2002-03-29 10:34:34

by Amit S. Kale

Subject: announce: kgdb 1.5 with reworked buggy smp handling

Hi,

kgdb 1.5 at http://kgdb.sourceforge.net/

smp handling is completely reworked. Previous kgdb had a bug
which caused it to hang when a processor spun with
interrupts disabled and another processor enters kgdb. kgdb
now uses nmi watchdog for holding other processors while
a machine is in kgdb.
--
Amit S. Kale
Linux kernel source level debugger http://kgdb.sourceforge.net/
[29th March - kgdb-1.5 with reworked smp handling.]
Translation filesystem http://trfs.sourceforge.net/

2002-03-29 11:00:24

by Keith Owens

Subject: Re: announce: kgdb 1.5 with reworked buggy smp handling

On Fri, 29 Mar 2002 16:01:36 +0530,
"Amit S. Kale" <[email protected]> wrote:
>kgdb 1.5 at http://kgdb.sourceforge.net/
>
>smp handling is completely reworked. Previous kgdb had a bug
>which caused it to hang when a processor spun with
>interrupts disabled and another processor enters kgdb. kgdb
>now uses nmi watchdog for holding other processors while
>a machine is in kgdb.

IA64 disabled spin loops ignore NMI :(.

2002-03-29 15:00:28

by Amit S. Kale

Subject: Re: announce: kgdb 1.5 with reworked buggy smp handling

Keith Owens wrote:
>
> On Fri, 29 Mar 2002 16:01:36 +0530,
> "Amit S. Kale" <[email protected]> wrote:
> >kgdb 1.5 at http://kgdb.sourceforge.net/
> >
> >smp handling is completely reworked. Previous kgdb had a bug
> >which caused it to hang when a processor spun with
> >interrupts disabled and another processor enters kgdb. kgdb
> >now uses nmi watchdog for holding other processors while
> >a machine is in kgdb.
>
> IA64 disabled spin loops ignore NMI :(.

Thanks for the info.

Isn't there any way get into an interrupt disabled spinning
processor on an ia64 smp machine?
--
Amit S. Kale
Linux kernel source level debugger http://kgdb.sourceforge.net/
[29th March - kgdb-1.5 with reworked smp handling.]
Translation filesystem http://trfs.sourceforge.net/

2002-03-30 00:17:53

by Keith Owens

Subject: Re: announce: kgdb 1.5 with reworked buggy smp handling

On Fri, 29 Mar 2002 20:33:19 +0530,
"Amit S. Kale" <[email protected]> wrote:
>Keith Owens wrote:
>>
>> On Fri, 29 Mar 2002 16:01:36 +0530,
>> "Amit S. Kale" <[email protected]> wrote:
>> >kgdb 1.5 at http://kgdb.sourceforge.net/
>> >
>> >smp handling is completely reworked. Previous kgdb had a bug
>> >which caused it to hang when a processor spun with
>> >interrupts disabled and another processor enters kgdb. kgdb
>> >now uses nmi watchdog for holding other processors while
>> >a machine is in kgdb.
>>
>> IA64 disabled spin loops ignore NMI :(.
>
>Thanks for the info.
>
>Isn't there any way get into an interrupt disabled spinning
>processor on an ia64 smp machine?

Only via an INIT interrupt. But that is also used to initialise a cpu
at startup, INIT interrupts go through PAL and SAL and have side
effects. I don't use them for kdb, they are too heavy.

For more background, see the thread "Replacements for local_irq_xxx()"
https://external-lists.vasoftware.com/archives//linux-ia64/2001-May/thread.html