Re: [PATCH v3 2/5] xen/common: Guard iommu symbols with CONFIG_HAS_PASSTHROUGH

[Julien Grall <julien@xen.org>]

Hi Jan,

On 17/05/2021 12:16, Jan Beulich wrote:
> On 14.05.2021 20:53, Connor Davis wrote:
>> --- a/xen/common/memory.c
>> +++ b/xen/common/memory.c
>> @@ -294,7 +294,9 @@ int guest_remove_page(struct domain *d, unsigned long gmfn)
>>       p2m_type_t p2mt;
>>   #endif
>>       mfn_t mfn;
>> +#ifdef CONFIG_HAS_PASSTHROUGH
>>       bool *dont_flush_p, dont_flush;
>> +#endif
>>       int rc;
>>   
>>   #ifdef CONFIG_X86
>> @@ -385,13 +387,17 @@ int guest_remove_page(struct domain *d, unsigned long gmfn)
>>        * Since we're likely to free the page below, we need to suspend
>>        * xenmem_add_to_physmap()'s suppressing of IOMMU TLB flushes.
>>        */
>> +#ifdef CONFIG_HAS_PASSTHROUGH
>>       dont_flush_p = &this_cpu(iommu_dont_flush_iotlb);
>>       dont_flush = *dont_flush_p;
>>       *dont_flush_p = false;
>> +#endif
>>   
>>       rc = guest_physmap_remove_page(d, _gfn(gmfn), mfn, 0);
>>   
>> +#ifdef CONFIG_HAS_PASSTHROUGH
>>       *dont_flush_p = dont_flush;
>> +#endif
>>   
>>       /*
>>        * With the lack of an IOMMU on some platforms, domains with DMA-capable
>> @@ -839,11 +845,13 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
>>       xatp->gpfn += start;
>>       xatp->size -= start;
>>   
>> +#ifdef CONFIG_HAS_PASSTHROUGH
>>       if ( is_iommu_enabled(d) )
>>       {
>>          this_cpu(iommu_dont_flush_iotlb) = 1;
>>          extra.ppage = &pages[0];
>>       }
>> +#endif
>>   
>>       while ( xatp->size > done )
>>       {
>> @@ -868,6 +876,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
>>           }
>>       }
>>   
>> +#ifdef CONFIG_HAS_PASSTHROUGH
>>       if ( is_iommu_enabled(d) )
>>       {
>>           int ret;
>> @@ -894,6 +903,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
>>           if ( unlikely(ret) && rc >= 0 )
>>               rc = ret;
>>       }
>> +#endif
>>   
>>       return rc;
>>   }
> 
> I wonder whether all of these wouldn't better become CONFIG_X86:
> ISTR Julien indicating that he doesn't see the override getting used
> on Arm. (Julien, please correct me if I'm misremembering.)

Right, so far, I haven't been in favor to introduce it because:
    1) The P2M code may free some memory. So you can't always ignore the 
flush (I think this is wrong for the upper layer to know when this can 
happen).
    2) It is unclear what happen if the IOMMU TLBs and the PT contains 
different mappings (I received conflicted advice).

So it is better to always flush and as early as possible.

Cheers,

-- 
Julien Grall