Re: [Xen-devel] [PATCH] reduce side effects of handling '*' debug key

["Jan Beulich" <JBeulich@xxxxxxxxxx>]

>>> On 15.12.10 at 11:45, Keir Fraser <keir@xxxxxxx> wrote:
> On 14/12/2010 14:46, "Jan Beulich" <JBeulich@xxxxxxxxxx> wrote:
> 
>> NMI watchdog should be suppressed, and softirqs should be handled at
>> least in the non-IRQ handler portion (they obviously must not be
>> processed in IRQ context).
> 
> Why do you add calls to watchdog_disable() --- The process_softirqs() call
> you add is intended to be sufficient to keep timers being handled in a
> timely fashion, is it not?
> 
> I can see why you remove console_start_log_everything(), since that is
> handled from handle_keypress(), but why do you add calls to
> console_start_sync()?

I didn't actually pay attention to the duplication with
handle_keypress(). Instead, I was simply following what the 'd'
handler does, noting that calling console_start_sync() makes
calling console_start_log_everything() pointless.

> Personally I think that if you have justifiable reason to place the console
> in synchronous mode, and protect yourself from the NMI watchdog, these calls
> should be added to handle_keypress() for all keyhandlers to enjoy.

Protection from the NMI watchdog shouldn't be needed for
handlers not taking long, so I'm not certain it'd be useful to add
generally.

Similarly, for handlers not printing much, forcing the console
into synchronous mode doesn't seem appropriate to me.
Basically, each handler should know for itself (and
handle_keypress() should, as it does, only do what all
handler want in common).

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel