[Forum] Try only to realise the truth | ANN.lu |
|
List of all comments to this article |
Try only to realise the truth : Comment 42 of 335 | ANN.lu |
Posted by Joe "Floid" Kanowitz on 04-Jul-2004 19:11 GMT | In reply to Comment 15 (Anonymous): He clearly said that even on non cache coherency platform LinuxPPC can work.
So let's try to hack stuff and for Linux to disable the cache coherency in your AOne...
It's the other way around, sooort of... if "DrBombcrater" is correct on Amiga.org, which may or may not be the case. You *want* cache coherency where you can get it, so if the chipset won't handle it transparently (bug or design decision), you tap it on the shoulder when you're done and make sure the cache lines agree with 'reality.' Since nothing other than the driver should be peeking at the DMA addresses, you probably don't need to hold a lock.
If this understanding is right, and my further understanding is right (danger, danger.. ;)), 2.4 didn't have a generalized API to say "Look, I'm on a platform that works this way, conduct a 'shoulder-tap' after every ____." 2.6 appears to, though I've helpfully lost the page for the project.
This probably gets more confusing when you're trying to deal with SMP (disclaimer: I know squat about the PPC bus protocols and nearly squat about x86), but SMP is always confusing.
---
Here's a thread about some mostly-unrelated confusion with the Linux DMA API, which demonstrates yet-more pitfalls you can... er, fall into. Note a few posts down, where someone says "So maybe the only advantage would be to allow more portable generic device drivers. Luckily I don't need this right now :)" -- unfortunately, this sort of portability is what Mai et al. assumed they could rely on, and it seems everyone (Mai, bPlan/Genesi, the whole Amiga befunge) got the unpleasant surprise.
If the problem really is known, there's no reason OS4 shouldn't work (we'll know the day we can test it... and/or maybe the day the driver API is explained), because it doesn't have the, er, obligation... to work with existing code (like the Linux device drivers as they stand). Meanwhile, no matter who did the work to refactor the whole API and perhaps all the drivers (assuming it's been done), such dramatic changes don't usually get patched on the main tree of a 'stable' series like 2.4. (So the A1-Linux or YellowDog trees might have them... more likely they have their own hacks while waiting for 2.6 to present a clearer path... but BenH's builds, the official Marcelo tree, or anything put out for the Pegasos might not. From what Ole-Egil was saying on the Debian list, it sounds like work has only begun to get things 'recognized' in Linus/Morton 2.6, even if the APIs to make it easy -- if(Articia_S && !April) { define DMA_WORKS_LIKE_THIS }* -- do exist... because the legwork's been left to hobbyists... and Hyperion employees** on unpaid time. ;))
*Psuedocode. Please do not make fun of my weak knowledge of preprocessors and C. ;)
**And this is Eyetech's problem, not Hyperion's -- their OS should work, after all -- so seeing what's gone on, I can understand how Ross must've felt getting 'fan mail' about it. At the same time, if he doesn't want to get stuck looking like project lead for being founder (everyone disclaims being 'lead,' last I checked, so there's a good chance the project doesn't have one... and it's a thankless position, because even if everything does work, it's not like Intel will hire you and put you on pension for it), maybe the project site should make a show of advertising for one... not that there'll be any takers (hey, y'never know, stranger people read KernelTrap, though there's that conflict of interest in showing off a "bug"), but just to show that no one's "in charge." ;) |
|
List of all comments to this article (continued) |
|
- User Menu
-
- About ANN archives
- The ANN archives is powered by #AmigaZeux. It was updated daily (news last: 22-Oct-2004; comments last: 18-May-2005).
ANN.lu was created, previously owned and maintained by Christian Kemp, www.ckemp.com.
- Contribute
- Not possible at this time!
- Search ANN archives
- Advanced search
- Hosting
- ANN.lu was hosted by Dreamhost. Sign up through this link, mention "ckemp" as referrer and he will get a 10% commission on any account you purchase.
Please show your appreciation for any past, present and future work on ANN.lu by making a contribution via PayPal.
|