From: Jiri Slaby <jslaby@suse.cz>
To: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Cc: mchehab@infradead.org, linux-media@vger.kernel.org,
linux-kernel@vger.kernel.org, jirislaby@gmail.com,
Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Subject: Re: [PATCH v2 -resend#1 1/1] V4L: videobuf, don't use dma addr as physical
Date: Mon, 28 Feb 2011 16:45:48 +0100 [thread overview]
Message-ID: <4D6BC32C.70408@suse.cz> (raw)
In-Reply-To: <201102281615.02193.laurent.pinchart@ideasonboard.com>
On 02/28/2011 04:14 PM, Laurent Pinchart wrote:
> Hi Jiri,
>
> On Monday 28 February 2011 16:07:43 Jiri Slaby wrote:
>> On 02/28/2011 11:53 AM, Laurent Pinchart wrote:
>>> On Monday 28 February 2011 10:37:02 Jiri Slaby wrote:
>>>> mem->dma_handle is a dma address obtained by dma_alloc_coherent which
>>>> needn't be a physical address in presence of IOMMU. So ensure we are
>>>> remapping (remap_pfn_range) the right page in __videobuf_mmap_mapper
>>>> by using virt_to_phys(mem->vaddr) and not mem->dma_handle.
>>>
>>> Quoting arch/arm/include/asm/memory.h,
>>>
>>> /*
>>>
>>> * These are *only* valid on the kernel direct mapped RAM memory.
>>
>> Which the DMA allocation shall be.
>>
>>> * Note: Drivers should NOT use these.
>>
>> This is weird.
>>
>>> They are the wrong
>>>
>>> * translation for translating DMA addresses. Use the driver
>>> * DMA support - see dma-mapping.h.
>>
>> Yes, ACK, and vice versa. DMA addresses cannot be used as physical ones.
>>
>>> */
>>>
>>> static inline unsigned long virt_to_phys(const volatile void *x)
>>> {
>>>
>>> return __virt_to_phys((unsigned long)(x));
>>>
>>> }
>>>
>>> Why would you use physically contiguous memory if you have an IOMMU
>>> anyway ?
>>
>> Sorry, what? When IOMMU is used, dma_alloc_* functions may return "tags"
>> as a DMA address, not a physical address. So using these DMA "addresses"
>> directly (e.g. in remap_pfn_range) is a bug.
>
> What I mean is that videobuf-dma-contig is meant to be used by drivers that
> require physically contiguous memory. If the system has an IOMMU, why would
> drivers need that ?
Aha. They actually need not but they would need do the mapping
themselves which they currently do not.
IOW the vbuf-dma-contig allocator is used unconditionally in the few
drivers I checked.
BUT Even if they need only one page and use vbuf-dma-contig, which I
don't see a reason not to, it will cause problems too.
regards,
--
js
suse labs
next prev parent reply other threads:[~2011-02-28 15:45 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-02-28 9:37 [PATCH v2 -resend#1 1/1] V4L: videobuf, don't use dma addr as physical Jiri Slaby
2011-02-28 10:53 ` Laurent Pinchart
2011-02-28 15:07 ` Jiri Slaby
2011-02-28 15:14 ` Laurent Pinchart
2011-02-28 15:45 ` Jiri Slaby [this message]
2011-02-28 14:53 ` Konrad Rzeszutek Wilk
2011-02-28 15:47 ` Jiri Slaby
2011-02-28 18:20 ` Mauro Carvalho Chehab
2011-03-21 22:43 ` Mauro Carvalho Chehab
2011-03-21 22:55 ` Jiri Slaby
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4D6BC32C.70408@suse.cz \
--to=jslaby@suse.cz \
--cc=jirislaby@gmail.com \
--cc=konrad.wilk@oracle.com \
--cc=laurent.pinchart@ideasonboard.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-media@vger.kernel.org \
--cc=mchehab@infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).