How to map device addresses into user space

[John Baldwin]

On Monday, January 07, 2013 12:25:33 PM Dr. Rolf Jansen wrote:
> Am 07.01.2013 um 13:11 schrieb John Baldwin:
> > On Monday, January 07, 2013 08:52:01 AM Dr. Rolf Jansen wrote:
> >> Am 03.01.2013 um 14:45 schrieb Dr. Rolf Jansen:
> >>> ...
> >>> 
> >>> So, how can I map device addresses into user space on FreeBSD?
> >> 
> >> Many Thanks to everybody, who responded.
> >> 
> >> I got it working. I wrote a PCI device driver, which in addition to
> >> open/close/read/write responds also to ioctl and mmap requests from user
> >> space. In its pci_attach routine, it assembles a custom address_space
> >> struct containing the addresses of the two BAR's of my NI PCI-DAQ board
> >> and two huge allocated memory regions for DMA...
> > 
> > Note that if your BARs are not contiguous you can allow mapping of things
> > in the middle.  This is a potentital security hole if your driver can be
> > opened by non-root.
> 
> I can understand this. Actually, on the present model of the DAQ board, the
> BARs are contiguous, and in the present incarnation, my driver can only be
> addressed by root. Anyway, I want to do it correctly, that means, I would
> like to avoid right now potential holes bubbling up in the future, when
> using my driver with other DAQ boards, and in different usage scenarios.

Note that the BARs being contiguous is a property of your BIOS or OS, not your 
board. :)  The firmware/OS are free to assign whatever ranges to your BARs 
that they wish.

> > A more robust approach is to define a virtual address space for
> > your device. This is also a lot easier to program against in userland.
> > For example, you could define offset 0 as mapping to BAR 0 and the next
> > chunk
> 
> > of virtual address space (the offset) map to BAR 1, etc.:
> I did this, but id didn't work out. The 2 BARs are contiguous, and the 2
> DMA regions are also, but there is huge gap between BAR and DMA. And
> depending on which mapping is done first, either the DMAs or the BARs
> receive invalid paddr. So, I guess, defining a virtual address space is a
> little bit more involved than simply getting the offsets straight as you
> lined out.
> 
> I am still learning things, so please bear with me, if it is a dumb
> question. How do I define a virtual address space for the device?

Ah, I'll explain a bit.  When you mmap a section of a file, the OS is treating 
the file as if it were its own section of virtual memory, where files contents 
were addressed by the offset.  That is, virtual address 0 of the file 
corresponds to the data at file offset 0, and virtual address N corresponds to 
the data at file offset N.

Now, think of your /dev/foo device as a file.  When you mmap() your /dev/foo, 
the OS has assumes it has a similar virtual memory that is addressed by the 
offset into the file.  However, for a character device, the OS needs the 
driver to describe this virtual address space.  That is what you are doing 
with d_mmap() and the 'offset' parameter.  You are telling it what physical 
page backs each "virtual page".  It is important to note that when using 
d_mmap(), these mappings are immutable.  Once you provide a mapping for a 
given virtual page, the OS will cache it forever (or until reboot).

In other words, you can write your d_mmap() to give the appearance that your 
/dev/foo "file" contains BAR 0 followed by BAR 1, etc.

> > static int
> > mydaq_mmap(...)
> > {
> > 
> > 	struct mydaq_softc *sc;  // you should be storing things like the
> > 	pointers
> > 	
> >                             // to your BARs in a softc
> > 	
> > 	sc = dev->si_drv1;
> > 	if (offset <= rman_get_size(sc->bar0res))
> > 	
> > 		return (rman_get_start(sc->bar0res) + offset);
> > 	
> > 	offset -= rman_get_size(sc->bar0res);
> > 	if (offset <= rman_get_size(sc->bar1res))
> > 	
> > 		return (rman_get_start(sc->bar1res) + offset);
> > 	
> > 	offset -= rman_get_size(sc->bar1res);
> > 	if (offset <= dmaAIBufSize)
> > 	
> > 		return (vtophys(sc->dma[0]) + offset);
> > 	
> > 	offset -= dmaAIBufSize;
> > 	if (offset <= dmaAOBufSize)
> > 	
> > 		return (vtophys(sc->dma[1]) + offset);
> > 	
> > 	/* Invalid offset */
> > 	return (-1);
> > 
> > }
> 
> I guess, you hacked this quickly together, and I picked the concept, so
> this is fine. In order to make this working, the addresses should be
> assigned to the *paddr parameter, and a status code should be returned. In
> order to check things, of course I applied these minor corrections, but it
> doesn't work for either the BAR block or the DMA block, depending on which
> is mapped first.

Yes, it needs the changes you made.  Can you post your updated d_mmap()
routine along with how you are mapping it in userland?  With what I posted
above you would do something like this in userland:

const char *bar0, *bar1;

int fd = open("/dev/foo");

bar0 = mmap(NULL, <size of BAR 0>, PROT_READ, MAP_SHARED, fd, 0);
bar1 = mmap(NULL, <size of BAR 1>, PROT_READ, MAP_SHARED, fd,
    <size of BAR 0>);

etc.

-- 
John Baldwin