Hey,
This is a bit off topic, but i was wondering whether direct hardware in linux was possible, for anything that is (PCI-Eth/Audio, DGA, etc)..

Is everything possible through ports, just out/in to certain ports, or through memory access?

(This isnt vital, just had an interest on the subject, so decided to ask. The linux forum is too quiet anyway )

Thanks in advance,
Gunblade

It is possible to do direct hardware access, but it's not easy You have to use IOCTL's to set permissions, and quite possibly run the program as root. If you write a kernel module, it should be easier (though you do risk killing the kernel if you're not careful)

Now thats a bloody quick answer

So the best idea for direct access is making a kernel module?
Is there not another way of making a program execute with the permissions of the kernel?

Thanks,
Gunblade

Yes.



Not that I'm aware of. You can use IOCTL's (fairly difficult IIRC) to request read/write permission for specific ports and possibly memory ranges, but it's not easy. They aren't well documented, and you may still have to use the files in /dev for some hardware. Keep in mind that Linux (and most other modern OSes) are designed to keep hardware access from userspace to a minimum.

x11/x* is not kernel module but it's popular.

it is user-mode device driver.
so it doesn't care about kernel like openbsd, freebsd, linux, hurd, macosx, ...

--

Does X11 directly access the hardware? I *know* that you can use IOCTL's to gain access to specific areas of the hardware, but full, unrestricted access is limited to kernel space. IOCTL's are a pain to use; which is why I suggest making a kernel module if you need direct hardware access.

no it isn't.
x11/x* use linux's system-call.
these 250 system-call(x86) units can access to the abstract hardware.

But, in unix(monolithic) environment, it's meaningless to classify system programming and application programming, the mechanism is different, everything is file/(c programming dmr, ken)
..

--

1.) Check out functions ioperm and iopl. With these functions (syscalls) you can get right for access to IO ports (see in/out assembly instructions).

2.) When communicating to the device, you likely need a contiguous memory place in order to serve the DMA transfer. Since you are in user space thus you are in virtual-address space. That is the virtual (or linear) addresses you see are not equal to physical ones the device needs, not speaking about that the memory you see to be virtually contiguous can physically be fregmented which is not what you want.
Here is a kernel patch I used when writing NIC drivers for MenuetOS. With this you can reach your aims, however I was able to allocate merely 4K (one page), but fortunately it was enough for me and at least I could supply DMA with it. Maybe it have been improved since then.
http://www.gelato.unsw.edu.au/IA64wiki/UserLevelDrivers

Endre.

thank you very much Endre,
that link proved very useful

And yeah, so now i see, its either a kernel device driver, or ioctl.
So thank you again


gunblade