Setting up hypervisor_callback and testing

Setting up hypervisor_callback and testing

[Ivan Kelly]

im setting up my hypervisor callback with:
hypervisor_set_callbacks(CS_SELECTOR, (unsigned long)hypervisor_callback,
			 CS_SELECTOR, (unsigned long)failsafe_callback);
hypervisor_callback and failsafe_callback are as yet very simple assembly
funcations that just shutdown the domain.
<SNIP>
_hypervisor_callback:
        call    _hypervisor_shutdown
        ret

_failsafe_callback:
        call    _hypervisor_shutdown
        ret
</SNIP>

Im trying to test it with:
<SNIP>
_xen_cb_test:
        call    _hypervisor_save_kernel_stack
        mov     esi, _hypervisor_shared_info
        mov     1(esi), 0
        push    0
        push    __HYPERVISOR_xen_version
        call    _xen_op
        add     esp, 4*2
        ret
</SNIP>

The callback doesn't seem to be getting called at all. At the moment im only
trying to get very basic callback stuff working, so i've left out the save and
iret stuff. Could anyone point out where im going wrong?
Regards
Ivan Kelly

Re: Setting up hypervisor_callback and testing

[Simon Kagstrom]

At Wed, 15 Feb 2006 11:49:05 +0000,
Ivan Kelly wrote:
> 
> im setting up my hypervisor callback with:
> hypervisor_set_callbacks(CS_SELECTOR, (unsigned long)hypervisor_callback,
> 			 CS_SELECTOR, (unsigned long)failsafe_callback);
> hypervisor_callback and failsafe_callback are as yet very simple assembly
> funcations that just shutdown the domain.
> <SNIP>
> _hypervisor_callback:
> [...]
> </SNIP>
> 
> The callback doesn't seem to be getting called at all. At the moment im only
> trying to get very basic callback stuff working, so i've left out the save and
> iret stuff. Could anyone point out where im going wrong?

Have you enabled event delivery? Look in the Mini-OS source kernel.c
and os.h, where "__sti" is defined:

#define __sti()                                                         \
do {                                                                    \
        vcpu_info_t *_vcpu;                                             \
        barrier();                                                      \
        _vcpu = &HYPERVISOR_shared_info->vcpu_info[smp_processor_id()]; \
        _vcpu->evtchn_upcall_mask = 0;                                  \
        barrier(); /* unmask then check (avoid races) */                \
        if ( unlikely(_vcpu->evtchn_upcall_pending) )                   \
                force_evtchn_callback();                                \
} while (0)

the events should be masked on startup I think, so maybe that's your
problem.

// Simon

Re: Setting up hypervisor_callback and testing

[Ivan Kelly]

On Wed, Feb 15, 2006 at 01:33:45PM +0100, Simon Kagstrom wrote:
> #define __sti()                                                         \
> do {                                                                    \
>         vcpu_info_t *_vcpu;                                             \
>         barrier();                                                      \
>         _vcpu = &HYPERVISOR_shared_info->vcpu_info[smp_processor_id()]; \
>         _vcpu->evtchn_upcall_mask = 0;                                  \
>         barrier(); /* unmask then check (avoid races) */                \
>         if ( unlikely(_vcpu->evtchn_upcall_pending) )                   \
>                 force_evtchn_callback();                                \
> } while (0)
> 
> the events should be masked on startup I think, so maybe that's your
> problem.
Ok, Ive setup up this. I hadn't the shared info mapped properly and took me a
while to figure out. now when i run the interrupt test i get:
(XEN) (file=/usr/src/xen/xen-2.0/xen/include/asm/mm.h, line=157) Error pfn
00000000: ed=fc59e6c0, sd2
Then the code seems to return to the entry point, and run again continuously.
I think it's trying to jump to an address that doesnt exist even though i have
callback_event set to _hypervisor_callback.
-Ivan

Re: Setting up hypervisor_callback and testing

[Ivan Kelly]

> Ok, Ive setup up this. I hadn't the shared info mapped properly and took me a
> while to figure out. now when i run the interrupt test i get:
> (XEN) (file=/usr/src/xen/xen-2.0/xen/include/asm/mm.h, line=157) Error pfn
> 00000000: ed=fc59e6c0, sd2
> Then the code seems to return to the entry point, and run again continuously.
> I think it's trying to jump to an address that doesnt exist even though i have
> callback_event set to _hypervisor_callback.
Solved my problem now. Not sure what solved it, but i was calling the
set_callbacks hypercall without failsafe_addr set. sorted now.
-ivan