
+		if (is_console)
+			console_atomic_cpu_unlock(flags);
 
 		if (ier & UART_IER_MSI)
 			value |= UART_MCR_MDCE | UART_MCR_FCM;

diff --git a/drivers/tty/serial/8250/8250_mtk.c b/drivers/tty/serial/8250/8250_mtk.c
index f7d3023f860f..f26140ebd67a 100644
--- a/drivers/tty/serial/8250/8250_mtk.c
+++ b/drivers/tty/serial/8250/8250_mtk.c

@@ -213,12 +213,37 @@ static void mtk8250_shutdown(struct uart_port *port)
 
 static void mtk8250_disable_intrs(struct uart_8250_port *up, int mask)
 {
-	serial_out(up, UART_IER, serial_in(up, UART_IER) & (~mask));
+	struct uart_port *port = &up->port;
+	unsigned long flags;
+	unsigned int ier;
+	bool is_console;
+
+	is_console = uart_console(port);
+
+	if (is_console)
+		console_atomic_cpu_lock(flags);
+
+	ier = serial_in(up, UART_IER);
+	serial_out(up, UART_IER, ier & (~mask));
+
+	if (is_console)
+		console_atomic_cpu_unlock(flags);
 }
 
 static void mtk8250_enable_intrs(struct uart_8250_port *up, int mask)
 {
-	serial_out(up, UART_IER, serial_in(up, UART_IER) | mask);
+	struct uart_port *port = &up->port;
+	unsigned long flags;
+	unsigned int ier;
+
+	if (uart_console(port))
+		console_atomic_cpu_lock(flags);
+
+	ier = serial_in(up, UART_IER);
+	serial_out(up, UART_IER, ier | mask);
+
+	if (uart_console(port))
+		console_atomic_cpu_unlock(flags);
 }
 
 static void mtk8250_set_flow_ctrl(struct uart_8250_port *up, int mode)

diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
index 2164290cbd31..2c2a5abc1baf 100644
--- a/drivers/tty/serial/8250/8250_port.c
+++ b/drivers/tty/serial/8250/8250_port.c

@@ -757,7 +757,7 @@ static void serial8250_set_sleep(struct uart_8250_port *p, int sleep)
 			serial_out(p, UART_EFR, UART_EFR_ECB);
 			serial_out(p, UART_LCR, 0);
 		}
-		serial_out(p, UART_IER, sleep ? UART_IERX_SLEEP : 0);
+		serial8250_set_IER(p, sleep ? UART_IERX_SLEEP : 0);
 		if (p->capabilities & UART_CAP_EFR) {
 			serial_out(p, UART_LCR, UART_LCR_CONF_MODE_B);
 			serial_out(p, UART_EFR, efr);

@@ -1429,7 +1429,7 @@ static void serial8250_stop_rx(struct uart_port *port)
 
 	up->ier &= ~(UART_IER_RLSI | UART_IER_RDI);
 	up->port.read_status_mask &= ~UART_LSR_DR;
-	serial_port_out(port, UART_IER, up->ier);
+	serial8250_set_IER(up, up->ier);
 
 	serial8250_rpm_put(up);
 }

@@ -1459,7 +1459,7 @@ void serial8250_em485_stop_tx(struct uart_8250_port *p)
 		serial8250_clear_and_reinit_fifos(p);
 
 		p->ier |= UART_IER_RLSI | UART_IER_RDI;
-		serial_port_out(&p->port, UART_IER, p->ier);
+		serial8250_set_IER(p, p->ier);
 	}
 }
 EXPORT_SYMBOL_GPL(serial8250_em485_stop_tx);

@@ -1681,7 +1681,7 @@ static void serial8250_disable_ms(struct uart_port *port)
 	mctrl_gpio_disable_ms(up->gpios);
 
 	up->ier &= ~UART_IER_MSI;
-	serial_port_out(port, UART_IER, up->ier);
+	serial8250_set_IER(up, up->ier);
 }
 
 static void serial8250_enable_ms(struct uart_port *port)

@@ -1697,7 +1697,7 @@ static void serial8250_enable_ms(struct uart_port *port)
 	up->ier |= UART_IER_MSI;
 
 	serial8250_rpm_get(up);
-	serial_port_out(port, UART_IER, up->ier);
+	serial8250_set_IER(up, up->ier);
 	serial8250_rpm_put(up);
 }
 

@@ -2124,14 +2124,7 @@ static void serial8250_put_poll_char(struct uart_port *port,
 	struct uart_8250_port *up = up_to_u8250p(port);
 
 	serial8250_rpm_get(up);
-	/*
-	 *	First save the IER then disable the interrupts
-	 */
-	ier = serial_port_in(port, UART_IER);
-	if (up->capabilities & UART_CAP_UUE)
-		serial_port_out(port, UART_IER, UART_IER_UUE);
-	else
-		serial_port_out(port, UART_IER, 0);
+	ier = serial8250_clear_IER(up);
 
 	wait_for_xmitr(up, BOTH_EMPTY);
 	/*

@@ -2144,7 +2137,7 @@ static void serial8250_put_poll_char(struct uart_port *port,
 	 *	and restore the IER
 	 */
 	wait_for_xmitr(up, BOTH_EMPTY);
-	serial_port_out(port, UART_IER, ier);
+	serial8250_set_IER(up, ier);
 	serial8250_rpm_put(up);
 }
 

@@ -2447,7 +2440,7 @@ void serial8250_do_shutdown(struct uart_port *port)
 	 */
 	spin_lock_irqsave(&port->lock, flags);
 	up->ier = 0;
-	serial_port_out(port, UART_IER, 0);
+	serial8250_set_IER(up, 0);
 	spin_unlock_irqrestore(&port->lock, flags);
 
 	synchronize_irq(port->irq);

@@ -2816,7 +2809,7 @@ serial8250_do_set_termios(struct uart_port *port, struct ktermios *termios,
 	if (up->capabilities & UART_CAP_RTOIE)
 		up->ier |= UART_IER_RTOIE;
 
-	serial_port_out(port, UART_IER, up->ier);
+	serial8250_set_IER(up, up->ier);
 
 	if (up->capabilities & UART_CAP_EFR) {
 		unsigned char efr = 0;

@@ -3282,7 +3275,7 @@ EXPORT_SYMBOL_GPL(serial8250_set_defaults);
 
 #ifdef CONFIG_SERIAL_8250_CONSOLE
 
-static void serial8250_console_putchar(struct uart_port *port, int ch)
+static void serial8250_console_putchar_locked(struct uart_port *port, int ch)
 {
 	struct uart_8250_port *up = up_to_u8250p(port);
 

@@ -3290,6 +3283,18 @@ static void serial8250_console_putchar(struct uart_port *port, int ch)
 	serial_port_out(port, UART_TX, ch);
 }
 
+static void serial8250_console_putchar(struct uart_port *port, int ch)
+{
+	struct uart_8250_port *up = up_to_u8250p(port);
+	unsigned long flags;
+
+	wait_for_xmitr(up, UART_LSR_THRE);
+
+	console_atomic_cpu_lock(flags);
+	serial8250_console_putchar_locked(port, ch);
+	console_atomic_cpu_unlock(flags);
+}
+
 /*
  *	Restore serial console when h/w power-off detected
  */

@@ -3311,6 +3316,32 @@ static void serial8250_console_restore(struct uart_8250_port *up)
 	serial8250_out_MCR(up, UART_MCR_DTR | UART_MCR_RTS);
 }
 
+void serial8250_console_write_atomic(struct uart_8250_port *up,
+				     const char *s, unsigned int count)
+{
+	struct uart_port *port = &up->port;
+	unsigned long flags;
+	unsigned int ier;
+
+	console_atomic_cpu_lock(flags);
+
+	touch_nmi_watchdog();
+
+	ier = serial8250_clear_IER(up);
+
+	if (atomic_fetch_inc(&up->console_printing)) {
+		uart_console_write(port, "\n", 1,
+				   serial8250_console_putchar_locked);
+	}
+	uart_console_write(port, s, count, serial8250_console_putchar_locked);
+	atomic_dec(&up->console_printing);
+
+	wait_for_xmitr(up, BOTH_EMPTY);
+	serial8250_set_IER(up, ier);
+
+	console_atomic_cpu_unlock(flags);
+}
+
 /*
  *	Print a string to the serial port trying not to disturb
  *	any possible real use of the port...

@@ -3327,24 +3358,12 @@ void serial8250_console_write(struct uart_8250_port *up, const char *s,
 	struct uart_port *port = &up->port;
 	unsigned long flags;
 	unsigned int ier;
-	int locked = 1;
 
 	touch_nmi_watchdog();
 
-	if (oops_in_progress)
-		locked = spin_trylock_irqsave(&port->lock, flags);
-	else
-		spin_lock_irqsave(&port->lock, flags);
-
-	/*
-	 *	First save the IER then disable the interrupts
-	 */
-	ier = serial_port_in(port, UART_IER);
+	spin_lock_irqsave(&port->lock, flags);
 
-	if (up->capabilities & UART_CAP_UUE)
-		serial_port_out(port, UART_IER, UART_IER_UUE);
-	else
-		serial_port_out(port, UART_IER, 0);
+	ier = serial8250_clear_IER(up);
 
 	/* check scratch reg to see if port powered off during system sleep */
 	if (up->canary && (up->canary != serial_port_in(port, UART_SCR))) {

@@ -3358,7 +3377,9 @@ void serial8250_console_write(struct uart_8250_port *up, const char *s,
 		mdelay(port->rs485.delay_rts_before_send);
 	}
 
+	atomic_inc(&up->console_printing);
 	uart_console_write(port, s, count, serial8250_console_putchar);
+	atomic_dec(&up->console_printing);
 
 	/*
 	 *	Finally, wait for transmitter to become empty

@@ -3371,8 +3392,7 @@ void serial8250_console_write(struct uart_8250_port *up, const char *s,
 		if (em485->tx_stopped)
 			up->rs485_stop_tx(up);
 	}
-
-	serial_port_out(port, UART_IER, ier);
+	serial8250_set_IER(up, ier);
 
 	/*
 	 *	The receive handling will happen properly because the

@@ -3384,8 +3404,7 @@ void serial8250_console_write(struct uart_8250_port *up, const char *s,
 	if (up->msr_saved_flags)
 		serial8250_modem_status(up);
 
-	if (locked)
-		spin_unlock_irqrestore(&port->lock, flags);
+	spin_unlock_irqrestore(&port->lock, flags);
 }
 
 static unsigned int probe_baud(struct uart_port *port)

@@ -3405,6 +3424,7 @@ static unsigned int probe_baud(struct uart_port *port)
 
 int serial8250_console_setup(struct uart_port *port, char *options, bool probe)
 {
+	struct uart_8250_port *up = up_to_u8250p(port);
 	int baud = 9600;
 	int bits = 8;
 	int parity = 'n';

@@ -3414,6 +3434,8 @@ int serial8250_console_setup(struct uart_port *port, char *options, bool probe)
 	if (!port->iobase && !port->membase)
 		return -ENODEV;
 
+	atomic_set(&up->console_printing, 0);
+
 	if (options)
 		uart_parse_options(options, &baud, &parity, &bits, &flow);
 	else if (probe)

diff --git a/drivers/tty/serial/8250/Kconfig b/drivers/tty/serial/8250/Kconfig
index d1b3c2373fa4..1671d2882ba7 100644
--- a/drivers/tty/serial/8250/Kconfig
+++ b/drivers/tty/serial/8250/Kconfig

@@ -9,6 +9,7 @@ config SERIAL_8250
 	depends on !S390
 	select SERIAL_CORE
 	select SERIAL_MCTRL_GPIO if GPIOLIB
+	select HAVE_ATOMIC_CONSOLE
 	help
 	  This selects whether you want to include the driver for the standard
 	  serial ports.  The standard answer is Y.  People who might say N

diff --git a/include/linux/serial_8250.h b/include/linux/serial_8250.h
index 5db211f43b29..aa011f668705 100644
--- a/include/linux/serial_8250.h
+++ b/include/linux/serial_8250.h

@@ -7,6 +7,7 @@
 #ifndef _LINUX_SERIAL_8250_H
 #define _LINUX_SERIAL_8250_H
 
+#include <linux/atomic.h>
 #include <linux/serial_core.h>
 #include <linux/serial_reg.h>
 #include <linux/platform_device.h>

@@ -125,6 +126,8 @@ struct uart_8250_port {
 #define MSR_SAVE_FLAGS UART_MSR_ANY_DELTA
 	unsigned char		msr_saved_flags;
 
+	atomic_t		console_printing;
+
 	struct uart_8250_dma	*dma;
 	const struct uart_8250_ops *ops;
 

@@ -180,6 +183,8 @@ void serial8250_init_port(struct uart_8250_port *up);
 void serial8250_set_defaults(struct uart_8250_port *up);
 void serial8250_console_write(struct uart_8250_port *up, const char *s,
 			      unsigned int count);
+void serial8250_console_write_atomic(struct uart_8250_port *up, const char *s,
+				     unsigned int count);
 int serial8250_console_setup(struct uart_port *port, char *options, bool probe);
 int serial8250_console_exit(struct uart_port *port);
 

-- 
2.20.1

On Tue, Aug 03, 2021 at 03:18:51PM +0206, John Ogness wrote:

Hi,

This is the next part of our printk-rework effort (points 3 and
4 of the LPC 2019 summary [0]).

Here the concept of "atomic consoles" is introduced through  a
new (optional) write_atomic() callback for console drivers. This
callback must be implemented as an NMI-safe variant of the
write() callback, meaning that it can function from any context
without relying on questionable tactics such as ignoring locking
and also without relying on the synchronization of console
semaphore.

As an example of how such an atomic console can look like, this
series implements write_atomic() for the 8250 UART driver.

This series also introduces a new console printing mode called
"sync mode" that is only activated when the kernel is about to
end (such as panic, oops, shutdown, reboot). Sync mode can only
be activated if atomic consoles are available. A system without
registered atomic consoles will be unaffected by this series.

When in sync mode, the console printing behavior becomes:

- only consoles implementing write_atomic() will be called

- printing occurs within vprintk_store() instead of
  console_unlock(), since the console semaphore is irrelevant
  for atomic consoles

For systems that have registered atomic consoles, this series
improves the reliability of seeing crash messages by using new
locking techniques rather than "ignoring locks and hoping for
the best". In particular, atomic consoles rely on the
CPU-reentrant spinlock (i.e. the printk cpulock) for
synchronizing console output.

If console is runtime suspended, who will bring it up?
Does it mean that this callback can't be implemented on the consoles that
do runtime suspend (some of 8250 currently, for example)?

-- 
With Best Regards,
Andy Shevchenko

On Tue, Aug 03, 2021 at 03:19:01PM +0206, John Ogness wrote:

Implement an NMI-safe write_atomic() console function in order to
support synchronous console printing.

Since interrupts need to be disabled during transmit, all usage of
the IER register is wrapped with access functions that use the
printk cpulock to synchronize register access while tracking the
state of the interrupts. This is necessary because write_atomic()
can be called from an NMI context that has preempted write_atomic().

...

+static inline void serial8250_set_IER(struct uart_8250_port *up,
+				      unsigned char ier)
+{
+	struct uart_port *port = &up->port;
+	unsigned long flags;
+	bool is_console;

+	is_console = uart_console(port);
+
+	if (is_console)
+		console_atomic_cpu_lock(flags);
+
+	serial_out(up, UART_IER, ier);
+
+	if (is_console)
+		console_atomic_cpu_unlock(flags);

I would rewrite it as

	if (uart_console()) {
		console_atomic_cpu_lock(flags);
		serial_out(up, UART_IER, ier);
		console_atomic_cpu_unlock(flags);
	} else {
		serial_out(up, UART_IER, ier);
	}

No additional variable, easier to get the algorithm on the first glance, less
error prone.

+}

+static inline unsigned char serial8250_clear_IER(struct uart_8250_port *up)
+{
+	struct uart_port *port = &up->port;
+	unsigned int clearval = 0;
+	unsigned long flags;
+	unsigned int prior;
+	bool is_console;
+
+	is_console = uart_console(port);
+
+	if (up->capabilities & UART_CAP_UUE)
+		clearval = UART_IER_UUE;
+
+	if (is_console)
+		console_atomic_cpu_lock(flags);
+
+	prior = serial_port_in(port, UART_IER);
+	serial_port_out(port, UART_IER, clearval);
+
+	if (is_console)
+		console_atomic_cpu_unlock(flags);

Ditto.

+	return prior;
+}

...

+		is_console = uart_console(port);
+
+		if (is_console)
+			console_atomic_cpu_lock(flags);
 		up->ier = port->serial_in(port, UART_IER);
+		if (is_console)
+			console_atomic_cpu_unlock(flags);
+

I'm wondering why you can't call above function here?

...

+		is_console = uart_console(p);
+		if (is_console)
+			console_atomic_cpu_lock(flags);
 		ier = p->serial_in(p, UART_IER);
+		if (is_console)
+			console_atomic_cpu_unlock(flags);

Ditto.

...

+	is_console = uart_console(port);
+
+	if (is_console)
+		console_atomic_cpu_lock(flags);
+
+	ier = serial_in(up, UART_IER);
+	serial_out(up, UART_IER, ier & (~mask));
+
+	if (is_console)
+		console_atomic_cpu_unlock(flags);

Ditto.

...

+	if (uart_console(port))
+		console_atomic_cpu_lock(flags);
+
+	ier = serial_in(up, UART_IER);
+	serial_out(up, UART_IER, ier | mask);
+
+	if (uart_console(port))
+		console_atomic_cpu_unlock(flags);

Ditto.

Looking into above note, that uart_console(port) can give different results
here, AFAIR.

-- 
With Best Regards,
Andy Shevchenko

On 03. 08. 21, 16:07, Andy Shevchenko wrote:

On Tue, Aug 03, 2021 at 03:19:01PM +0206, John Ogness wrote:

quoted

Implement an NMI-safe write_atomic() console function in order to
support synchronous console printing.

Since interrupts need to be disabled during transmit, all usage of
the IER register is wrapped with access functions that use the
printk cpulock to synchronize register access while tracking the
state of the interrupts. This is necessary because write_atomic()
can be called from an NMI context that has preempted write_atomic().

...

quoted

+static inline void serial8250_set_IER(struct uart_8250_port *up,
+				      unsigned char ier)
+{
+	struct uart_port *port = &up->port;
+	unsigned long flags;
+	bool is_console;

quoted

+	is_console = uart_console(port);
+
+	if (is_console)
+		console_atomic_cpu_lock(flags);
+
+	serial_out(up, UART_IER, ier);
+
+	if (is_console)
+		console_atomic_cpu_unlock(flags);

I would rewrite it as

	if (uart_console()) {
		console_atomic_cpu_lock(flags);
		serial_out(up, UART_IER, ier);
		console_atomic_cpu_unlock(flags);
	} else {
		serial_out(up, UART_IER, ier);
	}

No additional variable, easier to get the algorithm on the first glance, less
error prone.

Yes, the original is terrible.

Another option:

bool locked = console_atomic_cpu_lock(flags, uart_console());
serial_out(up, UART_IER, ier);
console_atomic_cpu_unlock(flags, locked);


Which makes console_atomic_cpu_lock to lock only if second parameter is 
true and return its value too.

BTW I actually don't know what console_atomic_cpu_lock does to think 
about it more as I was not CCed, and neither lore sees the other patches:
https://lore.kernel.org/linux-mips/20210803131301.5588-1-john.ogness@linutronix.de/

thanks,
-- 
js
suse labs

On 2021-08-05, Jiri Slaby [off-list ref] wrote:

On 03. 08. 21, 16:07, Andy Shevchenko wrote:

quoted

On Tue, Aug 03, 2021 at 03:19:01PM +0206, John Ogness wrote:

quoted

Implement an NMI-safe write_atomic() console function in order to
support synchronous console printing.

Since interrupts need to be disabled during transmit, all usage of
the IER register is wrapped with access functions that use the
printk cpulock to synchronize register access while tracking the
state of the interrupts. This is necessary because write_atomic()
can be called from an NMI context that has preempted write_atomic().

...

quoted

+static inline void serial8250_set_IER(struct uart_8250_port *up,
+				      unsigned char ier)
+{
+	struct uart_port *port = &up->port;
+	unsigned long flags;
+	bool is_console;

quoted

+	is_console = uart_console(port);
+
+	if (is_console)
+		console_atomic_cpu_lock(flags);
+
+	serial_out(up, UART_IER, ier);
+
+	if (is_console)
+		console_atomic_cpu_unlock(flags);

I would rewrite it as

	if (uart_console()) {
		console_atomic_cpu_lock(flags);
		serial_out(up, UART_IER, ier);
		console_atomic_cpu_unlock(flags);
	} else {
		serial_out(up, UART_IER, ier);
	}

Some locations have more than just 1 line of code in between
lock/unlock. I agree this looks better, but am unsure how much
copy/paste code is acceptable.

quoted

No additional variable, easier to get the algorithm on the first
glance, less error prone.

Yes, the original is terrible.

Another option:

bool locked = console_atomic_cpu_lock(flags, uart_console());
serial_out(up, UART_IER, ier);
console_atomic_cpu_unlock(flags, locked);

Which makes console_atomic_cpu_lock to lock only if second parameter
is true and return its value too.

I am not sure how common such semantics for lock/unlock functions
are. But since this pattern, using uart_console(), will most likely be a
common pattern for atomic consoles, I can see how this will be useful.

I will choose one of these 2 suggestions for v2. Thanks.

BTW I actually don't know what console_atomic_cpu_lock does to think 
about it more as I was not CCed, and neither lore sees the other patches:
https://lore.kernel.org/linux-mips/20210803131301.5588-1-john.ogness@linutronix.de/

Only the lkml mailing list saw the full series:

https://lore.kernel.org/lkml/20210803131301.5588-1-john.ogness@linutronix.de/

John Ogness

On Tue 2021-08-03 15:18:51, John Ogness wrote:

Hi,

This is the next part of our printk-rework effort (points 3 and
4 of the LPC 2019 summary [0]).

Here the concept of "atomic consoles" is introduced through  a
new (optional) write_atomic() callback for console drivers. This
callback must be implemented as an NMI-safe variant of the
write() callback, meaning that it can function from any context
without relying on questionable tactics such as ignoring locking
and also without relying on the synchronization of console
semaphore.

As an example of how such an atomic console can look like, this
series implements write_atomic() for the 8250 UART driver.

This series also introduces a new console printing mode called
"sync mode" that is only activated when the kernel is about to
end (such as panic, oops, shutdown, reboot). Sync mode can only
be activated if atomic consoles are available. A system without
registered atomic consoles will be unaffected by this series.

When in sync mode, the console printing behavior becomes:

- only consoles implementing write_atomic() will be called

- printing occurs within vprintk_store() instead of
  console_unlock(), since the console semaphore is irrelevant
  for atomic consoles

I am fine with the new behavior at this stage. It is a quite clear
win when (only) the atomic console is used. And it does not make any
difference when atomic consoles are disabled.

But I am not sure about the proposed terms and implementation.
I want to be sure that we are on the right way for introducing
console kthreads.

Let me try to compare the behavior:

1. before this patchset():

	/* printk: store immediately; try all consoles immediately */
	int printk(...)
	{
		vprintk_store();
		if (console_try_lock()) {
			/* flush pending messages to the consoles */
			console_unlock();
		}
	}

	/* panic: try hard to flush messages to the consoles and avoid deadlock */
	void panic()
	{
		/* Ignore locks in console drivers */
		bust_spinlocks(1);

		printk("Kernel panic ...);
		dump_stack();

		smp_send_stop();
		/* ignore console lock */
		console_flush_on_panic();
	}


2. after this patchset():

   + same as before in normal mode or when there is no atomic console

   + in panic with atomic console; it modifies the behavior:

	/*
	 * printk: store immediately; immediately flush atomic consoles;
	 *         unsafe consoles are not used anymore;
	 */
	int printk(...)
	{
		vprintk_store();
		flush_atomic_consoles();
	}

	/* panic: no hacks; only atomic consoles are used */
	void panic()
	{
		printk("Kernel panic ...);
		dump_stack();
	}


3. After introducing console kthread(s):

	int printk(...)
	{
		vprintk_store();
		wake_consoles_via_irqwork();
	}

	+ in panic:

	    + with atomic console like after this patchset?
	    + without atomic consoles?

	+ during early boot?


I guess that we will need another sync mode for the early boot,
panic, suspend, kexec, etc.. It must be posible to debug these states
even wihtout atomic console and working kthreads.

Best Regards,
Petr

On (21/08/05 17:47), Petr Mladek wrote:
[..]

3. After introducing console kthread(s):

	int printk(...)
	{
		vprintk_store();
		wake_consoles_via_irqwork();
	}

	+ in panic:

	    + with atomic console like after this patchset?
	    + without atomic consoles?

	+ during early boot?

I guess I'd also add netconsole to the list.