Re: [Xen-devel] [PATCH v3 3/8] xen: delay allocation of grant table sub structures

[Wei Liu <wei.liu2@xxxxxxxxxx>]

On Wed, Sep 06, 2017 at 05:36:54PM +0200, Juergen Gross wrote:
> On 06/09/17 17:32, Wei Liu wrote:
> > On Wed, Sep 06, 2017 at 05:15:46PM +0200, Juergen Gross wrote:
> >>>> +grant_table_init(struct domain *d)
> >>>> +{
> >>>> +    struct grant_table *gt = d->grant_table;
> >>>> +    unsigned int i, j;
> >>>> +
> >>>> +    if ( gt->nr_grant_frames )
> >>>> +        return 0;
> >>>> +
> >>>
> >>> EBUSY here? I think we should catch the cases when this is called
> >>> multiple times.
> >>
> >> No. The call of grant_table_init() from
> >> domain_unpause_by_systemcontroller() can't be masked, otherwise I
> >> would have to make struct grant_table public again. Multiple calls
> >> are okay.
> > 
> > For domain_unpause_by_systemcontroller, isn't it already guarded by
> > d->creation_finished to ensure there is only one call to
> > grant_table_init?
> > 
> > Or do you mean if gnttab_table_init fails the system administrator will
> > somehow tries to unpause the domain again hence calling grant_table_init
> > again?
> > 
> 
> No. It might have been called already due to e.g. gnttab_setup_table()
> being called as a result of xc_dom_gnttab_seed() during creation of the
> domU. The call from domain_unpause_by_systemcontroller() is just a
> safety net for cases where gnttab_setup_table() wasn't used (e.g. in
> case of a xenstore stubdom).
> 

Hmm... OK. In that case I think the multiple call paths is justified.
You can have my Rb on this patch.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel