[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH RFC] x86/altp2m: fix display frozen when switching to a new view early

>>> On 04.10.18 at 17:52, <george.dunlap@xxxxxxxxxx> wrote:
> On 10/04/2018 04:45 PM, Jan Beulich wrote:
>>>>> On 04.10.18 at 17:34, <george.dunlap@xxxxxxxxxx> wrote:
>>> On 10/04/2018 04:20 PM, Jan Beulich wrote:
>>>>>>> On 04.10.18 at 16:56, <rcojocaru@xxxxxxxxxxxxxxx> wrote:
>>>>> The biggest problem here is p2m->logdirty_ranges. This patch will
>>>>> (justly) not work, because struct rangeset is only forward-declared in
>>>>> xen/rangeset.h, so an incomplete type here:
>>>>> -void p2m_init_altp2m_ept(struct domain *d, unsigned int i)
>>>>> +int p2m_init_altp2m_ept(struct domain *d, unsigned int i)
>>>>>  {
>>>>>      struct p2m_domain *p2m = d->arch.altp2m_p2m[i];
>>>>>      struct p2m_domain *hostp2m = p2m_get_hostp2m(d);
>>>>>      struct ept_data *ept;
>>>>> +    if ( !p2m->logdirty_ranges )
>>>>> +        p2m->logdirty_ranges = rangeset_new(d, "log-dirty",
>>>>> +                                            RANGESETF_prettyprint_hex);
>>>>> +    if ( !p2m->logdirty_ranges )
>>>>> +        return -ENOMEM;
>>>>> +
>>>>> +    *p2m->logdirty_ranges = *hostp2m->logdirty_ranges;
>>>>> +
>>>>>      p2m->ept.ad = hostp2m->ept.ad;
>>>>> +    p2m->max_mapped_pfn = hostp2m->max_mapped_pfn;
>>>>> +    p2m->default_access = hostp2m->default_access;
>>>>> +    p2m->domain = hostp2m->domain;
>>>>> +
>>>>> +    p2m->global_logdirty = hostp2m->global_logdirty;
>>>>>      p2m->min_remapped_gfn = gfn_x(INVALID_GFN);
>>>>>      p2m->max_remapped_gfn = 0;
>>>>>      ept = &p2m->ept;
>>>>>      ept->mfn = pagetable_get_pfn(p2m_get_pagetable(p2m));
>>>>>      d->arch.altp2m_eptp[i] = ept->eptp;
>>>>> +
>>>>> +    return 0;
>>>>> +}
>>>>> But that's not even the biggest problem: even if that would compile, it
>>>>> would still be wrong, because logdirty_pages has pointers of its own,
>>>>> which means that two bitwise-copied distinct rangesets can still point
>>>>> to the same data and thus be vulnerable to race conditions and wanting
>>>>> synchronization.
>>> Yes, so "deep copy" means if a structure has pointers, you copy the
>>> structures pointed to; and if that structure has pointers, you copy
>>> those, all the way down.  After a deep copy, any operations on the
>>> structure should be operating on completely separate bits of memory and
>>> pointers.
>>>>> Furthermore there's no rangeset_copy() function in sight in rangeset.h
>>>>> (though there is a rangeset_swap()).
>>>>> Would you like me to add a rangeset_copy() function (presumably another
>>>>> intermediary patch) and proceed in that manner?
>>>> Roger recently has posted a patch adding rangeset_merge(), which I think
>>>> is more general than your rangeset_copy(). That said, I'm in no way
>>>> convinced copying (and then keeping in sync) the range sets across the
>>>> altp2m-s is the best approach. It may well be that the optimal solution is
>>>> somewhere in the middle between sharing everything and copying
>>>> everything.
>>> Er, you mean maybe we should share logdirty ranges and copy other
>>> things?  Or do you actually mean somehow to share bits of the logdirty
>>> range structure?
>> The former, of course. I'm sorry for the ambiguity.
>>> I think we can reasonably start with a simple-and-correct approach, and
>>> try to come up with an optimization later if we decide it's necessary.
>>> The two basic simple-but-correct approaches would be:
>>> 1. Share logdirty_ranges.  This would mean not duplicating the structure
>>> and keeping it in sync; but it would mean grabbing the main p2m lock on
>>> every resolv_misconfig().
>>> 2. Duplicate the structure and keep it in sync.  This  means not
>>> grabbing the main p2m lock on every resolv_misconfig(); but it does mean
>>> duplicating memory, as well as grabbing the lock of every altp2m
>>> structure every time logdirty_ranges changes.
>>> As I've said before, I think #2 is the most promising, since
>>> resolv_misconfig will be called (potentially) for each entry in the p2m
>>> table, but logdirty_ranges only changes once or twice during the entire
>>> lifetime of the guest.
>> So perhaps some r/w lock wants to be introduced?
> There will also be locking order issues to consider if we do that.
> What's your main reason for not wanting #2?

Duplicating and keeping in sync data for which no separate allocations
are needed seems fine to me. Duplicating allocations, otoh, seems not
only a waste of resources to me, but I also expect error handling to
become ugly: You'll need to undo all prior syncing if e.g. the allocation
for altp2m #5 fails.


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.