Does this PCI API look good enough for actual use?

[Owen]

Said device driver can still spatter crap all over your kernel using DMA.

Incidentally, that graph is only a partial truth: Both Intel and nVIDIA have to deal with the fact that the Vista and later display model is basically "the hardware interface to an ATi card", since ATi have been designing the DirectX 10 and later APIs (With results which have made everyone except ATi unhappy).

Remember: Graphics card manufacturers are under the most pressure of any company to get their software to market and squeeze out all the performance possible. Both ATi and nVIDIA's track records on Linux (and Solaris, and BSD...) are much better.

[NickJohnson]

Both ATi and nVIDIA's track records on Linux (and Solaris, and BSD...) are much better.

Really? At least the last time I tried, which was about a year ago, the nvidia drivers on linux were hardly stable - they wouldn't let you properly set the resolution, had crappy performance, and would occasionally bring down the entire system.

[Owen]

The only time I've had the nVIDIA drivers crash my system, I needed a BIOS update (It turns out I'd been running my system for about a year with a BIOS which didn't support my processor!)

Oh, and when my graphics card was at 90degC. But I don't blame them for that!

[Ready4Dis]

Incidentally, that graph is only a partial truth: Both Intel and nVIDIA have to deal with the fact that the Vista and later display model is basically "the hardware interface to an ATi card", since ATi have been designing the DirectX 10 and later APIs (With results which have made everyone except ATi unhappy).

Are you kidding me? Both companies (and others) have their inputs. Some special features may have been added due to ATI's input, but overall stability (not just features) has nothing to do with it. Also, NVIDIA was the first one with a DX10 card and the first with Microsoft certified DX 10 drivers, so I hardly think ATI had a jump on them in that respect as you've suggested. AMD just writes better (more stable, but typically not as optimized although they have gotten a lot better as of late) drivers (ATI's where known to be crap, and stop putting the little i in ATI, it's no longer a little i, it grew up). Anyways, I allow my drivers access to PCI info directly, the driver is responsible for it's own behavior, although I do agree with the reasons given for not trusting them. There is no 'right' way, it depends on your goals and your OS. For a bare minimum PCI API your functions look ok. My OS is pretty minimum when it comes to PCI,so I won't comment to much on the 'correct' way to do things, but I don't like having my kernel or boot loader setting up my PCI space like was suggested (putting them all into native). I would much prefer having a PCI bus driver that would do the setup and search for a compatible driver, and the driver should be responsible for setting the device up properly from there on. Each device (well, anything that requires it) driver requests a lock on resources, which can help take care of things that use a default location. I do need to have it so you can tell it what order to load (so something that is at a fixed location can be loaded before something that is dynamic just in case).

Again, a lot depends on your goals and your OS. If you trust your drivers, then give them access, if you don't, then don't. There are trade offs either way you go, just be aware of them and live with them. Most stuff was covered in previous posts, but if you plan on using on current/future hardware, I think you can ignore some of the special case stuff and focus on something that will work fine with hardware post 2000/ If you are targetting older hardware, then you may want to read up on some more of the quirks of the past and decide the best route.

[Combuster]

Most stuff was covered in previous posts, but if you plan on using on current/future hardware, I think you can ignore some of the special case stuff and focus on something that will work fine with hardware post 2000/ If you are targetting older hardware, then you may want to read up on some more of the quirks of the past and decide the best route.

If only one of the peevees mentioned wasn't a year-and-a-half old... (Damn you, Intel) You'll even find one of the more nasty ones in Virtual PC (Damn you, S3)

[jal]

Really? At least the last time I tried, which was about a year ago, the nvidia drivers on linux were hardly stable - they wouldn't let you properly set the resolution, had crappy performance, and would occasionally bring down the entire system.

You sure that you didn't use the open source nVidia drivers (Nouveau, iirc), as opposed to the closed source ones by nVidia themselves? nVidia has very stable Linux drivers, where ATI has a very crappy reputation in that area.


JAL