| Message ID | 20250414031450.42237-1-cuiyunhui@bytedance.com |
|---|---|
| State | New |
| Headers | show |
| Series | [v3] serial: 8250: fix panic due to PSLVERR | expand |
On Mon, 14 Apr 2025, Ilpo Järvinen wrote: > On Mon, 14 Apr 2025, Yunhui Cui wrote: > > > When the PSLVERR_RESP_EN parameter is set to 1, the device generates > > an error response if an attempt is made to read an empty RBR (Receive > > Buffer Register) while the FIFO is enabled. > > > > In serial8250_do_startup(), calling serial_port_out(port, UART_LCR, > > UART_LCR_WLEN8) triggers dw8250_check_lcr(), which invokes > > dw8250_force_idle() and serial8250_clear_and_reinit_fifos(). The latter > > function enables the FIFO via serial_out(p, UART_FCR, p->fcr). > > Execution proceeds to the dont_test_tx_en label: > > ... > > serial_port_in(port, UART_RX); > > This satisfies the PSLVERR trigger condition. > > > > Because another CPU(e.g., using printk()) is accessing the UART (UART > > is busy), the current CPU fails the check (value & ~UART_LCR_SPAR) == > > (lcr & ~UART_LCR_SPAR), causing it to enter dw8250_force_idle(). > > > > To fix this, all calls to serial_out(UART_LCR) and serial_in(UART_RX) > > should be executed under port->lock. Additionally, checking the readiness > > via UART_LSR should also be done under port->lock. > > > > Panic backtrace: > > [ 0.442336] Oops - unknown exception [#1] > > [ 0.442343] epc : dw8250_serial_in32+0x1e/0x4a > > [ 0.442351] ra : serial8250_do_startup+0x2c8/0x88e > > ... > > [ 0.442416] console_on_rootfs+0x26/0x70 > > > > Fixes: c49436b657d0 ("serial: 8250_dw: Improve unwritable LCR workaround") > > Link: https://lore.kernel.org/all/84cydt5peu.fsf@jogness.linutronix.de/T/ > > Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com> > > As Andy mentioned, this change looks it would benefit from splitting to > multiple parts. > > However, this change brings back some memories from a few years back. > Back then, there was a reporter who had issues issues related to > dw8250_force_idle() or writing some of the registers (IIRC). I ended up > looking into finding a better solution to the write-while-BUSY problem > which entirely replaced dw8250_force_idle() that is quite hacky and seems > unreliable on fundamendal level. > > Sadly, once I had posted a patch for testing, the reporter went dead > silent so the patch was left rotting as I had no time to try to reproduce. > > Perhaps the patch I created back then would be useful for addressing this > problem you're facing (the patch is attached). I've rebased the patch on > top of the tty-next now (but I did no testing beyond compiling). There are > a few further thoughts / missing bits mentioned in the comments within the > patch itself (I did not try to updated them now, so the comments may have > rotten too). After some further thinking. I realized you're interested in the opposite case (empty rx), whereas my patch focused on addressing (rx getting refilled constantly) so it doesn't seem that useful for your case. The patch shows though the direction I'd like to head with idle forcing approach. -- i. > > --- > > drivers/tty/serial/8250/8250_dw.c | 8 +++++ > > drivers/tty/serial/8250/8250_port.c | 46 ++++++++++++++++++----------- > > 2 files changed, 36 insertions(+), 18 deletions(-) > > > > diff --git a/drivers/tty/serial/8250/8250_dw.c b/drivers/tty/serial/8250/8250_dw.c > > index af24ec25d976..e97200ff30e3 100644 > > --- a/drivers/tty/serial/8250/8250_dw.c > > +++ b/drivers/tty/serial/8250/8250_dw.c > > @@ -13,6 +13,7 @@ > > #include <linux/delay.h> > > #include <linux/device.h> > > #include <linux/io.h> > > +#include <linux/lockdep.h> > > #include <linux/mod_devicetable.h> > > #include <linux/module.h> > > #include <linux/notifier.h> > > @@ -112,6 +113,13 @@ static void dw8250_force_idle(struct uart_port *p) > > struct uart_8250_port *up = up_to_u8250p(p); > > unsigned int lsr; > > > > + /* > > + * Serial_in(p, UART_RX) should be under port->lock, but we can't add > > + * it to avoid AA deadlock as we're unsure if serial_out*(...UART_LCR) > > + * is under port->lock. > > + */ > > + lockdep_assert_held_once(&p->lock); > > + > > serial8250_clear_and_reinit_fifos(up); > > > > /* > > diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c > > index 3f256e96c722..21bbd18195f5 100644 > > --- a/drivers/tty/serial/8250/8250_port.c > > +++ b/drivers/tty/serial/8250/8250_port.c > > @@ -1328,6 +1328,7 @@ static void autoconfig_irq(struct uart_8250_port *up) > > unsigned int ICP = 0; > > unsigned long irqs; > > int irq; > > + u16 lsr; > > > > if (port->flags & UPF_FOURPORT) { > > ICP = (port->iobase & 0xfe0) | 0x1f; > > @@ -1357,9 +1358,10 @@ static void autoconfig_irq(struct uart_8250_port *up) > > /* Synchronize UART_IER access against the console. */ > > uart_port_lock_irq(port); > > serial_out(up, UART_IER, UART_IER_ALL_INTR); > > + lsr = serial_in(up, UART_LSR); > > + if (lsr & UART_LSR_DR) > > + serial_port_in(port, UART_RX); > > uart_port_unlock_irq(port); > > - serial_in(up, UART_LSR); > > - serial_in(up, UART_RX); > > serial_in(up, UART_IIR); > > serial_in(up, UART_MSR); > > serial_out(up, UART_TX, 0xFF); > > @@ -2137,19 +2139,16 @@ static void wait_for_xmitr(struct uart_8250_port *up, int bits) > > static int serial8250_get_poll_char(struct uart_port *port) > > { > > struct uart_8250_port *up = up_to_u8250p(port); > > - int status; > > + int status = NO_POLL_CHAR; > > u16 lsr; > > > > serial8250_rpm_get(up); > > > > + uart_port_lock_irqsave(port, &flags); > > lsr = serial_port_in(port, UART_LSR); > > - > > - if (!(lsr & UART_LSR_DR)) { > > - status = NO_POLL_CHAR; > > - goto out; > > - } > > - > > - status = serial_port_in(port, UART_RX); > > + if ((lsr & UART_LSR_DR)) > > + status = serial_port_in(port, UART_RX); > > + uart_port_unlock_irqrestore(port, flags); > > out: > > serial8250_rpm_put(up); > > return status; > > @@ -2264,13 +2263,16 @@ int serial8250_do_startup(struct uart_port *port) > > * Clear the FIFO buffers and disable them. > > * (they will be reenabled in set_termios()) > > */ > > + uart_port_lock_irqsave(port, &flags); > > serial8250_clear_fifos(up); > > > > /* > > * Clear the interrupt registers. > > */ > > - serial_port_in(port, UART_LSR); > > - serial_port_in(port, UART_RX); > > + lsr = serial_port_in(port, UART_LSR); > > + if (lsr & UART_LSR_DR) > > + serial_port_in(port, UART_RX); > > + uart_port_unlock_irqrestore(port, flags); > > serial_port_in(port, UART_IIR); > > serial_port_in(port, UART_MSR); > > > > @@ -2380,9 +2382,10 @@ int serial8250_do_startup(struct uart_port *port) > > /* > > * Now, initialize the UART > > */ > > - serial_port_out(port, UART_LCR, UART_LCR_WLEN8); > > > > uart_port_lock_irqsave(port, &flags); > > + serial_port_out(port, UART_LCR, UART_LCR_WLEN8); > > + > > if (up->port.flags & UPF_FOURPORT) { > > if (!up->port.irq) > > up->port.mctrl |= TIOCM_OUT1; > > @@ -2428,15 +2431,16 @@ int serial8250_do_startup(struct uart_port *port) > > } > > > > dont_test_tx_en: > > I don't see this in the tty-next branch? > > ~/linux/tty-next$ git grep dont_test_tx_en | cat - > ~/linux/tty-next$ > >
diff --git a/drivers/tty/serial/8250/8250_dw.c b/drivers/tty/serial/8250/8250_dw.c index af24ec25d976..e97200ff30e3 100644 --- a/drivers/tty/serial/8250/8250_dw.c +++ b/drivers/tty/serial/8250/8250_dw.c @@ -13,6 +13,7 @@ #include <linux/delay.h> #include <linux/device.h> #include <linux/io.h> +#include <linux/lockdep.h> #include <linux/mod_devicetable.h> #include <linux/module.h> #include <linux/notifier.h> @@ -112,6 +113,13 @@ static void dw8250_force_idle(struct uart_port *p) struct uart_8250_port *up = up_to_u8250p(p); unsigned int lsr; + /* + * Serial_in(p, UART_RX) should be under port->lock, but we can't add + * it to avoid AA deadlock as we're unsure if serial_out*(...UART_LCR) + * is under port->lock. + */ + lockdep_assert_held_once(&p->lock); + serial8250_clear_and_reinit_fifos(up); /* diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c index 3f256e96c722..21bbd18195f5 100644 --- a/drivers/tty/serial/8250/8250_port.c +++ b/drivers/tty/serial/8250/8250_port.c @@ -1328,6 +1328,7 @@ static void autoconfig_irq(struct uart_8250_port *up) unsigned int ICP = 0; unsigned long irqs; int irq; + u16 lsr; if (port->flags & UPF_FOURPORT) { ICP = (port->iobase & 0xfe0) | 0x1f; @@ -1357,9 +1358,10 @@ static void autoconfig_irq(struct uart_8250_port *up) /* Synchronize UART_IER access against the console. */ uart_port_lock_irq(port); serial_out(up, UART_IER, UART_IER_ALL_INTR); + lsr = serial_in(up, UART_LSR); + if (lsr & UART_LSR_DR) + serial_port_in(port, UART_RX); uart_port_unlock_irq(port); - serial_in(up, UART_LSR); - serial_in(up, UART_RX); serial_in(up, UART_IIR); serial_in(up, UART_MSR); serial_out(up, UART_TX, 0xFF); @@ -2137,19 +2139,16 @@ static void wait_for_xmitr(struct uart_8250_port *up, int bits) static int serial8250_get_poll_char(struct uart_port *port) { struct uart_8250_port *up = up_to_u8250p(port); - int status; + int status = NO_POLL_CHAR; u16 lsr; serial8250_rpm_get(up); + uart_port_lock_irqsave(port, &flags); lsr = serial_port_in(port, UART_LSR); - - if (!(lsr & UART_LSR_DR)) { - status = NO_POLL_CHAR; - goto out; - } - - status = serial_port_in(port, UART_RX); + if ((lsr & UART_LSR_DR)) + status = serial_port_in(port, UART_RX); + uart_port_unlock_irqrestore(port, flags); out: serial8250_rpm_put(up); return status; @@ -2264,13 +2263,16 @@ int serial8250_do_startup(struct uart_port *port) * Clear the FIFO buffers and disable them. * (they will be reenabled in set_termios()) */ + uart_port_lock_irqsave(port, &flags); serial8250_clear_fifos(up); /* * Clear the interrupt registers. */ - serial_port_in(port, UART_LSR); - serial_port_in(port, UART_RX); + lsr = serial_port_in(port, UART_LSR); + if (lsr & UART_LSR_DR) + serial_port_in(port, UART_RX); + uart_port_unlock_irqrestore(port, flags); serial_port_in(port, UART_IIR); serial_port_in(port, UART_MSR); @@ -2380,9 +2382,10 @@ int serial8250_do_startup(struct uart_port *port) /* * Now, initialize the UART */ - serial_port_out(port, UART_LCR, UART_LCR_WLEN8); uart_port_lock_irqsave(port, &flags); + serial_port_out(port, UART_LCR, UART_LCR_WLEN8); + if (up->port.flags & UPF_FOURPORT) { if (!up->port.irq) up->port.mctrl |= TIOCM_OUT1; @@ -2428,15 +2431,16 @@ int serial8250_do_startup(struct uart_port *port) } dont_test_tx_en: - uart_port_unlock_irqrestore(port, flags); /* * Clear the interrupt registers again for luck, and clear the * saved flags to avoid getting false values from polling * routines or the previous session. */ - serial_port_in(port, UART_LSR); - serial_port_in(port, UART_RX); + lsr = serial_port_in(port, UART_LSR); + if (lsr & UART_LSR_DR) + serial_port_in(port, UART_RX); + uart_port_unlock_irqrestore(port, flags); serial_port_in(port, UART_IIR); serial_port_in(port, UART_MSR); up->lsr_saved_flags = 0; @@ -2492,6 +2496,7 @@ void serial8250_do_shutdown(struct uart_port *port) { struct uart_8250_port *up = up_to_u8250p(port); unsigned long flags; + u16 lsr; serial8250_rpm_get(up); /* @@ -2518,7 +2523,6 @@ void serial8250_do_shutdown(struct uart_port *port) port->mctrl &= ~TIOCM_OUT2; serial8250_set_mctrl(port, port->mctrl); - uart_port_unlock_irqrestore(port, flags); /* * Disable break condition and FIFOs @@ -2526,6 +2530,7 @@ void serial8250_do_shutdown(struct uart_port *port) serial_port_out(port, UART_LCR, serial_port_in(port, UART_LCR) & ~UART_LCR_SBC); serial8250_clear_fifos(up); + uart_port_unlock_irqrestore(port, flags); #ifdef CONFIG_SERIAL_8250_RSA /* @@ -2538,7 +2543,12 @@ void serial8250_do_shutdown(struct uart_port *port) * Read data port to reset things, and then unlink from * the IRQ chain. */ - serial_port_in(port, UART_RX); + uart_port_lock_irqsave(port, &flags); + lsr = serial_port_in(port, UART_LSR); + if (lsr & UART_LSR_DR) + serial_port_in(port, UART_RX); + uart_port_unlock_irqrestore(port, flags); + serial8250_rpm_put(up); up->ops->release_irq(up);
When the PSLVERR_RESP_EN parameter is set to 1, the device generates an error response if an attempt is made to read an empty RBR (Receive Buffer Register) while the FIFO is enabled. In serial8250_do_startup(), calling serial_port_out(port, UART_LCR, UART_LCR_WLEN8) triggers dw8250_check_lcr(), which invokes dw8250_force_idle() and serial8250_clear_and_reinit_fifos(). The latter function enables the FIFO via serial_out(p, UART_FCR, p->fcr). Execution proceeds to the dont_test_tx_en label: ... serial_port_in(port, UART_RX); This satisfies the PSLVERR trigger condition. Because another CPU(e.g., using printk()) is accessing the UART (UART is busy), the current CPU fails the check (value & ~UART_LCR_SPAR) == (lcr & ~UART_LCR_SPAR), causing it to enter dw8250_force_idle(). To fix this, all calls to serial_out(UART_LCR) and serial_in(UART_RX) should be executed under port->lock. Additionally, checking the readiness via UART_LSR should also be done under port->lock. Panic backtrace: [ 0.442336] Oops - unknown exception [#1] [ 0.442343] epc : dw8250_serial_in32+0x1e/0x4a [ 0.442351] ra : serial8250_do_startup+0x2c8/0x88e ... [ 0.442416] console_on_rootfs+0x26/0x70 Fixes: c49436b657d0 ("serial: 8250_dw: Improve unwritable LCR workaround") Link: https://lore.kernel.org/all/84cydt5peu.fsf@jogness.linutronix.de/T/ Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com> --- drivers/tty/serial/8250/8250_dw.c | 8 +++++ drivers/tty/serial/8250/8250_port.c | 46 ++++++++++++++++++----------- 2 files changed, 36 insertions(+), 18 deletions(-)