Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 7 Aug 2023 23:46:05 +0100
From:   "Russell King (Oracle)" <linux@armlinux.org.uk>
To:     Matthew Wilcox <willy@infradead.org>
Cc:     Marek Szyprowski <m.szyprowski@samsung.com>,
        linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org,
        Andrew Morton <akpm@linux-foundation.org>,
        Mike Rapoport <rppt@kernel.org>
Subject: Re: [PATCH] arm: dma-mapping: fix potential endless loop in
 __dma_page_dev_to_cpu()
Message-ID: <ZNF0LUrtoCp+aS5U@shell.armlinux.org.uk>
References: <CGME20230807152704eucas1p1bbe08af4559a7d2984198fe8ba487a2e@eucas1p1.samsung.com>
 <20230807152657.1692414-1-m.szyprowski@samsung.com>
 <ZNFstXmztIriaKOX@casper.infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZNFstXmztIriaKOX@casper.infradead.org>
Sender: Russell King (Oracle) <linux@armlinux.org.uk>
Precedence: bulk

On Mon, Aug 07, 2023 at 11:14:13PM +0100, Matthew Wilcox wrote:
> On Mon, Aug 07, 2023 at 05:26:57PM +0200, Marek Szyprowski wrote:
> > diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
> > index 70cb7e63a9a5..02250106e5ed 100644
> > --- a/arch/arm/mm/dma-mapping.c
> > +++ b/arch/arm/mm/dma-mapping.c
> > @@ -718,7 +718,7 @@ static void __dma_page_dev_to_cpu(struct page *page, unsigned long off,
> >  			folio = folio_next(folio);
> >  		}
> >  
> > -		while (left >= (ssize_t)folio_size(folio)) {
> > +		while (left && left >= (ssize_t)folio_size(folio)) {
> >  			set_bit(PG_dcache_clean, &folio->flags);
> >  			left -= folio_size(folio);
> >  			folio = folio_next(folio);
> 
> I've been thinking about this and I think this is the right fix for the
> wrong reason.  I don't understand how it can produce the failure you
> saw, but we shouldn't be calling folio_next() if left is zero, let alone
> calling folio_size() on it.  So I'd rather see this fix:
> 
> 		while (left >= (ssize_t)folio_size(folio)) {
> 			set_bit(PG_dcache_clean, &folio->flags);
> 			left -= folio_size(folio);
> +			if (!left)
> +				break;

Given that set_bit() involves atomics, wouldn't it be better if this
had been written as:

		while (left >= folio_size(folio)) {
			left -= folio_size(folio);
			set_bit(PG_dcache_clean, &folio->flags);
			if (!left)
				break;
> 			folio = folio_next(folio);
> 		}

That likely means that folio_size() will only be evaluated once per
loop rather than twice. I may be wrong though, I didn't check the
generated code.

Also, I'm wondering what that ssize_t cast is doing there - "left"
is a size_t, which is unsigned. folio_size() returns a size_t, so
is also unsigned. Why convert folio_size() to a signed number to
then be compared with an unsigned number?

Or did "left" get converted to ssize_t along with the folio
conversion?

Even if it did, how could "left" be negative (except through casting
a large positive number as "size" that in 2's complement would be
negative after casting to "left") ?

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!