linux-kernel - Re: [PATCH 1/1] Userspace I/O (UIO): Add support for userspace DMA

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: <20081204180809.GB3079@local>
Date:	Thu, 4 Dec 2008 19:08:09 +0100
From:	"Hans J. Koch" <hjk@...utronix.de>
To:	Peter Zijlstra <peterz@...radead.org>
Cc:	edward_estabrook@...lent.com, linux-kernel@...r.kernel.org,
	hjk@...utronix.de, gregkh@...e.de, edward.estabrook@...il.com,
	hugh <hugh@...itas.com>, linux-mm <linux-mm@...ck.org>,
	Thomas Gleixner <tglx@...utronix.de>
Subject: Re: [PATCH 1/1] Userspace I/O (UIO): Add support for userspace DMA

On Thu, Dec 04, 2008 at 09:39:02AM +0100, Peter Zijlstra wrote:
> On Wed, 2008-12-03 at 14:39 -0700, edward_estabrook@...lent.com wrote:
> > From: Edward Estabrook <Edward_Estabrook@...lent.com>
> > 
> > Here is a patch that adds the ability to dynamically allocate (and
> > use) coherent DMA from userspace by extending the userspace IO driver.
> > This patch applies against 2.6.28-rc6.
> > 
> > The gist of this implementation is to overload uio's mmap
> > functionality to allocate and map a new DMA region on demand.  The
> > bus-specific DMA address as returned by dma_alloc_coherent is made
> > available to userspace in the 1st long word of the newly created
> > region (as well as through the conventional 'addr' file in sysfs).  
> > 
> > To allocate a DMA region you use the following:
> > /* Pass this magic number to mmap as offset to dynamically allocate a
> > chunk of memory */ #define DMA_MEM_ALLOCATE_MMAP_OFFSET 0xFFFFF000UL
> > 
> > void* memory = mmap (NULL, size, PROT_READ | PROT_WRITE , MAP_SHARED,
> > fd, DMA_MEM_ALLOCATE_MMAP_OFFSET); u_int64_t *addr = *(u_int64_t *)
> > memory;
> > 
> > where 'size' is the size in bytes of the region you want and fd is the
> > opened /dev/uioN file.
> > 
> > Allocation occurs in page sized pieces by design to ensure that
> > buffers are page-aligned.
> > 
> > Memory is released when uio_unregister_device() is called.
> >
> > I have used this extensively on a 2.6.21-based kernel and ported it to
> > 2.6.28-rc6 for review / submission here.
> > 
> > Comments appreciated!
> 
> Yuck!
> 
> Why not create another special device that will give you DMA memory when
> you mmap it? That would also allow you to obtain the physical address
> without this utter horrid hack of writing it in the mmap'ed memory.
> 
> /dev/uioN-dma would seem like a fine name for that.

I don't like to have a separate device for DMA memory. It would completely
break the current concept of userspace drivers if you had to get normal
memory from one device and DMA memory from another. Note that one driver
can have both.

But I agree that it's confusing if the physical address is stored somewhere
in the mapped memory. That should simply be omitted, we have that information
in sysfs anyway - like for any other memory mappings. But I guess we need
some kind of "type" or "flags" attribute for the mappings so that userspace
can find out if a mapping is DMA capable or not.

Thanks,
Hans

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/