Re: [PATCH] iommu: Improve the performance for direct_mapping

[Yong Wu <yong.wu@mediatek.com>]

On Thu, 2020-11-26 at 15:19 +0000, Robin Murphy wrote:
> On 2020-11-20 09:06, Yong Wu wrote:
> > Currently direct_mapping always use the smallest pgsize which is SZ_4K
> > normally to mapping. This is unnecessary. we could gather the size, and
> > call iommu_map then, iommu_map could decide how to map better with the
> > just right pgsize.
> > 
> >  From the original comment, we should take care overlap, otherwise,
> > iommu_map may return -EEXIST. In this overlap case, we should map the
> > previous region before overlap firstly. then map the left part.
> > 
> > Each a iommu device will call this direct_mapping when its iommu
> > initialize, This patch is effective to improve the boot/initialization
> > time especially while it only needs level 1 mapping.
> > 
> > Signed-off-by: Anan Sun <anan.sun@mediatek.com>
> > Signed-off-by: Yong Wu <yong.wu@mediatek.com>
> > ---
> >   drivers/iommu/iommu.c | 20 ++++++++++++++++++--
> >   1 file changed, 18 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> > index df87c8e825f7..854a8fcb928d 100644
> > --- a/drivers/iommu/iommu.c
> > +++ b/drivers/iommu/iommu.c
> > @@ -737,6 +737,7 @@ static int iommu_create_device_direct_mappings(struct iommu_group *group,
> >   	/* We need to consider overlapping regions for different devices */
> >   	list_for_each_entry(entry, &mappings, list) {
> >   		dma_addr_t start, end, addr;
> > +		size_t unmapped_sz = 0;
> >   
> >   		if (domain->ops->apply_resv_region)
> >   			domain->ops->apply_resv_region(dev, domain, entry);
> > @@ -752,10 +753,25 @@ static int iommu_create_device_direct_mappings(struct iommu_group *group,
> >   			phys_addr_t phys_addr;
> >   
> >   			phys_addr = iommu_iova_to_phys(domain, addr);
> > -			if (phys_addr)
> > +			if (phys_addr == 0) {
> > +				unmapped_sz += pg_size; /* Gather the size. */
> >   				continue;
> > +			}
> 
> I guess the reason we need to validate every page is because they may 
> already have been legitimately mapped if someone else's reserved region 
> overlaps - is it worth explicitly validating that, i.e. bail out if 
> something's gone wrong enough that phys_addr != addr?

I'm not sure the history about why to validate every page. this
direct_mapping is called very early, normally after alloc_default_domain
and _attach_device. the "phys_addr != addr" looks impossible.

If there is a normal flow that may cause "phys_addr != addr", then
something go wrong, Could we give a warning like adding a
WARN_ON_ONCE(phys_addr != addr)? and it should be in a another patch.

> 
> Other than the naming issue (I agree that map_size is a far, far better 
> choice), I don't have any strong opinions about the rest of the 
> implementation - I've written enough variations of this pattern to know 
> that there's just no "nice" way to do it in C; all you can do is shuffle 
> the clunkiness around :)

:). I will send a v2.
Thanks.

> 
> Robin.
> 
> >   
> > -			ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
> > +			if (unmapped_sz) {
> > +				/* Map the region before the overlap. */
> > +				ret = iommu_map(domain, start, start,
> > +						unmapped_sz, entry->prot);
> > +				if (ret)
> > +					goto out;
> > +				start += unmapped_sz;
> > +				unmapped_sz = 0;
> > +			}
> > +			start += pg_size;
> > +		}
> > +		if (unmapped_sz) {
> > +			ret = iommu_map(domain, start, start, unmapped_sz,
> > +					entry->prot);
> >   			if (ret)
> >   				goto out;
> >   		}
> > 
> 
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-mediatek