On Tue, 23 Oct 2007, Boaz Harrosh wrote:Yes, that would be one clean situation. Well, I'd personally actually prefer to *not* have the count be passed down explicitly, because it's just too error prone. So I'd much rather see the count implicit in the list: whether it's in an explicit header structure (that is the *only* thing passed down) or whether it's embedded in the list itself is not important. Since the list itself has to have the "next pointer" for chaining, and thus already has "embedded information" in it, it actually does make sense in my opinion to just embed the end-of-list information too. And the end result right now is pretty simple, with "sg_next()" being really simple to use, and there being no way to screw things up by getting the count and the sg pointer out of sync. My biggest complaint right now is that a lot of users of the sg *filling* functions were mindlessly converted, so we have code like cryptoloop.c: sg_set_page(&sg_in, in_page); cryptoloop.c: sg_in.offset = in_offs; cryptoloop.c: sg_in.length = sz; which is just really stupid, and we should have a function for that. But worse is code like this: ub.c: sg_set_page(sg, virt_to_page(sc->top_sense)); ub.c: sg->offset = (unsigned long)sc->top_sense & (PAGE_SIZE-1); ub.c: sg->length = UB_SENSE_SIZE; which again was converted "line by line" and we actually *do* have a function to do the above three lines as sg_set_buf(sg, sc->top_sense, UB_SENSE_SIZE); where that *single* line is just tons shorter but more importantly, more readable, than the mess that is a brute-force conversion. So I think the SG stuff looks ok now, but I think we have a lot of "fix up the rough edges" to go! (The above is not the only case. Just grep for "sg_set_page", and you'll see several examples of this kind of hard-to-read code. Basically, I don't think it's ever a good idea to initialize the SG entries one by one, and even when we have a hard page/offset/size thing, we should not set them one by one, and we should probably extend sg_set_page() to always take offset and length too, since setting one without the other two is never really sensible!) Linus -
| Linus Torvalds | Re: Dual-Licensing Linux Kernel with GPL V2 and GPL V3 |
| Artem Bityutskiy | [RFC PATCH 06/26] UBIFS: add superblock and master node |
| Joe Perches | [PATCH 001/148] include/asm-x86/acpi.h: checkpatch cleanups - formatting only |
| Linus Torvalds | Re: LSM conversion to static interface |
git: | |
| Alexey Dobriyan | Re: [GIT]: Networking |
| Gerrit Renker | [PATCH 27/37] dccp: Integration of dynamic feature activation - part 2 (server side) |
| Christoph Lameter | Network latency regressions from 2.6.22 to 2.6.29 |
