Sergei Shtylyov wrote:
> Hello.
>
> Marc St-Jean wrote:
>
> >> > diff --git a/arch/mips/pmc-sierra/msp71xx/msp_serial.c
> >>b/arch/mips/pmc-sierra/msp71xx/msp_serial.c
> >> > new file mode 100644
> >> > index 0000000..3b956e9
> >> > --- /dev/null
> >> > +++ b/arch/mips/pmc-sierra/msp71xx/msp_serial.c
> >> > @@ -0,0 +1,185 @@
> >>[...]
> >> > +#ifdef CONFIG_KGDB
> >> > +/*
> >> > + * kgdb uses serial port 1 so the console can remain on port 0.
> >> > + * To use port 0 change the definition to read as follows:
> >> > + * #define DEBUG_PORT_BASE KSEG1ADDR(MSP_UART0_BASE)
> >> > + */
> >> > +#define DEBUG_PORT_BASE KSEG1ADDR(MSP_UART1_BASE)
> >> > +
> >> > +int putDebugChar(char c)
> >> > +{
> >> > + volatile uint32_t *uart = (volatile uint32_t *)DEBUG_PORT_BASE;
> >> > + uint32_t val = (uint32_t)c;
> >> > +
> >> > + local_irq_disable();
> >> > + while (!(uart[5] & 0x20)); /* Wait for TXRDY */
> >> > + uart[0] = val;
> >> > + while (!(uart[5] & 0x20)); /* Wait for TXRDY */
> >> > + local_irq_enable();
>
> >> Gah, why you decided to put local_irq_enable() there?! KGDB expects
> >>interrupts to be *disabled* while it has control, else some subtle state
> >>corruptions will ensue, and it will eventually lock up. Please remove
> >>these 2 calls completely.
>
> > Hmmm, this has been working for several months. I'll remove, retest and
> > resubmit.
>
> I should probably have said "may". From my experience with KGDBoE
> though
> (well, it's somewhat different KGDB implementation :-) it locks up pretty
> quickly because of local_irq_enable() or spin_unlock_irq(). Nevertheless,
> enabling interrupts while KGDB has control is undesirable. And look at the
> other KGDB serial code inarch/mips/ -- nobody else does this.
A search showed that some other architectures do use local_irq_disable()/enable()
and our code may have been derived from it. None of the MIPS platforms do and
the kgdb implementation seams to be arch specific so I'll drop them.
> > Are you aware if this is the case for the "putchar" used by early_printk
> > as well?
>
> No idea, ask Ralf. ;-)
I searched putchar implementations and there only seems to be one more for the
cobalt. It also doesn't twiddle the irg state so I'll drop it there as well.
I tested and seems well without the local_irq_disable()/enable(), I'll resubmit
the path.
Thanks,
Marc
