On Mon 2024-10-28 14:28:35, John Ogness wrote:
> On 2024-10-25, Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> wrote:
> >> +/*
> >> + * Only to be used directly by the console write callbacks, which may not
> >> + * require the port lock. Use serial8250_clear_IER() instead for all other
> >> + * cases.
> >> + */
> >> +static void __serial8250_clear_IER(struct uart_8250_port *up)
> >> {
> >> if (up->capabilities & UART_CAP_UUE)
> >> serial_out(up, UART_IER, UART_IER_UUE);
> >
> >> serial_out(up, UART_IER, 0);
> >> }
> >>
> >> +static inline void serial8250_clear_IER(struct uart_8250_port *up)
> >> +{
> >> + __serial8250_clear_IER(up);
> >
> > Shouldn't this have a lockdep annotation to differentiate with the
> > above?
>
> Yes, but the follow-up patch adds the annotation as a clean "revert
> patch". I can add a line about that in the commit message.
>
> >> +static void serial8250_console_byte_write(struct uart_8250_port *up,
> >> + struct nbcon_write_context *wctxt)
> >> +{
> >> + const char *s = READ_ONCE(wctxt->outbuf);
> >> + const char *end = s + READ_ONCE(wctxt->len);
> >
> > Is there any possibility that outbuf value be changed before we get
> > the len and at the end we get the wrong pointer?
>
> No. I was concerned about compiler optimization, since @outbuf can
> become NULL. However, it can only become NULL if ownership was
> transferred, and that is properly checked anyway. I will remove the
> READ_ONCE() usage for v4.
I agree that we do not need READ_ONCE() here.
Just to be sure that I understand it correctly.
The struct nbcon_write_context passed by *wctxt should be created on
stack of the caller. Only this process/interrupt context could change
it.
Namely, it might happen when nbcon_enter_unsafe() fails. It is done
later in this function by the code:
while (!nbcon_enter_unsafe(wctxt))
nbcon_reacquire_nobuf(wctxt);
and this function does not access *s or *end after this code.
Other CPUs could not change the structure in parallel
=> READ_ONCE() is not needed.
Just for completeness. The buffer could not disappear.
wctxt->outbuf always points to a static buffer.
Also the content of the buffer could not change if we read it only
after successful nbcon_enter_unsafe(). Only the panic CPU is allowed
to takeover the ownership in this case and it would use another static
buffer.
Best Regards,
Petr
PS: I do not have anything more to add for this patch. It seems to
work work as expected.
