Re: [PATCH 2/3] x86_64: install emergency handler for double fault

[Almudena Garcia]

Hi:

The cpus are started sequentially: don't start the next until the current has 
finished its configuration. 

But your idea is good 

Thanks 

El lunes 19 de junio de 2023, Luca Dariz escribió:
> Il 17/06/23 23:12, Samuel Thibault ha scritto:
> > Luca Dariz, le jeu. 15 juin 2023 23:49:30 +0200, a ecrit:
> >> diff --git a/i386/i386/ktss.c b/i386/i386/ktss.c
> >> index 1d880167..52f3722c 100644
> >> --- a/i386/i386/ktss.c
> >> +++ b/i386/i386/ktss.c
> >> @@ -61,6 +61,7 @@ ktss_fill(struct task_tss *myktss, struct 
> >> real_descriptor *mygdt)
> >>    /* Initialize the master TSS.  */
> >>   #ifdef __x86_64__
> >>    myktss->tss.rsp0 = (unsigned long)(exception_stack+1024);
> >> +  myktss->tss.ist1 = (unsigned long)(exception_stack+1024);
> > 
> > Shouldn't we use a different stack, to avoid overwriting information
> > from the first fault?
> 
> you're right, otherwise it will be less useful in case a double fault 
> happens before scheduling starts (which then overwrites rsp0 with the 
> pcb stack on every context switch).
> 
> Thinking about the smp case, are the cpu started in parallel or 
> sequentially? There might also be a need to set a cpu-specific first 
> rsp0, if there is the possibility to have early interrupts concurrently.
> 
> 
> Luca
> 
> 
>

-- 
Enviado desde mi dispositivo Sailfish