Re: 2.6.18-rc6-mm1: GPF loop on early boot

[Ingo Molnar <mingo@elte.hu>]

* Jeremy Fitzhardinge <jeremy@goop.org> wrote:

> Ingo Molnar wrote:
> >btw., what's the connection of %gs based PDA to Xen and 
> >paravirtualization in general - %esp based current is just as 
> >Xen-friendly, or am i wrong? I guess there must be some connection, 
> >given that you are working on this ;)
> >  
> 
> Yep.  The goal is to put the Xen VCPU structure into the PDA, so that it 
> can be easily accessed.  At present, masking events (ie, cli), is 
> something along the lines of
> 
>    xen_shared_info->vcpu[smp_processor_id()].mask = 1
> 
> which comes out to something like 20 bytes of code, and is probably too 
> awkward to inline.  If the vcpu is in the PDA, it would come out to:
> 
>    movb $1, %gs:xen_vcpu_mask
> 
> which has the added benefit of not needing a register.

take a look at lockdep: amongst other things it adds the TRACE_IRQFLAGS 
/ irqflags.h infrastructure, which could easily be modified to allow the 
easy replacement of cli/sti/pushf/popf. I wrote it with the side-goal of 
paravirtualization. I.e. instead of the DISABLE_INTERRUPT / 
ENABLE_INTERRUPT duplication you do in -mm currently please just enhance 
irqflags.h to cover the needs of paravirtualization too. (most of which 
should be the moving of cli/sti into the assembly callbacks)

into that the PDA would plug in a natural way: current->hardirqs_enabled 
is basically just an alias for 
!xen_shared_info->vcpu[smp_processor_id()].mask.

	Ingo