| 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: Mon, 21 Nov 2022 13:50:51 -0500
From: "Gabriel L. Somlo" <gsomlo@...il.com>
To: Jiri Slaby <jirislaby@...nel.org>
Cc: linux-kernel@...r.kernel.org, linux-serial@...r.kernel.org,
gregkh@...uxfoundation.org, kgugala@...micro.com,
mholenko@...micro.com, joel@....id.au,
david.abdurachmanov@...il.com, florent@...oy-digital.fr,
geert@...ux-m68k.org, ilpo.jarvinen@...ux.intel.com
Subject: Re: [PATCH v5 12/14] serial: liteuart: add IRQ support for the RX
path
Hi Jiri,
On Mon, Nov 21, 2022 at 09:54:34AM +0100, Jiri Slaby wrote:
> On 18. 11. 22, 15:55, Gabriel Somlo wrote:
> > Add support for IRQ-driven RX. Support for the TX path will be added
> > in a separate commit.
> >
> > Signed-off-by: Gabriel Somlo <gsomlo@...il.com>
> > Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>
> > ---
> >
> > Changes from v4:
> > - using dev_err() instead of a combo of pr_err() and pr_fmt()
> > - dropped "get irq" comment in probe()
> >
> > > Changes from v3:
> > > - add shadow irq register to support polling mode and avoid reading
> > > hardware mmio irq register to learn which irq flags are enabled
> > > - this also simplifies both liteuart_interrupt() and liteuart_startup()
> >
> > drivers/tty/serial/liteuart.c | 76 +++++++++++++++++++++++++++++++----
> > 1 file changed, 69 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/tty/serial/liteuart.c b/drivers/tty/serial/liteuart.c
> > index 8a6e176be08e..678c37c952cf 100644
> > --- a/drivers/tty/serial/liteuart.c
> > +++ b/drivers/tty/serial/liteuart.c
> > @@ -7,6 +7,7 @@
> > #include <linux/bits.h>
> > #include <linux/console.h>
> > +#include <linux/interrupt.h>
> > #include <linux/litex.h>
> > #include <linux/module.h>
> > #include <linux/of.h>
> > @@ -46,6 +47,7 @@ struct liteuart_port {
> > struct uart_port port;
> > struct timer_list timer;
> > u32 id;
> > + u8 irq_reg;
> > };
> > #define to_liteuart_port(port) container_of(port, struct liteuart_port, port)
> > @@ -76,6 +78,19 @@ static void liteuart_putchar(struct uart_port *port, unsigned char ch)
> > litex_write8(port->membase + OFF_RXTX, ch);
> > }
> > +static void liteuart_update_irq_reg(struct uart_port *port, bool set, u8 mask)
> > +{
> > + struct liteuart_port *uart = to_liteuart_port(port);
> > +
> > + if (set)
> > + uart->irq_reg |= mask;
> > + else
> > + uart->irq_reg &= ~mask;
> > +
> > + if (port->irq)
> > + litex_write8(port->membase + OFF_EV_ENABLE, uart->irq_reg);
> > +}
> > +
> > static void liteuart_stop_tx(struct uart_port *port)
> > {
> > }
> > @@ -129,13 +144,27 @@ static void liteuart_rx_chars(struct uart_port *port)
> > tty_flip_buffer_push(&port->state->port);
> > }
> > +static irqreturn_t liteuart_interrupt(int irq, void *data)
> > +{
> > + struct liteuart_port *uart = data;
> > + struct uart_port *port = &uart->port;
> > + u8 isr;
> > +
> > + spin_lock(&port->lock);
> > + isr = litex_read8(port->membase + OFF_EV_PENDING) & uart->irq_reg;
> > + if (isr & EV_RX)
> > + liteuart_rx_chars(port);
> > + spin_unlock(&port->lock);
> > +
> > + return IRQ_RETVAL(isr);
> > +}
> > +
> > static void liteuart_timer(struct timer_list *t)
> > {
> > struct liteuart_port *uart = from_timer(uart, t, timer);
> > struct uart_port *port = &uart->port;
> > - liteuart_rx_chars(port);
> > -
> > + liteuart_interrupt(0, port);
>
> Are you sure spin_lock() is safe from this path? I assume so, but have you
> thought about it?
I checked and at that point `in_serving_softirq()` is true.
*However*, after studying spin_lock() and friends for a while, I'm
not quite clear on whether that unequivocally translates
to "yes, we're safe" :)
As such, I'm inclined to switch to `spin_lock_irqsave()` and
`spin_unlock_irqrestore()` even in the interrupt handler, which is
explicitly stated to be "safe from any context":
https://www.kernel.org/doc/html/v4.15/kernel-hacking/locking.html#cheat-sheet-for-locking
The alternative could be to set `TIMER_IRQSAFE` in `timer_setup()`,
but no other tty driver seems to be doing that, so I'd be a bit off
the beaten path there... :)
Please do let me know what you think about this, particularly if you
consider going the spin_lock_irqsave-everywhere-just-to-be-safe route
overkill... :)
> > mod_timer(&uart->timer, jiffies + uart_poll_timeout(port));
> > }
> > @@ -161,19 +190,46 @@ static unsigned int liteuart_get_mctrl(struct uart_port *port)
> > static int liteuart_startup(struct uart_port *port)
> > {
> > struct liteuart_port *uart = to_liteuart_port(port);
> > + unsigned long flags;
> > + int ret;
> > - /* disable events */
> > - litex_write8(port->membase + OFF_EV_ENABLE, 0);
> > + if (port->irq) {
> > + ret = request_irq(port->irq, liteuart_interrupt, 0,
> > + KBUILD_MODNAME, uart);
>
> Just asking: cannot the irq be shared?
Given the way LiteX gateware is currently generated, each
irq-triggering device is given its own separate line. I don't think
setting the IRQF_SHARED flag actually *hurts* anything (no difference
in behavior while testing), but I don't think it's needed ATM.
> > + if (ret) {
> > + dev_err(port->dev,
> > + "line %d irq %d failed: switch to polling\n",
> > + port->line, port->irq);
>
> That is, it should be only dev_warn(), or?
Makes sense, will use dev_warn() in v6.
Please LMK what you think about spin_lock[_irqsave] (and IRQF_SHARED),
and I'll send out v6 with all the necessary chances right after that.
Thanks much,
--Gabriel
> > + port->irq = 0;
> > + }
> > + }
>
> thanks,
> --
> js
> suse labs
>
Powered by blists - more mailing lists