linux-kernel - RE: [RFC 1/1] swiotlb: Reduce calls to swiotlb_find

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID:
 <SN6PR02MB41578519BA2E432E1B2154FBD4D12@SN6PR02MB4157.namprd02.prod.outlook.com>
Date: Sat, 29 Jun 2024 15:53:07 +0000
From: Michael Kelley <mhklinux@...look.com>
To: Petr Tesařík <petr@...arici.cz>
CC: "robin.murphy@....com" <robin.murphy@....com>, "joro@...tes.org"
	<joro@...tes.org>, "will@...nel.org" <will@...nel.org>, "jgross@...e.com"
	<jgross@...e.com>, "sstabellini@...nel.org" <sstabellini@...nel.org>,
	"oleksandr_tyshchenko@...m.com" <oleksandr_tyshchenko@...m.com>, "hch@....de"
	<hch@....de>, "m.szyprowski@...sung.com" <m.szyprowski@...sung.com>,
	"iommu@...ts.linux.dev" <iommu@...ts.linux.dev>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"xen-devel@...ts.xenproject.org" <xen-devel@...ts.xenproject.org>
Subject: RE: [RFC 1/1] swiotlb: Reduce calls to swiotlb_find_pool()

From: Michael Kelley <mhklinux@...look.com> Sent: Thursday, June 27, 2024 8:05 AM
> 
> From: Petr Tesařík <petr@...arici.cz> Sent: Thursday, June 27, 2024 12:21 AM
> 
> [...]
> 
> > > @@ -187,10 +169,13 @@ static inline bool is_swiotlb_buffer(struct device *dev, phys_addr_t paddr)
> > >  	 * This barrier pairs with smp_mb() in swiotlb_find_slots().
> > >  	 */
> > >  	smp_rmb();
> > > -	return READ_ONCE(dev->dma_uses_io_tlb) &&
> > > -		swiotlb_find_pool(dev, paddr);
> > > +	if (READ_ONCE(dev->dma_uses_io_tlb))
> > > +		return swiotlb_find_pool(dev, paddr);
> > > +	return NULL;
> > >  #else
> > > -	return paddr >= mem->defpool.start && paddr < mem->defpool.end;
> > > +	if (paddr >= mem->defpool.start && paddr < mem->defpool.end)
> > > +		return &mem->defpool;
> >
> > Why are we open-coding swiotlb_find_pool() here? It does not make a
> > difference now, but if swiotlb_find_pool() were to change, both places
> > would have to be updated.
> >
> > Does it save a reload from dev->dma_io_tlb_mem? IOW is the compiler
> > unable to optimize it away?
> >
> > What about this (functionally identical) variant:
> >
> > #ifdef CONFIG_SWIOTLB_DYNAMIC
> > 	smp_rmb();
> > 	if (!READ_ONCE(dev->dma_uses_io_tlb))
> > 		return NULL;
> > #else
> > 	if (paddr < mem->defpool.start || paddr >= mem->defpool.end);
> > 		return NULL;
> > #endif
> >
> > 	return swiotlb_find_pool(dev, paddr);
> >
> 
> Yeah, I see your point. I'll try this and see what the generated code
> looks like. It might take me a couple of days to get to it.
> 

With and without CONFIG_SWIOTLB_DYNAMIC, there's no meaningful
difference in the generated code for x86 or for arm64.  

I'll incorporate this change into v2.

Michael