site_archiver@lists.apple.com Delivered-To: darwin-dev@lists.apple.com Any words of wisdom that I should be aware of? Thanks Scott _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.appl... I'm writing a kext driver for a pci-e hardware device and have a few questions. The driver needs to setup for a DMA in/out of user space using a scatter/gather list located in host memory. I've reviewed the example code and several kext drivers that do similar things and this operation seems very simple using IODMACommand compared to an existing Linux driver. The example (AppleSamplePCI) uses gen64IOVMSegments/gen32IOVMSegments to prepare a IOMemoryDescriptor for DMA and seems to be able to map everything into one contiguous physical segment. Seems like magic as the Linux driver need to map a passed user memory area into many physical pages. The hardware can handle both flavors but needs to read the scatter/gather list from host memory. This will be a Intel only deployment. Does it really work this way? What does the actual address translation to physical pages? Something has to re-map a memory access from the pci-e hardware to the correct physical pages. What about the offset to the user data within a page? Is this also handled transparently with IODMACommand? Is there any advantage to mapping to one vs several contiguous physical segment? That is, if I request one physical segment or n-segments when each is a page-size. This email sent to site_archiver@lists.apple.com