Re: [Xen-devel] [v3][PATCH 1/1] xen:vtd: missing RMRR mapping while share EPT

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 25.07.14 at 10:45, <tiejun.chen@xxxxxxxxx> wrote:
> On 2014/7/25 16:07, Jan Beulich wrote:
>>>>> On 25.07.14 at 08:47, <tiejun.chen@xxxxxxxxx> wrote:
>>>>   So I think it would be prudent to at least try to detect a clash here
>>>> and return an error.
>>>
>>> Could we check if this pte already present like this?
>>>
>>> @@ -1870,6 +1872,16 @@ static int rmrr_identity_mapping(struct domain *d,
>>>            if ( iommu_use_hap_pt(d) )
>>>            {
>>>                ASSERT(!iommu_passthrough || !is_hardware_domain(d));
>>> +            /* get last level pte */
>>> +            pg_maddr = addr_to_dma_page_maddr(d, ((paddr_t)gfn << 
>>> PAGE_SHIFT_4K), 0);
>>> +            page = (struct dma_pte *)map_vtd_domain_page(pg_maddr);
>>> +            pte = page + address_level_offset(addr, 1);
>>> +            if ( dma_pte_present(*pte) )
>>> +            {
>>> +                printk(XENLOG_ERR VTDPREFIX
>>> +                       "Overlapping RMRRs at %"PRIx64".\n", 
>>> (paddr_t)base_gfn);
>>> +                return -1;
>>> +            }
>>>                if ( set_mmio_p2m_entry(d, base_pfn, _mfn(base_pfn)) )
>>>                    return -1;
>>>            }
>>
>> I think you'd be better off using P2M functions here, even more so
>> considering that you're in an iommu_use_hap_pt() guarded code
>> section.
> 
> Are you pointing p2m->get_entry?

Sort of. For the check to really be meaningful the get and set
would need to be done without the respective lock dropped
intermediately.

Jan


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