Vincent Hanquez wrote:
> On 12/05/10 16:57, Gihan Munasinghe wrote:
>> The sleep is still needed, if not libxl_* calls fails sometimes. I'll do
>> more debugging with in the calls it self and see, but for now the
>> sleep() seems to do the trick.
>
> diff -Naur libxl/libxl.c libxl-patch/libxl.c
> --- libxl/libxl.c    2010-05-11 09:37:50.000000000 +0100
> +++ libxl-patch/libxl.c    2010-05-12 16:24:42.000000000 +0100
> @@ -538,12 +538,12 @@
>      shutdown_path = libxl_sprintf(ctx, "%s/control/shutdown", dom_path);
>
>      xs_write(ctx->xsh, XBT_NULL, shutdown_path, req_table[req], 
> strlen(req_table[req]));
> -    if (/* hvm */ 0) {
> +    if (/* hvm */ 1) {
>          unsigned long acpi_s_state = 0;
>          unsigned long pvdriver = 0;
>
> you can't just switch from 0 to 1, otherwise pv domains will just 
> fails (although i see we're not testing xc_get_hvm_param return values 
> either).

I though PV domain would have been captured by..
 xc_get_hvm_param(ctx->xch, domid, HVM_PARAM_CALLBACK_IRQ, &pvdriver);
        if (!pvdriver  .. )
>
> the if (/* hvm */ 0) is because the function never properly tested if 
> the domain is hvm or not. nowadays you have is_hvm(domid) function 
> that can handily replace the hardcoded value.
>
Yes having is_hvm is much better than the hard coded values I have 
changed that bit of code see attached patch.
I tested with hvm with and with out pv drivers. seems to work. Would be 
appreciate if someone can test in full pv domains.

Thanks

--
Gihan