On Wed, Oct 24 2007, Robert P. J. Day wrote:Completely disagree, it's well defined. The asm/ header adds the arch private stuff, like the actual structure definition. That absolutely has to reside there. The linux/ header adds manipulation headers for that structure. You'd want to put that in all the asm/ headers? I didn't think so. So I'll repeat - arch code may use the asm/ header, if they just need the structure definition. Drivers should use the linux/ header, since they should also use the accessor functions. Right now there's still a lot of open coding of for (i = 0; i < sg_nents; i++) sg_table[i] ... usage which should go away eventually and use a sg = sg_table; do { ... } while ((sg = sg_next(sg)) != NULL); construct, in which case there's no way around using the linux/ header. arch IOMMU code will want to be using the linux/ header as well, since they will be browsing the list also. That switch should happen when the driver actually needs it, not needlessly. The problems seen in the last few days have been code that actually use sg accessors AND don't use the linux/ header, and that problem only showing up on archs where linux/scatterlist.h doesn't get magically included from some other file. That is of course a bug, you should include the header with the functions that you use. It has nothing to do with asm/ vs linux/ include messiness. -- Jens Axboe -
| Ingo Molnar | [patch 12/13] syslets: x86: optimized copy_uatom() |
| Greg Kroah-Hartman | [PATCH 017/196] aoechr: Convert from class_device to device |
| Yinghai Lu | Re: 2.6.26, PAT and AMD family 6 |
| Jan Engelhardt | intel iommu (Re: -mm merge plans for 2.6.23) |
git: | |
| Gerrit Renker | [PATCH 27/37] dccp: Integration of dynamic feature activation - part 2 (server side) |
| David Miller | [GIT]: Networking |
| David Miller | Re: [PATCH] pkt_sched: Destroy gen estimators under rtnl_lock(). |
| Natalie Protasevich | [BUG] New Kernel Bugs |
