Messages in this thread | | | Subject | Re: [openib-general] Suggestions for how to remove bus_to_virt() | From | Ralph Campbell <> | Date | Fri, 14 Jul 2006 15:27:07 -0700 |
| |
On Thu, 2006-07-13 at 08:46 +0300, Muli Ben-Yehuda wrote: > On Wed, Jul 12, 2006 at 05:40:13PM -0700, David Miller wrote: > > From: Roland Dreier <rdreier@cisco.com> > > Date: Wed, 12 Jul 2006 17:11:26 -0700 > > > > > A cleaner solution would be to make the dma_ API really use the device > > > it's passed anyway, and allow drivers to override the standard PCI > > > stuff nicely. But that would be major surgery, I guess. > > > > Clean but expensive, you should not force the rest of the kernel > > to eat the cost of something you want to do when it's totally > > unnecessary for most other users. > > > > For example, x86 never needs to do anything other than a direct > > virt_to_phys translation to produce a DMA address, no matter what > > bus the device is on. It's a single simple integer adjustment > > that can be done inline in about 2 or 3 instructions at most. > > It's possible that even x86 will support multiple IOMMUs in the future > - for example, the Calgary IOMMU support we recently added to x86-64 > could be modified to work on plain x86 as well. > > I like the idea of a per-device DMA-API implementation, but only if it > can be done in a way that is zero cost to the majority of the users of > the API. We already have dynamic dma_ops on x86-64 to support nommu, > swiotlb, gart and Calgary cleanly, extending it to use a per-device > dma-ops isn't too difficult. > > Cheers, > Muli
A per-device DMA-API would solve my problem. It would be a fairly invasive changeset though. The basic idea would be to add a struct dma_mapping_ops * to struct device and change all the inline dma_* routines to something like:
static inline dma_addr_t dma_map_single(struct device *hwdev, void *ptr, size_t size, int direction) { BUG_ON(!valid_dma_direction(direction)); return hwdev->dma_ops ? hwdev->dma_ops->map_single(hwdev, ptr, size, direction) : dma_ops->map_single(hwdev, ptr, size, direction); }
Note that the current design only supports one IOMMU type in a system. This could support multiple IOMMU types at the same time.
Another possibility is to only do this for the infiniband subsystem. The idea would be to replace calls to dma_* with ib_dma_* which would be defined as above.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |