Re: [PATCH v2 05/18] IOMMU: have iommu_{,un}map() split requests into largest possible chunks

[Jan Beulich <jbeulich@xxxxxxxx>]

On 30.11.2021 16:24, Roger Pau Monné wrote:
> On Fri, Sep 24, 2021 at 11:45:57AM +0200, Jan Beulich wrote:
>> --- a/xen/drivers/passthrough/iommu.c
>> +++ b/xen/drivers/passthrough/iommu.c
>> @@ -260,12 +260,38 @@ void iommu_domain_destroy(struct domain
>>      arch_iommu_domain_destroy(d);
>>  }
>>  
>> -int iommu_map(struct domain *d, dfn_t dfn, mfn_t mfn,
>> +static unsigned int mapping_order(const struct domain_iommu *hd,
>> +                                  dfn_t dfn, mfn_t mfn, unsigned long nr)
>> +{
>> +    unsigned long res = dfn_x(dfn) | mfn_x(mfn);
>> +    unsigned long sizes = hd->platform_ops->page_sizes;
>> +    unsigned int bit = find_first_set_bit(sizes), order = 0;
>> +
>> +    ASSERT(bit == PAGE_SHIFT);
>> +
>> +    while ( (sizes = (sizes >> bit) & ~1) )
>> +    {
>> +        unsigned long mask;
>> +
>> +        bit = find_first_set_bit(sizes);
>> +        mask = (1UL << bit) - 1;
>> +        if ( nr <= mask || (res & mask) )
>> +            break;
>> +        order += bit;
>> +        nr >>= bit;
>> +        res >>= bit;
>> +    }
>> +
>> +    return order;
>> +}
> 
> This looks like it could be used in other places, I would at least
> consider using it in pvh_populate_memory_range where we also need to
> figure out the maximum order given an address and a number of pages.
> 
> Do you think you could place it in a more generic file and also use
> more generic parameters (ie: unsigned long gfn and mfn)?

The function as is surely isn't reusable, for its use of IOMMU
specific data. If and when a 2nd user appears, it'll be far clearer
whether and if so how much of it is worth generalizing. (Among other
things I'd like to retain the typesafe parameter types here.)

>> @@ -284,14 +316,18 @@ int iommu_map(struct domain *d, dfn_t df
>>          if ( !d->is_shutting_down && printk_ratelimit() )
>>              printk(XENLOG_ERR
>>                     "d%d: IOMMU mapping dfn %"PRI_dfn" to mfn %"PRI_mfn" 
>> failed: %d\n",
>> -                   d->domain_id, dfn_x(dfn_add(dfn, i)),
>> -                   mfn_x(mfn_add(mfn, i)), rc);
>> +                   d->domain_id, dfn_x(dfn), mfn_x(mfn), rc);
>> +
>> +        for ( j = 0; j < i; j += 1UL << order )
>> +        {
>> +            dfn = dfn_add(dfn0, j);
>> +            order = mapping_order(hd, dfn, _mfn(0), i - j);
>>  
>> -        while ( i-- )
>>              /* if statement to satisfy __must_check */
>> -            if ( iommu_call(hd->platform_ops, unmap_page, d, dfn_add(dfn, 
>> i),
>> -                            0, flush_flags) )
>> +            if ( iommu_call(hd->platform_ops, unmap_page, d, dfn, order,
>> +                            flush_flags) )
>>                  continue;
>> +        }
> 
> Why you need this unmap loop here, can't you just use iommu_unmap?

Good question - I merely converted the loop that was already there.
Looks like that could have been changed to a simple call already
before. I'll change it here, on the assumption that splitting this
out isn't going to be a worthwhile exercise.

Jan