Message-ID: <1606302214.26323.141.camel@mhfsdcap03>
Date:   Wed, 25 Nov 2020 19:03:34 +0800
From:   Yong Wu <yong.wu@...iatek.com>
To:     Will Deacon <will@...nel.org>
CC:     <youlin.pei@...iatek.com>, <anan.sun@...iatek.com>,
        Nicolas Boichat <drinkcat@...omium.org>,
        <srv_heupstream@...iatek.com>, <chao.hao@...iatek.com>,
        Joerg Roedel <joro@...tes.org>, <linux-kernel@...r.kernel.org>,
        Krzysztof Kozlowski <krzk@...nel.org>,
        Tomasz Figa <tfiga@...gle.com>,
        <iommu@...ts.linux-foundation.org>,
        <linux-mediatek@...ts.infradead.org>,
        Matthias Brugger <matthias.bgg@...il.com>,
        Robin Murphy <robin.murphy@....com>,
        <linux-arm-kernel@...ts.infradead.org>
Subject: Re: [PATCH] iommu: Improve the performance for direct_mapping

On Tue, 2020-11-24 at 11:05 +0000, Will Deacon wrote:
> On Tue, Nov 24, 2020 at 05:24:44PM +0800, Yong Wu wrote:
> > On Mon, 2020-11-23 at 12:32 +0000, Will Deacon wrote:
> > > On Fri, Nov 20, 2020 at 05:06:28PM +0800, Yong Wu wrote:
> > > > +				unmapped_sz = 0;
> > > > +			}
> > > > +			start += pg_size;
> > > > +		}
> > > > +		if (unmapped_sz) {
> > > > +			ret = iommu_map(domain, start, start, unmapped_sz,
> > > > +					entry->prot);
> > > 
> > > Can you avoid this hunk by changing your loop check to something like:
> > > 
> > > 	if (!phys_addr) {
> > > 		map_size += pg_size;
> > > 		if (addr + pg_size < end)
> > > 			continue;
> > > 	}
> > 
> > Thanks for your quick review. I have fixed and tested it. the patch is
> > simple. I copy it here. Is this readable for you now?
> > 
> > 
> > --- 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 map_size = 0;
> >  
> >  		if (domain->ops->apply_resv_region)
> >  			domain->ops->apply_resv_region(dev, domain, entry);
> > @@ -752,12 +753,21 @@ 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)
> > -				continue;
> > +			if (!phys_addr) {
> > +				map_size += pg_size;
> > +				if (addr + pg_size < end)
> > +					continue;
> > +				else
> 
> You don't need the 'else' here  ^^^
> 
> > +					addr += pg_size; /*Point to End */
> 
> addr = end ?
> 
> That said, maybe we could simplify this further by changing the loop bounds
> to be:
> 
> 	for (addr = start; addr <= end; addr += pg_size)
> 
> and checking:
> 
> 	if (!phys_addr && addr != end) {
> 		map_size += pg_size;
> 		continue;
> 	}
> 
> does that work?

It works but I think we can not check iommu_iova_to_phys(domain, end).
We should add a "if", like:

for (addr = start; addr <= end; addr += pg_size) {
...
	if (addr < end) {
		phys_addr = iommu_iova_to_phys(domain, addr);
		if (!phys_addr) {
			map_size += pg_size;
			continue;
		}
	}
...


If you don't like this "if (addr < end)", then we have to add a "goto".
like this:


for (addr = start; addr <= end; addr += pg_size) {
 	phys_addr_t phys_addr;
 
	if (addr == end)
		goto map_last;

	phys_addr = iommu_iova_to_phys(domain, addr);
	if (!phys_addr) {
		map_size += pg_size;
		continue;
	}

map_last:
	if (!map_size)
		continue;
	ret = iommu_map(domain, addr - map_size,
			addr - map_size, map_size, entry->prot);
	if (ret)
		goto out;
}

Which one is better?

> 
> Will
> 
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@...ts.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-mediatek

Powered by blists - more mailing lists