Is iommu_num_pages() broken ?

Is iommu_num_pages() broken ?

[Benjamin Herrenschmidt]

Hi folks !

I stumbled upon this today:

static inline unsigned long iommu_num_pages(unsigned long addr,
					    unsigned long len,
					    unsigned long io_page_size)
{
	unsigned long size = (addr & (io_page_size - 1)) + len;

	return DIV_ROUND_UP(size, io_page_size);
}
 

That doesn't look right to me...

The powerpc iommu code at least uses that with an addr which may not be
page aligned (ie, result of sg_virt() which include the offset).

The above code will align the start before adding the len which is wrong
and will result in potentially missing a page or am I missing
something ?

Shouldn't it be something like

static inline unsigned long iommu_num_pages(unsigned long addr,
					    unsigned long len,
					    unsigned long io_page_size)
{
	unsigned long start = addr & (io_page_size - 1);
	unsigned long end = addr + len;

	return DIV_ROUND_UP(end - start, io_page_size);
}

?

Cheers,
Ben.

Re: Is iommu_num_pages() broken ?

[Andreas Schwab]

Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:

> static inline unsigned long iommu_num_pages(unsigned long addr,
> 					    unsigned long len,
> 					    unsigned long io_page_size)
> {
> 	unsigned long size = (addr & (io_page_size - 1)) + len;
>
> 	return DIV_ROUND_UP(size, io_page_size);
> }
>  
>
> That doesn't look right to me...
>
> The powerpc iommu code at least uses that with an addr which may not be
> page aligned (ie, result of sg_virt() which include the offset).
>
> The above code will align the start before adding the len which is wrong

addr & (io_page_size - 1) computes the offset into the page pointed to
by addr.  Looks right to me.

addr & ~(io_page_size - 1) would round addr down to the start of the
page.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Is iommu_num_pages() broken ?

[Benjamin Herrenschmidt]

On Sun, 2012-08-05 at 08:39 +0200, Andreas Schwab wrote:
> addr & (io_page_size - 1) computes the offset into the page pointed to
> by addr.  Looks right to me.
> 
> addr & ~(io_page_size - 1) would round addr down to the start of the
> page.

Ah that's right, for some reason I'm so used to seeing ~ before such
constructs that my brain must have imagined it :-)

Sorry for the noise. Next time I'll have coffee first then re-check ...

Cheers,
Ben.